Paarifia .4

2...]. r .

. 6.. .25;

i3 9

.- 1 ..

.mERuFifllivnW
47.33....»

. A
)1 [am
«urnrvnmn.

.5. . 1qu
.fihuhn‘li i7.
. , .. Sfififihs...»
‘ ‘ ‘ ‘ , 4.4.5.394“... v...
.531... . .5 #93:...
1' I.
.

11.4.“... a'413. s

r. fv 4L .1. z.

01:1! uilvlatzi:
1' .3... 61‘.

 

L.
2 7 s!
{L furrséif...
1.22:..31; ..

L.vt‘1l
‘ ‘vVu
I.

.. 3!...
11.1.1!

53‘,

 

2.. ...
itts 9341!; u
211:1... .3 I.

1

 

 

 

 

 

This is to certify that the
dissertation entitled

THE RELATIONSHIP BETWEEN TEAM COMPOSITION AND
TEAM EFFECTIVENESS: A MULTl-LEVEL STUDY

presented by

JOSHUA M. SACCO

has been accepted towards fulfillment
of the requirements for the

Ph.D. degree in Psychology

 

7 A

 

Major Professor’s Signature

/«L/// 7/0”

Date

MSU is an Affirmative Action/Equal Opportunity Institution

 

 

 

LIBRARY
Michigan State
University

 

 

 

PLACE IN RETURN Box to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIFIC/DatoDuepes-sz

 

 

THE RELATIONSHIP BETWEEN TEAM COMPOSITION AND TEAM
EFFECTIVENESS: A MULTI-LEVEL STUDY
By

Joshua M. Sacco

A Dissertation
Submitted to
Michigan State University
In partial fulfillment of the requirements
For the degree of
Doctor of Philosophy
Department of Psychology

2002

Abstract
THE RELATIONSHIP BETWEEN TEAM COMPOSITION AND TEAM
EFFECTIVENESS: A MULTI-LEVEL STUDY
By

Joshua M. Sacco

The research reported here examines from a personnel selection standpoint team
composition and person-team fit in relation to aspects of team effectiveness in a sample
of 3,454 fast-food restaurant customer service teams consisting of 190,156 employees.
Team composition was examined with regard to ability, three dimensions of the Five- ,
Factor Model of personality (Agreeableness, Extraversion, and Conscientiousness; Costa
& McCrae, 1990; Digman, 1990) and demographic variables (race, sex, and age); person-
team fit is examined with regard to all of these variables except ability. Team financial
performance and team turnover rates were used as indicators of team effectiveness.

Team composition and person-team fit were studied in relation to team performance, and
turnover at both the team and individual level. The results indicated that mean levels of
ability were positively, though weakly related to changes in controllable profit, though
none of the ability or personality predictors had any meaningful relationship with
turnover. Latent growth curve analyses indicated that racial diversity at the start of data
collection was negatively associated with changes in profitability over time such that
practically meaningful reductions in profitability are predicted even from even very small
increases in racial diversity. A series of multilevel survival analyses indicated that the

age, race, and sex-based fit of an employee within a store composition predicted the

likelihood of turnover, and that this effect was markedly stronger earlier on in employees’
tenure. Similarly, age, race, and sex diversity was positively associated with restaurant-
level turnover rates. The practical, theoretical, and methodological implications of these

results are discussed.

Acknowledgements

This dissertation would not have been as successful as it was, or possibly completed
at all were it not for the contributions of a number of people. First, Tara Lynn Victor
provided an enormous amount of support and encouragement while I spent several weeks
straight at the library night afier night, and a similar amount of time wading through
mounds of data and obscure statistical procedures. Second, Neal Schmitt has made
monumental contributions to my professional growth and approach to problems in 1/0
psychology. His contributions to the thinking and skills demonstrated in this dissertation
cannot be overstated, and range from my greatly improved ability to write clearly and
concisely to the use of structural equation modeling to answer complex research
questions. Much of this learning has taken place on Neal’s time, in his office or working
on various projects together throughout my career as a graduate student. Neal, being
accessible, thoughtful and interested when I unexpectedly popped in to chat for an hour
has had a huge impact on my development. Thanks are also due to Rick DeShon for his
help throughout the years on various statistical issues that also helped lay the groundwork
for the approaches studied here. Ann Marie Ryan provided me with a number of
opportunities to conduct field research that helped me get interested in projects like this.
Dan Ilgen, by reading an earlier version of this paper for his Motivation class, and by
providing insightful comments during the proposal and final defense, made a number of

important points that shaped my thinking about this project.

iv

Table of Contents

LIST OF TABLES ............................................................................................................. vi
LIST OF TABLES ........................................................................................................... viii
CHAPTER 1
INTRODUCTION ............................................................................................................... 1
What Are Teams? .................................................................................................... 3
The Input-Process-Output Approach To Teams ..................................................... 5
Theoretical Model Of Team Effectiveness ............................................................. 7
KSAO — Performance Linkages at the Individual Level ......................................... 8
Individual Level KSAOs, Team Composition, and Team Performance ............... 18
Person-Environment Fit in Organizational Research ............................................ 39
Teams as a Context ............................................................................................... 46
The Present Study .................................................................................................. 75
CHAPTER 2
METHOD ......................................................................................................................... .79
Sample and Procedures ......................................................................................... 79
Measures ................................................................................................................ 81
CHAPTER 3
RESULTS ......................................................................................................................... 91
The Relationship Between Mean Levels of Ability and Personality with
Changes in Controllable Profit ........................................................................ 92
The Effects of Individual Standing on Ability and Personality Variables on
Turnover .......................................................................................................... 94
The Relationships Between Demographic Diversity and Profitability and
Turnover Rates ................................................................................................ 97
CHAPTER 4
DISCUSSION ................................................................................................................. 121
Team Ability and Personality Composition and Team Performance .................. 122
Relationships Between Ability and Personality and Turnover ........................... 126
Effects of Demographic Similarity on Turnover ................................................. 130
Methodological Implications for the Dynamic Studies of Teams ....................... 132
Theoretical and Methodological Implications for Relational Demography and
PO Fit ............................................................................................................ 134
Conclusion ........................................................................................................... 1 3 5
REFERENCES ................................................................................................................ 136
APPENDICES ................................................................................................................. 234

List of Tables

Table 1. Summary of Hypotheses and Planned Analytic Methods ................................. 160
Table 2. Descriptive statistics, internal consistency estimates, uncorrected and

corrected correlations between the CR1 and [PI scales (N = 135) ............................ 163
Table 3. Test-retest correlations of CR1 and IPI scales at a one-week interval (N =

29) ............................................................................................................................. 164
Table 4. Race and Sex Composition of the Sample. ....................................................... 165
Table 5. Age composition of the sample (Valid N = 251,355) ....................................... 166
Table 6. Categories of Turnover and Incidence .............................................................. 167
Table 7. Descriptives statistics and intercorrelations among restaurant-level

variables. ................................................................................................................... 168
Table 8. Hierarchical regressions for mean ability levels as a predictor of controllable

profit (N =2,206). ...................................................................................................... 169
Table 9. Hierarchical regressions for mean conscientiousness levels as a predictor of

controllable profit (N =2,207). .................................................................................. 170
Table 10. Hierarchical regressions for mean agreeableness levels as a predictor of

controllable profit (N =2,207). .................................................................................. 171
Table 11. Hierarchical regressions for mean extraversion and extraversion2 levels as ,

a predictor of controllable profit (N =2,205). ........................................................... 172
Table 12. Hierarchical regressions for mean conscientiousness X manager tenure

levels as a predictor of controllable profit (N =2,206). ............................................. 173
Table 13. Hierarchical regressions for mean agreeableness X manager tenure levels

as a predictor of controllable profit (N =2,206). ....................................................... 174
Table 14. Hierarchical regressions for mean ability X manager tenure levels as a

predictor of controllable profit (N =2,206). .............................................................. 175
Table 15. Hierarchical regressions for coefficient of variability in conscientiousness

as a predictor of controllable profit (N =2,029). ....................................................... 176
Table 16. Correlations among ll-month aggregates of restaurant level demographic

diversity measures and turnover rate (N = 2496-2498). ........................................... 177

Table 17. Multiple regression analyses for the prediction of 11-month aggregate of
turnover rates with ll-month aggregates of age, sex, and demographic diversity... 178
Table 18. Means, standard deviations, and intercorrelations for racial diversity across

the 11 time periods (N = 2,388). ............................................................................... 179
Table 19. Means, standard deviations, and intercorrelations for age diversity across

the 11 time periods (N = 2,388). ............................................................................... 180
Table 20. Means, standard deviations, and intercorrelations for controllable profit

across the 11 time periods (N = 2,388). .................................................................... 181
Table 21. Intercorrelations for controllable profit and racial diversity across the 11

time periods (N = 2,388). .......................................................................................... 182
Table 22. Intercorrelations for controllable profit and age diversity across the 11 time

periods (N = 2,388) ................................................................................................... 183
Table 23. Example of the period-person data structure used for the multilevel

survival analyses. ...................................................................................................... 184

vi

Table 24. Results of HnLM examining the effect of individual sex X restaurant sex

composition interaction on turnover probabilities. ................................................... 185
Table 25. Results of HnLM examining the effect of individual sex X restaurant sex
composition X time interaction on turnover probabilities. ....................................... 186
Table 26. Results of HnLM examining the effect of individual race X restaurant race
composition interaction on turnover probabilities. ................................................... 187
Table 27. Predicted turnover hazards as a function of employee race and store racial
composition. .............................................................................................................. l 88
Table 28. Results of HnLM examining the effect of individual race X restaurant race
composition X time interaction on turnover probabilities ........................................ 190
Table 29. Predicted turnover hazards as a function of employee race and store racial
composition and time. ............................................................................................... 191
Table 30. Results of HnLM examining the effect of individual age X restaurant age
composition interaction on turnover probabilities. ................................................... 193
Table 31. Predicted turnover hazards as a function of employee age X store age
composition interaction. ............................................................................................ 1 94
Table 32. Results of HnLM examining the effect of individual age X restaurant age
composition X time interaction on turnover probabilities. ....................................... 196
Table 33. Predicted turnover hazards as a function of employee age X store age
composition X time interaction ................................................................................. 197
Table 34. Results of HnLM examining the effect of individual age X restaurant age
composition X extraversion interaction on turnover probabilities. ........................... 199
Table 35. Results of HnLM examining the effect of individual race X restaurant race
composition X extraversion interaction on turnover probabilities ............................ 200
Table 36. Results of HnLM examining the effect of individual sex X restaurant sex
composition X extraversion interaction on turnover probabilities. ........................... 201
Table 37. Predicted turnover hazards as a function of employee age X store age
composition X extraversion interaction. ................................................................... 202
Table 38. Summary of Hypotheses and Results ............................................................. 204

vii

List of Figures

Figure 1. Hackman’s IPO approach to teams .................................................................. 207
Figure 2. Theoretical model underlying the prediction of team performance ................ 208
Figure 3. Theoretical model underlying the prediction of turnover. ............................... 209
Figure 4. Representation of a simple latent growth curve structural equation model ..... 210
Figure 5. Trends in racial diversity for a sample of 15 restaurants. ................................ 211
Figure 6. Latent growth curve model for racial diversity. N = 2388 restaurants ........... 212
Figure 7. Plot of standardized LISREL estimates of racial diversity slope factor

loadings ..................................................................................................................... 213
Figure 8. Trends in age diversity for a sample of 15 restaurants. ................................... 214
Figure 9. Plot of standardized LISREL estimates of age diversity slope factor

loadings. .................................................................................................................... 21 5
Figure 10. Latent growth curve model for age diversity. N = 2388 restaurants. ............ 216
Figure 11. Trends in controllable profit for a sample of 15 restaurants. ......................... 217

Figure 12. Latent growth curve model for controllable profit. N = 23 88 restaurants. 218
Figure 13. Plot of standardized LISREL estimates of controllable profit slope factor

loadings ..................................................................................................................... 219
Figure 14. Combined latent growth curve model for racial diversity and controllable .

profit. N = 2388 restaurants. ..................................................................................... 220
Figure 15. Predicted trajectories for controllable profit based on several different

levels of racial diversity ............................................................................................ 221
Figure 16. Predicted yearly total impact of racial diversity on controllable profit ......... 222
Figure 17. Predicted yearly controllable profit for several hypothetical racial

compositions. ............................................................................................................ 223
Figure 18. Combined latent growth curve model for age diversity and controllable

profit. N = 2388 restaurants. ..................................................................................... 224
Figure 19. Predicted cumulative survival function based on logistic regression

analysis of person-period dataset with linear and quadratic time factors. ................ 225
Figure 20. Predicted turnover hazard as a function of sex and two different store sex

compositions. ............................................................................................................ 226
Figure 21. Predicted turnover hazard as a function of sex and two different store

compositions and time ............................................................................................... 227
Figure 22. Predicted turnover hazard as a firnction of race and three different

restaurant race compositions. .................................................................................... 228
Figure 23. Predicted turnover hazard as a function of race, three different restaurant

race compositions and time. ...................................................................................... 229
Figure 24. Predicted turnover hazard as a function of age and three different

restaurant age compositions. . ................................................................................... 230
Figure 25. Predicted turnover hazard as a function of age, three different restaurant

age compositions, and time . ..................................................................................... 231
Figure 26. Predicted turnover hazard as a function of race, two different restaurant

racial compositions, and two levels of extraversion ................................................. 232
Figure 27. Schematic summary of results ....................................................................... 233

viii

INTRODUCTION

Over the last few decades there has been a dramatic shifi from individual-based work
towards the reliance on work teams to accomplish organizational goals (e.g., ll gen,
Major, Hollenbeck, & Sego, 1995). In fact, numerous authors suggest that successful
teams are critical for achieving organizational objectives (1] gen, 1999). This has
generated a huge amount of theoretical and empirical literature on the nature of team
effectiveness; however, from a personnel selection perspective we still know relatively
little about the impact of team composition on team effectiveness. This is despite the fact
that personnel psychologists exert a strong influence on team composition and the huge
body of literature that has successfully identified methods for predicting individual
outcomes (i.e., job performance; Banick & Mount, 1991; Hunter & Hunter 1984).

The few studies that have examined the prediction of team effectiveness used small
samples and have not taken full advantage of the more organizationally relevant criteria
that teams often produce as compared to that of individuals. Teams often produce output
that can be easily measured and has immediately obvious relevance to the organization
(e.g., financial performance; Schneider, Smith, & Sipe, 2000); however, the only
selection-oriented studies to date have focused on team performance ratings. The ability
to predict more organizationally relevant criteria would substantially enhance the utility
of personnel selection (Schneider et al., 2000). This is underscored by research
indicating that performance ratings are subject to a host of factors not directly related to
the construct of job performance including measurement instability, rater idiosyncrasies
and rater perspective (e. g., Conway, 1996; Gregarus & Robie, 1998; Mount et al., 1998;

Scullen, Mount, & Goff, 2000; Viswesvaran, Ones, & Schmidt, 1996).

A number of authors have recognized two additional important shifis in today’s
workplace. First, the United States economy is now largely service—oriented which
represents a significant change from the manufacturing emphasis of only a few decades
ago; service jobs are estimated to represent 41% of the American workforce (National
Research Council, 1999). Researchers have acknowledged that this is likely to lead to an
emphasis on a different Skill set requirement for service workers (Schmitt & Chan, 1998),
yet research on this issue has appeared only recently (Klimoski & Jones, 1995). Second,
the workforce is becoming increasingly diverse with regard to race, sex, and age. Many
authors have discussed this trend and its numerous practical, theoretical, and research
implications. The research reported here examines these issues by studying the
composition of diverse teams in a service setting from a personnel selection perspective.

In particular, the large sample research reported here looks at the relationship
between team composition and team effectiveness in a customer service-oriented setting.
Team composition is broadly defined in terms of ability and personality constructs
typically used in personnel selection, as well as the demographic variables of race, sex,
and age. Team effectiveness is measured by financial outputs and turnover within the
team. The use of financial performance as criteria is especially important in the present
research because each team works in a fast-service restaurant that can be considered its
own organization. Hence, this research has the potential to answer many questions
relating to whether the knowledge we have gained in personnel selection directly relates
to organizational effectiveness (Ostroff & Bowen, 2000; Schneider, 1996; Schneider et
al., 2000). Team composition is hypothesized as the main predictor of team performance,

while the fit within a given personality or demographic composition (i.e. mix) is used as a

predictor of turnover as well as performance. In addition, two potential moderators of the
effects of demographic and personality-based fit are studied: Time and Extraversion.

This paper is organized as follows. First, teams are defined, and the dominant
perspective in teams research is discussed. Second, a theoretical model is described that
underlies the proposed relationships examined in this research. This model serves as an
organizing framework allowing us to understand how the constructs and variables
described above might be related. Third, the theoretical and empirical underpinnings of
the specific linkages in the model are discussed in turn, and finally, hypotheses tested in
the present study are developed.

What Are Teams?

It is important to explain what constitutes a team and to differentiate it from similar
concepts. For the purposes of this paper a team will be defined as a collection of
individuals working in a differentiated manner on an interdependent task in a work
setting towards an organizationally relevant goal. This definition includes elements from
several influential definitions (Guzzo & Dickson, 1996; Guzzo & Shea, 1992; Hackrnan,
1987; Morgan, Glickman, Woodward, Blaiwes, & Salas, 1986; Salas, Dickinson,
Converse, & Tannenbaum, 1992) and provides a straightforward way of determining
whether something is a team. This definition is composed of five key points. First, a

team consists of a collection of individuals; that is, one individual does not constitute a

 

team. While pairs of individuals are often called dyads, and while there have been
studies on the effects of size on group processes (see Morgan & Lassiter, 1992; Nieva,
Flieshman, & Rieck, 1978), Guzzo & Shea (1992) note that size is not a primary concern

in the team research. Second, a team’s members must be working in a differentiated

 

m; in other words, though there may be overlap in duties there must be some
qualitative difference between the duties being performed by the team members. Third,
the task required of the team members must be interdependent in that at least some aspect
of each team member’s functioning impacts the rest of the team. Fourth, in order to be a
team these individuals must exist within a work setting. This is because the focal context
of the present research is on work in an organization. Last, there must be an
organizationally relevant goal. The team must exist to accomplish a specific task or set
of tasks impacting the organization.

It is also instructive to differentiate tea_m_s from @1133. Whereas teams exist at work
and are generally studied by industrial/organizational or management scholars, the latter
are generally studied by social psychologists (also note that researchers use the term
‘workgroup’ to refer to teams as well, and that the two terms will be used interchangeably
here). Teams are a specific type of group; that is, groups can occur anywhere (i.e., not
necessarily at work) and do not necessarily have specific goals. As such, groups are most
frequently studied in laboratory settings and naturalistic settings other than at work. This
reflects different theoretical and practical orientations of the two camps, as social
psychologists are often concerned with more basic group process than are team
researchers, who are interested in specific types of groups in work settings (i.e., teams;
Driskell & Salas, 1992). This frequently results in substantively different research
questions that are asked using groups versus teams. For instance, much of the group
research examines collections of individuals that work together for only very short
periods of time (generally only a few hours total) and on tasks that bear little resemblance

to work tasks. A number of authors have indicated that this makes it difficult to

generalize from studies of groups to studies of teams because social processes that unfold
over time are absent, and the nature of the tasks are often highly artificial. Further,
groups often consist of college students participating in research for extra credit, a
population that frequently is far narrower than the ones studied by team researchers.
Thus, although group research can inform our knowledge of team functioning one must
carefully consider whether the nature of a given group and the context in which it exists
allows us to make valid inferences about teams (e.g., Cohen & Bailey, 1997; Sundstrom,
McIntyre, Halfhill, & Richards, 2000).
The Input-Process-Output Approach To Teams

The prevailing approach to team research and theory development follows the input-
process-output (IPO) model frequently attributed to McGrath (1984). This approach
suggests that team process mediates the relationship between team inputs and outputs.
Hackman’s (1987; Hackman & Morris, 1975) elaboration on this notion is depicted in
Figure 1. As can be seen in the figure, team inputs are thought to exist at multiple levels;
that is, they can be properties of individuals, workgroups, or of the environment in which
the team exists. Examples of individual level factors described in Hackman’s original
formulation include skills, personalities, and attitudes. Workgroup-level factors include
attributes that can be ascribed to the team as a whole, including size, the structure of the
group and cohesiveness. Environment-level factors are influences in the broader context
in which the group operates (e.g., reward structure).

Process, which is responsible for converting inputs into outputs, was originally
conceptualized as “all the observable interpersonal behavior” (Hackman & Morris, 1975,

p. 49) that exists within the workgroup over a period of time. This notion has since been

expanded by a number of researchers. For instance, Gladstein (1984) broadened this
view by incorporating prior thinking describing team process as having either a task or
team preservation function. For example, weighing individual inputs appropriately
would be considered a task function, while fostering smooth interpersonal
communication and relationships would be considered a preservation function. Later
work has focused on more specific processes as coordination (e. g., Guastello &
Guastello, 1998; Zalesny, Salas & Prince, 1995), cooperation (Swezey & Salas, 1992;
Tjosvold, 1995; Tjosvold & Tjosvold, 1994), task monitoring (see Cannon-Bowers,
Tannenbaum, Salas, & Volpe, 1992), strategy formation (e. g., Tesluk & Mathieu, 1999)
and communication (e. g., Sanderson & Benda, 1998). Nonetheless, process is still
frequently broadly conceptualized as anything that turns inputs into outputs.

Team outputs are multidimensional and consist of three components: (1)
performance, (2) affective outcomes such as team member satisfaction, and (3) the
continuance of the team (i.e., or capability of team members to work together in the

fiiture). These three outputs have been collectively termed tgam effectiveness.

 

Sundstrom et al. (2000) indicate that about half of the studies they reviewed used a global
indicator of effectiveness such as performance or productivity. Other outcomes that were
used less commonly include cohesion, satisfaction, attitudes, and behavioral outcomes
such as turnover, absenteeism, accidents, and prosocial behavior.

The IPO approach to teams has received considerable empirical support, both in
terms of tests of the overall linkages as well as studies focusing on only one linkage (e.g.,
that between team process and effectiveness; see Cohen & Bailey, 1997; Sundstrom et

al., 2000). Despite the fact that a number of researchers have expanded and/or modified

this conceptualization, the current thinking is that the basic model still holds and is a
useful way of approaching teams theory and research. Indeed, a number of researchers
have stated that this model is widely used or that it underlies all team research (e. g.,
Klimoski & Jones, 1995; Morgan & Lassiter, 1992; cf. Sundstrom, de Meuse, & Futrell,
1990).
Theoretical Model Of Team Effectiveness

Now that we have clarified the nature of teams and team effectiveness, theoretical
models underlying the research reported here are introduced in Figures 2 and 3 for team
performance and turnover, respectively. As can be seen in the figures there are constructs
and linkages that are thought to exist at both the individual and team level (represented
by solid lines that go from the individual to the team level), as well as interactions
between the two (i.e., cross-level interaction effects; Rousseau, 1977; 1985), and a
relationship among non-equivalent constructs at two levels (represented by a dashed
line). The following sections briefly describe these models. Although not all of the
constructs in these models are operationalized in the present study (those that are
measured are underlined) it is nonetheless important to clearly explicate the mechanisms
via which hypothesized relationships supposedly occur. Similarly, these models are not
presented as being exhaustive, but rather are meant to convey certain aspects of the
relationships that are studied here.

Figure 2 presents the theoretical model of team performance underlying the research
reported here. The first link (A) in the model suggests that individual performance is a
function of KSAOs and motivation. Next, KSAOs form the basis of team composition

(link B), and the fit between these KASOs and team composition, a cross-level effect, or

interaction between two variables at different levels of analysis (Rousseau, 1985), is
thought to impact individual affective outcomes (link C). Link D suggests that these
affective outcomes impact individual motivation, and thus, individual performance.
Further, individual level motivation, in turn, should impact team processes (Link E).
Moving to effects at the team level, composition is thought to impact team process (Link
F). In other words, the nature of the team in terms of its members KSAOs should impact

team processes. These team processes exert cross-level effects, termed transformation,

 

the manner in which individual performance is converted into team performance (Link
G).

Figure 3 presents the theoretical model of turnover underlying this research. First,
note that two linkages are identical to those in Figure 2, namely the relationships between
KSAOs and composition, and the fit cross-level effects of the two on affective outcomes
(links B and C, respectively). The first new Link (I) simply suggests that there are
individual KSAOs that predict individual level turnover. Link J indicates that turnover
can be conceptualized at the team level as well as the individual level (i.e., individual
turnover and team viability). The last link in the model, Link K, indicates that team
composition should impact team viability.

This paper is organized around the predictor constructs of interest within the broader
context of these two models. Taking each in turn, we review the relevant empirical and
theoretical work to determine how each construct might fit into these models, formulating
hypotheses along the way.

KSAO — Performance Linkages at the Individual Level

Personnel selection is devoted to the study of individual differences that can be used to
select people for jobs (e. g., Murphy, 1996). A number of different constructs and testing
methods are used in personnel selection including cognitive ability, personality, job
preferences, biodata assessments, assessment centers, situational judgment tests, and
work samples. The present study examines cognitive ability and Conscientiousness, two
constructs frequently used in personnel selection. In addition, two other constructs,
Extraversion, and Agreeableness are used because they have been shown to be especially
pertinent in team-based customer-service environments. Accordingly, the following
sections describe conceptualizations of individual job performance and explain how the
predictors used in this study have been shown to fit into these models.
The Construct of Job Performa_n_c_:_e

Campbell, McCloy, Oppler, and Sager (1993) suggest that up until the 1990s
individual job performance was viewed as a unitary construct, with any data indicating
otherwise being attributed to statistical artifacts or unreliability. In response to this
perspective, as well as the laments of many researchers about ‘the criterion problem’
(Austin & Villanova, 1992; Landy & Farr, 1983; Smith, 1976), Campbell et a1. and
Borman and Motowidlo (1993, 1997) introduced multidimensional conceptualizations of
the job performance construct. Campbell et al.’s formulation consists of eight factors that
can be thought of as facets of Borman and Motowidlo’s two-factor model of task and
contextual performance. Of these two models, the latter has by far earned the most
attention in the empirical literature. Task performance is described as the technical core
of job requirements for a specific job. For example, the specific task components of a

software engineer’s job might be writing efficient and correct software code, as well as

debugging existing code. Contextual performance, on the other hand, is seen as anything
that contributes to organizational effectiveness aside from task performance. Thus,
contextual behaviors are seen as less role-prescribed, relatively constant across jobs, as
not contributing to the specific technical focus of the organization, and as being preceded
by dispositional or volitional as opposed to ability-based constructs. Aspects of a
software engineer’s contextual performance might be organizational citizenship
behaviors such as volunteering for organizational events, helping other coworkers, or
behavior that helps to maintain cohesion within the organization, for instance.

The existing literature indicates that this approach to understanding job performance
has empirical and theoretical merit as well as practical utility. Empirically, a number of
studies support the distinction between contextual and task performance (e.g., Conway,
1996, 1999; Johnson, 2001; Motowidlo & Van Scotter, 1994; Van Scotter & Motowidlo,
1996). Second, these two domains of performance have been used by many researchers
to make theoretical and practical contributions in personnel selection (e. g., Borman &
Motowidlo, 1997; Hattrup, O’Connell, & Wingate, 1998; Hattrup, Rock, & Scalia, 1997;
Motowidlo, Borman, & Schmit, 1997). This distinction is pertinent to the present study
because the precursors of the two aspects of job performance are thought to be different,
as is the relative importance of the two (LePine & Van Dyne, 2001). Specifically,
cognitive ability is thought to be the main predictor of task performance. On the other
hand, contextual performance should be very important in team-based work
environments in general, as well as those emphasizing customer service interactions in
particular. This is because behaviors that are critical in teamwork such as interpersonal

facilitation and cooperation are thought to be among the core components of contextual

10

performance (e.g., LePine, Hanson, Borman & Motowidlo, 2000; Van Scotter &
Motowidlo, 1996). In sum, this suggests that cognitive ability should be helpful in
predicting aspects of team (task) performance, though other constructs that predict team
process (i.e., contextual) performance should be examined as well. The existing literature
suggests that personality is likely to be an important predictor of contextual performance
(e. g., Hogan, Rybicki, Motowidlo, & Bonnan, 1998; Lepine & Van Dyne; Murphy &
Shiarella, 1998; Hurtz & Donovan, 2000). Accordingly, the following sections describe
in more detail the nature of cognitive ability and several personality constructs used in the
present research.

Cognitive Ability — Performance Linkagg Cognitive ability represents the totality of
intellectual resources an individual can bring to bear. It is well known that specific
abilities have a strong tendency to exhibit positive manifold; that is, they are all
positively correlated and the nature of these correlations suggests a single underlying
factor (Ackerman & Humphreys, 1990; Carroll, 1993; Hunter, 1986; Jensen, 1980).
There is a vast literature supporting the notion that cognitive ability consistently predicts
individual job performance for all jobs (Hunter & Hunter, 1984; Hunter, 1986; Ree &
Earles, 1992; Schmidt & Hunter, 1998; Schmidt, Ones, & Hunter, 1992). The main
mediator of this relationship is thought to be job knowledge (e. g., Hunter, 1986).
Although the prediction of job performance is a firnction of job complexity, with better
prediction seen in more complex jobs, the true relationship between these two constructs
is estimated to be substantial (93:56, .51, .40, & .23 for high, medium, semi-skilled, and

unskilled jobs, respectively; Schmidt & Hunter, 1998). Thus, with regard to the

11

theoretical model presented in Figure 2, it is easy to see how individuals higher in
cognitive ability will achieve better individual performance outcomes.

Personality. Highly influential work by Mischel (1968) and Onion and Gottier (1965)
substantially curtailed research on personality in Industrial/Organizational psychology for
nearly two decades. Mischel called into question the consistency of behavior across
situations whereas the latter authors concluded that personality had very little or no
relationship to outcomes of interest to I/O psychologists (cf. Ghiselli, 1966). Hough &
Schneider (1996) point out that by the 19803 there was a revival of interest in the study of
personality (e.g., Kenrick & Funder, 1998) among psychologists in general and among
[/0 psychologists in particular. This has been attributed to the increased consensus
surrounding the dimensionality of normal (i.e., nonclinical) personality (e. g., Digman,
1990). In particular, a five-factor model (FFM) of personality has emerged that many
researchers accept as adequately capturing normal personality variation (see Costa &
McCrae, 1995; Digrnan & Inouye, 1986; Digrnan, 1990; Goldberg, 1990). The positive
poles of these five constructs are commonly referred to as Conscientiousness, Emotional
Stability, Agreeableness, Extraversion, and Openness to Experience. While there is still
some debate as to the appropriateness of this generalization (e. g., Hough, 1992; Block,
1995), it has undoubtedly focused research in personnel selection that was previously
plagued by the ‘shotgun approach’ to research that used dozens of labels for different
personality traits that typically yielded near-zero relationships with job performance.
More recently researchers have tended to focus more on the five-factor model of
personality constructs to test relationships driven by theory (Driskell, Salas, & Hogan,

1987) and careful job analysis (Banick et al., 2001). This is important because it is

12

thought that relationships between ‘non-cognitive’ predictor constructs such as
personality are thought to be context specific (e.g., Murphy, 1996).

Social and legal pressures to reduce racial differences in selection rates have also
likely impacted the importance of personality. Because racial minorities have been found
to score considerably lower on cognitive ability tests (e.g., Black-White standardized
difference z 1.00; Roth et al., 2001), there has been an increasing emphasis on other
predictors such as personality because they frequently exhibit differences that are either
near zero or substantially smaller (Schmitt et al., 1997). Similarly, because cognitive
ability is so well studied, many researchers have also examined the validity of alternative
predictors after controlling for the effects of ability (i.e., their incremental validity; e.g.,
Clevenger et al., 2001 ). This reflects the practical necessity of demonstrating the benefits
of adding constructs to a test battery containing cognitive ability tests. Consequently, the
hypotheses in this research that focus on team performance will examine the incremental
validity of personality as well.

Conscientiousness. Despite some disagreement, conscientiousness is generally
thought to consist of two facets, dependability and perseverance. Aspects of
dependability include the tendency to be careful, responsible, orderly, cautious, organized
and efficient (Digman, 1990; Goldberg, 1990; 1993). The other facet, perseverance, has
led some to characterize conscientiousness as the ‘will to achieve’ (e.g., Digrnan, 1989).
Perseverance consists of facets that come to mind when one thinks of achievement
motivation (Weiner, 1979; 1985); that is, diligence, hardworking, and showing a general
interest in achievement and performing tasks well. Although a conceptual distinction

exists between these two facets, Mount and Barrick’s (1995) meta analysis found that the

13

 

two had very similar patterns of relationships across a range of different types of job
performance criteria.

Of the five factors, conscientiousness has received the most attention in the personnel
selection literature. The attention conscientiousness has received is likely due to several
large-scale meta-analyses conducted in the last decade reporting that conscientiousness
consistently predicts job performance across a range of job types and criteria (Banick &
Mount, 1991; Barrick, Mount, & Judge, 2001; Hurtz & Donovan, 2000; Mount &
Barrick, 1995; Tett, Jackson, & Rothstein, 1991). Despite widespread concerns about the
impacts that faking might have on validities (e. g., Banick & Mount, 1996; Ellingson,
Sackett, & Hough, 1999; McFarland & Ryan, 2000; Rosse, Stecher, Miller, & Levin, .
1998, Viswesvaran & Ones, 1999), the estimated mean corrected validity across all
criteria in Barrick and Mount’s (1991) first two meta analyses was .22 and .31,
respectively. The traits described above imply that Conscientiousness is a motivational
rather than ability-based construct in that it captures what an individual wants or prefers
to do, rather than what an individual can do (i.e., ‘will-do’ versus ‘can-do’; also, see
Banick, Stewart, & Piotrowski, 2002). Consistent with this notion, the estimated true
validities of conscientiousness in predicting criteria Mount and Barrick (1995) labeled as
‘can-do’ versus ‘will-do’ were .22 and .45, respectively. Further, measures of
conscientiousness are typically uncorrelated with cognitive ability measures, meaning
that the forrner’s variance in predicting a criterion of interest is incremental. This speaks
to the separate criterion Space assessed by measures of conscientiousness as opposed to

ability measures.

14

With regard to customer service and teamwork, the results of several recent meta
analyses are particularly relevant. First, Frei and McDaniel’s (1998) meta analysis
indicated that the average uncorrected correlation between conscientiousness and several
commercially available customer service tests was .42 based on a total sample size of
over 90,000. These customer service scales, in turn, exhibited a substantial correlation
with job performance (p=.50) in customer-service-related jobs (N=6,495). Mount,
Banick and Stewart (1998) also conducted a meta analysis that narrowly focused on the
validity of F FM constructs measured by the Personal Characteristics Inventory in team
environments and jobs requiring interpersonal interactions (N z 1,500, k=11 studies). The
results indicated that conscientiousness predicted overall individual job performance in
these jobs (p=.26), overall job performance in team jobs (p =.21) and dyadic service jobs
(p =.29), and supervisory ratings of interactions with others (p =.17, and p =.23 in the
two different types of jobs, respectively). These results for the prediction of customer
service job performance are consistent with another meta analysis that examined scales
explicitly measuring FFM constructs (Hurtz & Donovan, 2000; p=.29, N = 1,369).
While the focus of the present study is on team-based jobs in customer service
environments, the research described above makes it clear how conscientiousness relates
to individual performance in these jobs specifically, and to job performance in general.
This is consistent with link A in the model presented in Figure 2.

Agreeableness. The second personality construct studied here is Agreeableness,
typically associated with being trusting, cooperative, good-natured, flexible, caring,
cheerful, and tolerant (Mount & Barrick, 1995; Costa & McCrae, 1995; Digrnan, 1990).

Banick and Mount’s (1991) meta-analysis indicated there was little relationship between

15

this construct and job performance both broadly, and across the different job types and
rating criteria they examined (p=.07). This, and perhaps the lack of an expected
theoretical relationship between agreeableness and job performance, has likely led to it
being used less often as a predictor in personnel selection. However, the two meta
analyses described above provide some support that Agreeableness should be relevant to
team and customer-service based settings. Frei & McDaniel (1998) reported that
agreeableness was highly related to valid, commercially available customer service scales
(7- =61, N=l4,786). Costa & McCrae (1995) also reported nontrivial relationships
between facets of Agreeableness with J. Hogan and R. Hogan’s (1986) Customer Service
Orientation scale, with correlations ranging from .21 to .52. Similarly, Agreeableness
consistently predicted overall job performance in the Mount et al. (1998) study (p =.21),
though this effect was substantially higher for team-based jobs (p =.33) as opposed to
dyadic service jobs (p =.13). The Hurtz and Donovan (2000) study mentioned earlier
reported an estimated relationship in between these two (p=.19) though their sample of
studies did not allow them to break this estimate down according to the type of customer
service job studied. These results are not inconsistent with the earlier large-scale meta
analysis of Banick and Mount (1991) because these types of jobs were not heavily
represented or examined as a moderator in the earlier research. These results are also
logical because these specific types of jobs heavily emphasize positive interpersonal
interactions. The likelihood of positive interactions, in turn is a core component of
Agreeableness’ definition as a construct. So much so, in fact, that failure to find a
relationship in these contexts would seriously call into question either the theoretical or

operational specifications of one or both of the constructs used in these studies. In other

16

words, people who are agreeable, by definition, should be more likely to respond
positively to customer complaints (i.e., to be caring), to be friendly (i.e., to be cheerful),
to be open to other perspectives that are inevitable in team settings (i.e., to be flexible and
cooperative). Indeed, Pugh (2001) reported that employees’ positive emotion evokes
similar responses in customers. This suggests that agreeableness fits into the model with
regard to job performance as Figure 2 suggests.

Extraversion. The third construct studied here, is very similar to Eysenck’s (1947)
Extraversion/Introversion and Tellegen’s (1982) Positive Affectivity. Extraversion is
associated with being sociable, talkative, gregarious, assertive, active, and energetic.
Hence, this construct can be well captured by the extent to which someone is likely to_ be
active in social situations. Like Agreeableness, Extraversion showed a very small
relationship with overall job performance in the original meta-analysis by Banick and
Mount (1991; p =.13) though it showed stronger relationships with individual job
performance in team jobs (Mount et al., 1998; p=.22). In contrast to Agreeableness,
however, Extraversion exhibited a substantially larger relationship with performance in
team settings (p =.22) as opposed to dyadic customer service settings (p=.07). Similarly,
Extraversion exhibited a small relationship with the commercially available customer
service tests studied by Frei and McDaniel (mean 7' =07, N=28,502; though these
authors also report that excluding a particular test which specifically excludes
extraversion substantially increases this relationship to 7' =22). These authors suggested
that talkative customer service employees might unnecessarily extend interactions due to
their gregariousness, though they emphasize the need for firture research on this issue.

Hurtz and Donovan’s estimate of the validity for customer service jobs was closer to the

17

lower bound of these figures (p=.l l)._Thus, the extant literature suggests that
extraversion is important for individual performance in teams though its importance in
customer service jobs seems to be an open question.
Summgg of Individual-Level Predictor-Job Performance Relationshgis.
The model depicted in Figure 2 was designed to provide an overall framework that
depicts construct relationships underlying the hypotheses examined in this research. The
literature described above indicates that the predictor constructs largely relate to
performance given the context of this research (i.e., customer service teams). In
particular, cognitive ability and conscientiousness were described as having consistent
relationships to job performance in all jobs. Agreeableness and Extraversion, on the -
other hand, do not seem to be universally related to job performance, but nonetheless are
related to job performance in team and customer service settings. These findings provide
the basis for the following section, which examines the relationships between team
composition on these constructs and team performance.

Individual Level KSAOs, Team Composition, and Team Performance

Link B of the models presented in Figures 2 and 3 indicate that individual KSAOS

may have an equivalent construct at the team level called composition. Clearly, one can
have composition in terms of ability, personality and demographics, for instance, though
in this figure the type of composition is not specified because the intent of this figure is to
formulate a general framework. To reiterate, the emphasis on the importance of team
composition in this research is because this is the aspect of the IPO model over which
personnel selection practitioners exert the most control; that is, they determine the

particular KSAOs measured in a given selection context (Klimoski & Jones, 1995;

18

Morgan & Lassiter, 1992). Team composition generally refers to the makeup of a team
as a whole as a function of the individual team members’ attributes. This statement
reveals two critical considerations with regard to team composition: (1) The attributes
considered, and (2) the way in which these attributes combine to yield something
meaningfully descriptive of the team as a whole. These two issues are discussed in turn
below.

First, one must consider team composition with regard to what. Researchers have
studied the effects of almost every individual difference one can imagine as elements of
team and group composition. Team researchers have examined composition with regard
to functional background in top management and product development teams (e.g.,
Ancona & Caldwell, 1992) and cognitive ability and/or personality in a variety of
different types of teams (e. g., Barrick, Stewart, Neubert, & Mount, 1998; Barry &
Stewart, 1997; Bond & Shin, 1997; Devine, 1999; Lepine et al., 1997; Neuman, Wagner,
and Christiansen, 1999; Neuman & Wright, 1999; Tziner & Eden, 1985). Group
researchers, on the other hand, have studied composition variables including group size,
group member opinions, ability and personality, and attitudes (e.g., see Bettenhausen,
1991; Cohen & Bailey, 1997). The criteria of interest in these studies are almost
invariably aspects of group process or performance. Group and organizational scholars
have examined composition as it pertains to a wide array of demographic variables
including age, tenure, sex and race, generally studying performance or turnover as the
primary outcomes of interest. This high volume of research speaks to the acceptance of

team composition as an important determinant of team effectiveness. Because the

19

emphasis here is on the prediction of team effectiveness from a personnel selection
standpoint, KSAO composition is the main focus in this section.

A number of researchers have noted the importance of carefully specifying
composition models describing the specific theoretical form of the relationships of
supposedly equivalent constructs at multiple levels of analysis (e.g., Chan, 1998;
Kozlowski & Klein, 2000; Rousseau, 1985). Failing to do so has been shown to result in
a host of problems that cloud valid inference including disconnects between the level of
theory and analysis (Chan, 1998), and misleading data analysis results (Glick & Roberts,
1984; Rousseau, 1985). This specification involves articulating the conceptual meaning
of the higher level construct which should in turn dictate the way in which the lower level
data are aggregated to represent an indicator of the higher level construct (Chan, 1998;
Rousseau, 1985). This is not to say that there also could not be meaningful measures of
the focal construct at the level at which it is said to exist, but merely that there are
oftentimes practical or theoretical constraints that dictate otherwise. Here, aggregation is
necessary because KSAOS are measured at the individual level in personnel selection.
Thus, we must describe a composition theory that specifies the form of the relationships
between individual KSAOs and team composition. Chan presented a typology of
composition models that articulates a number of possible ways in which composition
might be conceptualized and operationalized. Again, properly conceptualizing
composition models allows one to choose an aggregation method that is consistent with
the conceptualization.

It appears to be accepted as a fact among team researchers that the nature of the task

is critical in understanding how individual performance relates to team performance

20

(McGrath, 1984). Assuming that KSAOs are related to individual performance, this
allows researchers to formulate composition models specifying the way in which
aggregated KSAOs might relate to team performance. Of several task typologies
(Herold, 1978; McGrath; 1984; Shaw; 1971), Steiner’s (1972) has received the most
attention. Steiner described three major task types: Additive, Conjunctive, and
Disjunctive. Additive tasks are those in which the sum of team members’ performances
equals the performance of the team; conjunctive tasks are those where the lowest-
perforrning team member defines the overall performance of the team. Conversely,
disjunctive tasks are those in which the best team member’s performance defines the
team’s performance. These three task types suggest that the average, minimum, and ,
maximum team member KSAOs would be appropriate representations of team KSAOs,
respectively, if one were interested in predicting team performance. The task of fast-
service restaurant teams, which are the focus of this study, fits most closely into the
additive model, though it is not strictly additive (see the Method section for more details).
This suggests that the mean of team member KSAOs represents the most appropriate
operationalization of team performance. First, the scope of the team task is relatively
large and the task is highly interdependent and thus generally speaking no single team
member will have a large overall impact. Second the task does not require specialized
skills and individual members exhibit horizontal redundancy in that team members can
perform each other’s duties (see Lepine et al., 1997). While some research has
emphasized the ability of a specific team member, this is only appropriate when a
specific team member occupies an especially important role within the team (e.g., team

leader; Lepine et a1. 1997) or if the ability of one team member can dramatically impact

21

the overall performance of the team (e.g., Laughlin, 1980). The nature of the task
suggests this is not the case here.

The most common ways to index the entire team’s standing on a KSAO would be
with the mean and the variance. As Chan (1998) and others (e.g., Neuman & Wright,
1999) indicate, multiple operationalizations could be appropriate given theories to
support those operationalizations; that is, operationalizations might differ according to
the specific KSAO under consideration, or the focus of a particular theory that specifies
the process via which the operationalization might relate to another construct of interest
(e. g., Kozlowski & Klein, 2000). Hence, the sections below discuss the way in which the
mean KSAO level of team members should relate to team performance; the importance
of the variance of team member KSAOs is discussed later.

Ability Composition and Performance.

There is research and theory to support the notion that the mean level of cognitive
ability within a team represents a meaningful construct. We will call this construct gag
gggnitive resources Cl‘ CR). Williams & Stemberg (1988) labeled the same notion as
‘group intelligence’ described as “the functional intelligence of a group of people
working in a unit.” (p. 356). There are a number of reasons to think that TCR should
relate to team performance (Arthur, Tubre, & Hanson, 2000). First, TCR should be
related to the quality of the individual performance exhibited within a team. This is
because teams that have a high TCR have members who are likely to exhibit high
performance because ability is consistently related to performance. Second, teams high
in TCR should also exhibit processes that allow them to better translate individual output

into team performance (link F in Figure 2; Lepine et al., 1997). This is because a number

22

of process variables are thought to have cognitive components, including task
monitoring, coordination, and integration of individual performance products. This is
consistent with research reported by Stevens and Carnpion (1994), who developed a test
designed to measure teamwork KSAs (i.e., a test to predict effective team process). This
test’s estimated true correlation with a general cognitive ability measure approached
unity, and test scores consistently predicted supervisors’ and peers’ ratings of individuals’
teamwork skills. Further, the cognitive ability subtests also tended to consistently predict
supervisory ratings of individuals’ teamwork skills, though the correlations with peer
ratings were not significant.

Rather than a direct, linear effect from team process to performance, Figure 2 further
suggests that team process moderates the relationship between individual and team
performance (Weingert, 1997). Recall the earlier discussion in which team process was
described as the transformation of team inputs into outputs. Link G implies that teams
with better process yield better team outputs than teams with less effective process for a
given level of individual performance. In other words, individual performance should
make a more direct contribution to team performance when team process is optimal. This
is because team process, by definition, ensures that the individuals focus on producing
the correct inputs, that the inputs are coordinated, and that they fit together according to
the demands of the task. This is especially true if a given task is highly interdependent.

This is consistent with the results of a classic study of team composition effect
conducted by Tziner & Eden (1985). These researchers examined the effects of 2-person
tank crews’ cognitive ability composition on commanders’ ratings on tank performance.

These authors concluded that the performance of crews composed of two high ability

23

members was substantially higher than their individual abilities alone would have
yielded. This effect is consistent with the logic described above; that is, high ability
teams developed effective coordination processes that enabled their overall product to be
superior to that of the individual members.

Research by Lepine et al. (1997) is also consistent with the Transformation link in
Figure 2. These researchers examined the effects of leader cognitive ability and that of
the staff in a cognitively complex laboratory simulation where staff cognitive ability was
operationalized as the lowest-ability member (their task was conjunctive; hence, this was
an appropriate operationalization). Studying hierarchical decision-making teams, they
found that these two ability levels had a multiplicative relationship with team decision
accuracy. In particular, staff cognitive ability was positively related to accuracy when
leader cognitive ability was high, but there was a slight negative relationship when the
opposite was true. This is consistent with the model in Figure 2 because the leader in
their task is in charge of integrating and transforming the recommendations of the
individual staff members into a team performance product. Hence, in this situation,
leaders cognitive ability can be conceptualized as a strong indicator of the quality of team
process (as the task of integrating the recommendations is cognitively complex), thus
supporting link G in Figure 2. Similarly, several laboratory (Devine, 1999; Williams &
Stemberg, 1988) and field studies (Banick et al., 1998; suggest that mean team level
cognitive ability relates to team performance. Thus, the following hypothesis is

proposed:

H1. Mean cognitive ability will be positively related to
team performance.

24

Personality gt the Team Level

Numerous authors have indicated that the personality characteristics of team
members should relate to team process and effectiveness (e. g., Fleishman & Zacarro,
1992; Hogan, Raza, & Driskell, 1988; Moynihan & Peterson, 2001). Though much
laboratory research has been conducted on this issue, this research has used so many
different personality traits and tasks that Jackson (1992) concluded that no firm
conclusions could be drawn (also see Levine & Moreland, 1990; Morgan & Lassiter,
1992). This research is also hard to characterize because it lacked the organizing
framework provided by the F FM (Banick et al., 1998a). Thus, the interested reader is
referred elsewhere for reviews of this research (Jackson 1992).

Examining personality at the team level by aggregating individual personality scores
does not necessarily imply that such scores described the team’s personality (Lepine et
al., 1997). Rather, it merely suggests that aggregating individual personalities tells us
something meaningful about the people in the team itself (Lepine et al., 1997). This is
likely to be fruitful to the extent that personality is related to either individual
performance or team process in that either of these relationships implies that team
personality and team performance might be related. The specific ways in which
aggregated personality might relate to team performance is discussed below.
(firscientiousness Composition and Team Performance. Earlier in this paper the nature
of conscientiousness at the individual level was described. The discussion now turns to
the notion of conscientiousness at the group level. One of the steps in adequately
specifying a composition model is theory relating the composed construct to other

constructs of interest. The personality levels of a group as a whole have been cited by

25

numerous authors (e. g., Driskell et al., 1987) as likely determinants of team process and
outcomes, and several authors have indicated this should be the case for
conscientiousness in particular (Banick et al., 1998; Barry & Stewart, 1997; Bond &
Shiu, 1997). The mean level of conscientiousness should be related directly to team
performance and process, and one would expect there to be indirect effects as well. In
terms of direct effects, Barry and Stewart (1997) argue that a conscientious team should
have less process loss (Steiner, 1972) due to free riding and social loafing, two
phenomena commonly examined by social psychologists in which group members
withhold effort from group tasks (F orsyth, 1999). There is a large body of research
indicating that these effects do indeed occur, and due to the task-oriented nature of people
who are conscientious, team processes should be less likely to suffer from these effects
when a team’s members are high in conscientiousness. Similarly, teams with high levels
of conscientiousness by their nature should attend more diligently to the importance of
team process. This is because teams such as these are composed of individuals who are
by their nature interested in efficiency and producing work that is of high quality. Thus,
there are several reasons to believe that a team high in conscientiousness should display
more effective team process. Given the research described earlier, conscientious teams
should also exhibit better performance via the indirect effects on individual performance
shown in Figure 2. Similarly, the individual work products should be more effectively
compiled and coordinated into effective team performance.

Very few studies have examined the role of conscientiousness at a team level. Barry
and Stewart (1997) hypothesized that the proportion of highly conscientious members of

a student group would positively relate to task focus, cohesion, group process and

26

performance; however, none of these hypotheses were supported. This may be because
they reduced the variance of their team level conscientiousness variable by examining a
proportion of members exceeding a particular threshold. Equally odd, individual level
conscientiousness scores were not associated with individual performance. These
researchers mentioned several potential problems with the measures and criteria they
used as other possible explanations for the unexpected results.

Barrick et al. (1998) examined the relationship between the mean level of
conscientiousness in a sample of 51 assembly and maintenance teams in manufacturing
plants. In this research, the correlation between mean conscientiousness score and
supervisory ratings of team job performance was significant (r=.26), which was
consistent with the hypotheses. Further, as would be expected by the typically near-zero
correlation between conscientiousness and cognitive ability, the incremental validity of
conscientiousness was very similar to its zero-order validity. Neuman et al. (1999) also
examined the team mean conscientiousness scores as a predictor of small teams’
performance in automotive stores. Their results were consistent with those of Banick
and colleagues, though the correlation obtained in this sample was somewhat higher
(r=.40) for performance ratings from two different sources.

Neuman and Wright (1999) reported similar results in their sample of 79 human
resources service teams; however, these researchers examined the minimum of
conscientiousness scores within the team rather than the mean. This was because the
teams’ task was described as conjunctive and required distributed expertise (1] gen et al.,
1995), implying that the team’s performance should be a function of the lowest-

perforrning member.

27

In summary, only 3 studies have examined the relationship between mean
conscientiousness scores and team performance. All three utilized small samples (though
fairly large in the teams research literature), and 2 reported evidence of the hypothesized
relationships between conscientiousness and ratings of team performance. A fourth
study, which examined the team minimum on conscientiousness due to the nature of the
team task, also reported a significant relationship between this index and team
performance. One of these studies reported that conscientiousness added incremental
validity above that provided by cognitive ability. Thus, while the research in this area
certainly suggests that conscientiousness does indeed predict team performance,
additional research is needed. Specifically, the research reported here examines team
financial performance, as well as other aspects of team performance that should be
closely linked to the overall conscientiousness of a team. The value of using criteria
specifically targeted to predictor constructs has been widely acknowledged; however,
none of the research described above has either used financial performance or specifically
targeted performance outcomes as criteria. While some may argue that financial
performance is not an outcome that is specifically targeted towards any predictor
constructs, given that the end goal of business is to make money, and that the major role
of personnel selection is to select employees for businesses, it seems reasonable to say
that profitability is a very appropriate (though admittedly stringent) criterion for

personnel selection research. Thus, it is hypothesized that:

H2. The team mean conscientiousness score will be
positively related to team performance.

28

H3. The team mean conscientiousness score will be
positively related to team performance after controlling for
the effects of team cognitive ability.

Agreeableness Composition and Team Performance. Based on the results of the meta-
analysis by Barrick et al. (1998) one would expect that teams with higher levels of
agreeableness would perform better. This is because such teams should experience more
positive interpersonal interactions, thus allowing their teamwork to result in better
translations of individual outputs into team performance. Research conducted by George
(1990) provides some indirect support for this notion. In her research, sales teams’
negative affective tone was strongly and negatively related to prosocial behavior within
the work team (r = -.57). While it is easy to see how negative affective tone might be
negatively related to high levels of agreeableness in a team, unexpectedly, this construct
was not related to team performance. Similarly, positive affective tone was unrelated to
team performance as well. Thus, this study provides conflicting messages about the
importance of team level agreeableness.

The Barrick et al. (1998) and Neuman et al. (1999) studies discussed earlier, however,
provide stronger support for the notion that mean levels of agreeableness should be
related to team performance. These researchers found that the mean level of
agreeableness was positively related to team performance (r=.34, and r=.41,
respectively). These results are consistent with those of Neuman and Wright (1999) who
reported that minimum team agreeableness on a conjunctive task was positively related to
team performance and team interpersonal interactions. In both of these studies
agreeableness was essentially unrelated to cognitive ability, and thus these constructs

provided validity that was incremental to that of ability.

29

In contrast, Barry and Stewart (1997) found that mean levels of agreeableness in
MBA student workgroups were unrelated to their measures of group process and
performance. One explanation for this effect may lie in the single rating of overall
performance that served as the criteria in this study. Based on the results described

above, the following hypotheses were proposed:

H4. Mean level of team Agreeableness will relate to team
performance

H5. Mean level of team agreeableness will relate to team
performance after controlling for cognitive ability.

The impact of the minimum level of agreeableness within a team will also be
examined. This is because it has been suggested that a single disagreeable team member
can seriously disrupt the functioning of a team (Barrick et al., 1998). However, because
the teams studied here are fairly large (i.e., generally more than 10 employees), it is not
clear whether the impact of a single disagreeable team member will be substantial (i.e.,

detectable). Thus, these relationships will be examined in an exploratory fashion.

Extraversion Composition and Team Performance. The form of the relationship between
extraversion and team performance might be less straightforward than that proposed for
agreeableness and conscientiousness. There are several theoretical and empirical

findings suggesting that extraversion should impact team performance via an indirect
effect through team process. First, recall from the earlier discussion of several meta
analyses that extraversion did not consistently relate to overall job performance unless the
job was in a team setting. This may be because the criterion space in team settings

includes team process relevant contributions whereas this is not the case for individually

30

or dyadically based jobs. Second, extraverts by their nature prefer interpersonal
interactions (Moynihan & Peterson, 2001), and communication has been shown to be an
important part of team process; hence, as team process is related to team performance,
one would expect that Extraversion should be related to team performance. Third, Jung
and Sosik (1999) recently found that a group’s overall preference for group work
predicted team performance in group decision-making tasks after controlling for several
other relevant variables. Campion, Medsker, and Hi ggs (1993) found that the same
preference predicted satisfaction within the group. Preferring group as opposed to
individual work settings is emblematic of preferences that extraverts might hold in work
settings.

Indeed, there are several studies that have examined empirically the relationship
between team extraversion and performance in work settings. Barrick et a1. (1998) found
that the mean levels of extraversion were unrelated to team performance. In contrast,
team extraversion was positively related to team social cohesion yet negatively related to
team conflict. Similarly, neither Neuman and Wright (1999) nor Neuman et al. (1999)
reported a significant relationship with their operationalization of team extraversion and
team performance in human resource teams. These results are not consistent with those
of the Barrick et al. (1998) meta analysis, nor are they consistent with the theoretical
underpinnings described above that suggest extraverts should perform better in team
settings.

In contrast, Barry and Stewart (1997) hypothesized and found nonlinear relationships
between extraversion and task focus and performance. In particular, the proportion of

high extraverts in a team had a U-shaped relationship to task focus, which was contrary to

31

their expectations; the prediction of performance showed an opposite pattern, which was
expected. These results indicate that teams that are either relatively introverted or
extroverted exhibited lower task performance than those with moderate amounts of
extraversion. Levels of extraversion that are too low may be detrimental to team
performance because members may not interact, while teams very high in extraversion
may socialize too much and thus do relatively little work. This is consistent with what
one might expect in a large group of extraverts and also might explain the null results
obtained by three of the studies described above. (Interestingly, this is one of few cases
where increased supplemental fit within a group is posited to have negative consequences

for the organization.) Hence, it is hypothesized that:

H6. Mean levels of team extraversion will exhibit a
nonlinear (inverted U-shaped) relationship with team
performance.

The hypotheses described above all relate to the team’s average ability or personality
as a predictor of team performance. The research of Lepine et al. (1997) described earlier
suggests that characteristics of team members with strong influences on team
performance (e.g., a team leader) interact with team KSAO levels to predict performance.
In particular, team ability or personality was more strongly related to team performance
with a high-ability leader. While there are no team members in fast-service restaurant
teams that have as much structural influence as a team leader, the manager of a restaurant
can be thought of as occupying a similar position. This is because the manager has a
great deal of control over many aspects of the team’s work environment that are likely to
shape team effectiveness. Thus, effective managers might develop work processes that

allow team members’ abilities to be more strongly expressed in terms of performance

32

(e.g., Schneider et al., 2000). For instance, managers with longer tenure might have
stronger HR systems in place (Ostroff & Bowen, 2000) that remove barriers to the
expression of abilities and skills. Conversely, less effective managers might create work
environments that do not facilitate the translation of team abilities into performance.
Because job experience is consistently related to performance, we expected that the
managerial tenure would impact team KSAO-performance relationships in the following

manner:

H7. Team manager tenure will moderate the relationship
between mean team KSAO levels and team performance
such that there is a stronger relationship between KSAO
levels and performance for teams with high-tenure
managers than with those with low tenure managers.

Turnover at the Individual Level

Turnover is commonly defined as an employee leaving an organization. Researchers
usually do not include promotions, dernotions, or intra-organizational transfers as
turnover, although these job changes may be of importance in their own right. Turnover
has numerous implications for individuals and organizations. For instance, a substantial
amount of research and theory has addressed the psychological and economic well being
associated with employee turnover (Leana & Feldman, 1994), as well as that of those
who remain on the job after large waves of turnover (e. g., Brockner, 1986; 1987; 1988;
1994; Gowan, Riordan & Gatewood, 1999). Turnover also impacts organizations in
terms of increased training costs, lowered effectiveness, demoralization, and lost
knowledge. Given the array of impacts that have been associated with turnover, it is not
surprising that there is a large body of literature examining the prediction of turnover or

intent to turnover including individual difference, contextual, and attitudinal variables.

33

There are also numerous theories that seek to explain why individuals leave jobs. A
review and integration of these theories is beyond the scope of this paper and is presented
by Griffeth & Horn (1995). These authors present an integration of these theories based
on available theory and meta-analytic evidence. Job satisfaction, commonly
conceptualized as an employee’s global attachment to the job, is one of the central
antecedents of turnover in this integrated model, and the main pathway is via withdrawal
cognitions (i.e., thinking about turning over). Indeed, several recent meta-analyses have
indicated that the estimated true correlation between global job satisfaction and turnover
is approximately -.20 (Link H in Figure 3; Tett & Meyer, 1993; Griffeth & Horn, 1994;
Griffeth, Horn, & Gaertner, 2000). While this correlation might sound trivial, it is
important to remember that turnover is a dichotomous variable, which limits the degree to
which other variables can correlate with it. Similarly, even relatively weak prediction of
turnover can have substantial implications for an organization because of its wide-
ranging impact.

@srtructs Used in Personnel Selectiow Predictors of Tumover.

Despite the large body of literature devoted to examining the impact of individual-
level predictors of turnover there has been relatively little emphasis on constructs
typically used in personnel selection. This may be the case either because there is little
theoretical justification for such research, or because turnover is not typically used as a
criterion in personnel selection. In personnel selection, cognitive ability by far has
received the most attention; however, it is not obvious why one would expect there to be
a relationship between cognitive ability and turnover. Indeed, in a recent meta-analysis,

Griffeth et al. (2000) reported the average corrected correlation between cognitive ability

34

and turnover to be only -.07. Only seven studies in their meta analysis examined this
relationship, as opposed to 67 that examined job satisfaction. While job performance has
been shown to consistently relate to turnover, and cognitive ability to job performance,
the relationship between the former set of variables may be nonlinear (Trevor, Gerhart &
Boudreau, 1997), and thus any relationship between cognitive ability and turnover is
likely to be complex. Thus, the following sections discuss ways in which the personality
constructs studied here might impact turnover. These discussions are preceded by a brief
discussion of the distinction between voluntary and involuntary turnover.

Voluntary versus Involuntary Turnover.

Researchers often draw a distinction between voluntary turnover (VT) and
involuntary turnover (IVT). VT is widely regarded as the appropriate criterion when
examining motivational antecedents of turnover because these approaches are chiefly
interested in what leads an employee to choose to leave a job. The rationale for treating
these two phenomena differently is highlight in the following:

An instance of voluntary turnover, or a quit, reflects an
employee’s decision to leave an organization, whereas an
instance of involuntary turnover, or a discharge, reflects an
employer’s decision to terminate the employment
relationship. . .To treat quits, discharges, and total turnover
as synonymous ignores the markedly different etiologies
and effects of these phenomena. In an organization with
high quit rates, for various reasons employees find it more
attractive to leave than to stay. In an organization with
high discharge rates, however, presumably incorrect hiring
decisions are remedied through termination (Shaw, Delery,
Jenkins, & Gupta, 1998, p. 511-512).

Consistent with this argument, in their organizational-level analyses, these

researchers found that the two types of turnover were distinct (though related), and had

35

different precursors. Despite the near consensus of the important distinction between VT
and IVT there has been disagreement and difficulty in drawing an operational distinction
between the two. For instance, in a review of the turnover literature, Griffeth and Horn
(1994) point out that turnover due to pregnancy is classified as VT or IVT by different
researchers, as is leaving a job at the insistence of one’s spouse. Similarly, in practice it
has proven very difficult to disentangle the two types of turnover. This is true when
using personnel records’ reasons for turnover due to a host of issues relating to litigation,
unemployment benefits, the reputation of a terminated employee, for instance. Similarly,
employees’ own accounts of reasons for turnover may also be inaccurate for a variety of
reasons including social desirability, embarrassment, and a lack of clarity, among others.
Nonetheless, IVT still remains a topic worthy of study. This is because organizations
incur costs of turnover regardless of which party terminates employment. Though one
might think IVT would yield long-term organizational benefits (as terminated employees
are supposedly either performing poorly or perhaps are engaging in misconduct at work)
it would be optimal if such employees could be identified before they are hired.
Choosing an appropriate employee (i.e., one who is less likely to be involuntarily
terminated) should reduce costs to the organization associated with hiring and training
replacements. Similarly, changing jobs also causes disruption in employees’ lives, so
achieving a better fit would seem to benefit both parties in the employment relationship.
Poor job performance or other misconduct that leads to involuntary termination might be
at least in part due to underlying motivational components that can be identified during
selection. Thus, it would seem that a motivational approach to turnover might also

examine IVT though this has largely been ignored in the turnover literature.

36

For instance, Barrick and Mount (1994) located only four studies examining
antecedents of involuntary turnover (LaRocco, Puch & Gunderson, 1977; Stumpf &
Dawley, 1981; Wanous, Stumpf, & Bedrosian, 1979; Wells & Muchinsky, 1985). The
authors indicated these studies tended to focus on demographics and job performance as
antecedents, reporting consistent demographic differences, as well as a tendency for IVT
to be more prevalent among poor performers. None of these studies examined constructs
typically used in personnel selection. Barrick and Mount filled this void by examining
the role of demographics, conscientiousness and supervisory ratings of job performance,
and job involvement as predictors of IVT during downsizing. This research found that
the relationship between conscientiousness and IVT was moderate (r=-.33), and the .
results suggested that this effect was completely mediated by job performance.

The results described above are promising in that they suggest that conscientiousness
should be related to IV T in at least certain situations; however, the context of the present
research is very different. First, the focus here is teams. A number of researchers have
suggested that personality characteristics are more likely to be relevant in team-based
work settings as opposed to jobs that are individually based. Second, from a selection
standpoint it is important to identify employees likely to turnover involuntarily due to
reasons other than downsizing. In downsizing many employees who would otherwise be
retained are terminated, generally due to an organization’s need to cut costs or adverse
market conditions (Cameron, 1998); however, the termination of an employee outside the
context of downsizing clearly suggests something problematic about that employee’s job
performance or work behavior more broadly. Based on the traits associated with

conscientiousness, one would expect individuals high in conscientiousness to be less

37

likely to be terminated outside of a downsizing event (Rosse & Noel, 1996). Hence, in
non-downsizing contexts, conscientiousness should be more closely linked to IVT. First,
conscientious employees should be less likely to engage in gross misconduct, poor
performance, or other unreliability that would lead to IVT. For example, Mount and
Barrick (1995) reported a mean corrected correlation between conscientiousness and
criteria they termed ‘employee reliability’ to be .41; the corrected correlation with criteria
they termed ‘effort’ was .51. Similarly, conscientiousness seems to embody traits that
would lead one to remain on the job. In other words, conscientious individuals are risk-
averse, should be less likely to act impulsively or recklessly (Baron, 1996), and have a
desire to work hard. Indeed, these characteristics of conscientiousness are the basis for
its inclusion in most personality-based integrity tests (e.g., Schmidt, Ones, & Hunter,
1992). It is somewhat surprising, then, that Banick and Mount’s (1991) meta analysis
reported a corrected correlation between conscientiousness and turnover/retention to be
only .12. The only recent research I have located that examines the relationship between
conscientiousness and turnover reported a correlation of -.23 in a sample of
approximately 300 long-haul truckers (Barrick & Mount, 1996); I have found no research
examining the role between conscientiousness and IVT. Based on this, the following

hypotheses are proposed:

H8. Conscientiousness will be negatively related to VT.

H9. Conscientiousness will be negatively related to IVT.

Though it is predicted that conscientiousness will relate to turnover, there is relatively

little reason to believe that the two other personality constructs studied here, extraversion

38

and agreeableness, relate to turnover at the individual level. Thus, the available evidence
suggests that Link I in Figure 3 does not apply to all the constructs studied here.
Turnover at Higher Levels of Analysis

The discussion thus far has examined the notion of turnover at the individual level.
Since psychologists tend to study individuals, it is not surprising that psychologists have
paid relatively little attention to turnover at the group or team level (i.e., turnover rates).
Organizations often examine the performance of their units, and thus it is clear why the
turnover rate within a given unit is of interest. Clearly, the turnover rate for an
organizational unit is the aggregation of individual turnover occurring within that unit
(Figure 3, link J); however, there may be group—level predictors of turnover rates that do
not exist at the individual level (George & Bettenhausen 1990; Rousseau, 1985).
Understanding such relationships would not only further our ability to predict
organizationally relevant phenomena, but also to gain insight into the processes that
underlie turnover in teams. The approach taken in the research reported here is that an
employee’s fit within a given team is a potential precursor to both individual level
turnover and turnover rates within teams. Thus, the following section broadly describes
the research on fit within various work environments, becoming more detailed at the team
or workgroup level of analysis.

Person-Environment Fit in Organizational Research

Individuals are embedded in a number of contexts that shape their behavior. At work,
a person can be embedded in all of the following environments: A dyad, team, unit,
department, division, organization and vocation. A great deal of research has been

conducted examining the influence of the fit between employees and these environments.

39

Most of this research in the organizational sciences has specifically examined person-
organization (PO) fit. PO fit is often described as having two key components:
Complementary and supplementary fit (Kristof, 1996). In complementary fit, the
organization and/or the employee provides what the other needs. Supplementary fit
occurs when there is a similarity between the organization and the person. Fit is thus

defined as one or both of these two types of matches, which can occur on a variety of

 

dimensions including values, attitudes, personality, goals, ability and demographics
(Kristof, 1996). The impacts of PO fit have been examined throughout the relationship
an employee has with an organization (i.e., job choice, socialization, work behaviors and
attitudes, and turnover).

There are several frameworks that provide theoretical support for the notion that PO
fit should be related to positive outcomes at work. One of the most well known theories
is Holland’s (1985) theory of vocational choice. This theory states that over time, people
choose work environments that are compatible with their orientations. The Theory of
Work Adjustment (Lofquist & Dawis 1969) goes one step further by predicting that the
fit between the person and the work environment directly impacts job tenure. Two other
influential models are more specific in their predictions. Schneider’s (1987) attraction-
selection-attrition (ASA) framework states that people are attracted to, selected into, and
turnover from jobs as a result of the similarity between their own personality and that of
the other organizational members. Over time, this should result in organizations that are
relatively homogeneous with regard to personality. In social psychology, the similarity-
attraction paradigm (Berscheid & Walster, 1978; Bryne, 1971) states that similarity of all

types leads to liking and attraction, which in turn leads to a host of positive outcomes.

40

The cumulative body of empirical research strongly supports these theories as providing
powerful rationales underlying the notion that the fit between an individual and his or her
work environment predicts important individual and organizational outcomes. Clearly,
some of these theories are more important than others with regard to different types of fit.
Thus, a brief overview of the PO fit literature is below, while more thorough explications
of the relevant theories and research are given when the constructs studied here are
discussed.

A consistent finding is that both complementary and supplementary fit are related to
individual behavior in organizations. With regard to job search, for instance, attraction to
organizations has been correlated with P0 fit on preferences for reward structure,
individual personality to organizational climate, and value congruence between job
seekers and potential organizations. From the organization’s perspective, perceived
(rather than actual) PO fit between applicants and interviewers has also been shown to be
a predictor of interview outcomes; applicants who were perceived as having better PO fit
with the organization were more likely to receive higher interview ratings. PO fit has
also shown positive individuals’ attitudes at work. This has been found in a number of
studies examining effects of value congruence on attitudes such as motivation, job
satisfaction, work group cohesion, and organizational commitment. Similar findings
have been reported in several studies examining the congruence between employees and
their supervisors. Dipboye (1994) suggests that person-job fit is so important that it is at
least in part responsible for the entrenchment of unstructured interviews in organizational
settings despite a large body of literature indicating that structured interviews are far

superior (e.g., McDaniel, Whetzel, Schmidt, & Mauer, 1994).

41

As one of the primary outcomes of interest in this study is turnover, it is helpful to
review the effects of fit on turnover in more detail. The reasoning underlying this
research is that positive outcomes associated with P0 fit such as those described above
lead to an increased propensity to remain on the job. Similarly, a lack of PO fit is thought
to produce dissatisfaction, strain and other negative outcomes. Hence, turnover should
result from a M of PO fit. Indeed, Chatrnan (1991) reported that PO fit based on values
measured by the Organizational Culture Profile (O’Reilly, Chatrnan, & Caldwell, 1991)
predicted auditors’ satisfaction and intentions to remain with their firms. O’Reilly et al.
(1991) reported similar findings with actual turnover in the same industry, as did
Vendenberghe (1999) in the health care industry. Bretz and Judge (1994) reported that
PO fit with regard to several important aspects of fit predicted both satisfaction and job
tenure. Kristof-Brown and Kay Stevens (2001) examined the role of goal congruence, a
study that was conducted within the context of students completing a group assignment
for a class requirement. Their results indicated that the perceived congruence between
each student and the team predicted student satisfaction with the team Similarly,
Vancouver and Schmitt (1991) found that the fit between teachers’ and principals’ goal
congruence predicted satisfaction, organizational commitment and intention to remain on
the job. Chan (1997) examined something he termed cognitive misfit, the lack of fit
between the cognitive demands of a job and the preferred cognitive style of employees in
that particular job. His results indicated that cognitive misfit between the employee and
the job was related to turnover after controlling for the effects of job performance. Other
research examined the fit between job creativity demands and the environmental

conditions that facilitate creativity (Shalley, Gilson, & Blum, 2000). Job satisfaction and

42

intent to remain on the job was higher when there was a match between the job demands
and the supporting environment. Again, the underlying theme of these studies is that
misalignment between the individual and environment (e. g., organizational values or
goals, or job demands) results in a variety of negative outcomes such as stress, low levels
of job commitment, and satisfaction, and that these outcomes, in turn, produce turnover.
Indeed, Intent to remain on the job is an important cognition, in part because it is
moderately related to actual turnover (p=.38; Griffeth et al. 2000). These results, taken
together, suggest that the fit between an employee and the job or the organization exerts
consistent effects on turnover and related outcomes.

The gravitational hypothesis (McCormick, DeNisi, & Staw, 1979; McCormick, .
J eanneret, & Mecham, 1972) addresses fit more specifically by suggesting that people
gravitate towards jobs throughout their life based on a fit between their own abilities and
job complexity. This is thought to occur both because people should choose jobs that are
commensurate with their abilities, and because employers select people whose abilities
can meet the demands of a given job. Given that there are often educational requirements
for entry into jobs, and that cognitive ability is frequently used to select employees, it is
logical to think that a person’s ability would thus be central in determining how people
move from job to job over time. There is some research supporting the notion that this
indeed occurs.

Wilk, Desmaris, & Sackett (1995) examined job mobility over a five-year period for
people who had finished high school. These researchers found that cognitive ability
predicted job change at the end of the study after controlling for initial job complexity.

Wilk and Sackett (1996) extended this study by examining ability-job complexity

43

changes in two large national databases spanning ten and twelve years. These researchers
found significant positive relationships between ability and change in objective and
subjective job complexity over time. These results are consistent with studies reporting
that job performance-turnover relationships might be non-linear. For instance, Trevor et
al. (1997) reported an inverted parabolic relationship between job performance and
turnover, and that salary growth and promotions impacted the strength of this
relationship. Because ability is a consistent and strong predictor of job performance,
ability should have a similar relationship to turnover; however, the form of this
relationship is likely to be different depending on whether one examines VT or IVT. In
particular, based on the gravitational hypothesis and the results described above, high
ability employees in low complexity jobs should have higher rates of VT but not IVT as
compared to low-ability employees. The jobs examined here are very low on job

complexity. Hence, the following hypothesis is proposed:

H10. High ability employees will be more likely to exhibit
VT as opposed to IVT after controlling for
Conscientiousness.

The research described above indicates it is possible to examine many types of fit
rather than fit within an organization per se. Indeed, fit has been studied in a wide variety
of contexts including organizations, jobs, teams, groups, leaders, dyads and supervisors.
A number of authors have indicated that the boundaries between these different types of
fit are often blurred, though they all fall under the more general rubric of person-
environment fit. What makes these results especially intriguing for teams is that they
study the fit between the person and relatively distal contexts as compared to the team

context. That is, the context provided by a team should be much more immediate to the

everyday experiences of an employee than the overall organizational context. This is
based on the notion of “bond strength”, which states that constructs at the same or
adjacent levels of analysis (e. g., a person and team) should be more strongly related than
those that are more distal (person and organization; Simon, 1973). There may be several
levels of hierarchy separating an individual and an organization (e. g., team, sub-unit, and
department) whereas persons and teams are adjacent levels in an organizational hierarchy
(dyads are not normally considered to be an organizational level). This is similar to
House et al.’s (1995) and Tosi’s (1992) notion of ‘tight coupling’, which states that units
across different levels of analysis that are more interdependent will influence each other
more strongly. Individuals actually make up a team, suggesting the two are tightly ,
coupled; hence, one would expect to find strong relationships between teams and
individual behavior. In fact, where team based work systems predominate, the team
context should be the main context experienced by an employee. For instance, Hackman
(1992) describes this context as the “social world” of an employee, impacting the ambient
stimuli to which a given employee is exposed. He argues that these stimuli, in turn,
affect employees’ cognitions, affect, and behavior. Bedeian, Kemery and Mossholder
(1989) and McGrath (1998) also argue that team composition should be viewed as a
context that can shape individual behavior.

This suggests that person-team fit should be strongly related to individual and team
outcomes (the ‘Fit’ and k link in Figure 3), especially given the consistent relationships
between a variety of variables and PO fit. Similarly, it is also important to examine how
team characteristics may serve as a context that influences not only team-relevant

outcomes, but also relationships between individual level construct relationships

45

(Bedeian et al.1989; Moreland & Levine, 1992; Moreland, Levine, & Wingert, 1992).
For instance, Van Der Vegt, Emans, and Van De Vliert (2001) examined the effects of a
match between team-level outcome interdependence and the extent to which tasks were
interdependent at the individual level. These researchers hypothesized that individual
task interdependence would more strongly relate to job and team satisfaction and job and
team commitment when there was outcome interdependence at the team level (i.e., that
the team level variable moderates the individual level relationships). The results
generally supported these hypotheses: A match between the rewards in a team and the
extent to which those rewards are obtainable via task interdependence yielded positive
affective outcomes. Again, while not examining the composition-based fit, this research
nonetheless broadly suggests that fit between individual level and team level constructs
leads to positive outcomes and that this can be examined as an effect that moderates
relationships at the individual level of analysis.
Teams as a Context

The foregoing discussion shows why the fit of an individual within a team should be
related to important outcomes. Indeed, several authors have recently focused on the idea
of person-team fit in personnel selection (Werbel & Gilliland, 1999; Werbel & Johnson,
2001) and person-group fit more generally (House et al., 1995; Morgan & Lassiter,
1992). Werbel and Johnson suggest that positive interpersonal interactions are very
important in team based work environments because they should positively impact both
team process and performance. This suggests one should emphasize supplemental fit as a
predictor of effective team process and outcomes. In terms of process, these authors, as

well as others (e.g., George, 1990) posit that increased supplementary fit will positively

46

impact team cohesiveness, or the bonds between team members. Cohesiveness, in turn,
has been related to team performance (Gully, Devine, & Whitney, 1995), and is also
expected to relate to turnover within teams (Griffeth & Horn, 1994). Supplemental fit is
also expected to facilitate cooperation as well, another critical aspect of team process
(Werbel & Gilliland, 1999) that has been linked to team performance. When is
complementary fit important? Theory suggests that complementary fit within a team is
important when the task requires specialized knowledge and skills (i.e., if there are a
number of specialized roles within the team). This is likely to be the case when a task is
complex (Morgan & Lassiter, 1992). For instance, in cross-functional teams one might
seek complementary fit on functional backgrounds to increase the team’s breadth of
expertise, and thus supposedly the effectiveness of the team as a whole. Indeed, a
number of studies examining cross-functional teams indicate that heterogeneity with
regard to skills or functional background is beneficial in this regard (e. g., Ancona &
Caldwell, 1992; Voiers, 1956, cited in Jackson, 1992). Complementary fit also tends to
be emphasized when examining the compatibility between job demands and employees’
abilities or environmental facilitators of those demands (e. g., Chan, 1997, Shalley et al.,
2000). In contrast, I have located no research suggesting that the personality-based
complementary fit within large, undifferentiated teams is critical. Accordingly, the
present research largely emphasizes supplemental fit.

Several authors argue that fit is inherently multilevel (Schmitt & Chan, 1998; Werbel
& Gilliland, 2000) and that selection should be evaluated from a multi-Ievel perspective
(Ployhart, 1998; Schneider, 2000), an idea that is central to the research reported here.

For instance, House et al. (1995) suggest that a lack of fit will result in decreased

47

influence across levels of analysis (e. g., from the team to the individual). This supports
the notion that person-team fit might be related to turnover because there will be weaker
bonds holding the individual to the team. At the team level, one might think about misfit
as team-level diversity on important individual attributes of interest. This is appropriate
if there is good reason to believe that the variability within a team on a given attribute
tells us something meaningful about the team itself that can be related to other key team
level constructs. This is likely to be true when the focus is on supplemental fit because
variability within a team implies supplemental misfit. On the other hand, one can
construe person-team fit as a cross-level effect where the team level mean (or other
appropriately conceptualized operationalization of team standing) moderates
relationships between individual level variables of interest (Rousseau, 1985). This
approach would be helpful if one were interested in treating teams as a context shaping
relationships among individual level variables (Bedeian et al., 1989; Markham, 1988).
Very few studies have empirically studied teams in such a way. One exception is
research reported by Kidwell, Mossholder, and Bennett (1997) who found that increased
work group cohesiveness strengthened the relationship between job satisfaction and
courtesy. Underscoring how little work has been done in this area is that Kidwell et al.’s
study was more a demonstration of statistical techniques rather than a substantive focus
on teams as a context. Treating person-team supplemental fit as a cross-level effect is
conceptually closely tied to examining the effects of team level diversity; in fact, the two
can be conceptually equivalent given correctly specified composition models.
Accordingly, the following sections discuss person-team fit as both team-level diversity

and a cross-level effect.

48

Personality-Based Team Context

There are several reasons to think that the fit of personalities within a team should
impact team process and team effectiveness. First, a particular mixture of personalities
may lead to competition or conflict in a team. For instance, it is easy to imagine how a
team with little or no variability on the personality trait of dominance might have
difficulties. If all the team members are very similar on dominance, leaders might not
emerge in the team, or everyone might be competing for a leadership position. In either
case, one would expect there to be negative effects on team process and performance;
however, more commonly it is thought that similarity in personalities should lead to
better interpersonal relationships. For instance, there is a substantial amount of research
supporting the notion that personality similarity is related to more positive exchanges
between leaders and team members (e. g., Askanasay & O’Connor, 1997; Engle & Lord,
1997; Phillips & Bedeian, 1994), job satisfaction (Meglino, Ravlin, & Adkins, 1989;
1991) and performance ratings (Antonioni & Park, 2001). Examining job search
preferences, Judge and Cable (1997) linked the congruence between job seeking students’
personalities to attraction to organizations. The theory most often cited to support this
notion is the similarity-attraction paradigm (Berscheid & Walster, 1978; Bryne, 1971)
which states that similarity (e. g., racial, attitudinal, etc.) leads to liking and attraction.
This is because similarity is thought to lead to positive reinforcement of one’s attitudes
and beliefs. This reinforcement, in turn, leads to smoother and desirable social
interactions. There is an overwhelming amount of research conducted over several
decades in a variety of contexts supporting this perspective. This suggests that people

should find it less rewarding and more dissatisfying to work with people that are

49

dissimilar to them. This is also consistent with Schneider’s (1987) ASA model discussed
earlier.

Research conducted by Ferris, Youngblood, and Yates (1985) also provides some
broad support for the notion that the similarities within a team (i.e., supplementary fit)
predict organizationally relevant outcomes. Like Kidwell et al. (1997), these researchers
examined the group as a context that shapes individual level outcomes. Their results
indicated that the overall F FM-based personality fit between a sample of newcomer flight
attendants and a sample of incumbents moderated the relationships between performance
and attendance and turnover. In particular, performance was unrelated or negatively
related to these outcomes where person-group fit was low, whereas the opposite was
found where fit was high. Several caveats are worth mentioning, however; first, person-
group fit did not predict turnover or absenteeism. This is inconsistent with the broader
notion that person-group fit is associated with positive outcomes (i.e., attendance and
remaining on the job). Second, the referent group was a small sample of flight attendants
taken from the larger pool of flight attendants as a whole. Since there was little data or
discussion of the representativeness of this sample of flight attendants, it is unclear
exactly what the referent group represented in this research. Third, these researchers
calculated a distance score for each newcomer flight attendant across all the FFM
personality constructs. Thus, the nature of this measure was unclear because it consisted
of five different distance scores. Nonetheless, this study is still somewhat informative
because it broadly suggests that groups’ personalities might provide a context that shapes

relationships among variables or constructs at the individual level.

50

There is theoretical support for the notion that team process and thus performance
might decline as a function of personality misfit. Shaw and Barret-Power (1998) argue
that ‘cognitive diversity’ impedes a group’s ability to integrate their behaviors, and that
this is increasingly likely to occur on tasks where multiple perspectives are not critical
(like the task of the teams studied here). This should also be true when there are multiple
preferences for how work is conducted (e. g., levels of conscientiousness). Overcoming
the effects of this diversity is said to have a “cognitive cost” which makes this difficult
(p. 1315). The literature on shared mental models is also consistent with this notion.
Shared mental models are schemas that are Shared within the team that are in some way
relevant to the team task at hand. These schemas can relate to either how the task should
be performed (i.e., taskwork) or how team members should interact to achieve the task
(i.e., teamwork). Accordingly, Rentsch and Hall (1994) state that shared mental models
can manifest themselves as expectations and guidelines that structure team interaction
and the compilation of individual inputs into outputs.

These authors present a theoretical model that suggests several different ways in
which schema dissimilarity can disrupt teamwork and taskwork and thus impact team
performance and the desire of the team to work together in the future. More concretely,
however, it is easy to give examples of how schema dissimilarity might lead to less
efficient team processes. For instance, if two individuals have different schemas
regarding the notion of cooperation they might encounter serious communication and
coordination problems if they are asked to cooperate. The effects of this similarity might
be expected to cascade if it happens at the team level. These authors also assert that

schema similarity is more likely to occur when the team members show supplemental fit

51

in terms of their values, personalities, and preferences. This is important because the
extent to which effective mental models are shared have been theoretically (Salas et al.,
1992) and empirically (e. g., Mathieu et al., 2000) linked to impact performance and are
considered by some to be critical determinants of coordinated team performance (e.g.,
Rouse, Canon-Bowers & Salas, 1992). To the extent that Conscientiousness,
Agreeableness, or Extraversion represents a preferred way of behaving at work in team
settings, one might expect that the sharedness of these preferences should relate to
important outcomes. J ehn, Northcraft and Neale (1999) reported some results consistent
with this notion in that values for the way in which work was conducted in a sample of
moving employees predicted intent to remain on the job and actual group performance;
however, these researchers used D-scores and analyzed their data at the individual level,
so these results must be interpreted cautiously.

CLnscientiousness. Of the three personality constructs examined here, conscientiousness
is clearly the most directly related to tasks at work; it is easy to translate the traits
associated with conscientiousness to specific behaviors one might exhibit at work. For
example, someone who is detail oriented would tend to complete tasks at work to higher
standards than someone who is low on conscientiousness (i.e., is not detail oriented).
These behaviors, in turn, are likely to be at least in part due to an underlying cognitive
preference. For example, one might reasonably infer that someone who generally
describes herself as detail oriented prefers to ensure that the details of tasks are
completed. To the extent that this is true, then, one might expect that the fit between an
individual’s standing on conscientiousness and that of her team might lead to important

outcomes. This is because a lack of fit would imply that work tasks are being done in a

52

manner that is discordant with the preferences of the individual: In our example of detail
orientation either too much or too little attention is paid to the details of work tasks.
Either situation can cause dissatisfaction in a worker that does not fit in this way. For
instance, an employee that is substantially lower than the team-level mean on
conscientiousness might be pressured by coworkers to put more effort into the details or
work tasks. On the other hand, the work products of coworkers he or she considers
sloppy might frustrate someone markedly higher on conscientiousness. At an individual
level, dissatisfaction may result. If there is large variance in conscientiousness at the
team level (i.e., an overall lack of supplemental fit within the team), overall process
might suffer as a result because high conscientious individuals might have to redo the
work of low conscientiousness individuals (Barrick et al., 19983).

Several studies have examined or reported the impact on conscientiousness
similarity in team and work units. Barrick et al. (1998), in the study of manufacturing
and maintenance teams described earlier, reported that the variance of conscientiousness
scores within a team was negatively related to team performance (r=-.33); however, this
effect was not hypothesized. Neuman et al. (1999) also examined the same index in an
exploratory manner and found no relationship between conscientiousness diversity and
team performance. These researchers did not provide a detailed description of the extent
to which the team members were interdependent; hence, it is difficult to develop possible
explanations for these null results.

Day and Bedeian (1995) examined the effects of conscientiousness similarity on the
job satisfaction, job performance and tenure in a sample of Black nurses. Their sample

represented approximately one-third the entire sample of nurses in a medical center and

53

similarity scores were calculated for each nurse. These similarity scores were designed to
index the extent to which each nurse was similar to the larger sample of nurses within the
entire medical center. These researchers found that higher similarity was associated with
lower organizational tenure and higher job satisfaction. The former finding was contrary
to what one would expect based on the research and theory described above. This null
result may have occurred because in their particular context nurses may not have
necessarily worked together. Perhaps conscientiousness similarity is not consistently
related to outcomes in predictable ways when employees do not work together on team
tasks.

Thus, to summarize, the existing literature paints an unclear picture of the relationship
between person-team conscientiousness fit on individual and group outcomes; however,
based on the reasoning put forth above, it is expected that conscientiousness fit at the

team level would indeed predict team performance. In particular:

H11. Increased team level diversity in conscientiousness
should be associated with lower team performance.

Recall the earlier discussion describing how person-team fit can also be examined as
a cross-level effect as well as a team-level diversity effect. The individual level outcome
examined in the present study is turnover. Based on the general findings in the PO fit
literature, as well as the reasoning described above, it was expected that
conscientiousness fit would relate to increased VT but not IVT. Hence, it is hypothesized

that:

H12. Increased diversity in conscientiousness should be
associated with higher team voluntary turnover (but not
involuntary turnover).

54

H13. Increased conscientiousness-based person-team fit
should be associated with a lower likelihood of individual
level voluntary turnover (but not involuntary turnover).

Agreeableness. Two of the three studies described above (Barrick et al., 1998a;
Neuman et al., 1999) found no relationship between Agreeableness diversity and team
performance. Banick et al. did report, however, that increased Agreeableness diversity
was associated with lower levels of cohesion and communication, and higher levels of
conflict. Day and Bedeian (1995) found that increased Agreeableness similarity was
associated with higher job performance (r=. l 7). Like Conscientiousness, these results are
mixed; however, it is less clear why one might expect the diversity or fit with regard to
agreeableness to impact team process or performance. This is because people that are
agreeable by their nature are supposed to be understanding and tolerant. One would thus
expect people that are high in agreeableness not to react negatively to having to work
with people that are low in agreeableness. Hence, given the lack of theoretical basis and
conflicting empirical results, no hypotheses regarding agreeableness fit are proposed and
exploratory analyses will be conducted.

Extraversion. The existing research on extraversion diversity in teams is somewhat
clearer than that on Agreeableness. Barrick et al. (1998) hypothesized and found that it
would positively relate to team social cohesion; however, there was no explicit discussion
of why they expected this to occur. Neuman et al. (1999) also reported that the variance
in team extraversion was positively related to team performance ratings (r=.26), though
again, their research was exploratory. In contrast, Day and Bedeian (1995) found that
extraversion similarity was related to job satisfaction, though not to performance.

However, these researchers used difference scores to index similarity, an approach that

55

has been widely criticized (see below). Perhaps a mix of extraversion is necessary within
a team because otherwise interpersonal conflict will arise as the extraverts ‘jockey’ for
social interactions. In other words, at the team level, supplemental fit might actually lead
to negative effects on team outcomes because of the nature of Extraversion as a construct.
On the other hand, it is somewhat difficult to see how an extravert might prefer to work
with introverts. Hence, the relationship between extraversion fit will be examined in an
exploratory fashion.

Demogr_aphics. Jackson et al. (1995) differentiate between readily detectable (‘surface
level’) and underlying (‘deep level’) qualities that can be used to examine the effects of
diversity. The discussion thus far has focused on deep level diversity and composition
effects. The following sections discuss the literature surrounding the effects of surface
level diversity, laying the groundwork for hypotheses that focus on this aspect of person-
team fit.

It is widely acknowledged that the demographics of the American workforce are
rapidly changing. Over the last few decades there have been marked changes with regard
to the racial, gender, and age composition of the workforce. In particular, there is an
increasing diversity in these three demographic variables in the workforce (National
Research Council, 1999). Not surprisingly, there is a substantial amount of psychological
literature devoted to the effects of this increased diversity (i.e., the demographic mix)
within teams and other organizational units (Williams & O’Reilly, 1998).

This literature was sparked by Pfeffer’s (1983) seminal article that coined the term
grganizational demogr_aphy. Pfeffer argued that the distribution of demographics within a

given organizational unit (e. g., organization, department, workgroup, etc.) impacts the

56

quality of social integration within that unit. Focusing on tenure, Pfeffer argued that the
distribution of demographics within an organizational unit (e.g., organization,
department, workgroup, etc.) impacts the amount of conflict within that unit. This
disruption, in turn, was theorized to impact performance, innovation, turnover, and power
distributions. In the years following Pfeffer’s article researchers have examined the
effects of a wide array of demographic variables including race, age, sex, job tenure and
functional background on outcome variables such as unit performance, turnover, and
unit-processes such as social integration, social network strength, and conflict (e.g.,
O’Reilly, Williams, & Barsade, 1998). An offshoot of this thinking was research on the
effects of demographic similarity at the individual level, referred to as relational
demcgraphy (Tsui, Egan, & O’Reilly, 1992; Tsui & O’Reilly, 1989). The idea
underlying this research is that demographic similarity positively impacts the social
relationships between individuals. Conversely, demographic dissimilarity is said to
disrupt social relationships between individuals. These putative effects of dissimilarity
are thought to negatively impact affect and attitudes, which in turn are hypothesized to
impact performance and turnover. Thus, despite the focus on work units and individuals,
respectively, the underlying processes and theoretical rationales for both relational and
organizational demography are very similar.

There are several major theories in social psychology that support the notion that the
mix of demographic characteristics within a group or team impact the outcomes
described above. First, there is the similarity-attraction paradigm, described earlier. One
would expect this theory to apply to demographics to the extent that demographics are

salient indicators of similarity in a particular environment. Two theories that speak to

57

this point are categorization (T ajfel, 1981; Turner, 1987) and social identity theory
(Turner, 1982). Social categorization theory states that individuals have a strong need to
maintain high self-esteem. In doing so, they categorize others according to salient
characteristics including race, sex, and membership in groups. The need to maintain
positive self regard leads people to strongly identify with their own group and to
maximize perceived differences between groups. The end result, which has received an
overwhelming amount of empirical support, is that individuals tend to react more
positively in interactions with people in the same group, even when group distinctions are
arbitrary (Sherif, 1961). Thus, even if demographics are not related to underlying
attributes on which people might differ, it is still possible that people assume that they are
(e. g., McGrath, 1998; Milliken & Martins, 1996). Before this large body of research is
reviewed, however, it is important to first discuss several important theoretical and
methodological limitations that pervade this research. This is because these limitations
make it difficult to draw unambiguous conclusions from these studies, and so they should

be considered in any review of this research.

 

Methodological and Conceptual Issues in Relational Demogr_aphy Research. Failure
to pay careful attention to issues relating to levels of analysis can lead to a host of
conceptual and data analytic problems (e. g., Klein & Kozlowski, 2000; Ostroff, 1993;
Roberts & Burstein, 1980; Roberts, Hulin, & Rousseau, 1978; Rousseau, 1985).
Relational demography research largely suffers from the several problematic ways in
which researchers conceptualize and thus operationalize demographic diversity. Like the
other composition variables discussed earlier, demographics are measured at the

individual level yet the key predictor of interest is demographic dissimilarity between an

58

individual and a referent organizational unit. In reviewing the relational demography
literature, Riordan (2000) outlines the three major approaches to operationalizing this
dissimilarity.

The most common approach calculates a difference (D) score for each individual that
is said to represent the average demographic distance between an individual and the
members of the focal organizational unit of interest. D-scores have been widely
criticized on a host of conceptual and methodological grounds (e.g., Edwards, 1994;
2002; Johns, 1981; Riordan & Shore, 1997; Riordan, 2000) yet they are still commonly
used. The second approach which Riordan termed the ‘interaction approach’, involves
disaggregating an index of group composition to the individual level and performing
moderated regression analyses to test the significance of the interaction between group
composition and individual level demographic characteristics (e.g., Chattophaday, 1999;
Ferris, Judge, Chachere & Liden, 1991; Flynn & Shore, 1994, cited in Riordan, 2000;
Mellor, 1995; Riordan & Holliday-Wayne, 1998, cited in Riordan, 2000; Riordan &
Shore, 1997; Riordan & Weatherly, 1999, cited in Riordan, 2000). Although Riordan
(2000) criticizes this approach as lacking in power, in fact, tests of these interaction terms
are biased upwards because the individual level sample size severely overestimates the
number of groups that exist (i.e., by several fold; Bryk & Raudenbush, 1992). The
interaction approach, as well as the D-score approach, introduces correlated errors of
prediction that are not accounted for in ordinary least squares regression which can yield
unpredictable results (Bryk & Raudenbush, 1992; Harman, 1990; Hofmann, 1997;
Kennedy, 1998; Kenny & Judd, 1986). For instance, when nonindependence due to

groups is ignored decisions about hi gher-level variables (e.g., group racial composition)

59

will be more prone to Type I errors, whereas those concerning lower-level variables (e.g.,
individual race) will exhibit more Type II errors (Bliese, 2000; 2002; Heck & Thomas,
2000). Bryk and Raudenbush (1992) also note that ignoring independence likely violates
the constant variance assumption. Clearly, when forming an interaction term the true
results are muddled to say the least. Indeed, it has been well documented that
disaggregation is associated with a host of problems, both conceptual and analytical (e. g.,
Glick & Roberts, 1984; Harman, 1971; Langbein & Lichtrnan, 1978; Nezleck &
Zyzniewski, 1998; Rousseau, 1985; cf. Roberts et al., 1978). Thus, in research using the
first two approaches (e.g., Barsade et al., 2000; Chatrnan, et al., 1998; Chatrnan & Flynn,
2001; Chattopadhyay, 1999; Day & Bedeian, 1995; Ferris etal., 1991; Flynn & Shore,
1994; Green, Anderson, & Shivers, 1996; Iverson & Buttigieg, 1997; Jackson et al.,
1991; Mellor, 1995; Mueller, Finley, Iverson, & Price, 1999; O’Reilly et al., 1984;
O’Reilly et al., 1989; Pelled, 1996; Riordan & Holliday—Wayne, 1998; Riordan & Shore,
1997; Riordan & Weatherly, 1999; Tsui, Egan, & O’Reilly, 1992; Tsui & O’Reilly, 1989;
Turban & Jones, 1988; Wagner et al., 1984; Westphal & Zajac, 1995; Wiersema & Bird,
1993; Zenger & Lawrence, 1989) it is difficult to determine the extent to which the
reported results accurately describe the true nature of the relationships of interest. Yet,
these approaches continue to be used despite the acknowledgment by influential authors
in the area that demographic effects can occur at both the individual and group level
(Jackson et al., 1995) and that attribute dissimilarity is a multi-level construct (Jackson,
May, & Whitney, 1995; Jackson, Stone, & Alvarez, 1992, Figure 1). Theoretically, the
first two approaches misspecify the level at which the hypothesized processes occur

(Rousseau, 1985). This is because these approaches examine effects at the individual

60

level despite theory focusing on the dissimilarity between an individual and a group.
Hence, applying the results to any level of analysis is difficult because model
misspecification effects are pervasive (e. g., Klein, Dansereau, & Hall, 1994; Roberts et
al., 1985; Rousseau, 1985).

A third flawed approach is the use of perceived demographic similarity which asks
individuals to indicate how demographically similar they perceive themselves in relation
to a work unit (e. g., Kirchmeyer, 1995; Ragins & Comwell, 2001; Riordan & Weatherly,
1999, cited in Riordan, 2000). While perceived demographic similarity may have some
limited interest in its own right, it should not be treated as the same as actual
demographic similarity for several reasons. First, demographic similarity is a broadterm
and it is likely that individuals have very different internal conceptualizations of what
that might mean (Riordan, 2000). Second, because it is so broad it cannot easily be
meaningfully compared to other indices of demographic similarity. Third, because
demographic information is often readily available and clearly observable (McGrath,
Berdahl, & Arrow, 1996) it seems that this information should be gathered instead or that
actual and perceived similarity should be compared, at the very least. Perceptual
measures such as this are generally only used when the characteristics of interest are not
outwardly observable or when there is clearly articulated reasoning explaining the
interest in the perceptual approach to similarity. Not only are all three approaches
described above seriously flawed, they also yield results that do not exhibit convergent
validity (Riordan, 1997) which makes drawing conclusions across studies difficult.

These limitations reflect a lack of conceptual clarity and methodological

sophistication. This is because dissimilarity between an individual and a group is not

61

solely an individual or group-level phenomenon; it is a cross-level effect, referring to the
demographic fit between an individual and a group. That is, the composition of the group
is thought to moderate the relationship between the individual level demographic variable
and the individual-level outcome of interest. Given a significant interaction effect, its
form, in turn, will reveal whether the results are consistent with theory. This approach is
consistent with the fundamental assumptions of relational demography stating that
demographic dissimilarity is the key predictor of interest. Data analytic tools that allow
for modeling both individual and group level variables do exist (Bryk & Raudenbush,
1992), though they are not as widely understood and used as ordinary least squares
regression, the analytic technique of choice in much industrial/organizational psychology
and management research. Alternatively, if the outcome of interest is at the group level it
is reasonable to examine the relationships between demographic diversity and the
outcome variable of interest as long as the outcome variable is either (1) meaningfully
measured at the group level, or (2) if aggregation from the individual level is theoretically
justified given the data at hand. Modeling relational demography as both a cross-level
and a group level diversity effect is consistent with the notion that different processes,
outcomes, and relationships among variables can occur at different levels of analysis
(Ostroff, 1993; Rousseau, 1985). Note, however, that it is also possible to operationalize
team level diversity in different ways, and that the theory underlying a given study should
guide the approach.

The literature generally reports that increased diversity or dissimilarity leads to a host
of negative outcomes in terms of communication, satisfaction, cohesion, and social

integration in the workplace, and that these in turn impact turnover and the performance

62

of organizational units. Despite the criticism leveled above, it is still instructive to review
this research to understand the overall tone of the findings and the types of questions that
are asked. Further, some (albeit very few) studies do not use the flawed data analysis
methods described above. Thus, the following sections discuss this literature as it relates
to team and work-unit composition.

1% Race plays an especially important role in personnel selection because there is
often pressure not only to remain within legal guidelines but to also racially diversify
workforces beyond legal requirements (Arvey & F aley, 1988; Hartigan & Wigdor, 1989;
Schmidt, 1993; Schmitt, 1989). This issue has generated an extensive body of literature
(e. g., de Corte, 1999; Hough, Oswald, & Ployhart, 2001; Roth, Bobko, Switzer & Dean,
2001; Sackett & Ellingson, 1997; Schmitt et al., 1997). Thus, from a practical
perspective it is especially important to examine the effects of team racial composition.

Several studies have examined the effects of unit racial composition on aspects of
unit effectiveness. Two of the very few studies in this area that have appropriately
modeled relational demographic variables examined the effects of unit racial diversity
and supervisor-subordinate racial dissimilarity on job satisfaction. Wharton, Rotolo, and
Bird (2000) found that race heterogeneity affected mean levels of job satisfaction within
university departments even though race itself was not significantly related to job
satisfaction at the individual level. Similarly, Wesolowski and Mossholder (1997) found
that the interaction between supervisor and subordinate race predicted subordinates’ job
satisfaction such that subordinates with supervisors of different races exhibited less job
satisfaction. Mueller et al. (1999) reported that increased teachers’ racial diversity in

schools was negatively associated with job satisfaction; however, in school settings it

63

seems likely that there are so many other possible influences such as resource
availability, which is known to be strongly correlated with race in educational settings,
that it may be difficult to draw firm conclusions from this study. Pelled, Eisenhardt, and
Xin (1999) reported that racial diversity in process improvement teams was positively
and significantly related to mean group-level perceptions of conflict. In contrast, Pelled
(1996) conducted another study in which racial dissimilarity was unrelated to several
different types of perceived conflict. Pelled speculated that she might not have obtained
the expected results because her small teams were so highly integrated (she gives a
hypothetical example of a team consisting of one Black, one Asian, one Hispanic, and
one White) that employees of different races were less likely to feel like outsiders. Tsui
et al. (1992) reported that higher racial distance was related to lower individual
psychological commitment and intent to remain with the organization and absence.
Chatrnan and Flynn (2001) also reported that demographic similarity (calculated as an
aggregate of race, sex, and citizenship) was negatively associated with group process. In
particular, these researchers found that demographic diversity was negatively associated
with individuals’ perceptions of COOperative group norms and satisfaction; however, the
results of these last three studies are ambiguous because all used the D-score approach.

Riordan and Shore (1997) studied the attitudinal and perceptual outcomes associated
with racial diversity. These researchers reported that racial diversity of work units was
related to lower perceived work group productivity and advancement opportunities,
commitment and cohesiveness; however, racial diversity was assessed for three different
races (Whites, Blacks, and Hispanics) coded as a single variable and group racial

composition was split into 3 categories, disaggregated, and used in calculating an

64

interaction term for moderated regression. Hence, although this is one of the most widely
cited studies in this area it is unclear what conclusions can be properly drawn.

Several authors have suggested that demographic diversity might relate to team
performance (e. g., Shaw & Bennet-Power, 1998) though only a few studies have
specifically examined racial diversity. Timmerman (2000) examined the effects of racial
diversity in two different types of teams that exhibit varying levels of task
interdependence: Professional baseball and basketball teams. The theory underlying this
approach is that more coordination is required for a sport like basketball, which is more
interdependent than baseball. Hence, diversity effects were expected for basketball as
opposed to baseball because any negative effects of diversity should be more disruptive
(i.e., should lead to more process loss) in a sport that requires more coordination.
Similarly, the higher level of coordination should be more difficult to achieve in the face
of any negative effects that might result from racial diversity. Data fi'om 38 years of
baseball and basketball teams were used, and racial diversity predicted team basketball
performance (i.e., winning percentage) after controlling for aggregated individual level
indices of performance (e. g., field goal percentage). Consistent with expectations, the
same effect was not found in baseball. This suggests that racial diversity disrupts team
cooperation or coordination, which in turn impacts performance. This finding is
consistent with those of Hoffman (1985) who reported that the proportion of Black
supervisors was positively associated with communication via formal meetings. As
formal communication may be a substitute for more informal communication, these
results suggest that increased racial diversity leads to weaker social network ties related

to social communication (Shaw, 1981). A number of studies suggest that weaker social

65

network ties lead to turnover in organizations and groups (e. g., F eeley & Barnett, 1997;
McPherson, Popielarz, & Drobnic, 1992) implying that racial diversity might lead to
turnover.

Several authors have compared racially homogeneous and non-homogeneous teams.
Watson, Kumar, and Michaelson (1993) compared such teams of college students who
had group assignments for a management course. These researchers found that
homogenous groups had higher ratings of group performance and several process
variables, though these effects faded at the end of the 17-week assignment. Baugh and
Graen (1997) also compared the same types of teams in a state government agency.
Racially non-homogenous teams perceived themselves as less effective than racially
homogenous teams, though this result did not hold for external evaluations of team
performance. These researchers noted their small sample size (N =31 teams) might have
accounted for the lack of significance. At the least, this research suggests that racial
diversity impacts team members’ attitudes.

Another aspect of effectiveness, turnover, has also been examined as a potential
outcome of racial dissimilarity. Chattopadhyay (1999) used D-scores to examine the
effects of racial dissimilarity on organizational citizenship behaviors (OCBs). Racial
dissimilarity was negatively related to three self-report facets of OCBs. Further, race
interacted with racial dissimilarity in this regard; that is, this effect occurred for Whites
working in minority-dominated groups but not when the opposite was true, a result also
reported by Tsui et al. (1992). This effect was found to be mediated by organization-
based self-esteem rather than peer relationships, suggesting that the Whites who worked

in these settings felt undervalued. This is also consistent with research reported by

66

Hinds, Carley, Crackhart, and Wholey (2000) who reported that students choosing work
group members on long-term projects preferred members of the same race. Turnover
from a group should result if dissatisfaction with the group members surpasses a certain
threshold.

Although I have found no studies of the effects of racial diversity on team
performance in organization settings, Richard (2000) studied this issue for the entire set
of employees working in banks. Using mail surveys, this author obtained racial
composition information from the banks’ human resource managers and compared it to
revenue data and two other indicators of bank performance. The zero-order correlation
between racial diversity and market performance was significant and negative (F-.32),
though after entering relevant control variables into the regression equation this effect
was largely eliminated. It is important to note that this study had a fairly small sample
size (N=63 banks), 75% of which was drawn from California. Not only is California
widely considered to be among the most liberal states in the country (“State's Diversity
Becoming Its Strength”, 2001), it is also the most racially diverse. Hence, these null
results are not necessarily very surprising. One might expect to find more positive results
in a more representative sample. It is also possible that the results might have been
different had the unit of analysis been work teams rather than the entire bank as a whole.

In sum, the results reported above suggest that racial person-team fit has several
important behavioral and attitudinal outcomes; however, due to serious methodological
and conceptual problems that underlie almost all of the studies conducted to date further
research is needed to more accurately assess these effects. Nonetheless, these results

taken together, and the substantial theory underlying this research, suggest that racial

67

misfit within organizational units leads to a number of important outcomes. Similarly,
because a number of researchers suggest that demographics might be proxies for other
deeper-level variables, this research will control for ability or any of the personality
constructs that are related to demographics and turnover. Hence, the following

hypotheses are proposed:

H14. The racial fit of the individual within the team will
predict individual turnover such that misfit is associated
with a higher likelihood of turnover. Based on work by
Tsui et a1 (1992), this relationship will be more evident in
Whites working in teams with higher minority
representations rather than vice-versa. This effect will be
evident after controlling for the relevant personality
constructs and ability.

H15. Increased team racial diversity will be related to
increased team turnover rates. This effect will be evident
after controlling for the relevant personality constructs and
ability.

H16. Increased racial diversity will be negatively related to
team performance. This effect will be evident after
controlling for the relevant personality constructs and
ability.

Age. A number of studies have examined the effects of the mix of ages or age
dissimilarity in work units on a variety of psychological and group process variables, and
turnover. Most of these studies used the D-score approach. Tsui et al. (1992) studied
entire organizational units (e. g., an entire hospital) and reported that age dissimilarity
predicted intent to remain with the organization but not psychological commitment. This
odd result suggests that perhaps one or both measures lacked construct validity or that the

entire organizational unit represented a focus that was too broad. In contrast, Zenger &

Lawrence (1989) found that age dissimilarity was negatively related to communication

68

outside of the project team, and that dissimilarity was similarly related to intra-team
communication. Two other studies in top-management (Wagner et al., 1984) and work
teams (O’Reilly et al., 1989) found that age similarity was negatively related to turnover;
however, the latter study reported that age similarity was not related to social integration,
a finding that is conceptually at odds with the reported turnover relationship.

Other researchers have used different approaches to study age dissimilarity.
Kirchmeyer (1995) used the perceptual approach to examine what he termed
‘generational similarity’, which was a combination of perceived age, education and
lifestyle. This author reported that it was not significantly related to either organizational
commitment or turnover. Jackson et al. (1991) reported that group age variability .
predicted group turnover rates. At the individual level using the D-score approach,
however, these researchers reported that age dissimilarity was unrelated to turnover.
Riordan and Weatherly (1999; cited in Riordan, 2000) reported that perceived age
similarity was related to group performance, and that age similarity at the individual level
was related to OCBs. The Timmerman (2000) study reported earlier on professional
basketball and baseball teams reported results that were similar to those with regard to
race; that is, age variability impacted basketball but not baseball team performance after
controlling for average individual level performance. In this context, however, age
effects might be overstated in relation to organizational contexts because age is such a
salient issue in sports.

These studies, taken together, suggest that an employee’s age-based fit within a team
might impact team process and team performance. However, these studies also illustrate

the fragmented nature of the approaches (and thus, results) in this literature. For instance,

69

using the D-score, some studies indicated dissimilarity was related to turnover but not to
hypothesized intervening variables. This is especially surprising because turnover is
dichotomous whereas the intervening variables are continuous; thus, correlations with the
latter should be substantially easier to obtain. Similarly, Jackson et al. (1991) modeled
relational age at both the individual and group levels but reported peculiar results:
Relationships between age and turnover at the group level but not at the individual level.
These seemingly contradictory results suggest these researchers might have omitted
relevant variables that were perhaps ‘deeper’ psychological or motivational correlates of
race and sex. One cannot tell, however, because the D-score approach is seriously
flawed. Overall, it is clear that additional research with a clearer conceptualization. of age

dissimilarity would be informative. Thus, the following hypotheses are proposed:

H17. The age-based fit of the individual within the team
will predict individual turnover such that misfit is
associated with a higher likelihood of turnover. This effect
will be evident after controlling for the relevant personality
constructs and ability.

H18. Increased team age diversity will be related to
increased team turnover rates. This effect will be evident
after controlling for the relevant personality constructs and
ability.

H19. Increased age diversity will be negatively related to
team performance. This effect will be evident after
controlling for the relevant personality constructs and
ability.
Sex. A number of authors have also examined the effects of sex diversity on
individual and group outcomes in work settings. As a whole, these results are mixed.

Several authors (Pelled et al., 1999; Riordan & Shore, 1997) examined group level sex

heterogeneity and found no relationships to a host of outcomes including task and

70

emotional conflict, work group productivity, commitment, and cohesiveness. On the
other hand, Harrison et al. (1998) reported that heterogeneity was negatively related to
group cohesion. Similarly, Baugh & Graen (1997) found that gender homogeneous
groups perceived their performance to be higher than non-homogeneous groups;
however, there was no difference between these groups on perceptions of the quality of
intra-workgroup working relationships or on external evaluations of workgroup
performance.

At the individual level, most researchers have used the D-score approach.
Chatappadaphay (1999) found dissimilarity as unrelated to peer relations and altruism,
and Tsui et al. (1992) reported that it was unrelated to intent to stay with the organization.
On the other hand, Riordan and Weatherly (1999; cited in Riordan, 2000) reported that
actual gender dissimilarity using disaggregated group—level indices was unrelated to
liking of peers or job satisfaction, but that perceived gender Similarity was positively
related to both outcomes. Other researchers (Flynn & Shore 1994, cited in Riordan,
2000; Riordan & Shore, 1997) using disaggregated group-level data to form interaction
terms, reported finding no relationships with perceived group productivity, perceived
advancement opportunities, perceived group communication, or satisfaction with several
facets of the work environment. Again, like the studies examining age heterogeneity, this
research reports mixed results using a variety of approaches, some of which are seriously
flawed. Hence, the present study aims to examine these same issues using a novel, yet

theoretically sound way of examining these issues. Thus, the following hypotheses will

be tested:

71

H20. The sex-based fit of the individual within the team
will predict individual turnover such that misfit is
associated with a higher likelihood of turnover. This effect
will be evident after controlling for the relevant personality
constructs and ability.

H21. Increased team sex diversity will be related to
increased team turnover rates. This effect will be evident
after controlling for the relevant personality constructs and
ability.

H22. Increased sex diversity will be negatively related to
team performance. This effect will be evident after
controlling for the relevant personality constructs and
ability.

The PotentialflVIoderating Effects of Time and Extraversion on Demographic- aLd

Personality-Based Fit Effects

 

The sections above provide the rationale underlying the expectation that in general,
demographic fit will be related to increased turnover at the team and individual level.
Riordan (2000) suggested the impact of demographic diversity will be initially important
but will fade over time as people have an opportunity to ‘see past’ surface-level features.
Shaw and Barret-Power (1998) also allude to this notion, in that their model of
demographic and cognitive diversity effects suggests that the former are more important
at the outset of group development whereas the latter are linked to later stages. Indeed,
this is consistent with Watson et al.’s (1993) research discussed earlier, which showed
that racial homogeneity was related to positive outcomes early in team development but
that these effects faded over time. Ancona and Caldwell (1998) reported similar results
in that the effects on several demographic diversity variables were weaker in groups that
had been established longer. The Chatrnan and Flynn (2001) study mentioned earlier

directly assessed whether demographic diversity effects varied as a function of time. In

72

their first study, the results indicated that the negative effects of diversity on perceptions
of cooperative group norms were mainly present early on in the group’s lifespan. In
another sample, negative effects of demographic diversity were only present among
newcomers. Although these results should be interpreted with caution because D-scores
were used, they still suggest that the negative effects of demographic differences might
be ameliorated by time. Similarly, Harrison et al. (1998) found that gender (though
neither racial nor age) diversity and affective diversity interacted with time such that the
former was more strongly related to group cohesion in groups that had not been together
very long. Conversely, underlying attitudes became a stronger predictor of group
cohesion as time together increased. This suggests that the interpersonal interactions that
naturally occur over time mitigate the effects of demographic diversity but that other
types of diversity (e. g., attitudinal) that are only apparent after extended interactions have
subsequent impacts. This is consistent with the underlying notion that individuals form
similarity judgments based on the available information at a given time; it is logical to
think that readily observable information such as demographics are used in people’s early
interactions because other information will be less available.

Chatrnan and Flynn (2001) indicate that time effects in relational demography
research are inconsistent with the similarity-attraction paradigm because similarity does
not change over time. In contrast, they argue that social categorization theory allows for
changes in what is salient in a given situation. As time goes on and group members know
each other more closely, other, ‘deeper-level’ similarity may become more important.
Indeed, Shaw and Barret-Power (1998) discuss the notion of ‘willingness to

communicate’ as a potential diversity management skill that might mitigate the negative

73

effects of diversity. Though these authors suggest this is likely to be most important with
regard to deep-level diversity it seems likely that this would be true of demographic
diversity as well.

Work by Jackson et al. (1992) and Arrow (1998) also suggests that demographic
effects should change over time. These researchers see socialization, the process via
which new employees become integrated into a group, as a critical mediator of the effects
of demographic dissimilarity. In particular, these authors present a number of specific
theoretical mechanisms via which dissimilarity should impact socialization, which in turn
suggests that demographic effects should vary over time because socialization is a
longitudinal process. While little empirical research to date has focused on the role
demographic dissimilarity may play in socialization, this work nonetheless provides a
framework that explains how well known organizational processes associated with
socialization might be impacted by demographic dissimilarity.

Consistent with these notions is research reported by Flynn, Chatrnan and Sparato
(2001) that studied Extraversion as a personality construct that may facilitate the extent to
which these putative effects occur. Using MBA student teams working on a semester
project, these researchers hypothesized that extraversion would mitigate negative
impressions formed of other students that were demographically dissimilar. The rationale
for this expectation was that individuals that are extraverted are more likely to get to
know their team members more quickly than those who are introverted; hence, as
interpersonal familiarity increases, similarity on demographic characteristics should
become less important. Demographic dissimilarity was calculated by averaging

dissimilarity of race, citizenship and sex at the individual level using the D-score

74

approach I critiqued earlier. The results were consistent with the hypotheses in that
increased extraversion slightly attenuated the negative effects of demographic
dissimilarity on impressions of their team members. However, a serious methodological
caveat with this research includes the D-score approach that averaged three different
demographic indices, further clouding inference. Nonetheless, these results and

approaches, taken together, suggest the following hypotheses:

H23. The fit between an individual’s demographics and
that of the team will be more strongly related to turnover at
lower levels of tenure.

H24. The effects of demographic fit within a team will
depend on an individual’s extraversion such that high
extraversion individuals that do not demographically fit
will be less likely to turnover than low extraversion
individuals who do not fit.

The hypotheses formulated above all pertain to the effects of time and Extraversion
on individual behavior. Team-level effects are not studied here because unlike laboratory

studies the teams studied here are examined well after they have initially formed.

The Present Study
The present study broadly examines team composition as a predictor of team
effectiveness in a large sample of fast-service restaurant teams. The goal is to contribute
to the existing literature in a number of ways. First, there are virtually no large-scale
studies of teams in the psychological literature. The benefits of large-sample studies have
been widely acknowledged because they yield more power and more accurate parameter
estimates. Second, this research examines financial performance as an indicator of team

effectiveness. Though teams are more likely to yield such organizationally relevant

75

output, I have located no research on team composition and selection that has used such
outcomes. Because each of the teams in this study can also be seen as an independent
organization, this research answers calls to examine whether personnel selection has
implications for organization-level effectiveness (Schneider, 1996; Schneider et al.,
2000). In other words, this research examines mesotheoretical models linking individual,
team and organizational effectiveness (House, Rousseau, & Thomas-Hunt, 1995;
Rousseau & House, 1994). In addition, examining the financial performance of teams
allows for extensions of existing theory and application of utility modeling (Zeidner &
Johnson, 1991) because the financial value of performance is directly predicted at a
higher level of analysis (Schneider et al., 2000). Third, this study examines multiple
predictor constructs and multiple conceptualizations of team diversity. Very little
research has formulated hypotheses incorporating both multiple predictors and multiple
conceptualizations of team composition (McGrath, 1998). The addition of racial
composition as a critical concern in personnel selection also adds another dimension not
seen in existing research.

This research also seeks to make several contributions to the relational demography
literature. First, there have been very few large sample studies in this area and there is a
need for relational demographic studies in service teams (Tsui, Egan, & Xin, 1995).
Second, this research seeks to clarify a number of theoretical and methodological
limitations that are pervasive in the existing relational demography literature.
Theoretically, the contribution is the conceptualization of demographic dissimilarity as
either a cross-level effect or one that can be examined as a team-level diversity effect.

Methodologically, this research relies on hierarchical linear modeling (Bryk &

76

Raudenbush, 1992) as a data analytic system that is both powerful and tied to an
appropriate conceptualization of demographic similarity. This allows the study of
individual and group effects using a unified analytic fi'amework. Conceptualizing and
analyzing the data in such a way is consistent with Riordan’s (2000) call for relational
demography research that is conceptually, methodologically, and analytically integrated.
This research takes Riordan’s notion even further by examining these effects at multiple
levels of analysis. Practically, this should enhance our understanding of relational
demographic effects by clearly specifying the level that these supposed effects might
occur. Last, this research examines the impact of time and extraversion as potential
moderators of demographic similarity effects. These represent potentially moderating _
effects that have only recently been incorporated into relational demography research,
although numerous authors have suggested that composition effects should be studied
over time (e. g., McGrath, 1998).

The research reported here is also one of the only studies to conceptualize and
examine team effectiveness within the context of person-team fit, a notion that should
relate to multiple aspects of team effectiveness including team performance and turnover.
This is consistent with Moynihan and Peterson’s (2001) ‘contingent configuration’
approach to the study of personality in teams. In other words, the nature of the
relationship between personality and team outcomes is thought to depend on the
personality dimension studied, the task-driven conceptualization of team personality
level, and the outcome of interest. Second, by examining person-team fit as both a team-
level diversity and cross-level effect, this research seeks to demonstrate how fit

researchers can expand the domain of questions they ask. In other words, fit can be

77

related to both team and individual level outcomes. Similarly, examining variability at
the team level is something that a number of authors have emphasized with regard to
team composition research (e. g., O’Connor, 1998). Third, the multilevel nature of this
research acknowledges arguments that it is unlikely that individual level phenomena will
be unaffected by higher levels (Klein et al., 1994; Klein & Kozlowski, 2000, Rousseau &
House, 1994; Schneider 1985). In particular, these authors suggest that contextual factors
should be modeled when the existing knowledge base permits. Given the large
knowledge base in personnel selection and team composition, it appears that the time is
ripe for such research.

Finally, this research is conducted in jobs that are characterized by very low pay and
high turnover. While it might be intuitively appealing to dismiss research using such
samples as lacking any practical import, a large number of jobs like this are inevitably
needed in our economy. Further, in light of the September 11'h attacks, airport security
screeners have come under increasing scrutiny as a possible weak link in our aviation
security system. These screening jobs have been widely characterized as having low pay
and very high levels of turnover. Thus, there is a clear practical need to further our
understanding of team performance and turnover in jobs such as these.

Table 1 lists the hypotheses formulated in the research reported here. These
hypotheses are divided into individual, team, and cross-level effects based on how they
were described above, and based on the criteria of interest. The analysis method for each
is also provided. Because of the complexity of the dataset used here, however, to fully
understand the methods used to test these hypotheses, it may be helpful to refer to the

Method section, presented below.

78

METHOD
Sample and Procedures

The sample for this study consisted of a total of 336,085 applicants and employees
from 3,454 fast-service restaurant franchises across the country; however, the sample size
varies widely depending on the analysis because there was a large amount of individual
and team level missing data. The sample sizes for each of the predictors are given in the
sections below. Each franchise is independently operated though there are standardized
operating guidelines across the organization.

Hiring decisions for the organization were made based on the test used here; hence,
this research uses a predictive validation design. Based on test scores the organization .
assigned candidates to one of six bands: Double Green, Green, Double Yellow, Yellow,
Double Red, and Red. The three colors were based on norms that were designed to split
the applicant sample into thirds. Restaurants were advised that candidates in the green
band could safely be offered jobs but that they should use additional information sources
to evaluate candidates in the yellow bands. Applicants in the red band were considered
‘not recommended’. The actual cut points, and thus selection ratio, however, were
determined by each individual restaurant.

Nature of the Team Task.

Crew members are entry level employees in these restaurants and are responsible for
the following tasks: Taking orders, preparing food, delivering food to the customer,
cleaning and maintaining the premises, and setting up and preparing equipment. In other
words, the crew is entirely responsible for the day-to-day tasks associated with operating

the restaurant. The objective of the team is to quickly provide customers with food of a

79

consistent quality, and in a pleasant manner that encourages them to return. The
restaurants also must be kept physically clean for health reasons as well as to facilitate
return business. Employees that directly interface with customers also have
responsibilities that are directly related to the financial performance of the store in that
they are required to ask customers if they would like to order certain accompaniment
items such as soda. This task is very highly interdependent in that no one or two
employees can complete any significant portion of the team goal on their own. Similarly,
teamwork is involved because crew members must assist other team members who
require work products or services (e. g., a cashier might have to ask another crew member
to prepare a certain item). Crew responsibilities can also overlap. For instance, if a
customer returns to the counter to ask for something additional or complain about an
item, an employee besides the original cashier might be able to remedy what might
otherwise be a less than satisfying dining experience. Similarly, because customers
might emphasize any number of different aspects of the dining experience (e. g., speed of
service, friendliness of service, quality of food), it is difficult to define this task as
conjunctive. For instance, if a majority of customers emphasize fast service an employee
who provides unfiiendly service might not seriously impact team performance. Thus, to
reiterate, the teams studied here do not neatly fall into any of the major task typology
categories, though the situation is most closely related to the additive model; accordingly,

the mean is used to index the team level of personality and ability standing.

80

Measures
Die—Mg)!

Personnel records from 1995-1996 were used to obtain data for individual employees
(N = 190,156). The dataset contains race, sex, hire date, termination date and termination
reason for all selected employees. In addition, the last five jobs held within the
organization are listed, as is the pay for each job, and the date at which the pay or job
codes changed. The test used to hire employees provides the indicators of personality
and ability studied here, and was first implemented in 1994, though it was not widely
used until 1995. Store level financial performance and the number of transactions were
available for 13 four-week periods beginning in mid 1995. In 1995, the first time period
is Period 8 (P8), followed by P9, and so on, until P13. In 1996, the periods begin with P1
and go up to P7. So, in order, the 13 time periods are as follows: P8, P9, P10, P11, P12,
P13, P1, P2, P3, P4, P5, P6, and P7. Individual level data were available from before P8
in 1995 and through P5 in 1996. Thus, the last two time periods, P6 and P7, are not used
in any of the analyses reported here.

Independent Variarbles.

Selection Test. The test used in the current study was an 89-item paper and pencil
measure (hereinafter called the CR1) designed for use in the fast service restaurant
industry. Test booklets contained pages in both English and Spanish. Applicants were
instructed to use the language in which they were most comfortable. Items were
translated and retranslated and examined by several language and psychometric experts
to ensure that they were parallel across the two languages. Test scores were available for

123,656 individuals in 2,991 restaurants. Of these, 53,661 were hired, representing an

81

overall selection ratio of 43.4%. Implementation of the test occurred unevenly
throughout the organization. Prior to using the test this organization used a test that
measured similar constructs. Discussions with informed personnel within the
organization indicate that implementation varied considerably because of the
decentralized nature of the individual restaurants. The test was based on a job analysis of
jobs in different types of fast service restaurants (described below in the section titled
‘validation study’). Test items were in either multiple choice or Likert format with
response options anchored as follows: (1=Definitely True; 2=Somewhat True; 3=Don’t
Know; 4=Somewhat False; 5=Definitely False). A job analysis identified the following
broad requirements for successful crew members: Work orientation, Fundamental Skills,
Teamwork Orientation and Service Orientation. Although there was not a direct mapping
of constructs specified by the test developers onto the F FM personality constructs used
here, there was some consistency between the original construct labels and these
constructs. In particular, Work Orientation items tended to resemble Conscientiousness,
Service Orientation items Agreeableness, Team Orientation items Extraversion. Original
construct linkages were evaluated and changed when necessary based on consultation
with several influential papers on the F FM constructs (e. g., Barrick & Mount, 1991;
Digrnan, 1990; Mount & Banick, 1995). Examples of items on the Conscientiousness
scale are “I consider myself a very dependable person” and “People say I am a very
reliable person.” Examples of Agreeableness items are “I often lose my patience with
others” and “I have a pretty quick temper” (both reverse coded). Examples of
Extraversion items are “I would prefer a job where I work by myself” and “I’m generally

not interested in joining group activities” (both reverse coded).

82

The level of the fundamental skills items was aimed at the low cognitive complexity
of these jobs. Hence, as the name of this scale implies, it was designed to assess whether
candidates possessed minimum cognitive skills that were deemed to be required based on
the job analysis. There were a total of 12 multiple-choice items on the fundamental skills
scale, with 3 items pertaining to each of 4 problems. However, only 11 of the items were
used because one of the items (#81) was not captured in the data archive file. Each
problem presented a table with numeric and/or text information. Questions relating to the
numeric tables required candidates to look up answers to the questions in the table.

Those relating to tables with text were more complex; these items asked candidates to
make logical inferences and mathematical computations based on the information
provided in the question and the table. Thus, these items seem to require mathematical,
verbal, and logical reasoning. Items were ordered so that they increased in difficulty.

Cronbach’s (1971) index of internal consistency reliability (or), for this scale was .67 It is

not uncommon to see ot’s like this on relatively short scales with dichotomously scored
items.

Construct Validity Study of Personality Items. A construct validity study of the
personality scales described above was conducted to evaluate the extent to which they
overlap with known construct-valid measures of the Big Five. This was done because
there exists relatively little independent evidence of the construct validity of these items;
however, based on item content it was expected that these scales would yield substantial
corrected correlations with the known construct-valid scales. The items used as the
referent, construct-valid measures of the Big Five were those on Goldberg’s International

Personality Item Pool (IPIP) website (http://ipip.ori.org/ipip/). The website presents a

83

large body of construct validity evidence for these items including correlations with a
range of other commonly used and construct valid Big Five measures. 10 IPIP items
were used to assess each of the Big Five constructs (see Appendix A).

The IPIP and CR1 items were administered to 135 undergraduate psychology students
who participated in exchange for extra course credit. The CRI scales were refined based
on item statistics (means, standard deviations, and inter-item correlations and corrected
item-total correlations), item content and their correlations with the [PI scales. As a
result, one item was dropped fi'om the original conscientiousness scale based on content
(“Many supervisors expect employees to work too hard”) and its corrected item-total
correlation. The entire extraversion scale remained unchanged. The original
agreeableness scale, however, posed more challenges in that it exhibited statistically
significant, yet relatively low uncorrected correlations with the IPI agreeableness items (r
= .23, p < .05). A shorter (IO-item), though substantially less internally consistent scale,
was thus formed that exhibited a much higher correlation with the IPI scale (r = .47,
p<.05, new or = .66, original ct = .82). This tradeoff was considered reasonable because
the only hypotheses concerning agreeableness were at the store level, which yield scale
scores that are considerably more reliable than what is indicated by internal consistency
estimates because they are averaged across multiple individuals. Further, the average
inter-item correlation was the same as on the original scale, indicating that the low
internal consistency estimate was primarily due to a relatively low number of items and
not their relationships with each other per se. Based on the applicant sample ct’s were as
follows: Conscientiousness (.74), Agreeableness (.62), and Extraversion (.57). Some of

these internal consistency estimates of reliability are less than optimal; indeed, they are

84

close to the lower bound specified by Nunnally (1978) for use in research settings (as
opposed to use in making individual selection decisions). This is likely due to the short
length of the agreeableness and extraversion scales, as the average inter-item correlations
are similar to those on the conscientiousness scale. Given the large sample size used in
this research this should not impact statistical power as adversely as otherwise might be
the case. These effects will also be mitigated because aggregated data are used in many
of the analyses (Rousseau, 1985).

The results of this construct validity study are summarized in Table 2, which presents
the means, standard deviations, uncorrected correlations (upper diagonal), internal
consistency estimates (diagonal), and correlations corrected using the internal consistency
estimates (lower diagonal). As can be seen in the table, the corrected correlations
between the CR1 and IPI scales are .81, .82, and .65 for Conscientiousness, Extraversion,
and the revised Agreeableness scale, respectively. This indicates that there is a
substantial amount of overlap between the scales for all three constructs, although the
strength of these results is somewhat less compelling for the Agreeableness scale.
Accordingly, the results for this scale may have to be interpreted with some caution.

To assess the test-retest reliability of these scales, 29 of these 135 subjects took both
the IPI and CR1 items twice separated by a one-week interval. The results of the test-
retest reliability analyses are presented in Table 3. As can be seen in the table, the test-
retest reliabilities, on the diagonal, range fiom .82 - .83 for the CR1 scales, and .83 - .91
for the IPI scales. This indicates that these items exhibit substantial stability over time,
although one would certainly expect these correlations to decrease as the time interval

between administrations increases.

85

Restaurant Level Test Saturation. Because the test was implemented unevenly across
the organization it was necessary to examine the proportion of employees who were
selected with the test used here. This proportion is hereinafter referred to as CRI
Saturation. After examining these data it was decided that the last time period (P5)
would be used for all the analyses including ability or one of the personality variables.
This was done because this time period had the highest CRI saturation, and because
restaurants that are among the first to implement a new test might differ on several
important dimensions that might impact the results. By studying change over the last
time period these effects should be mitigated, as more time has passed allowing the test to
be more widely implemented. It should be noted that examining only the last time period
as the outcome for the analyses using test scores would likely yield very similar results to
the analyses reported below if other time periods were used because the average
correlation between controllable profit at adjacent time periods was .87. For this time
period, the CRI saturation was 39% (SD = .20). Although this is clearly less than ideal,
the effects of test saturation will be modeled in this research. Specifically, saturation will
be used as a moderator variable because stronger relationships should be evident in stores
with higher CR1 saturations. This is because there should be a more accurate index of the
store-level standing on the constructs studied here as test saturation increases.

Validation Stuck. A concurrent validation study of this test was conducted in a
number of franchises in this organization using both individual and store-level outcomes
as criteria (N=883 employees in 101 restaurants). This serves as a kind of pilot study for
some of the relationships examined in this research. Performance data consisted of

supervisory ratings of individual crew members’ job performance. These performance

86

ratings were gathered for research purposes only, and confidentiality of responses was
assured and maintained throughout the process. Raters were trained on common rating -
errors and strategies for the avoidance of these errors. Because there are several
supervisors with information on each crew member’s job performance, care was taken to
ensure the supervisor with the most knowledge of a given employee’s performance was
used to make the ratings for that employee. Training and ratings were made on company
time at a regional meeting.

The performance instrument was developed specifically for the purposes of this
validation study in this organization. The instrument asked for a total of 10 job skill
ratings (e. g., dependability) and 12 work habit ratings (e. g., goes beyond required duties),
half of which were positive and half of which were negative. There was also a single
item asking for an overall rating of job performance. Each of the 22 dimensions as well
as the overall-rating item was accompanied by a definition and behaviorally anchored
rating scale. The results of a principal components analysis of these ratings yielded 4
interpretable factors, which correspond with some of the dimensions of the test used in
this study (see Appendix B for performance dimensions and factor loadings, Appendix C
for performance dimension intercorrelations and internal consistency estimates). These
results indicate the performance ratings gathered were of relatively high quality; many
researchers have noted that performance ratings often only yield a single interpretable
factor (Campbell et al., 1993).

Several store level criteria were also collected in this study, namely, pre-advertising
profit, cost of labor, and cost of sales across five periods in time. Team and Interpersonal

Skills, Overall Performance, and a composite of the performance dimensions averaged

87

across employees within a store exhibited significant negative correlations with these
store-level criteria (rs = -.22, -.21, and -.18, respectively). This is evidence of the
construct validity of these performance ratings in that higher mean performance ratings
were associated with lower costs of labor at the store level. These results are even more
impressive because not every employee was rated within each store. Of course, it is also
possible that store performance impacts mean job performance ratings at the individual
level so the direction of the relationships here cannot be conclusively known.

As in a traditional validation study, correlations were calculated between the
dimensions of the test and the job performance ratings. The results indicated a
statistically significant and practically meaningful relationship between a composite total
test score and a performance rating composite (F28; Appendix D presents scale-level
validities for the four performance factors). Average test scores within each store were
also examined in relation to the store-level criteria described above. The only
relationship that was significant was the correlation between Fundamental Skills and the
cost of labor (1 = -.27). This provides some evidence that the test used here predicts
store-level outcomes of interest.

Demographics. The following demographic variables were examined in this study: Race,
sex, and age. Race was coded according to racial groups identified by the Equal
Employment Opportunity Commission guidelines. Races that accounted for more than
1% of the total sample are: White (46.6%), Black (30.5%), Asian/Pacific Islander (2.8%),
Hispanic (18.2%), and Native American (2.0%). The sample was 51.3% female. Table 4
provides a breakdown of these data. These data indicate the sample was diverse with

regard to race and sex. Table 5 describes the sample’s age.

88

There are two common formulas used to index the composition of a group on
categorical variables such as race. A number of studies (e.g., Harrison et al., 1998) have

used Blau’s (1977) index of heterogeneity:

P = proportion of group members in a category

i = number of different categories in the team

The equation above indexes the heterogeneity in a given group for categorical data,
and it varies from 0 to l asymptotically, with higher values indicating higher
heterogeneity.

Other studies (e. g., Ancona & Caldwell, 1992) use Teachman’s (1980) index of

heterogeneity:
- 2(111 P.)

Where the elements have the same meaning as those in Blau’s (1977) index. A
comparison of these indices under a number of conditions for 5 or fewer groups indicated
that the correlation between them was above .95. This suggests that respective
relationships with external variables will be nearly identical. Thus, the Blau index was
used in analyses examining the team-level effects of racial diversity as a whole because it
is more commonly used. In analyses where the dissimilarity between an employee and
the racial composition as a whole is the focus, two variables are used to indicate the
proportions of three racial groups in a given team. (Asians and Native Americans will not

be examined in analyses using dummy coded variables to reduce the number of highly

89

correlated variables in the models. Because these groups represent such a tiny fi'action of
the sample, this should have a trivial effect, if any). Store-level composition with regard
to sex was captured as the mean of a dummy coded variable where male=0 and female=1.
Age composition was also calculated directly, as the mean or standard deviation of the
ages of the crew members within a given restaurant, as appropriate.

Store Manager Tenure. The tenure of the restaurant general manager (RGM) at the
end of data collection was provided.

Dependent Variables.

Dependent variables at the individual and team level were obtained from
organizational records. Extensive discussions were conducted with representatives of the
organization to ensure that criteria were both organizationally relevant and believed to be
accurate/reliable. All the variables described below met these two criteria.

Turnover and Tenure.

The available data contains each employee’s hire and termination date, and a reason
for termination. Upon the termination of employment (by either party) the restaurant
manager completed required paperwork giving a reason for termination. Discussions
with informed organizational personnel indicated that these reasons for turnover are
likely to describe actual turnover reasons. The organization separates these reasons into
two categories: Voluntary and involuntary. Following the recommendation of Horn &
Griffeth (1995), I further separated these into Avoidable and Unavoidable categories
(only the former will be used in testing the IVT hypotheses). These reasons and their
categorizations, as well as their incidence, are given in Table 6. Though it is difficult to

determine whether there might have been some pressure exerted on managers to report

90

turnover as voluntary when an employee was actually fired (i.e., to perhaps avoid legal
action), the fact that Avoidable Involuntary Turnover represents a sizeable portion of
overall turnover indicates that this type of voluntary turnover still occurred at a relatively

high rate.

Team Performance.

 

Controllable profit was used as the indicator of team performance in this study and
was available from 2,502 stores for the 13 periods. Again, for the analyses involving test
scores, the controllable profit at P5 was used, and the value at the previous time point
(P4) was used as a control variable. Thus, the change in profitability was examined. In
addition, the number of sales transactions for the period was also included as a controlto
help mitigate the effects of large-between store differences in the volume of transactions
processed. These two control variables control for a wide range of between-store
differences that might be related to the variables studied here. The client organization
indicated that controllable profit was of obvious interest to them and it would be an
appropriate, though stringent criterion to use in this research.

Control Variables.

 

Individual Pay. Because a number of turnover theories suggest that pay impacts turnover
intentions, analyses were conducted to determine whether pay was related to any of the
predictors used here. The data revealed that pay had near-zero relationships with the
variables studied here, and thus it was not considered in any of the analyses.
RESULTS
Table 7 provides descriptive statistics and intercorrelations for the variables used in

the team-level hypotheses reported here. As can be seen in the table, there was a

91

reasonable amount of variability in all the measures except for the aggregated personality
scales, which are highly elevated and have very little variance, both as compared to the
individual level student sample (see Table 2) and based on what one would expect given
that the scale ranges from one to five. This is typical of personality items used in
selection settings. Also of note is that of the average test scores, only ability exhibits a
significant positive (though small) association with controllable profit for Period 5, which
is the last period for which both individual and store level data were available. The
results of more detailed analyses, which take into account the CRI saturation and other
control variables, are described below.

The Relationship Between Mean Levels of Ability and Personality on Changes in -

Controllable Profit

Hypothesis 1 stated that mean levels of ability would predict the change in
controllable profit after taking into account the number of sales transactions and CR1 test
saturation. It was also thought that the interaction between test saturation and ability
would add to the prediction of controllable profit because one obtains a better indicator of
ability in stores with higher test saturation. This hypothesis was tested using moderated
regression where controllable profit for the prior period, sales transactions and test
saturation were entered first, followed by ability scores, and then by the ability X test
saturation interaction term at the last step. Table 8, which summarizes the results of these
analyses, shows that the three predictors entered at the first step accounted for a very
large proportion of the variability in profit, especially profit at the prior time period ([3 =
.64; overall step R2 = .854, p < .05). The incremental contribution of ability was

significant (B = .03; AR2 = .001 , p < .05), though the ability X CR1 saturation interaction

92

term was not significant (B = .03; AR2 = .000, p > .05). To put the effect for ability in
more easily understandable terms, a store that has an ability level .5 SDs above the mean
generated an additional change of $128.57 in profitability per period as compared to a
store whose employees scored at the mean on the ability test (this was calculated by
multiplying the unstandardized regression coefficient by ‘/2 the standard deviation for P5
controllable profit). Increasing ability scores by .5 SDs in every store throughout this
sample would thus yield an additional $4.18 million in controllable profit over the course
of a year ($128.57*2,502 restaurants*(52 weeks/4 weeks)). Thus, hypothesis 1 was
supported.

A similar analysis was conducted for Hypothesis 2, which substituted
conscientiousness for ability. The results in Table 9 indicate that this hypothesis was not
supported, as the increment of conscientiousness was not significant after accounting for
the control variables (B = -.01; AR2 = .000, p > .05). Similarly, the conscientiousness X
CR1 saturation interaction was non-significant as well (B = .04; AR2 = .000, p > .05).
Hypothesis 3, which stated that conscientiousness was expected to contribute
incrementally after controlling for the effects of ability, was not tested because the results
of Hypothesis 2 indicate that this effect will be non-significant.

Table 10 summarizes the results relating to Hypothesis 4, which was parallel in
structure to Hypothesis 2 except that agreeableness is the focal construct. As can be seen
in the table, agreeableness did not significantly predict a change in controllable profit
after the first block of variables were entered (B = -.02; AR’ = .000, p > .05); Consistent
with the results reported earlier, the interaction term between test scores and

agreeableness did not add to the prediction of change in controllable profit. Thus, neither

93

Hypothesis 4, nor Hypothesis 5, which pertained to the incremental contribution of
agreeableness, was supported

Table 11 summarizes the results relating to Hypothesis 6, which focused on
extraversion exhibiting an inverted-U relationship with change in controllable profit.
Because the interaction between CRI saturation and the other personality and ability
variables were non-significant in the analyses reported above, the product term was
dropped from this analysis. At Step 2, extraversion added to the prediction of
controllable profit but this effect was negative (B = -.02; AR2 = .0004, p < .05) such that
higher mean levels of extraversion were associated with an decrease in profitability. The
polynomial term, entered at Step 3, was not significant (B = .08; AR2 = .000, p > .05); '
hence, Hypothesis 6 was not supported.

Hypothesis 7 stated that manager tenure would interact with the mean levels on the
ability and personality variables within a store to predict change in profits. The results of
these analyses are summarized in Tables 12, 13, and 14. In each of the analyses, the
control variables were entered in the first step, followed by the ability or personality
predictor, manager tenure, and then the interaction term in the third step. There was no
evidence for a significant effect for interactions associated with conscientiousness (B =
.14; AR2 = .000, p > .05), agreeableness (B = .11; AR2 = .000, p > .05), or ability (B = -
.02; AR2 = .000, p > .05). Thus, these hypotheses were not supported.

The Effects of Individual Standing on Ability and Personality Variables on Turnover

These hypotheses were tested with Cox regression (Cox, 1972), which evaluates the
effects of a set of covariates on the instantaneous risk of encountering a particular event.

Numerous authors (e. g., Harrison, 2002) have suggested this approach because it uses

94

data from cases that are censored (i.e., those individuals that have not turned over by the
end of data collection). In addition, because time is treated continuously and is evaluated
with respect to risk for turnover, more information is gleaned as compared to approaches
that treat turnover as a dichotomous variable. Much like logistic regression, testing series
of Cox regression models yields differences in model fit distributed as x2 which can then
be tested for statistical significance. Parameter estimates can be anti-logged to yield
estimates of the changes in risk due to a given change in the covariate or predictor.

Hypothesis 8 suggested that conscientiousness would be negatively related to the risk
of voluntary turnover. Of those that had conscientiousness scores, 22,606 experienced
voluntary turnover during data collection, while 15,959 were censored.
Conscientiousness was a significant predictor of turnover risk (Ax2 = 37.13, B = .11, p <
.05.). Anti-logging the B parameter (hereinafter referred to as Exp(B)) yields a value of
1.1 1, indicating that a one-unit increase in conscientiousness results in a risk of turnover
that is 111% of the original risk. This does not support Hypothesis 8 because higher
conscientiousness scores were predicted to lower one’s likelihood of voluntary turnover.

Hypothesis 9 was identical Ito the previous hypothesis except that it specifed
involuntary rather than voluntary turnover. This analysis was based on 10,518 employees
with conscientiousness scores who experienced involuntary turnover, and 15,959 who
were censored. The results of this analysis did not support the hypothesis because
conscientiousness was unrelated to the risk of involuntary turnover (Ax2 = .47, B = .02, p
> .05.). (Because conscientiousness was not related to turnover it was not used as a

control in any subsequent hypotheses concerning demographic diversity and turnover).

95

Hypothesis 10 stated that high ability employees would be more likely to exhibit
voluntary as opposed to involuntary turnover. This hypothesis was tested using two
separate survival analyses — one for each type of turnover. For voluntary turnover (N =
22,596; censored N = 15,948), higher ability was associated with a slightly lower risk of
turnover (Ax2 = 23.28, B = -.013, SE = .003, p < .05., Exp(B) = .987). For involuntary
turnover (N = 10,511, censored N = 15,948), higher ability was also associated with a
slightly lower risk of turnover (Ax2 = 201.28, B = -.06, SE = .004, p < .05., Exp(B) =
.944). These results are not consistent with the hypothesis.

Hypothesis 11 stated that increased diversity in conscientiousness would be
associated with decreases in store performance. In this analysis, after the control
variables were entered, the coefficient of variation for conscientiousness was entered.
This coefficient is calculated by dividing the standard deviation by the mean as a way of
expressing the former controlling for the latter. As can be seen in Table 15, the
increment in this step was not significant (B = .01; AR2 = .000, p > .05). A similar
exploratory analysis was conducted with agreeableness, yielding similar results (B = .01;
AR2 = .000, p > .05).

Hypotheses 12 and 13 made predictions about the relationship between
conscientiousness diversity and voluntary and involuntary turnover rates. This
relationship was examined for Period 4 because there were no termination reasons
recorded for terminations that occurred during Period 5. These hypotheses were not
supported, as the correlation between the conscientiousness coefficient of variation and
voluntary and involuntary turnover rates was not significant (r = .02 and r = .01 ,

respectively, p > .05).

96

The Relationships Between Demographic Diversity and Profitability and Turnover
Rates
These hypotheses — 15, 16, 18, 19, 21, and 22 - were tested using latent grth curve
(LGC) methodology in structural equation modelling (Meredith & Tisak, 1990;
Stoolmiller, 1995; Willett & Sayer, 1994). LGC modelling uses a latent intercept and
slope representing the starting point and change over time, respectively, of a given
variable with repeated measurements. The slope can be linear or a higher order
polynomial (e.g., quadratic), depending on the hypothesized or observed pattern of
change over time. The mean of the slope and intercept parameters define the overall
trajectory for the entire sample on which they are calculated. The statistical significance
of their variances indicates whether there is a significant amount of inter-individual
variability in these trajectories. If one can identify a model that adequately describes the
latent form of the change pattern underlying the observed data, and if the slope and
intercept variances are significant, one can then seek substantive predictors of these two
parameters. In this way, LGCs make it possible to identify correlates of the starting point
(intercept) or change over time (slope) in a given set of individuals or to relate intercept
and slope parameters across different variables (e. g., changes in diversity and changes in
profitability). Analyses of the latter type, which are used to test the hypotheses in the
research reported here, have been termed cross-domain LGCS (Chan, 1998).
Although LGCS are traditionally used to model individual change over time, in the

present study a number of store-level changes were of substantive interest. In particular,
the hypotheses mentioned above identify demographics as potential correlates of profit

and turnover rates. The dynamic nature of these criteria make them well suited to the

97

LGC approach described above; however, rather than identifying a single static predictor
of changes in profit and turnover rates, the analyses conducted here also take into account
the dynamic nature of the predictors. Thus, for this study the intent was to estimate
separate LGC models for each of the five variables in these hypotheses: Age, sex, and
racial diversity, controllable profit and turnover rates. Once good-fitting models were
identified, the respective models were combined as appropriate to see if their slopes and
intercepts covary. This allows one to examine whether the starting point or rate of
change for the diversity variables is related to the starting point or rate of change in
profitability or turnover rates over time.

A graphical representation of a simple LGC model is depicted in Figure 4 (Chan, -
2002). The variables Y1 — Y4 represent a variable measured at four different points in
time. Associated with each of these variables is a disturbance or error term (a. - 8,),
which includes both the prediction and measurement error associated with each. In
addition, each has a factor loading on the intercept (An - A“) and slope (Ms — A43) latent
factors. In order to define the intercept as such, the factor loadings are constrained to
one. For the slope, two loadings have to be fixed at zero and one to facilitate model
identification, while the remaining loadings can be fixed to represent other forms of
growth (Meredith & Tisak, 1990). For example, a linear growth pattern would fix the
remaining two loadings at two and three, whereas a quadratic pattern would specify an
additional latent slope factor where the the third and fourth factor loadings are fixed at 4
and 9 (22 and 32, respectively). Alternatively, if no particular form of slope is expected,
or if no easily recognizable form is seen in the data, one could freely estimate the final

two slope parameters. As mentioned earlier, the intercepts and slopes also have a mean

98

(u. and us) and variance (0’, and 0’s), and a covariance among each other (05.). This
general model, discussed below, was implemented in the LGCS described here.

LISREL 8 (J Oreskog & Sérbom, 1993) was used to estimate the LGCS reported here.
To assess the adequacy of the various models tested, five fit indices were used to obtain
convergent evidence of model fit. This has become common practice in studies using
structural equation modeling because the original )6 fit index is highly sensitive to
sample size in that models with excellent fit will often yield significant )6 values falsely
indicating the hypothesized model does not fit if the sample is large enough. The indices
used were: The root mean square error of approximation (RMSEA; Steiger, 1990);
normed and nonnorrned fit index (NFI and NNFI, respectively; Bentler & Bonnett, 1980);
adjusted goodness of fit index (AGFI; J Oreskog & SOrbom, 1989); and standardized root
mean residual (SRMR; J Oreskog & S6rbom, 1986). All of these fit indices range fiom 0
to 1, with 1 indicating better fit except for SRMR and RMSEA. For the former category
of fit indices, values above .90 are considered good fit, whereas for the latter, values of
.10 or less are considered to be good fit and values of .05 or less are considered close fit.
In addition, modification indices (Mls), standardized residuals and standardized expected
change parameters for particular models were examined to determine the extent to which
fit might be improved by estimating or constraining specific parameters. Also considered
in these decisions were substantive concerns (e.g., whether fixing a freely estimated
parameter makes sense given the research design and hypotheses of interest).

Before any of the LGC models were tested, the bivariate correlations between the
three diversity variables and controllable profit and turnover rates were examined. This

was done because there is little to be gained by estimating LGC models for variables that

99

might not be correlated in the first place. This was the case for all the diversity variables
and turnover rates. In particular, the mean same-period correlation between each of the
11 4-week period turnover rates and racial, age, and sex diversity was .03, .01 and .01,
respectively. In addition, the average correlation among the monthly turnover rates was «-
.04, and the maximum correlation was only .02. This does not support the notion that
there is a readily identifiable latent variable driving the period turnover rates. Indeed, it
was not possible to achieve a good fit for a LGC model with these data, as several of the
fit indices were stuck below .90. Despite this, it nonetheless seemed worthwhile to
investigate whether turnover rates aggregated over a longer time period might yield more
positive results, especially given earlier studies (e.g., McElroy, Morrow, & Rude, 2001;
Shaw et al., 1998) that have done so. Table 16 shows the correlations among the
aggregated turnover rate and diversity variables. As can be seen in the table, the
correlations of interest are all significant, though relatively small. As predicted, higher
diversity of all three types was associated with higher turnover rates. To assess the joint
impact of these diversity variables on turnover rates, a multiple regression analysis was
conducted where all three diversity variables were entered in a single step with the
turnover rate serving as the dependent variable. As can be seen in Table 17, the partial
regression coefficients for race and sex diversity were significant (Bmdi, =.07; Bmg, = .08;
AR2 = .015, p<.05), though the partial effects of age diversity were no longer significant
(BM...-v =.02). Thus, hypotheses 15, 18, and 21 received bivariate support, although the
LGC methodology originally proposed to test the hypotheses was found to be

inappropriate due to the lack of dimensionality underlying the 4-week turnover rates.

100

Returning to the correlation matrices underlying the proposed LGC models, for sex
diversity and controllable profit the average same-period correlation was only .02, and
inspection of the scatter plots did not reveal any discernible patterns (e.g., nonlinearity).
Thus, hypothesis 22 was not supported. In contrast, there were larger and more
consistently significant relationships between age and racial diversity and controllable
profit. Thus, the following sections describe the development of the individual growth
models for these three variables. After each of these models is developed, two combined
models are discussed that examine whether the intercept and slope parameters of the
diversity variables are significantly correlated with those of controllable profit.

LGC Model for Racial Diversity

Recall that racial diversity is operationalized as 1— Z P,2 where P is the proportion

of members in 1 groups. This index was calculated for each of the 11 time periods. The
means, standard deviations and intercorrelations of these values were used as LISREL
input in fitting the initial model. These data, presented in Table 18, indicate that the
mean and standard deviation are almost constant across time. However, a plot of the
individual store trajectories showed substantial variability in the nature of change in
racial diversity over time. The change in this index over time for a random sample of 15
restaurants is presented graphically in Figure 5. As can be seen in the figure, there is no
easily identifiable function that might be used to represent the slope in these data; thus,
the last nine of the eleven slope parameters were freely estimated. Similarly, there
appears to be reasonable variability in the intercept and slopes across the stores. If this
variability can be explained, the data may be more interesting than the constant pattern of

means and standard deviations in Table 18 suggests. Another pattern worthy of note is

101

that the data in Table 18 exhibit serial autocorrelation in that the correlations are
substantially higher as the periods become closer in time. After parameters in an initial
model were estimated, and after the MIs were considered in light of these data, it was
decided that adjacent error covariances of the 11 indicator variables should be freely
estimated to improve model fit. In addition, the MI for the error covariance between P3
and P5 was 284, so this parameter was allowed to be freely estimated as well. This
yielded a final model depicted in Figure 6. The fit indices for this model suggested a
good fit to the data (NFI = .99, NNFI = .99, CFI = .99, AGFI = .93, RMSEA = .08,
SRMR = .05, x2 = 593.53, df = 41, p<.05). In this model the pattern of slope factor
loadings supports the notion that neither a linear nor quadratic slope would accurately-
represent these data. A plot of the standardized LISREL estimates of the slope factor
loadings is presented in Figure 7. These might be more appropriately called “shape
patterns” (Chan, 2002) or “linear splines” (Meredith & Tisak, 1990), but the original
“slope” designation is used here to be more broadly consistent with the LGC literature.
Chan, Ramey, Ramey, and Schmitt (2000) note that much like in regression, in LGC
analyses, one can test the incremental fit provided by estimating a freeforrn slope as
compared to a linear or quadratic slope. Thus, models identical to the one described
immediately above were tested, with the slope factor loadings constrained to either a
linear or quadratic form. The relative fit of these models was tested with the commonly
used )6 difference test, although differences in the practical fit indices were examined as
well. The model with a linear slope (i.e., one with Slope factor loadings fixed to 0 - 10
for the 11 time periods) fit the data relatively well as indicated by the practical fit indices

(NFI = .98, NNFI = .98, CFI = .98, AGFI = .87, RMSEA = .10, SRMR = .05). The x2

102

value for this model was 1132.44 with 50 degrees of freedom. Thus, the Ax2 was 538.91 ,
with Adf = 9 (p < .05). This indicates this straight-line trajectory model fits the data
significantly less well than the freeforrn slope described above. This is consistent with
the slight decrement in the practical fit indices. An additional model was estimated that
had two latent slope factors - one with a linear trajectory and another that was quadratic.
The practical fit indices for this model were nearly identical to that of the initial model
(NF I = .99, NNFI = .99, CFI = .99, AGFI = .92, RMSEA = .08, SRMR = .01), though the
)6 value was somewhat larger (x2 = 728.17, df = 46), a difference that was statistically
significant (Ax2 = 134.64, with Adf = 5, p < .05). This also supported the notion that the
freeforrn slope model provided the best fit to these data.

LGC Model for Age Diversity

Age diversity was operationalized as the standard deviation of employee age in days
(age was calculated when the employee was hired). LISREL could not calculate starting
values with the resulting large numbers in the fitted covariance matrix, and so this
variable was divided by 100, which eliminated the problem. The data input to LISREL,
after applying this transformation, are presented in Table 19. As with race diversity, the
means and standard deviations show a very consistent pattern across time. To better
understand the nature of the changes in age diversity over time a plot of a random sample
of 15 restaurants was examined (see Figure 8). Seeing a similarly unclear pattern as with
the racial diversity profile, the form of the slope parameter was again freely estimated.
This yielded a slope factor depicted in Figure 9. As in the previous models, based on the
MIs, the error covariances of adjacent time points were allowed to vary, as were those

between several variables only two time periods apart: P13 and P2, P11 and P13, and P3

103

and P5. The resultant model, shown in Figure 10, fit the data very well (NF I = .98, NNFI
= .98, CFI = .98, AGFI = .94, RMSEA = .08, SRMR = .08, x2 = 543.89, df = 39, p<.05).
The same model with a linear slope fit the data significantly less well (NFI = .97, NNFI =
.96, CFI = .97, AGF I = .89, RMSEA = .09, SRMR = .08, x2 = 1,085.77, df= 48; Ax’ =
541.88, with Adf = 9, p < .05). The model with two latent slopes — with linear and
quadratic shapes — also fit the data significantly less well as compared to the original
model (NFI = .98, NNFI = .98, CFI = .98, AGFI = .93, RMSEA = .08, SRMR = .04, x2 =
666.61, df= 44; Ax’ = 122.72, with Adf= 5, p < .05). This supports the use ofthe
freeforrn slope model for age diversity.

LGC Model for Controllable Profit

As with age diversity, LISREL was unable to obtain starting values due to the scale of
controllable profit. Thus, this variable was divided by 100 to facilitate the computation
of starting values. The LISREL input is provided in Table 20. As with the previous
variables, the trajectories of a random sample of 15 cases was examined (Figure 11). The
figure suggests that estimating a fiee—fonn slope would be appropriate in this situation.
After examining an initial model with error covariances constrained to zero, it was
decided that adjacent error covariances should be freely estimated. The resulting model,
shown in Figure 12, provided a good fit to the data without any additional modifications
(NFI = .98, NNFI = .98, CFI = .99, AGFI = .93, RMSEA = .08, SRMR = .02, x2 =
643.16, df = 42, p < .05). Both in relative and absolute terms, a linear growth model fit
the data very poorly (NFI = .88, NNFI = .88, CFl = .89, AGFI = .75, RMSEA = .21,
SRMR = .05, x2 = 4,536.89, df= 51; A36 = 3,893.73, with Adf= 9, p < .05), as did a

model with linear and quadratic slope factors (N F I = .94, NNFI = .93, CF I = .94, AGFI =

104

.90, RMSEA = .15, SRMR = .04, x2 = 2,406.88, df= 47; Ax’ = 1,763.72, with Adf= 5, p
< .05). This supports the use of the fieeform slope model for controllable profit (see
Figure 13).

Combined LGC Model for Racial Diversity and Controllable Profit

Table 21 presents the intercorrelations between racial diversity and controllable profit
over the 11 time periods. As can be seen in the table, these correlations are all significant,
and the average correlation is .13. This supports the notion that investigating a combined
model where the slopes and intercepts of these two variables are allowed to covary is
worthwhile. The data provided in Tables 18, 20, and 21 were used to estimate such a
model. However, there were estimation problems with the freely estimated factor
loadings of the racial diversity slope that prevented LISREL from achieving stable
starting values. Fixing the racial diversity slope factor loading for P1 and P2 to the
maximum likelihood LISREL (i.e., unstandardized) estimates obtained in the separate
LGC race model depicted in Figure 6 solved this problem. The resulting model fit well
but the M1 for the error covariance of racial diversity at P1 and P12 was 184. Because
this was a large M1 for variables that were only two time periods away, this parameter
was freely estimated. The final model allowed the four latent variables to covary, while
the separate racial diversity and profit LGC models were identical to those discussed
earlier except for the two fixed racial diversity slope factor loadings and the freely
estimated error covariance mentioned directly above.

The fit indices indicated this model fit the data well (NF I = .99, NNFI = .99, CFI =
.99, AGFI = .95, RMSEA = .05, SRMR = .03, x2 = 1,178.78, df= 201, p <.05). The

associated parameter estimates are presented in Figure 14. As can be seen in the figure,

105

the race intercept is significantly associated with both profitability parameters. In
particular, it is positively associated with the controllable profit intercept (standardized
effect size estimate = .11, p < .05), but negatively associated with the slope (standardized
effect size estimate = -.16, p < .05). This means that at the beginning of data collection
racially diverse stores also tended to be slightly higher in terms of profitability.
However, racial diversity was also associated with a negative change in profitability over
time.

To put these effects in more concrete terms, the impact of racial diversity on profit
trajectories was calculated for stores at -l, -.5, 0, +5, and +1 standard deviation (SD)

units from the mean of racial diversity using the equation below:

A — lo _
PRDSD, = [(.1 1RDSD, ) + P, ]+ [Z (1.31,, — .16RDSD, XPSDM )+ PM] (EQI)

i=1

Where:

R

Pmsm =the total predicted profitability across the 11 time periods for a
given standing on racial diversity in k SD units;

RDSD.K = the deviation fi'om the racial diversity mean in k SD units where
k = -1, -.5, 0, +5, and +1.0;

[—90 = the mean profit at the first time period

j = the 10 intervals separating the 11 time periods;

Ms,- = the completely standardized factor loading for the profitability slope
at time j+l;

PSD,.. = the SD of profit at time j+l;
F

1+! = the mean profit at time j+1.

106

This equation was used to calculate five predicted totals and trajectories for
controllable profit, each associated with one of the k values of racial diversity expressed
in SD units. As can be seen in the equation, the positive effect of racial diversity on the
profitability intercept is taken into account by the terms on the left of the summation. In
particular, the SD units for a given level of racial diversity is multiplied by .11, the
correlation between the racial diversity and profitability intercepts, which is then added to
the mean profit at time one (P8). So, for a restaurant 1 SD above the mean on racial
diversity the predicted starting point on profitability is (.1 1*] SD*$9,041/SD) + $19,921
= $20,915.

The terms to the right of the summation represent the predicted profit for a given
level of racial diversity in SD units for each of the 10 time intervals. In the first set of
parentheses after the summation, the predicted profit slope factor loading is calculated
based on the initial slope factor loading, the racial diversity SD, and the correlation
between the racial diversity intercept and profit slope (-.16). Thus, for the first interval
(i.e., in between P8 and P9), where the profit slope factor loading is .04, the predicted
profit slope factor loading for a restaurant 1 SD above the mean on racial diversity is (-
.16*lSD)+.O4 = -.12 profitability SD units. This value is then sealed by the profit SD for
a given time point, and added to the mean profit at that time point to obtain the predicted
profit for that time point given the standing on racial diversity. Thus, for the example
using +1 SD units above the racial diversity mean, the terms for the first and last time

periods are:

[(— .16)(ISD)+ -.1 IKSS,555/SD)+ $17,929

107

and
[(—.16)(ISD)+.04I$8,970/SD)+ $20,654

Summing across the 10 time intervals and taking into account the initial starting
point, discussed above, thus yields the total predicted profit for the 11 time points for the
given level of racial diversity in SD units.

Figure 15 presents the five predicted profitability trajectories associated with these
levels of racial diversity. The figure shows that stores higher in racial diversity start off
at higher levels of profitability, but that the effect on the slopes offsets this effect after the
first time period (P9). Stores at higher levels of racial diversity are projected to continue
at lower levels of profitability throughout the data collection period. Using equation 1, l
the total profits were calculated for the time period under study for these five levels of
racial diversity. Figure 16 shows the linear association between standardized racial
diversity levels and the predicted total predicted profits over the course of a year. A 1 SD
difference in racial diversity amounts to $14,656 per restaurant. To make the impact of
the racial diversity metric more concrete, several possible store compositions were
examined to assess the predicted effects of racial diversity on profitability over the course
of one year (the predicted values calculated using EQl were multiplied by 52/44 to scale
the 44-week predicted total to an annual predicted total). Figure 17 presents these results
graphically. Each pattern represents a different racial group and the smallest slice (e. g.,
one of the two small slices in the Chart 2 of Figure 17) represents one person. These
charts reinforce the notion that small differences in racial diversity are predicted to make

sizable differences in controllable profit (again, note that these charts show the predicted

108

effects of racial diversity, rather than those associated with the proportion of a particular
racial group). For instance, the difference between Charts 5 and 6 of Figure 17 is two
members from one racial group, but the difference in predicted annual profit between the
two compositions is almost $13,000. Similarly, Chart 6 has only one member from the
non-majority group in the restaurant whereas Chart 4 depicts a restaurant where half the
members are of the same race and the remainder is made up of three different races. This
translates into an estimated annual difference in profit of over $41,000. Applying these
numbers across the entire organization of over 2,000 stores results in differences that are
practically meaningful changes in controllable profit.

Combined LGC Model for Age Diversity and Controllable Profit

Table 22 presents the intercorrelations between age diversity and controllable profit
over the 11 time periods. As can be seen in the table, these correlations are generally
significant, though small. This suggests that investigating a combined model where the
slopes and intercepts of these two variables are allowed to covary is worthwhile. The
data provided in Tables 19, 20, and 22 provided the data used to estimate this model. As
with the combined model for racial diversity, there were estimation problems with the
freely estimated factor loadings of the age diversity slope that prevented LISREL from
achieving stable estimates, and so two loadings were fixed using their LISREL estimates
from the age diversity model. The combined model fit the data very well (N FI = .98,
NNFI = .98, CF] = .98, AGFI = .94, RMSEA = .05, SRMR = .05, x2 = 1,336.37, df=
200, p <.05). Figure 18 provides a representation of this model along with the
standardized effect size estimates. As can be seen in the figure, the age diversity

intercept was positively and significantly associated with the profitability intercept

109

(standardized effect size estimate = .10), though none of the other parameters of interest
were significant. Thus, hypothesis 19 was not supported.

Relationship Between Demographic Fit Within a Store and the Probability of Turnover

Hypotheses 14, 17, and 19 stated that the race, age, and sex-based fit of an individual
within a store are expected to predict individual turnover probabilities. Earlier, a number
of reasons were discussed as to why survival analysis is well suited toward the prediction
of such probabilities. Here, this approach is extended in a multilevel framework where
store composition is used as a moderator of the individual-level relationships between
these demographic variables and turnover. This was done using hierarchical nonlinear
modeling (HnLM) using the HLM software package (Bryk, Raudenbush & Congdon,
1996). At the individual level, the logit link function was used to model the relationship
between a given demographic characteristic and turnover within each store. The
variability in the individual-level parameters is then modeled as a function of store-level
composition variables. If these store-level composition variables do indeed predict the
variability in the individual level parameters (i.e., the extent to which a given
demographic variable is related to turnover probabilities), then the form of the
relationship will be inspected to determine whether it is consistent with the hypotheses.
Because continuous time multilevel survival analysis models are not as fully

developed in comparison to discrete time models, the latter approach was taken here.
Transforming continuous measurement of time (here it was measured in days; i.e., the
day hired and terminated) is referred to as coarsening in the event history modeling
literature. A number of researchers have noted that coarsening does not pose serious

estimation or bias problems as long as the coarsening occurs at random and no more than

110

roughly 5% of the events (i.e., turnover) occur at a given interval (Allison, 1984). Both
of these conditions are met here. For the purposes of this study, a period of two weeks
was chosen. This was done because this was a relatively high turnover job, and an
inspection of the survival curve which suggested that a longer interval would lose some
information about the survival probability over time. In addition, with this approach
there is actually relatively little coarsening because many employees’ terminations were
registered at two-week intervals (i.e., at the end of a pay period). In other words, if one
were to examine a bar chart plotting the number of terminations over time there would be
a series of small spikes spaced two weeks apart.

This approach resulted in 35 periods (i.e., 70 weeks) from which individual hire and -
termination data were available. If an employee did not turnover by the end of the 35‘”
period, his or her tenure is censored (i.e., turnover did not occur by the end of data
collection). Singer and Willett (1993; Willett & Singer, 1991) indicate that standard
logistic regression software yields maximum likelihood estimates that can be used to test
the hypotheses associated with survival analysis. However, to allow for this, the cases
must be restructured into a person-period format. An example of this data structure is
provided in Table 23. The table contains records for three hypothetical individuals who
worked in the same store for a total of 40 two-week periods for which data were
available. As can be seen in the table, there is one record for each of these two-week
periods. Based on the third column, one can see that employees 1 and 2 turned over after
working for three and two time periods, respectively; employee 3 worked for the entire
span of data collection and was censored because the turnover event had not occurred at

the end of data collection. In the fourth column the sex of each employee is entered on

111

all the records for that employee. In the last two columns, we have the store sex
composition calculated for each period the employee worked, and the product between
the individual’s sex and the store sex composition (the treatment of these last two
variables in the analysis is discussed later). The sex compositions are not the same for
employees 1 and 2 because the time periods are not the same — rather than representing
an absolute point in time, they represent, in sequence, the number of periods a given
employee worked (i.e., each employee’s time starts at one). This was done to ensure that
the effects of time are modeled consistently across employees. Before proceeding with
the multilevel survival analyses, however, a single level analysis was conducted to better
understand the form of the survival function. This was done by conducting a logistic
regression on the individual-level person-period dataset structured as described above (N
= 1,319,180 person—periods; N = 177,873 employees; censored N =29,515; these figures
also describe the sample size for the demographics-based multilevel analyses reported
below). In this analysis turnover was regressed on period and period’, and predicted
turnover hazards at each of the t periods were calculated using the following equation
(Tabachnick & Fidell, 1996):

A+BlPeri0d +BZPeri0d2
t t

e
P(Turn0ver)t = (EQ2)
A-I-BlPeriodt +32 Period!2

 

l+e

The results of this analysis are shown graphically in Figure 19 (P(Turnover) = -1.85 -

.04'(Period) +.001‘(Period2), -2 Log Likelihood = 924,926; all parameters significant at

p<.05). This figure plots cumulative survival as a function of time. As can be seen in the

112

figure, the survival curve closely resembles a quadratic function. The shape of the curve
and the significance of the quadratic term suggest that a quadratic term should be used in
the multilevel analyses.

One of the challenges alluded to above is how to handle the dynamic nature of the
store compositions. Fortunately, survival analysis at the individual level can easily
handle covariates that change over time. Similarly, multilevel survival analyses can
handle time-varying covariates, either at the individual or group level, though
implementations of the latter are extremely rare. Barber, Murphy, Axinn, and Maples
(2000) present the only discussion of which I am aware that deals with how to handle this
latter situation. Because this is not yet commonly done, a brief explanation of the
analysis is provided here, though the reader is referred to Barber et al. (2000) for a more
detailed discussion.

For this example the sex will be used because a dichotomous variable simplifies the
the explanation. Based on the hypotheses formulated earlier in this paper, one would
expect to find that the store-level sex composition (i.e., proportion of employees of a
given sex) moderates the relationship between sex of the employee and the probability of
turnover. Recall that effects such as these are commonly referred to as cross-level
interactions. In addition, it is necessary to include variables establishing the baseline

survival function. One way to do this is to enter periods and the periods2 to allow for a

113

quadratically shaped survival curve.l The conceptual model in a traditional two-level

form, as in Bryk and Raudenbush (1992), is as follows

L0git(p,jk) = .30). +fllPeri0da. + ,BzPeriod; + fluSexjk (EQ3)

fl... = 700 + 60,. (EQ4)
.31 = 7m (EQS)
fl. = 720 (EQ6)
[3,, = 730 + yMStoreSexlk + 5,, (EQ7)

Where:

,60, = the overall turnover probability in store k

,6, = the effects of time on the turnover probability for person j at time t

,3, = the effects of time2 on the turnover probability for person j at time t

,63, = the sex effect on the turnover probability for person j, which varies as a

function of the store sex composition in store k at time t

80K ,£,K = the unobserved random effects error terms (i.e., that vary across k stores)

 

' Willett and Singer (1993) recommend using a dichotomous variable representing each time period in
discrete time survival analyses. Their recommended approach has the benefit of describing the survival
function more precisely because a hazard is calculated for each time period under study. This poses
problems in the present research, however, because interactions with each of the 35 time periods would
require adding several dozen new variables to the HnLMs (over 200 in the case of the analyses of racial
similarity due to the use of dummy variables to represent the racial groups). This, combined with the shape
of the survival curve, which closely resembled a quadratic function, led to the decision to use a linear and
quadratic factor in the multilevel analyses. Other researchers have also used this approach (e. g., Barber ct
al., 2000; Homey, Osgood, and Marshall, 1995). Thus, parsimony dictated the use of the Period and
Period2 variables as opposed to the indicators.

114

Substituting terms to express these equations as a one-level model yields the

following:
Logit(prjk) = (700 + 80K ) +(710Peri0dij) +(720Peri0d5)
'I' (730 + 73ISt0resexik + 83K) (EQ8)

This single level equation can now be reformulated as a two-level HnLM where

StoreSex... is treated as an individual level covariate that does not vary across stores:

L0git(p,j,) = flak + ,8, Period 0' + ,BzPeriod: + ,B3kSexj + ,B4StoreSexU.

+ ,6, Storesex, Sex ,. (EQ9)
[30. = 700 + 8M (EQ10)
lJ’. = 7.0 (EQI 1)
A = 720 (EQIZ)
A. = 730 + 83x (EQ13)
fl. = 7.0 (EQ14)
fl. = 750 (EQ15)

Thus, as can be seen above, the main effect and corresponding cross-level interaction
between the time-varying StoreSex,.K covariate and the employees’ sex,- is now at the
individual level, without a corresponding random error component in EQ14 or EQ15.
This allows for proper estimation of this effect. The results of these models are described
below.

Effects of Store Sex Composition on the Relationship Between Sex and Turnover

115

Equations 9-15 above represent the initial model fitted to the data.2 Table 24
summarizes the results of this analysis. The first column provides the variable name (and
indicates the coding in the case of the dummy and composition variables), followed by
the gamma designation and associated random error components (when applicable), and
then the parameter estimate, standard error, the t coefficient and Exp (B). All the
substantive parameters of interest were significant predictors of survival probabilities. A
predicted turnover hazard was calculated for each time period for males and females
working in two different store sex compositions using the formula given in EQ2 which
was expanded to include all the coefficients included in this analysis. The results of these
calculations reveals the form of the sex X store sex composition interaction (see Figure .
20). As can be seen in the figure, shifting to a store with a higher female composition
slightly reduces the likelihood of turnover for a female employee. An opposite, though
somewhat larger effect is seen for males. Thus, hypothesis 20 was supported because
increased sex similarity reduces the hazard of turnover. Note that while the differences
between the curves on the chart are relatively small, the plotted hazards are not
cumulative, but rather are calculated separately for each time period. Cumulative
survival probabilities would reveal larger differences than those depicted in Figure 20.

The curves in Figure 20 are parallel because this model does not include any

interactions with time. Thus, additional analyses were conducted to see if the sex X store

 

2 Models with time and time2 as random effects were also estimated, which allows the particular shape of
the survival function to vary across stores. The results obtained in these analyses were very similar to those
reported here (where these variables were modeled as fixed). However, due to ill-conditioning, in some of
the analyses several thousand restaurants were dropped. While this effect could likely be mitigated by
centering the level 1 predictors, the intercept in the models for race and sex is inherently meaningful where
the level 1 predictors are in their raw score form, and so it was decided that the two time parameters should
be modeled as fixed to make better use of the entire dataset.

116

sex composition interaction might vary as a (linear) function of time. This was done
because some of the relational demography literature reviewed in the introduction
suggests that these effects occur earlier on in groups (Hypothesis 23). Because the
groups studied here were very dynamic, these effects were modeled as a function of
individual tenure rather than by looking at the start or end of a group. This analysis thus
added three cross-level effects to the model: Sex X period, store sex composition X
period, and sex X period X store sex composition. Table 25 summarizes the results of
these analyses. As can be seen in the table, all three interactions were significant. These
results are also presented graphically in Figure 21. As can be seen in the figure, the
curves are no longer parallel, resulting from the significant interactions with time.
Clearly, there is support for the notion that the sex X store sex composition interaction is
initially important but that this effect fades over time. This is evident after examining the
change in the vertical displacement of the two curves for males over time. At the end of
the first two week time period (P1), a male working in a 90% female store has a predicted
hazard rate of 16%, as compared to a predicted hazard of 11% working in a 10% female
store. By P20, however, the curves cross and the form of the relationship between sex
and store sex composition is reversed. A similar effect is seen for females, except that
the curves cross much earlier, indicating that sex similarity effects become a less
important predictor of turnover hazard more quickly as compared to males. This supports
Hypothesis 23.

Effects of Store Race Composition on the Relationship Between Race and Turnover

The model for race was the same as the one for sex except that an additional

individual level dummy variable was necessary because three race categories were

117

examined for these analyses: Blacks, Hispanics, and Whites. Similarly, another store
composition variable was necessary to represent the additional proportion which served
as another indicator of store racial composition. As with sex, only the two indicator
variables for individual employees’ race were allowed to vary across stores. The results
of these analyses are presented in Table 26. As can be seen in the table, all of the
coefficients are significant. These results, used in conjunction with an appropriately
expanded version of the formula presented in EQ2, are represented graphically in Figure
22. (Note that because some of the curves overlap, the underlying data are presented in
Table 27). The figure shows predicted turnover hazards at various points in time as a
function of individual race within three different store racial compositions. As can be
seen in the figure and the accompanying table, the turnover hazard for Whites is highest
in a store composition that is 10% White, 10% Hispanic, and 80% Black. In contrast, it
is lowest when the composition is 80% White, 10% Hispanic, and 10% Black. A similar
pattern is seen for Hispanics and Blacks in that their predicted turnover hazard is lowest
when their own race is most highly represented in the store composition. Thus, H14 was
supported.

As with sex, additional analyses were conducted to see if these effects vary as a
function of time. The results of these analyses are presented in Table 28. As can be seen
in the table, several interactions with time were statistically significant. These results
(with the non-significant coefficients set to zero in the calculation of predicted turnover
hazard), are presented graphically in Figure 23 (accompanying data are in Table 29). For
Whites, there is some evidence to support the hypothesized effect in that the interaction

between time, the individual level White dummy variable and the two store composition

118

variables were significant. Indeed, the figure shows that the difference in predicted
turnover hazard between the 80% White, 10% Black, 10% Hispanic and the other two
store compositions for Whites becomes smaller as time goes on. A similar effect is seen
for Hispanics, though this is only seen in the difference between the 80% Hispanic, 10%
Black and 10% White line and the 80% White, 10% Black and 10% Hispanic lines. The
curves for Blacks are similar in that the protective effects of a predominantly Black store
composition on turnover hazard actually reverses in pattern with one of the other
compositions at P16. Thus, these data provide some support for the notion that the
impact of racial similarity on turnover risk is higher when individuals are first hired as
compared to later on in their tenure. This supports Hypothesis 23.

Effects of Store Age Composition on the Relationship Between Age and Turnover

The HnLM for age was the same as sex except that the former was substituted for the
latter at both the individual level and the restaurant level. Age was continuously scaled to
years by dividing the age in days by 365. The results of this analysis are summarized in
Table 30. As can be seen in the table, all three interaction effects were significant.
Predicted turnover hazards were calculated at approximately the following values:
Average age (19 years old), 10tb percentile (16 years old) and 90"1 percentile (31 years
old). These predicted values are shown graphically in Figure 24, with the underlying data
presented in Table 31. As can be seen in the figure, at P1 a 31-year-old employee has a
predicted turnover hazard of roughly 15% in a store with an average age of 16, whereas
the predicted hazard for a 16-year-old employee in a store with the same average age is

slightly less than 12%. Conversely, in a store whose employees average 31-years old, the

119

pattern is reversed. This supports Hypothesis 17 because age similarity is related to a
reduced turnover hazard.

As with the previous two demographic variables supplemental analyses were
conducted where additional linear interaction terms with time were added. The
coefficients for this analysis are presented in Table 32; predicted values across time are
found in Table 33, which are depicted graphically in Figure 25. The data represented in
the figure show that the curve with the steepest slope is that of a 31-year old in a store
with an average age of 16 years old. Similarly, the differences in the predicted hazard
values for a store with an average age of 31 are largest earlier on. This supports the
notion that these demographic effects are stronger earlier on in employees’ tenure. This
analysis, along with those presented earlier, support the assertion of Hypothesis 23 in that
demographic similarity had a stronger effect on turnover hazard at lower tenure levels.

The Effect of Extraversion on the Demographic Similarity-Tumover Relationship

Hypothesis 24 stated that extraverts should be less susceptible to demographic
similarity effects because this trait should facilitate making judgments based on deeper
aspects of similarity rather than demographic similarity. For these analyses, the N was
48,440 (censored N = 10,211). These analyses were conducted using the HnLMs
described earlier except that extraversion was added at the individual level as a random
effect, while the following interaction effects were also added at the individual level as
fixed to accommodate their time-varying components: Extraversion X demographics;
extraversion X demographic composition; and extraversion X demographics X
demographic composition. The significance of this last interaction term is the focus of

this hypothesis. Like the results described earlier, separate analyses were conducted for

120

each of the three types of demographic similarity studied here to keep the number of
variables in each analysis manageable. These results are summarized in Tables 34, 35,
and 36 for age, race, and sex, respectively. As can be seen at the bottom of Table, 34 the
age X age composition X extraversion interaction was not significant. Thus, this
hypothesis was not supported for age. On the other hand, Table 35 indicates that
Hispanic X race composition X extraversion interactions were statistically significant. A
plot of the predicted hazards associated with two different levels of extraversion (3 and
5), two different races (Black and Hispanic), and two different restaurant racial
compositions is presented in Figure 26, with the accompanying data presented in Table
37. As can be seen in the figure and the accompanying table, higher levels of
extraversion were associated with an increased turnover hazard when in a restaurant
racial composition inconsistent with one’s own race. Thus, this hypothesis was not
supported for race. As can be seen in Table 36, the sex X sex composition X
extraversion interaction was not significant. Thus, hypothesis 24 was not supported for
any of the demographic variables.
DISCUSSION

The research reported here investigated several key relationships: Between mean
levels of personality and ability test scores and demographic diversity and financial
performance; between personality-based fit and turnover; and between demographic-
based fit and turnover. These questions were addressed using a very large sample of
applicant test scores and demographics from over 2,000 quick service restaurants across
the country. The results, summarized in Table 38 and more theoretically in Figure 27,

yielded evidence of a positive association between ability and team financial

121

performance. In contrast, the personality variables were not related to team financial
performance in the predicted direction, and neither the personality variables nor ability
had any meaningful relationship with turnover. The results concerning demographic fit
and diversity generally supported the hypotheses in that evidence of relationships to team
financial performance, team turnover rates, and individual turnover were found. The
following sections describe these results in more detail, and discuss key practical,
methodological, and theoretical implications of each.
Team Ability and Personality Composition and Team Performance

The hypothesized relationship between mean levels of team personality and ability is
depicted in Figure 2. In short, mean levels of ability and conscientiousness were
hypothesized to be positively related to team performance, whereas extraversion was
hypothesized to have an inverted-U relationship to performance. Team performance was
operationalized as a change in controllable profit for the restaurant in a given month. The
only support for any of these hypotheses was with regard to ability, which was positively
though weakly related to this change in profitability. Given the vast literature supporting
the notion that ability is related to individual level job performance, this result was fully
expected. What was surprising, however, was that this relationship was relatively weak,
only accounting for a change of perhaps several thousand dollars per restaurant over the
course of a year.

There are several possible explanations for this result. First, because of the large
amount of missing data (i.e., the overall low proportion of employees who had test
scores), the average level of ability might have been an unreliable indicator of the true

level of ability within a store. The data were not consistent with this explanation,

122

however, because test saturation did not moderate the ability — performance relationship.
If this explanation was in fact true, one would expect to find a stronger relationship in
stores with higher test saturations. Second, the ability measure was quite short, which led
to a somewhat less reliable indicator of ability than might be desired. This explanation
seems unlikely as well, however, because aggregating the ability scores of people within
a store substantially improves the reliability of the measure. Another explanation seems
more likely. In particular, these jobs may have been too simple to observe much of a
benefit associated with a higher mean level of ability. Given the strongly supported
notion that the ability-performance relationship declines as jobs become less complex,
this explanation seems very plausible. The low complexity of these jobs is exemplified
by the substantial amount of resources that are directed towards making fast-food jobs as
simple as possible. Thus, future research should examine this issue in more complex
jobs. In addition, it would be helpful to conduct studies in which measures of individual
job performance are available to see whether this relationship is at least in part mediated
by average levels of individual performance. This would address the issue of whether
having a high ability team merely enhances team performance via individual inputs, or
whether there is an additional compositional effect that accentuates the translation of
individual into team performance. Studies such as this would be especially helpful
because it is important to demonstrate the impact of personnel selection at higher levels
of analysis. With regard to the specific results obtained here, the fact that the standardized
incremental relationship between ability and financial performance was only .03 suggests
a relatively weak overall relationship between individual job performance and restaurant

performance. If this is in fact the case one has to seriously question the usefulness of

123

ability in contexts such as this, especially given the substantial adverse impact that will
invariably result.

Given the low level of job complexity studied here, motivational variables would be
expected to be more important than ability variables. Surprisingly, however, mean levels
of conscientiousness, the motivational variable that has received the most attention in
selection research over the last 10 years, were not related to the change in controllable
profit in the month-long period under study. The most obvious potential explanation for
this null result is that individuals’ conscientiousness scores were rendered somewhat
meaningless by faking. This is consistent with the highly elevated mean and relative lack
of variance in these scores. The scores were distorted both in comparison to the student
sample that had little motivation to fake, as well as in comparison to many incumbent
samples using similar items developed by the same consulting firm that developed the
test used here. Given that the conscientiousness scale exhibited strong construct validity
based on the correlations with the IPI items, one is left either with this conclusion or the
alternative that conscientiousness simply doesn’t matter at the team level. This latter
explanation seems difficult to justify, as it is easy to think of numerous ways in which
careless employees can reduce the profitability of a restaurant. Of course this explanation
is in conflict with predictive designs that report significant associations between
conscientiousness and job performance, but few other explanations seem plausible. On
the other hand, it is important to note that the few studies cited in the introduction that
have found mean levels of conscientiousness relate to team performance all used
concurrent rather than predictive designs. Last, as with ability, it is also important to note

that because the test saturation X conscientiousness interaction still did not yield a

124

relationship to controllable profit, it is difficult to ascribe these null results to missing
conscientiousness data at the individual level.

The same results were found for agreeableness and extraversion. These results were
contrary to the hypotheses, which were formulated based on empirical studies on teams,
and on meta analyses of research examining the role of these variables in customer
service or team-based work systems. Although there is certainly less evidence supporting
the validity of these constructs for the prediction of job performance at the individual
level as compared to conscientiousness, the existing research and theory strongly
supports the notion that they should relate to the outcomes studied here. The discussion
above concerning conscientiousness seems applicable here as well, in that the scores on
these scales showed strong evidence of faking vis-a-vis the means and standard
deviations reported in Table 7. Additional support for this notion can be found in that the
intercorrelations at both the restaurant and individual applicant level suggest a single
personality factor, whereas the individual level student sample correlations suggest
distinct personality factors. These results provide especially strong support for the notion
that at least some faking is going on because the student sample is actually much more
likely to be representative of the applicant sample in this job as compared to other jobs.
Similarly, it should be noted that the meta analyses cited in the introduction do not clearly
indicate whether the included studies are predictive or concurrent, so it is difficult to
assess whether the results relied upon would be expected to transfer to applicant settings.
The likelihood of and potential implications associated with faking have received

extensive attention elsewhere, and so they will not be reiterated here; however, the results

125

reported with the very large sample studied here clearly imply that this is indeed an
important problem.
Relationships Between Ability and Personality and Turnover

Figure 3, link I, shows that in general it was expected that there would be a
relationship between personality and ability and turnover at the individual level. There
was no support found for any of these hypotheses in that test scores were unrelated to
turnover probability. This is despite that fact that time to turnover was studied using
survival analysis rather than treating it is a dichotomous variable, and that this was a very
high turnover job, presumably allowing for ample variability in this measure. Similarly,
a distinction was made between voluntary and involuntary turnover, and also between -
avoidable and unavoidable turnover. It was thought that examining turnover at a granular
level would increase the likelihood that significant predictors would be identified, though
again, there were still null results.

Based on the person-job fit literature, it was thought that ability would be positively
related to turnover probabilities. This is because the jobs studied here are extremely
simple, suggesting that high ability employees would find a mismatch between their own
level of ability and that required on the job. The level of the ability items suggest that
one possible explanation for a lack of a relationship is that these items might exhibit
ceiling effects. Indeed, the consulting firm that created the items refers to them as
“firndamental skills” items because they are aimed at assessing a very low baseline of
cognitive skills. The test score distribution, however, provided very little evidence of
ceiling effects. This suggests that either there is a true null effect in these data or that

perhaps the ability items are seriously lacking in construct validity. Given the large

126

sample size, even a somewhat construct-deficient measure should have likely yielded at
least some detectable relationship to turnover if one in fact existed. Thus, it seems more
likely that the ability-turnover relationship is extremely small or nonexistent in these
data. Perhaps these jobs were so simple that it is plainly obvious to any high ability
employees that they do not fit in with the cognitive demands of the job, but that these
employees are taking the position anyway due to financial hardship that does not allow
them to turnover quickly.

Similarly, there was also no evidence that conscientiousness was related to any type
of turnover. This was true both at the individual level, and in terms of conscientiousness
diversity at the group level. As mentioned in the introduction, there is relatively little ~
research addressing this issue. It seems as if holding onto even menial jobs would be
associated with key aspects of conscientiousness as it is currently understood. Again,
given the very large sample size, appropriate analytic techniques, and construct validity
evidence on based on the research conducted in the student sample, it seems either that
the null hypothesis is true or that responses on the conscientiousness scale were so
distorted that it is impossible to find any meaningful correlates of these scores.

These results also make it difficult to draw any conclusions about the usefulness of
the involuntary and (avoidable and unavoidable) voluntary turnover. This is because null
relationships were found between the ability and the personality variables and all types of
turnover. While these distinctions are seen as important to some researchers, others have
criticized them for a number of reasons. Indeed, it seems as if one may be able to
generate plausible hypotheses that apply to all types of turnover, rendering useless the

distinctions that were attempted here. For instance, it is easy to see how a motivational

127

variable might lead one to turnover voluntarily due to dissatisfaction on the job, but also
to be terminated (i.e., to turnover involuntarily) because of a lack of motivation and thus
poor job performance.

Effects of Demographic Diversity on Team Performance and Team Turnover Rates

The research reported here tested several hypotheses concerning the effects of team
demographic diversity on team performance and turnover rates. The hypotheses
concerning turnover were supported for all three demographic variables in that higher
diversity was also associated with higher turnover rates. This is consistent with the
relational demography literature discussed earlier in the paper. Interestingly, no LGC
analyses of the trend in turnover rates were possible due to the lack of identifiable
dimensionality underlying the turnover rates over time. This lack of stability over time
calls into question the nature of turnover as a construct and is broadly consistent with the
large body of literature that suggests its etiology is complex and multifaceted. Thus,
while having obvious practical importance, this finding highlights the challenges
researchers may have in more theoretically-oriented studies of turnover rates.

The LGC analyses found that racial diversity was negatively associated with changes
in the trend in controllable profit over time. The fact that the racial diversity intercept
was positively associated with the controllable profit intercept suggests that the former
finding cannot be easily discounted by the explanation that some other third variable
leads to poor profitability. In fact, the LGC analyses control for the influence of the
controllable profit intercept on the relationship between racial diversity intercept and the

controllable profit slope. This means that it is also difficult to dismiss these findings with

128

the argument that racially diverse restaurants show declines in profitability simply
because (for some unknown reason) they start off so high to begin with.

These findings have a number of important implications. First, they are consistent
with relational demography theory, which states that increased demographic diversity
should be associated with negative outcomes at work. This clearly provides some strong
evidence for the relevance of relational demography theory to organizational science,
moving beyond studies that examine “softer” criteria such as communication or
attachment to an organization. It is important to note, however, that these results were
not obtained for sex or age diversity. Due to the sample size and nature of the measures
used here, these null results cannot be easily ascribed to a lack or power or traditional .
conceptions of measurement error. On the other hand, it is likely that there is some noise
in these indices of diversity because all the employees’ demographics were equally
weighted, regardless of whether the employee worked only a few hours a week as
compared to 30, for instance. Taking these results at face value, however, suggests two
important implications. First, it suggests that relational demography theory may be too
broad and that perhaps certain types of demographic diversity may be salient in different
situations. Second, it suggests that racial diversity is associated with something that is
much more powerful and organizationally relevant as compared to the two other types of
diversity studied here. Though it is easy to ascribe this to the overall relevance of race to
everyday life, future research should seek to identify why this might be the case.

This is the first study of which I am aware that has successfully linked racial diversity
to organizational profitability. If this finding can be replicated in other settings it

presents yet another quandary for organizations that seek to become more racially

129

diverse. This is underscored by the fact that the measures of racial diversity, like the
other measures in this study, underestimate the true effect Sizes due to missing
information about how many hours each employee worked at a given restaurant, and
more importantly, which employees actually worked together. Thus, while to some it
may seem politically or ethically correct to seek racially diverse workforces, results such
as these suggest that this is likely to negatively impact an organization’s bottom line.
Organizational decision makers are thus faced with a choice between two desirable goals
that are mutually exclusive to some extent. It is important to emphasize, however, that
these results speak to the impact of racial diversity per se, rather than representation of a
particular racial group.
Effects of Demographic Similarity on Turnover

The relational demography literature reviewed in the introduction suggested that
demographic similarity would be associated with a reduced likelihood of turnover. There
was strong support for this set of hypotheses for all three demographic variables studied
here — age, race, and sex. In particular, being similar to the restaurant’s composition on
each of these variables was associated with a lower likelihood of turnover. Like the
results discussed earlier, this supports the notion that demographic similarity is associated
with important outcomes at work. In contrast to the results obtained for turnover rates at
the restaurant level of analysis, these results were obtained using a cross-level approach.
These results taken together support the notion that relational demography and fit
research more broadly can be construed either as a cross-level or higher-level diversity
effect, depending on the outcome of interest. The multilevel approach to assessing these

issues as cross-level effects represents a useful yet somewhat novel approach to this

130

issue, though other researchers have recently explored approaches that bear some
similarities to the one used here (Davison, Kwak, Seok Seo, & Choi, 2002; Reise, 2000).
It should be noted that other fit studies may not necessarily find the same results using
both cross-level and diversity approaches with other variables; however, a careful
consideration of theory, research designs, and levels of analysis issues should assist
researchers in predicting when these two approaches would be expected to yield different
results.

The results also indicated that demographic similarity is related to higher turnover
hazards earlier in an employee’s tenure, and that this effect fades, and in some cases,
reverses direction over time. This is support for a potentially key qualifierjof relational
demography theory — that demo graphic similarity is perhaps most important at the initial
stages on the job. The theoretical implication of this finding is consistent with the
literature from social psychology discussed earlier suggesting that people make salient
categorizations based on the information available to them at the time. Practically, this
suggests that during the first few weeks or months on the job employees might be
considered to be at the highest risk for demographic-similarity related turnover and that
interventions might be directed towards this time frame.

In contrast, the hypotheses concerning the extraversion X demographic similarity
interaction on turnover were not supported. Based on several studies reviewed in the
introduction, it was expected that extraverts would be at a lower risk for turnover as a
firnction of demographic Similarity. These null results are somewhat surprising in light of
those concerning time in that if the mechanism underlying the influence of time is getting

to know one’s coworkers, being high on extraversion should accelerate this process.

131

Thus, either this may not be the underlying mechanism, or again, perhaps some faking
occurred that distorted the meaning of applicants’ extraversion scores. Future research
should be directed at clarifying this issue.
Methodological Implications for the Dynamic Studies of Teams

As was mentioned in the introduction, a number of authors have called for research
that takes into account the dynamic nature of teams. Team research poses serious
challenges as compared to individual level research because the composition of teams
changes over time. This was especially true given the high rate of turnover in the jobs
studied here. This presented serious challenges because there are no obvious approaches
for handling situations such as these. Two of the analytic techniques used here are
uniquely well suited toward handling this problem, though to my knowledge neither has
been applied in published teams research in this way. First, the LGC methodology
allowed the changes in demographic diversity to be easily captured and then related to
changes in controllable profit over time. This approach was especially valuable because
the zero-order correlations between the monthly racial diversity index and controllable
profit were actually positive. However, when analyzed using the LGC approach a much
different pattern emerged, underscoring the complexity involved in studying the
correlates of change over time. Similarly, because store-level outcomes such as
profitability are likely subject to a range of forces beyond the immediate control of the
store employees, studying changes over time undercuts most arguments that the obtained
results are due to a third, uncontrolled variable that varies widely across stores. Thus, the

usefulness of this approach in studies of dynamic team composition is clear, and

132

researchers should strongly consider this approach when studying dynamic team
processes.

Second, the multilevel survival analyses captured the dynamic nature of team
demographic composition by taking advantage of the fact that survival analysis can easily
handle time-varying covariates. The multilevel implementation of this approach is
somewhat more complex than traditional multilevel analyses but can be accommodated
in standard multilevel software such as MLwin, HLM, and SAS proc NLMIXED.
Researchers studying the effects of dynamic team compositions on dichotomous
outcomes should strongly consider the benefits of this approach.

Although this study applied new approaches to novel problems in I/O psychology, -
there also remains a key statistical issue that would benefit from additional research.
This issue pertains to the unknown impact of the restriction of range that occurred in the
individual level personality scores. There appears to be no published research on how
this is expected to impact the magnitude of the correlations observed in designs like the
one used here. Generating some data indicated that the effects are likely to be very
complex because of the wide variability in restaurant size and selection ratios used in
each restaurant. Thus, this issue is ripe for simulation or analytic studies. Specifically, it
would be helpful to know how this range restriction impacts studies such as this as a
function of the number level 2 units as well as the number of cases in each one, the
strength of the true correlation between the aggregated variable and the other level 2
variable, and the magnitude and variability of the range restriction. This would be

especially insightful because studies with these designs are likely to become increasingly

133

common in personnel selection, where restriction of range is often an issue in both
predictive and concurrent designs.
Theoretical and Methodological Implications for Relational Demography and PO Fit

One of the major themes throughout this paper was the emphasis on looking at the
effects of team composition on both individual and team-level outcomes. Although
similar results were obtained for the cross- and team-level analyses, the latter were only
evident when turnover rates were aggregated over a considerable time period. This
highlights the complexities involved in conducting multi-level longitudinal research, in
that different results can be obtained depending on the time scale and level of analysis.
This suggests that researchers should thoroughly evaluate the soundness of their choices
along these lines.

The results reported here also have implications for the way in which similarity and
PO fit are related. In particular, this research showed how similarity can be
conceptualized as fit between an individual and a composition in which that individual is
embedded. HnLMs were then used to see if the consistency between that composition
and the individuals’ attributes predicted an important outcome of interest. Taking similar
approaches would again help to clarify the confusion surrounding levels of analysis in the
relational demography literature, and also would provide researchers an altemative to the
complexities involved in polynomial regression analysis advocated by Edwards (e. g.,
2002), often the approach of choice for P0 fit studies. In addition, this research
demonstrated how fit can also be construed as a higher-level diversity effect. This

approach should be pursued in other studies in that enables PO fit and other types of fit

134

researchers to broaden the nomological network of variables surrounding the notion of
fit.
Conclusion

The research reported here found evidence that diversity is negatively related to
organizational and individual level outcomes. Because diversity is mandatory rather than
optional in most organizations, future research should investigate what can be done to
ameliorate these effects. First, research that measures the variables that are thought to
connect diversity with these outcomes would be helpful. Second, examining the efficacy
of programs or interventions targeting these intervening variables would be helpful.
Because the relationships were found to Operate at multiple levels of analysis researchers
and practitioners should consider leverage points at both levels as well. Not only will this
further our knowledge about the particular phenomena studied here, but it will also
enhance our ability to conceptualize and operationalize multilevel studies of complex

human behavior at work.

135

REFERENCES

Ackerrnan, P.L. & Humphreys, LG. (1990). Individual differences in industrial and
organizational psychology. In M.D. Dunnette & L.M. Hough (Eds), Handbook of
industrial and organizational psychology, 2nd edition (Vol 1, pp. 223-282).

 

Agresti, A. (1996). An introduction to categorical data analysis. New York: Wiley.

Allison, P. D. (1982). Discrete-time methods for the analysis of event histories. In S.
Leinhardt (Ed.). Sociological Methodology (pp. 61-98). Jossey Bass: San Francisco.

Allison, P. D. (1984). Event history analysis: Regression for longitudinal event data.
Beverly Hills, CA: Sage.

Ancona, D. & Caldwell, D. (1992). Demography and design: Predictors of new product
team performance. Organization Science, 3, 321-341.

Ancona, D. G., & Caldwell, D. F. (1998). Rethinking team composition fiom the outside
in, GruenfeldLDeborah H (Ed). (1998). Composition. Resgrch on managing groups and
teams Vol. 1. (pp. 21-37). Stamford, CT, US: JAI Press, Inc.

 

 

Antonioni, D., & Park, H. (2001). The effects of personality similarity on peer ratings of
contextual work behaviors. Personnel Psychology. 54(2), 331-360.

Arrow, H. (1998). Standing out and fitting in: Composition effects on newcomer
socialization. In Gruenfeld, D. H. (Ed.), Research on Managing Groups and Teams:
Composition (pp. 59-80). Stamford, CT: JAI Press.

Arthur, W., Tubre, T., & Hanson, A. (2000). g—based optimal pairing strategies in a team
training protocol. Paper presented at the Annual Society for Industrial/Organizational
Psychology, New Orleans, LA.

Arvey, R. D. & Faley, R. H. (1984). F‘aimess in Selecting Employees. Reading Mass:
Addison-Wesley.

Ashkanasy, N. M. & O’Connor, C. (1997). Value congruence in leader-member
exchange. Journal of Social Psychology, 137, 647-662.

Austin, J. T., & Villanova, P. (1992). The criterion problem: 1917-1992. Joumal of
Applied Psychology, 77(6), 836-874.

Barber, J. S., Murphy, S. A., Axinn, W. A., & Maples, J. (2001). Discrete-time
multilevel hazard analysis. Sociological Methodology. 31. 201- 235.

Baron, R. A. (1996). Interpersonal relations in organizations. In K. R. Murphy (Ed.).
Individual Differencesfiand Behavior in mgmizamons. (pp. 334-370). San Francisco:
J ossey-Bass.

136

Banick, M. R., & Mount, M. K. (1996). Effects of impression management and self-

deception on the predictive validity of personality constructs. Journal of Applied
Psychology, 81, 261-272.

Barrick, M. R., & Mount, M. K. (1996). Effects of impression management and self-

deception on the predictive validity of personality constructs. Journal of Applied
Psychology, 81(3), 261-272.

Barrick, M. R., Mount, M. K., & Judge, T. A. (2001). Personality and performance at the
beginning of the new millennium: What do we know and where do we go next.
International Journal of Selection and Assessment, 9(1-2), 9-30

Barrick, M. R., Stewart, G. L., & Piotrowski, M. (2002). Personality and job
performance: Test of the mediating effects of motivation among sales representatives.
Journal of Applied Psychology, Applied-Psychology.

Barrick, M. R., Stewart, G. L., Neubert, M. J ., & Mount, M. K. (1998). Relating member
ability and personality to work-team processes and team effectiveness. Journal of Applied
Psycholrgyfil 377-391.

 

Barrick, M.R. & Mount, MK. (1991). The Big Five personality dimensions and job
performance: A meta-analysis. Personnel Psychology, 44, 1-26.

Barry, 8., & Stewart, G. L. (1997). Composition, process, and performance in self-
managed groups: The role of personality. Journal of Aprfied Psycholgy, 82(1), 62-78.

Barsade, S. G., Ward, A. J ., Turner, J. D. F., & Sonnenfeld, J. A. (2000). To your heart's
content: A model of affective diversity in top management teams. Administrative Science
Quarterly. 45(4), 802-836.

 

Baugh, S. G., & Graen, G. B. (1997). Effects of team gender and racial composition on

perceptions of team performance in cross-firnctional teams. Group and Organization
Management, 22(3), 366-383.

Bedeian, A. G., Kemery, E. R., & Mossholder, K. W. (1989). Testing for cross-level
interactions: An empirical demonstration. Behavioral Science, 34(1), 70-78.

Bentler, P. M. & Bonnett, D. G. (1980). Significance tests and goodness of ,fit in the
analysis of covariance structures. Psychological Bulletin, 88, 588-606.

Berscheid, E. & Walster, H. (1978). Integersonal Attraction. Reading, MA: Addison-
Wesley.

Bettenhausen, K. L. (1991). Five years of groups research: What we have learned and
what needs to be addressed. Jourpal of Managementfl, 345-381

Blau, P. (1977). Inequality and Heterogeneity. New York: Free Press.

Bliese, P. D. (2000). Within-group agreement, non-independence, and reliability:
Implications for data aggregation and analysis, K. J. Klein & S. W. J Kozlowski (Eds).

137

(2000). Multilevel theory, research, and methods in organizations: Foundations
extensionsLand new directions. (pp. 349-381). San Francisco, CA, US: Jossey Bass Inc,
Publishers.

 

Bliese, P. D. (2002). Multilevel random coefficient modeling in organizational research:
Examples using SAS and S-PLUS. In F. Drasgow & N. Schmitt (Eds.), Measurinwd
Analyzing Behavior in Organizations: Advances in Measurement and Data Analysis. San
Fransisco: Jossey-Bass.

 

Block, J. (1995). A contrarian view of the five-factor approach to personality description.
Psychological Bulletin, 117, 187-215

Bond, M. H., & Shiu, W. Y. F. (1997). The relationship between a group‘s personality

resources and the two dimensions of its group process. Small Group Research, 28(2),
194-217.

Borman, W. C., & Motowidlo, S. J. (1997). Task performance and contextual

performance: The meaning for personnel selection research. Human Performance, 10(2),
99-109.

Borman, W.C. & Motowidlo, SJ. (1993). Expanding the criterion domain to include
elements of contextual performance. In N. Schmitt, W.C. Borman & Associates (Eds),
Personnel selection in organizations (pp. 71-98). San Francisco: Jossey-Bass Publishers.

Bowers, C. A., Jentsch, F., Salas, E., & Braun, C. C. (1998). Analyzing communication
sequences for team training needs assessment. Human Factors, Human-Factors.

Bretz, R. D., & Judge, T. A. (1994). Person-organization fit and the Theory of Work
Adjustment: Implications for satisfaction, tenure, and career success. Journal of
Vocational Behavior, 44(1), 32-54

Brockner, J ., & et al. (1986). Layoffs, equity theory, and work performance: Further
evidence of the impact of survivor guilt. Academy of Management Journal, 29(2), 373-
384.

Brockner, J ., Grover, S. L., & Blonder, M. D. (1988). Predictors of survivors' job
involvement following layoffs: A field study. Journal of Applied Psychology, 73(3), 436-
442.

Brockner, J ., Grover, 8., Reed, T., DeWitt, R., & et al. (1987). Survivors' reactions to
layoffs: We get by with a little help for our fiiends. Administrative Science Quarterly
32(4), 526-541.

Brockner, J ., Konovsky, M., Cooper Schneider, R., F olger, R., & et al. (1994). Interactive
effects of procedural justice and outcome negativity on victims and survivors of job loss.
fldemy of Management Journal, 37(2), 397-409.

Bryk, A. S., & Raudenbush, S. W. (1992). Hierarchical linear models: Application; and
data analysis methods. Thousand Oaks, CA: Sage.

138

Bryk, A., Raudenbush, S., & Congdon, R. (1996). Hierarchical linear and nonlinear
modeling with the HLM/2L and HLM/3L Programs. Chicago: SSI.

Bryne, D. E. (1971). The Attraction Paradigm. New York: Academic Press.

Cameron, K. S. (1998). Strategic downsizing: An extreme case. In L. L. Cummings &
B. M. Staw (Eds.), Research in Organizational Behavior. Volume 20 (pp. 185-229).
Greenwich, CT: JAI Press.

Campbell, J. P., Blake Glasser, M., & Oswald, F. L. (1996). The substantive nature of
job performance variability. In K. R. Murphy (Ed.). Individual Differences and Behavior
in Organizations. (pp. 258-299). San Francisco: Jossey-Bass.

Campbell, J .P., McCloy, R.A., Oppler, S.H. & Sager, CE. (1993). A theory of
performance. In N. Schmitt, W.C. Borman & Associates (Eds.), Personnel selection in
organizations (pp. 35-70). San Francisco: Jossey-Bass Publishers.

Campion, M. A., Medsker, G. J ., & Higgs, A. C. (1993). Relations between work group
characteristics and effectiveness: Implications for designing effective work groups.
Personnel PsychongL 46(4), 823-850.

Canon-Bowers, J. A., Tannenbaum, S. I., Salas, E., & Volpe, C. E. (1992). Defining
competencies and establishing team training requirements. In R. R. Guzzo & E. Salas
and Associates (Eds.), Team Effectiveness and Decision Maflg in Organizatioaa. (pp.
333-380). San Fransisco: Jossey-Bass.

Carroll, 1.3. (1993). Human cognitive abilities: A survey of factor;analytic studies.
New York: Cambridge University Press.

Chan, D. (1998). The conceptualization and analysis of change over time: An integrative
approach incporating longitudinal means and covariance structures analyis (LMACS) and
multiple indicator latent growth modelling (MLGM). Organizational Research Methods.
1, 421-483.

Chan, D. (1998). Functional relations among constructs in the same content domain at
different levels of analysis: A typology of composition models. Journal of Applied

Psychology, 83, 234-246.

Chan, D., Ramey, S., Ramey, C., & Schmitt, N. (2000). Modeling intraindividual
changes in children’s social skills at home and at school: A multivariate latent growth
approach to understanding between-settings differences in children’s social skills
development. Multivariate Behavioral ResearchL35, 365-396.

Chatrnan, J. A. (1991). Matching people and organizations: Selection and socialization in

public accounting firms. Administrative Science Qaarterly, 36(3), 459-484.

Chatrnan, J. A., & Flynn, F. J. (2001). The influence of demographic heterogeneity on the
emergence and consequences of cooperative norms in work teams. Academy of
Management Journal, 44(5), 956-974.

139

Chatrnan, J. A., Polzer, J. T., Barsade, S. G., & Neale, M. A. (1998). Being different yet
feeling similar: The influence of demographic composition and organizational culture on
work processes and outcomes. Administrative Science Quarterly, 43(4), 749-780.

Chattopadhyay, P. (1999). Beyond direct and symmetrical effects: The influence of
demographic dissimilarity on organizational citizenship behavior. Academy of
Management Journal,42, 273-287.

Clevenger, J., Pereira, G. M., Wiechmann, D., Schmitt, N., & Harvey, V. S. (2001).
Incremental validity of situational judgment tests. Journal of Applied Psychology, 86(3),
410-417.

Cohen, S. G., & Bailey, D. E. (1997). What makes teams work: Group effectiveness
research from the shop floor to the executive suite. Journal of Management, 23, 239-290

 

Conway, J. M. (1996). Additional construct validity evidence for the task/contextual
performance distinction. Human Performance. 9(4), 309-329.

Conway, J. M. (1999). Distinguishing contextual performance fiom task performance for
managerial jobs. Journal of Applied Psychology, 84(1), 3-13.

Costa, P.T, Jr. & McCrae, RR. (1988). Personality in adulthood: A six-year
longitudinal study of self-reports and spouse ratings on the NBC Personality Inventory.
Journal of Personality and Social Psychology, 54, 853-863.

Costa, P.T, Jr. & McCrae, RR. (1990). Personafityigdulthood. New York, NY, US:
The Guilford Press

Costa, P.T. & McCrae, RR. (1995). Domains and facts: Hierarchical personality
assessment using the Revised NEO Personality Inventory. Journal of Personality
Assessmenapi, 21-50.

 

Cox, D. R. (1972). Regression models and life tables. Journal of the Royal Statistical
Society, 34, 187-220.

Davison, M. L., Kwak, N., Seok Seo, Y., & Choi, J. (2002). Using Hierarchical Linear
Models to Examine Moderator Effects: Person-by-Organization Interactions.
Organizational Research Methods, 5, 231-254.

Day, D. V., & Bedeian, A. G. (1995). Personality similarity and work-related outcomes
among African-American nursing personnel: A test of the supplementary model of
person environment congruence. Jourgl of Vocational Behavior. 46(1), 55-70.

De Corte, W. (1999). Weighing job performance predictors to both maximize the quality
of the selected workforce and control the level of adverse impact. Journal of Applied
Psychologva84(5), 695-702.

 

Devine, D. J. (1999). Effects of cognitive ability, task knowledge, information sharing,
and conflict on group decision-making effectiveness. Small Group Research, 30, 608-634

140

Digman, J. M. (1990). Personality structure: Emergence of the five-factor model. Annual
Review of Psychology, 41 , 417-440.

Di grnan, J .M. (1989). Five robust trait dimensions: Development, stability and utility.
Journal of Personality, 57, 195-214.

Di grnan, J .M. & Inouye, J. (1986). Further specification of the five robust factors or
personality. Journal of Personality and Social Psychology. 50, 116-123.

Dipboye, R. L. (1994). Structured and unstructured selection interviews: Beyond the job-
fit model. In G. Ferris (ed.), Research in Personnel and Human Resources Management
12, Greenwich, CT: JAI Press, pp. 79-123.

Driskell, J. E. & Salas, E. (1992). Can you study real teams in contrived settings? The
value of small group research to understanding teams. In R. W. Swezey & E. Salas.
Teams: Their Training and Performance. (pp. 101-126). USA: Ablex.

Edwards J. (1994). The study of congruence in organizational behavior research: Critique
and a proposed alternative. Qgganizational Behavior and Human Decision Processes,
5 8,5 1-1 00.

Edwards, J. R. (2002). Alternatives to difference scores: Polynomial regression analysis
and response surface methodology. In F. Drasgow & N. Schmitt (Eds), Measuring and
Analyziag Behavior in Organizations: Advances in Measurement and Data Anafisis. San
Fransisco: Jossey-Bass.

Ellingson, J. E., Sackett, P. R., & Hough, L. M. (1999). Social desirability corrections in
personality measurement: Issues of applicant comparison and construct validity. Journal
of Applied Psychology, 84, 155-166.

Engle, E. M., & Lord, R. G. (1997). Implicit theories, self-schemas, and leader-member
exchange. Academyof Management Journal, 40(4), 988-1010.

Eysenck, H. J. (1947). Dimensions of Personality. New York: Praeger.

F eeley, T. J., & Barnett, G. A. (1997). Predicting employee turnover fi'om
communication networks. Human Communication Research, 23, 370-387

Ferris, G. R., Judge, T. A., Chachere, J. G., & Liden, R. C. (1991). The age context of
perforrnance-evaluation decisions. Psycholggy and Aging, 6(4), 616-622.

Ferris, G. R., Youngblood, S. A., & Yates, V. L. (1985). Personality, training
performance, and withdrawal: A test of the person-group fit hypothesis for organizational
newcomers. Journal of Vocational Behavior, 27(3), 377-388.

F leishman, E. A. & Zaccaro, S. J. (1992). Toward a taxonomy of team performance
functions. In R. W. Swezey & E. Salas. Teams: Their Training and Performance. (pp.
31-56). USA: Ablex.

141

Flynn, L. P. & Shore, L M. (1994). Gender and context: The moderating effect of work
group gender composition and occupation sex-type on work attitudes and behaviors. In
M. Schnake (Ed.), Proceedinga of the Southern Management Association.

Forsyth, D. R. (1999). Group Dynamics. USA: Wadsworth.

Frei, R. L., & McDaniel, M. A. (1998). Validity of customer service measures in

personnel selection: A review of criterion and construct evidence. Human Performance,
fl, 1-27

 

George, J. M. (1990). Personality, affect, and behavior in groups. Journal of Applied
thology, 75(2), 107-116.

George, J. M. (1996). Trait and state affect. In K. R. Murphy (Ed.). Individual
Differences and Behavior in Organizations. (pp. 145-174). San Francisco: Jossey-Bass.

George, J. M., & Bettenhausen, K. (1990). Understanding prosocial behavior, sales
performance, and turnover: A group-level analysis in a service context. Journal of
Applied Psychology. 75(6), 698-709.

Ghiselli, G. E. (1966). The Validity of Occuszional Aptitude Tests. New York: Wiley,

Gladstein, D. (1984). Groups in context: A model of task group effectiveness.
Administrative Science Quarterly, 29, 499-517.

Glick, W. H., & Roberts, K. H. (1984). Hypothesized interdependence, assumed
independence. Academy of Management Review, 9, 722-735

Goldberg, LR. (1990). An alternative “description of personality”: The Big-Five factor
structure. Journal of Personality and Social Psychology, 59, 1216-1229.

Goldberg, LR. (1993). The structure of phenotypic personality traits. American
Pchhologist, 48, 26-34.

Gowan, M. A., Riordan, C. M., & Gatewood, R. D. (1999). Test of a model of coping
with involuntary job loss following a company closing. Journal of Applied Psychology,
MU), 75-86.

Green, S. G., Anderson, S. E., & Shivers, S. L. (1996). Demographic and organizational
influences on leader-member exchange and related work attitudes. Organizational
Behavior and Human Decision Processes, 66(2), 203-214.

Greguras, G. J ., & Robie, C. (1998). A new look at within-source interrater reliability of
360-degree feedback ratings. Journal of Applied Psychology, 83(6), 960-968.

Griffeth, R. & Horn, P. (1994). Employee Turnover. USA: South-Western College
Publishing.

Griffeth, R. W., Hom, P. W., & Gaertner, S. (2000). A meta-analysis of antecedents and
correlates of employee turnover: Update, moderator tests, and research implications for
the next millennium. Journal of Management, 26(3), 463-488.

142

Guastello, S. J., & Guastello, D. D. (1998). Origins of coordination and team
effectiveness: A perspective from game theory and nonlinear dynamics. Journal of
Applied Psychologg 83(3), 423-437.

Guion, R.M. & Gottier, RF (1965). Validity of personality measure in personnel
selection. Personnel Psychology, 18, 135-164.

Gully, S. M., Devine, D. J ., & Whitney, D. J. (1995). A meta-analysis of cohesion and
performance: Effects of levels of analysis and task interdependence. Small Group
Research, 26(4), 497-520.

 

Guzzo, R.A. & Dickson, M.W. (1996). Teams in organizations: Recent research on
performance and effectiveness. Annual Review of Psychology, 46, 307-338.

Guzzo, R.A. & Shea, GP. (1992). Group performance and intergroup relations in
organizations. In M.D. Dunnette and L.M. Hough (Eds), Handbook of Industrial aad
Organizational Psychology (Vol 3, pp. 269-313). Palo Alto, CA: Consulting
Psychologists Press.

Hackman, J. R. (1975). Group tasks, group interaction process, and group performance
effectiveness: A review and proposed integration. In L. Berkowitz (Ed.). Advances in ‘
Experimental Social Psychology. New York: Academic Press.

Hackman, J. R. (1987). The design of work teams. In J. W. Lorsch (Ed.), Handbook of
Organizational Behavior (pp. 315-342). Englewood Cliffs, NJ: Prentice-Hall.

 

Hackman, J. R. (1992). Group influences on individuals in organizations. In Dunnette,
Marvin D. (Ed); Hough, Leaetta M. (Ed). (1992). Handbook of industrial and
opganizational psychology, Vol. 3 (2nd ed.). (pp. 199-267).

Hackman, J .R. & Morris, CG. (1975). Group task, group interaction process, and group
performance effectiveness: A review and proposed integration. In L. Berkowitz (Ed.),
Advances in Experimental Social Psychology, (Vol 8). New York: Academic Press.

Harman, M. T. (1971). Aggregation and Disagggegation in Sociolagy. Lexington, MA:
Health.

Harman, M. T. (1990). Aggregation and Disaggr_egation in the Social Sciences.
Lexington, MA: Lexington Books.

Harrison, D. A. (2002). Structure and timing in limited range dependent variables:
Regression models for predicting if and when. In F. Drasgow & N. Schmitt (Eds).
Measurirgand Analyzirg Behavior in Organizations. J ossey—Bass: San Francisco.

Harrison, D. A. (2002). Structure and timing in limited range dependent variables:
Regression models for predicting if and when. In. F. Drasgow & N. Schmitt (Eds),
Measuring and AnalyzingBehavior in Oganizations: Advances in Measurementana
Data Analysis. San Fransisco: Jossey Bass.

Harrison, D. A., Price, K. H., & Bell, M. P. (1998). Beyond relational demography: Time

143

and the effects of surface- and deep-level diversity on work group cohesion. Academy of
Management Journal, 41(1), 96-107.

Hartigan, J. A., & Wigdor, A. K. (1989). Fairness in employment testing: Validity
generalization, minority issues, and the General Aptitude Test Battery..

Hattrup, K., O' Connell, M. S., & Wingate, P. H. (1998). Prediction of multidimensional
criteria: Distinguishing task and contextual performance. Human Performance, 11, 305-
319

 

Hattrup, K., O' Connell, M. S., & Wingate, P. H. (1998). Prediction of multidimensional
criteria: Distinguishing task and contextual performance. Human Performance, 11(4),
305-319.

Hattrup, K., Rock, J ., & Scalia, C. (1997). The effects of varying conceptualizations of
job performance on adverse impact, minority hiring, and predicted performance. Journal
of Aprfled Psychology, 82, 656-664.

Heck, R. H., & Thomas, S. L. (2000). An introduction to multilevel modeling techniques.
Mahwah, NJ, US: Lawrence Erlbaum Associates, Inc., Publishers.

Hedeker, D. & Gibbons, R. D. (1996).. MIXREG: A computer program for mixed-effects
regression analysis for autocorrelated errors. Computer Methods and Programs in
Biomedicine, 49, 229-252.

Hedeker, D., Siddiqui, O., & Hu, RB. (2000). Random-effects regression analysis of
correlatedgrouped-time survival data. Statistical Methods in Medical Research, 9:161-
179

Herold, D. M .(1978). Improving the performance effectiveness of groups through a
task-congingent selection of intervention strategies. Academy of Management Review,
3, 315-325.

Hinds, P. J ., Carley, K. M., Krackhardt, D., & Wholey, D. (2000). Choosing work group
members: Balancing similarity, competence, and familiarity. Organizational Behavior
and Human Decision Processes, 81(2), 226-251.

 

Hofmann, D. A. (1997). An overview of the logic and rationale of hierarchical linear
models. Journal of Management, 23(6), 723-744.

Hofmann, D. A., & Gavin, M. B. (1998). Centering decisions in hierarchical linear
models: Implications for research in organizations. Jourrgrl of Management, 24, 623-641

Hogan, J ., Rybicki, S. L., Motowidlo, S. J ., & Borman, W. C. (1998). Relations between
contextual performance, personality, and occupational advancement. Human
Performance, 1 1(2-3), 189-207

Hogan, R., & Hogan, J. (1995). Hogan Personality Inventory Manual. Tulsa, OK:
Hogan Assessment Systems.

144

Hogan, R., Raza, S., & Driskell, J. E. (1988). Personality, team performance, and
organizational context. In P. Whitney & Ochsman, R. B. (Eds.), Psychology and
Productivity. New York: Plenum.

Holland, J. L. (1985). Making Vocational Choices: A Theory of Vocmnal Personalities
and Work Environments. (2“d Ed.). Englewood Cliffs, NJ: Prentice-Hall.

Homey, J ., Osgood, D. W., & Marshall, I. H. (1995). Criminal careers in the short-term:
Intra-individual variability crime and its relation to local life circumstances. American
Sociological Review,@, 655-673.

Hough, L. M. & Schneider, R. J. (1996). Personality traits, taxonomies, and applications
in organizations. In K. R. Murphy (Ed.). Individual Differences and Behavior in
Organizations. (pp. 31-88). San Francisco: Jossey-Bass.

Hough, L. M. (1992). The "Big Five" personality variables--construct confusion:
Description versus prediction. Human Performance, 5(1-2), 139-155.

Hough, L. M., Oswald, F. L., & Ployhart, R. E. (2001). Determinants, detection and
amelioration of adverse impact in personnel selection procedures: Issues, evidence and
lessons learned. International Journal of SelectiorLapd Assessment, 9(1-2), 152-194

House, R., Rousseau, D.M., & Thomas-Hunt, M. (1995). The meso paradigm: A
framework for the integration of micro and macro organizational behavior. In L.L.

Cummings & B.Staw (Eds.), Research in organizational behavior (Vol 17, pp. 71-114).
Greenwich, CT: JAI Press.

Hunter, J. E. (1986). Cognitive ability, cognitive aptitude, job knowledge, and job
perforrrrance. Journal of Vocational Behaviorj9, 340-362

Hunter, J .E. & Hunter, RF (1984). Validity and utility of alternative predictors of job
performance. Psychological Bulletin, 96, 72-98.

Hurtz, G. M., & Donovan, J. J. (2000). Personality and job performance: The Big Five
revisited. Journal of Applied Psychology, 85(6), 869-879.

Ilgen, D. R. (1999). Teams embedded in organizations: Some implications. American
Psychologist, 54, 129-139

Ilgen, D. R., Major, D. A., Hollenbeck, J. R., & Sego, D. J. (1995). Raising an Individual
Decision-Making Model to the Team Level: A New Research Model and Paradigm. In
R. A. Guzzo & E. Salas and Associates (Eds.), Team Efi’ectivenessfi and Decision Mafiag
in Organizations. San Fransisco: Jossey-Bass.

Iverson, R. D., & Buttigieg, D. M. (1997). Antecedents of union commitment: The
impact of union membership differences in vertical dyads and work group relationships.
Human Relations, 50(12), 1485-1510.

145

Jackson, S. E. (1992). Team composition in organizational settings: Issues in managing
an increasingly diverse work force. In S. Worchel, W. Wood, & J. A. Simpson (Eds.),
Group Process and Productivity. (pp. 138-177). Newburry Park, CA: Sage.

Jackson, S. E., & Ruderman, M. N. (Eds.). (1995). Diversipyin work teams: Research
paradigms for a changinglopkplace. Washington, DC, US: American Psychological
Association.

Jackson, S. E., Brett, J. F., Sessa, V. 1., Cooper, D. M., & et al. (1991). Some differences
make a difference: Individual dissimilarity and group heterogeneity as correlates of
recruitment, promotions, and turnover. Journal of Applied Psychology, 76(5), 675-689.

Jackson, S. B, May, K. E., & Whitney, K. (1995). Understanding the dynamics of
diversity in decision-making teams. In R. A. Guzzo & E. Salas and Associates (Eds.),
Team Effectiveness and Decision Making in Oganizations. San Francisco: Jossey-Bass.

Jackson, S. E., Stone, V. K., & Alvarez, E. B. (1992). Socialization amidst diversity: The
impact of demographics on work team oldtimers and newcomers. In L. L. Cummings &
B. M. Staw (Eds.), Research in Organizational Behavior. Volume 15 (pp. 45-110).
Greenwich, CT: J AI Press.

Jehn, K. A., Northcrafi, G. B., & Neale, M. A. (1999). Why differences make a
difference: A field study of diversity, conflict, and performance in workgroups.

Administrative Science Quarterly, 44(4), 741-763.
Jensen, AR. (1980). Bias in mental testing, New York: Free Press.

Johns, G. (1981). Difference score measures of organizational behavior variables: A
critique. Organizational Behavior and Human Decision Processes, 2, 443-463.

Johnson, J. W. (2001). The relative importance of task and contextual performance
dimensions to supervisor judgments of overall performance. Journal of Applied
Psychology, 86, 984-996.

J oreskog, K. & Séirbom, D. (1986). LISREL 6: Analysis of linear structural relationships
bmaximum likelihood fld least sguare methods. Chicago: Scientific Software
International.

Jéreskog, K. & Sérbom, D. (1989). LISREL 7: A Guide to the program and applications.
(2nd Edition). Chicago: Scientific Software International.

J6reskog, K. & Séirbom, D. (1993). LISREL 8 user’s reference Ede. Chicago:
Scientific Software International.

Jung, D. I., & Sosik, J. J. (1999). Effects of group characteristics on work group
performance: A longitudinal investigation. Group Dynamics, 3(4), 279-290.

Kennedy, P. (1998). A Guide to Econometrics. USA: MIT Press.

 

Kenny, D. A., & Judd, C. M. (1986). Consequences of violating the independence

I46

assumption in analysis of variance. Psychological Bulletin, 99(3), 422-431.

 

Kenrick, D. T., & Funder, D. C. (1988). Profiting from controversy: Lessons from the
person-situation debate. American Psychologist, 43, 23-34

Kidwell, R. E., Jr., Mossholder, K. W., & Bennett, N. (1997). Cohesiveness and
organizational citizenship behavior: A multilevel analysis using work groups and
individuals. Journal of Management, 23(6), 775-793.

Kirchmeyer, C. (1995). Demographic similarity to the work group: A longitudinal study
of managers at the early career stage. Journal of Organizational Behavior, 16(1), 67-83.

Klein, K. J. & Kozlowski, S. W. J. (2000). Multilevel theoryLresearch, and methods in
prganizations: Foundations, extensions and new directions. San Francisco, CA, US:
Jossey Bass Inc, Publishers

 

Klein, K. J ., Dansereau, F., & Hall, R. J. (1994). Levels issues in theory development,
data collection, and analysis. Academy of Management Review, 19, 195-229.

Klimoski, R. & Jones, R.G. (1995). Staffing for effective group decision making: Key
issues in matching people and teams. In R.A. Guzzo, E. Salas, & Associated (Eds.),
T_eam effectivenegafnd decision mam in organizations (pp. 291-332). San Francisco:
Jossey-Bass.

Kozlowski, S. W. J ., & Klein, K. J. (2000). A multilevel approach to theory and research
in organizations: Contextual, temporal, and emergent processes, K. J. Klein & S. W. J
Kozlowski (Eds). (2000). Multilevel theory,research, and methods in organization;
Foundations, extensions and new direction_s. (pp. 3-90). San Francisco, CA, US: Jossey
Bass Inc, Publishers

 

Kreft, I. & de Leeuw, J. (1998). IntroducingMultilevel Modeling. Thousand Oaks, CA:
Sage.

Kristof Brown, A. L., & Stevens, C. K. (2001). Goal congruence in project teams: Does
the fit between members' personal mastery and performance goals matter? Journal of
Applied Psychology, 86(6), 1083-1095.

Kristof, A. L. (1996). Person-organization fit: An integrative review of its
conceptualizations, measurement, and implications. Personnel Psychology. 49. 1-49

Landy, F. J. & Farr, J. L. (1983). The measurement of work performance. New York:
Academic Press.

Langbein, L. I. & Lichtrnan, A. J. (1978). Ecological Inference. Beverly Hills, CA:
Sage.

 

LaRocco, J. M. (1983). Job attitudes, intentions, and turnover: An analysis of effects
using latent variables. Human Relations, 36(9), 813-825.

Laughlin, P. R., & Adamopoulos, J. (1980). Social combination processes and individual

147

learning for six-person cooperative groups on an intellective task. Journal of Personality
and Social Psychology, 38(6), 941-947.

Leana, C. R. & Feldman, D. C. (1994). The psychology of job loss. In G. R. Ferris
(Ed.). Research in Personnel and Human Resources Management. Volume 12 (pp. 271-
302). New York: JAI Press.

LePine, J. A., & Van Dyne, L. (1998). Predicting voice behavior in work groups. Journal
of Applied Psychology, 83, 853-868.

LePine, J. A., & Van Dyne, L. (2001). Voice and cooperative behavior as contrasting
forms of contextual performance: Evidence of differential relationships with Big Five
personality characteristics and cognitive ability. Journal of Applied Psychology, 86, 326-
336.

Lepine, J. A., Hanson, M. A., Borman, W. C., & Motowidlo, S. J. (2000). Contextual
performance and teamwork: Implications for staffing. In G. R. Ferris (Ed.). Research in
Personnel and Human Resources Management. Volume 19 (pp. 53-90). New York: JAI
Press.

LePine, J. A., Hollenbeck, J. R., Ilgen, D. R., & Hedlund, J. (1997). Effects of individual
differences on the performance of hierarchical decision-making teams: Much more than
g. Journal of Applied Psychology. 82(5), 803-811.

Levine, J. M., & Moreland, R. L. (1990). Progress in small group research. Annual
Review of Psychology, 41, 585-634.

Lofquist, L. H. & Dawis, R. (1969). Adiustrnent To Worlg: A Psychogrgical View Of
Man's Problems In A Work-Oriented Society. New York: Appleton-Century-Crofts.

Markham, S. E. (1988). Pay-for-perforrnance dilemma revisited: Empirical example of
the importance of group effects. Journal of Applied Psychology, 73(2), 172-180.

Mathieu, J. E., Heffner, T. S., Goodwin, G. F ., Salas, E., & Cannon Bowers, J. A. (2000).
The influence of shared mental models on team process and performance. Journal of
Applied Psychology, 85(2), 273-283.

McCain, B. E., 0' Reilly, C., & Pfeffer, J. (1983). The effects of departmental
demography on turnover: The case of a university. Academy of Management Journal,
2__6_(4), 626-641.

McCormick E., DeNisi A., Staw J. (1979). Use of the position analysis questionnaire for
establishing the job component validity of tests. J ourn_al of Applied Psychology, 6_4_, 51-
56.

McCormick E., J eanneret P., Mecham R. (1972). A study of job characteristics and job
dimensions as based on the Position Analysis Questionnaire (PAQ), (Monograph) Journal
oprplied Psychology, 56,347-368.

McCormick, E. J ., DeNisi, A. S., & Shaw, J. B. (1979). Use of the position analysis

148

questionnaire for establishing the job component validity of tests. Journal of Applied
Psychology, Applied-Psychology.

McDaniel, M. A., Whetzel, D. L., Schmidt, F. L., & Maurer, S. D. (1994). The validity of
employment interviews: A comprehensive review and meta-analysis. Journal of Applied
Psychology, 79(4), 599-616.

McElroy, J. C., Morrow, P. C., & Rude, S. N. (2001). Turnover and organizational
performance: A comparative analysis of the effects of voluntary, involuntary, and
reduction in force turnover. Journal of Applied Psycholpgya86, 1294-1299.

 

McFarland, L. A., & Ryan, A. M. (2000). Variance in faking across noncogrritive
measures. Journal of App_lied Psychology, 85, 812-821.

McGrath, J. E. (1984). Groups: Interaction and Performance. Englewood Cliffs, NJ:
Prentice-Hall.

McGrath, J. E. (1998). A view of group composition through a group-theoretic lens. In
D. E. Gruenfeld (Ed.). Research on ManaflgTeams and Groups: Composition (pp. 255-
272). Greenwich, CT: JAI Press.

McGrath, J. E., Berdahl, & Arrow, H. (1996). Traits, expectation, culture, and clout: The
dynamics of diversity in work groups. In S. E. Jackson & M. N. Ruderrnan (Eds.),
Diversity in Work Teaans: Research Paradigms for a Changing Workplace (pp. 17-45).
Washington DC: American Psychological Association.

McPherson, J. M., Popielarz, P. A., & Drobnic, S. (1992). Social networks and
organizational dynamics. American Sociolog’cal Review, 57(2), 153-170.

Meglino B, Ravlin E, Adkins C. (1989). A work values approach to corporate culture: A

field test of the value congruence process and its relationship to individual outcomes.
Journal of Applied Psychology, 74, 424-432.

Meglino B., Ravlin E., Adkins C. (1992). The measurement of work value congruence: A
field study comparison. Journal of Management,l 8, 33--43.

Mellor, S. (1995). Gender composition and gender representation in local unions:
Relationships between women's participation in local office and women's participation in
local activities. Journal of Applied Psycholggy, 80(6), 706-720.

Meredith, W. & Tisak, J. (1990). Latent curve analysis. PsychometrikaLSS, 107-122.

Milliken, F. J. & Martins, L. L. (1996). Searching for common threads: Understanding
the multiple effects of diversity in organizational groups. Academy of Management
Review 21, 402-443.

 

Mischel, W. (1968). Persoglity and assessment. New York: Wiley.

Moreland, R. L., Levine, J. M. & Wingert, M. L. (1992). Creating the ideal group:
Composition effects at work.

149

Morgan, B. B. & Lassiter, D. L. (1992). Team composition and staffing. In R. W.
Swezey & E. Salas. Teams: Their Trainirg and Performance. (pp. 75-100). USA: Ablex.

Morgan, B.B, Glickman, A.S., Woodward, E.A., Blaiwes, A.S., et al. (1986).
Measurement of team behaviors in a Navy training environment. US Naval Training
Systems Center Technical Reports.

Morita, J. G., Lee, T. W., & Mowday, R. T. (1989). Introducing survival analysis to
organizational researchers: A selected application to turnover research. Journal of
Applied Psychology, 74, 280-292.

Morita, J. G., Lee, T. W., & Mowday, R. T. (1993). The regression-analog to survival
analysis: A selected application to turnover research. Academy of Management Journal,
E, 1430-1464

Motowidlo, S. J ., & Van Scotter, J. R. (1994). Evidence that task performance should be
distinguished from contextual performance. Journal of Applied Psychology, 79, 475-480.

Motowidlo, S. J ., Borman, W. C., & Schmit, M. J. (1997). A theory of individual
differences in task and contextual performance. Human Performance, 10, 71-83

Mount, M. K., Barrick, M. R., & Stewart, G. L. (1998). Five-Factor Model of personality
and performance in jobs involving interpersonal interactions. Human Performance, 11(2-
3), 145-165

Mount, M. K., Judge, T. A., Scullen, S. E., Sytsma, M. R., & Hezlett, S. A. (1998). Trait,
rater and level effects in 360-degree performance ratings. Personnel Psycholpgy. 51(3),
557-576.

Mount, M.K. & Barrick, MR. (1995). The Big Five personality dimensions:
Implications for research and practice in human resources management. In G. R. Ferris

(Ed.) Research in PemomeLand Human Resources Management (Volume 13) 153-200
New York: JAI.

Moynihan, L. M. & Peterson, R. S. (2001). A contingent configuration approach to
understanding the role of personality in organizational groups. Research in
Organizational Belmior, 23, 327-378.

Mueller, C. W., Finley, A., Iverson, R. D., & Price, J. L. (1999). The effects of group
racial composition on job satisfaction, organizational comrrritrnent, and career
commitment: The case of teachers. Work and Occupations 26(2), 187-219.

Mumarne, R. J., Singer, J .D., & Willett, J. B. (1989). The career paths of teachers:
Implications for teacher supply and methodological lessons for research. Educational
Researcher, 17, 22-30.

Murphy, K. R. (1996). Individual Differences and Behavior in Organizations: Much
More than g. In K. R. Murphy (Ed.). Individual Differences and Behavior in
Organizations. (pp. 3-30). San Fransisco: Jossey-Bass.

 

 

150

Murphy, K.R. & Shiarella, AH. (1998). Implications of the multidimensional nature of
job performance for the validity of seletion tests: Multivariate frameworks for studying
test validity: Errata. Personnel Psychology, 51, 822.

 

National Research Council (1999). The Changing Nature of Wopk. Washington DC:
National Academy Press.

Neuman, G. A., & Wright, J. (1999). Team effectiveness: Beyond skills and cognitive
ability. Journal of Applied Psychology, 84, 376-389.

Neuman, G. A., Wagner, S. H., & Christiansen, N. D. (1999). The relationship between
work-team personality composition and the job performance of teams. Group and
Organization Management, 24, 28-45

Nezlek, J. B., & Zyzniewski, L. E. (1998). Using hierarchical linear modeling to analyze
grouped data. Group Dmamics, 2(4), 313-320.

Nieva, V. F., Fleishman, E. A., & Rieck, A. M. (1978). Team dimensions: Their identity,
their measuremetn and their relationships. Washington DC: ARRO.

Nunnally, J. (1978). Psychometric Theory (2“d Ed). New York: McGraw Hill.

0' Reilly, C. A., 111, Williams, K. Y., & Barsade, S. (1998). Group demography and
innovation: Does diversity help?, Gruenfeld,Debopah H (Ed). (1998). Composition.
Research on managing groups and teams, Vol. 1. (pp. 183 207). Stamford, CT, US: JAI
Press, Inc.

O’Connor, K. M. (1998). Experiental diversity in groups: Conceptualizing and
measuring variation among teammates. In Gruenfeld, D. H. (Ed.), Research on
Managing Groups and Teams: Composition @p. 167-182) Stamford, CT: JAI Press.

O’Reilly, C. A., Williams, K. Y., & Barsdale, S. (1998). Group demography and
innovation: Does diversity help? In Gruenfeld, D. H. (Ed.), Research on Managing
Groups and Teams: Comosition (pp. 183-207) Stamford, CT: JAI Press.

O'Reilly, C. A., Caldwell, D. F., & Barnett, W. P. (1989). Work group demography,
social integration, and turnover. Administrative Science Quarterly, 34(1), 21-37.

O'Reilly, C. A., Chatrnan, J ., & Caldwell, D. F. (1991). People and organizational
culture: A profile comparison approach to assessing person-organization fit. Academy of
Management Jourfll, 34(3), 487-516.

 

 

Ostroff, C. (1993). Comparing correlations based on individual-level and aggregated
data. Journal of Applied PsychologyL78(4), 569-582.

Ostroff, C., & Bowen, D. E. (2000). Moving HR to a higher level: HR practices and
organizational effectiveness, K. J. Klein & S. W. J Kozlowfid (Eds). (2000). Multilevel

theory, research, and methods in organizations: Foundations, extensions, and new
directions. (pp. 211-266). San Francisco, CA, US: Jossey Bass Inc, Publishers.

151

Pelled, L. H. (1996). Demographic diversity, conflict, and work group outcomes: An
intervening process theory. Organization Science, 7(6), 615-631.

Pelled, L. H. (1996). Relational demography and perceptions of group conflict and
performance: A field investigation. International Journal of Conflict Management, 7(3),
230-246.

Pelled, L. H., Eisenhardt, K. M., & Xin, K. R. (1999). Exploring the black box: An
analysis of work group diversity, conflict, and performance. Administrative Science
Quarterly, 44(1), 1-28.

 

Pelled, L. H., Xin, K. R., & Weiss, A. M. (2001). No es como mi: Relational demography
and conflict in a Mexican production facility. J ournal of Occupational and Organizatiopal
Psychology, 74(1), 63-84.

Peters, L. H., & Sheridan, J. E. (1988). Turnover research methodology: A critique of
traditional designs and a suggested survival model alternative. In K. M. Rowland & G.
R. Ferris (Eds.), Research in Personnel and Human Resource Management (Volume 10).
Grennwich, CT: J AI Press.

Pfeffer, J. (1983). Organizational demography. Research in Organizational Behavior, 5, ‘
299-357.

Phillips, A. S., & Bedeian, A. G. (1994). Leader-follower exchange quality: The role of
personal and interpersonal attributes. Academy of Management Journala3 7(4), 990-1001.

Ployhart, R. E. (1998). A multilevel perspective on personnel selection. Paper presented
at the Annual Meeting of the Academy of Management, San Diego, CA.

Ployhart, R. E., & Hakel, M. D. (1998). The substantive nature of performance
variability: Predicting interindividual differences in intraindividual performance.
Personnel Psychology. 51(4), 859-901.

Pugh, S. D. (2001). Service with a smile: Emotional contagion in the service encounter.
Apademy of Management Journal, 44(5), 1018-1027.

Ragins, B. R., & Comwell, J. M. (2001). Pink triangles: Antecedents and consequences
of perceived workplace discrimination against gay and lesbian employees. Journal of
Applied Psychology, 86(6), 1244-1261.

Ree, M.J. & Earles, J .A. (1992). Intelligence is the best predictor of job performance.
Current Directions in Psychological Science, 1, 86-89.

Reise, S. P. (2000). Using multilevel logistic regression to evaluate person-fit in IRT
models. Multivariate Behavrml Research, 35, 543-568.

Rentcsh, J. R. & Hall, R. J. (1994). Members of great teams think alike: A model of team
effectiveness and schema similarity among team members. In M. M. Beyerlein & D. A.
Johnson. (Eds.), Advances in Interdisciplinary Studies of Work Teams: Theories of Self-
Managing Work Teams. Volume 1 (pp. 223-262). Greenwich, CT: JAI Press.

152

Richard, O. C. (2000). Racial diversity, business strategy, and firm performance: A
resource-based view. A_cademy of Management Journal,43(2), 164-177.

Riordan, C. M & Holliday-Wayne, J. (1998). Work group gender composition attitudes:
A test of competing hypotheses. Proceedings of the Southern Management Association.

Riordan, C. M. & Weatherly, E. W. (1999). Relational demography within groups: An
empirical test of a theoretical model. Paper presented at the annual meeting of the
Academy of Management, Chicago.

Riordan, C. M. (1997). Advancing relational demography theory: A construct validity
study of three measures of demographic similarity. In L. N. Doeler & J. B. Keys (Eds.).
Academy of Management Best Paper Proceedings.

Riordan, C. M. (2000). Relational demography within groups: Past developments,
contradictions, and new directions. In G. R. Ferris (Ed.), Research in Personnel aad
Human Resources Management. Volume 19 (pp. 131-174). New York: JAI.

Riordan, C. M., & Shore, L. M. (1997). Demographic diversity and employee attitudes:

An empirical examination of relational demography within work units. Journal of
Applied Psychology, 82(3), 342-358.

Roberts, K. H. & Burstein, K. (1980). Issues of aggregation. In K. H. Roberts & K.
Burstein (Eds.), New directions for methodology in the social sciences. (Volume 6). San
Fransisco: Jossey-Bass.

Roberts, K. H., Hulin, C. L., & Rousseau, D. M. (1978). Developingan interdisciplingy
science of organizations. San Fransisco: Jossey-Bass.

Rosse, J. G. & Noel, T. W. (1996). Leaving the organization. In K. R. Murphy (Ed.).
Individual Differences and Behavior in Organizations. (pp. 451-506). San Francisco:
J ossey-Bass.

Rosse, J. G., Stecher, M. D., Miller, J. L., & Levin, R. A. (1998). The impact of response
distortion on preemployment personality testing and hiring decisions. Journal of Applied
Psychomy, 83(4), 634-644.

Roth, P. L., Bobko, P., Switzer, F. S., III, & Dean, M. A. (2001). Prior selection causes
biased estimates of standardized ethnic group differences: Simulation and analysis.
Personnel Psychology. 54(3), 591-617.

Rousseau, D. M. (1985). Issues of Level in Organizational Research: Multi-Level and
Cross-Level Perpsectives. In L. L. Cummings & B. M. Staw (Eds.), Research in
Organizational Behavior. Volume 7 (pp. 1-37). Greenwich, CT: JAI Press.

Rousseau, D. M., & House, R. J. (1994). MESO organizational behavior: Avoiding the
fundamental biases. In C. L. Cooper & D. M. Rousseau (Eds.), Trends in Organizational
Behavior Volume 1 (pp. 13-30). London: Wiley.

 

153

Rousseau, D.M. (1977). Technological differences in job characteristics, employee
satisfaction, and motivation: A synthesis of job design research and sociotechnical
systems theory. Organizational Behavior and Human Decision Processes, 19, 18-42.

Rousseau, D.M. (1985). Issues of level in organizational research: Multilevel and cross-
level perspectives. In L.L. Cummings & B.Staw (Eds.), Research in organizational
behavior (Vol 7, pp. 1-37). Greenwich, CT: JAI Press.

Sackett, P. R., & Ellingson, J. E. (1997). The effects of forming multi-predictor
composites on group differences and adverse impact. Personnel Psychology, 50(3), 707-
721.

Salas, E., Dickinson, T. L., Converse, S. A., & Tannenbaum, S. I. (1992). Toward an
understanding of team performance and training. In R. W. Swezey & E. Salas. Teams:
Their Training and Performance. (pp. 3-30). USA: Ablex.

Schmidt, F. L. (1993). Personnel psychology at the cutting edge. In N. Schmitt & W. C.
Borman and Associates, Personnel Selection in Organizations. (pp. 497-516). San
Fransisco: Jossey-Bass.

Schmidt, F. L., & Hunter, J. E. (1998). The validity and utility of selection methods in
personnel psychology: Practical and theoretical implications of 85 years of research
findings. Psychological Bulletin, 124, 262-274

Schmidt, F. L., & Hunter, J. E. (1998). The validity and utility of selection methods in
personnel psychology: Practical and theoretical implications of 85 years of research
findings. Psychological Bulletin, 124(2), 262-274.

Schmidt, F. L., Ones, D. S., & Hunter, J. E. (1992). Personnel selection. Annual Review
of Psycholog , 43, 627-670.

 

Schmitt, N. & Chan, D. (1998). Personnel selection: A theoretical approach. Thousand
Oaks, CA: Sage.

Schmitt, N. (1989). Fairness in employment selection.

Schmitt, N., Rogers, W., Chan, D., Sheppard, L., & Jennings, D. (1997). Adverse impact
and predictive efficiency of various predictor combinations. J oumrl of Applied
Psycholggy, 82, 719-730.

Schneider, B. (1985). Organizational Behavior. Annufleview of Psycholggy, 36, 573-
611.

Schneider, B. (1987). The people make the place. Personnel Psycholgy, 40(3), 437-453.

Schneider, B. (1996). When individual differences aren’t. In K. R. Murphy (Ed.).
Individual Differencesapd Behavior in Organizations. (pp. 548-572). San Francisco:
Jossey-Bass.

Schneider, 8., Smith, D. B., & Sipe, W. P. (2000). Personnel selection psychology:

154

Multilevel considerations, K. J. Klein & S. W. J Kozlowski (Eds). (2000L Multilevel
theoglaresearch, and methods in organizations: Foundatiops, extensioris, and new
directions. (pp. 910120). San Francisco, CA, US: Jossey Bass Inc, Publishers.

Scullen, S. E., Mount, M. K., & Goff, M. (2000). Understanding the latent structure of
job performance ratings. Journal of Applied Psycholog, 85(6), 956-970.

Sessa, V. I. & Jackson, S. E. (1995). Diversity in decision-making teams: All differences
are not created equal. In M. M. Chemers, M. A. Costanzo, M. A. & S. Oskamp (Eds.),
Diversity in Opganizations: New Perspectives for a Changing Workplace. (pp. 157-190).
Thousand Oaks, CA: Sage.

Shaw, J. B., & Barrett Power, E. (1998). The effects of diversity on small work group
processes and performance. Human Relations, 51(10), 1307-1325.

 

Shaw, J. D., Delery, J. E., Jenkins, G. D., & Gupta, N. (1998). An organization-level
analysis of voluntary and involuntary turnover. Academy of Management Journal, 41 ,
511-525.

Shaw, ME. (1971). Group dynamics,the pchholggy of small goup behavior. New
York: McGraw-Hill.

Sherif, M., Harvey, O. J ., White, B. J ., Hood, W R., & Sherif, C. W. (1961). Intergroup
conflict and cooperation. The Robbers Cave Experiment. Norman, OK: Institute of
Group Relations.

Simon, H. A. (1973). The organization of complex systems. In H. H. Patee (Ed.).
Hierarchy Theopy (pp.l-27). New York: Braziller.

Singer, J. D. & Willett, J. B. (1991). Modeling the days of our lives: Using survival
analysis when designing and analyzing longitudinal studies of duration and the timing of
events. Psychological Bulletin, 110, 268-290.

Singer, J. D., & Willett, J. B. (1991). Modeling the days of our lives: Using survival
analysis when designing and analyzing longitudinal studies of duration and the timing of
events. Psychological Bulletin, 110(2), 268-290.

Smith, P. C. (1976). Behavior, results, and organizational effectiveness: The problem of
criteria. In M. D. Dunnette (Ed.), Handbgk of Industrial and Organizational Psychology
(pp. 745-775). Chicago: Rand McNally.

Steiger, J. H. (1990). Structural model evaluation and modification: An interval
estimation approach. Multivariate Behavioral Research, 25, 173-180.

Steinder, ID. (1972). Group process and productivity. New York: Acadmic Press.

Stevens, M. J ., & Campion, M. A. (1994). The knowledge, skill, and ability requirements
for teamwork: Implications for human resource management. Journal of Management,
29(2), 503-530.

155

Stoolmiller, M. (1995). Using latent growth curve models to study developmental
processes. In J. Gottman (Ed.), The Analysis of Changa (pp. 103-138). Hillsdale, NJ:
Erlbaum.

Stumpf, S. A., & Dawley, P. K. (1981). Predicting voluntary and involuntary turnover
using absenteeism and performance indices. Academy of Management Journal, 24(1),
148-163.

Sundstrom, E., de Meuse, K.P. & Futrell, D. (1990). Work teams: Applications and
effectiveness. American Pswholggist, 45, 120-133.

Sundstrom, E., McIntyre, M., Halflrill, T., & Richards, H. (2000). Work groups: From the
Hawthorne studies to work teams of the 19903 and beyond. Group Dynamics, 4(1), 44-
67.

Swezey, R.W. & Salas, E. (1992). Guidelines for use in tearn-training development. In
R.W. Swezey & E. Salas (Eds.), Teams: Their training and performance (pp. 219-245).
Stamford, CT: Ablex Publishing Corp.

Tabachnick, B. G. & F idell (1996). Using Multivariate Spatistics (3rd Ed.). New York:
Harper Collins. -

Tajfel, H. (1981). Human Groups and Social Categories: Studies in Social Psychology.
Cambridge, England: Cambridge University Press.

Tajfel, H. (1982). Social psychology of intergroup relations. Annual Review of
Psychology, 33, 1-39.

Teachman, J. D. (1980). Analysis of population diversity: Measures of qualitative
variation. Sociological Methods and Research,8, 341-362.

Tellegen, A. (1985). Structure of mood and personality and their relevance to assessing
anxiety, with an emphasis on self-report. In A.H. Tuma & J .D. Maser (Eds.), Anxiety
and the anxiety disorder (pp. 681-706). Hillsdale, NJ: Lawrence Erlbaum Associates,
Inc.

Tesluk, P. E., & Mathieu, J. E. (1999). Overcoming roadblocks to effectiveness:
Incorporating management of performance barriers into models of work group
effectiveness. Journal of Applied Psychology,84, 200-217.

Tett, R. P., Jackson, D. N., & Rothstein, M. (1991). Personality measures as predictors of
job performance: A meta-analytic review. Personnel Psychology, 44, 703-742

Timmerman, T. A. (2000). Racial diversity, age diversity, interdependence, and team
performance. Small Group Research, 31(5), 592-606.

T josvold, D. & Tjosvold, M. M. (1994). Cooperation, competition, and constructive
controversey: Knowledge to empower self-managing work teams. In M. M. eyerlein &
D. A. Johnson. (Eds.) Advances in Interdisciplinary Studies of Work Teams: Theories of
Self-Managing Wopk Teams. Volume 1 (pp. 119-144). Greenwich, CT: JAI Press.

156

Tjosvold, D. (1995). Cooperation theory, constructive controversy, and effectiveness:
Learning from crisis. In R. A. Guzzo & E. Salas and Associates (Eds.), Team
Effectiveness and Decision Making in Organizations. San Fransisco: J ossey-Bass.

Tosi, H. (1992). The Environment/Organization/Person Contingency Model: A Meso
Approach to the Study of Organizations. Greenwich, CT: JAI Press.

Trevor, C. O., Gerhart, B., & Boudreau, J. W. (1997). Voluntary turnover and job
performance: Curvilinearity and the moderating influences of salary grth and
promotions. Journal of Applied PsychongJZU), 44-61.

Tsui, A. S., & 0' Reilly, C. A. (1989). Beyond simple demographic effects: The
importance of relational demography in superior-subordinate dyads. Academy of
Manfiment Journal, 32(2), 402-423.

 

Tsui, A. S., Egan, T. D., & 0' Reilly, C. A. (1992). Being different: Relational
demography and organizational attachment. Administrative Science Quarterly, 37(4),
549-579.

Tsui, A. S., Egan. T. D., & Xin, K. R. (1995). Diversity in organizations: Lessons from
demography research. In M. M. Chemers, M. A. Costanzo, M. A. & S. Oskamp (Eds.), '
Diversity in Organizations: New Perspectives for a Chan 'n Wor lace. (pp. 191-219).
Thousand Oaks, CA: Sage.

Tsui, A. S., Xin, K. R., & Egan, T. D. (1995). Relational demography: The missing link
in vertical dyad linkage, Jackson, Susan E. (Ed); Ruderman, Marian N. (Ed). (1995).
Diversity in worlpteams: Research paradigms for a changing workplace. (pp. 97 129).
Washington, DC, US: American Psychological Assocaiation.

Turban, D. B., & Jones, A. P. (1988). Supervisor-subordinate similarity: Types, effects,
and mechanisms. Journal of Applied Psychology, 73(2), 228-234.

Turner, J. (1982). Toward a cognitive definition of the group. In H. Tajfel (Ed.). Social
Identity and Intergroup Relation_s. Cambridge: Cambridge University press.

Turner, J. (1987). Rediscoverinnghe Social Group: A Social Catagorization Theory.
Oxford, UK: Blackwell.

 

Tziner, A., & Eden, D. (1985). Effects of crew composition on crew performance: Does
the whole equal the sum of its parts? Journal of Applied Psychology. 70(1), 85-93.

Van der Vegt, G., Emans, B. J. M., & Van De Vliert, E. (2001). Patterns of
interdependence in work teams: A two-level investigation of the relations with job and
team satisfaction. Personnel Psychology, 54(1), 51-69.

Van Scotter, J. R., & Motowidlo, S. J. (1996). Interpersonal facilitation and job

dedication as separate facets of contextual performance. Journal of Applied Psychology,
fl, 525-531.

Vandenberghe, C. (1999). Organizational culture, person-culture fit, and turnover: A

157

replication in the health care industry. Journal of Organizational Behavior,
Organizational-Behavior.

Viswesvaran, C. & Ones, DS. (1999). Meta-analyses of fakability estimates:

Implications for personality measurement. Educational and Psychological Measurement,
52, 197-210.

Viswesvaran, C., Ones, D. S., & Schmidt, F. L. (1996). Comparative analysis of the
reliability of job performance ratings. Journal of Applied Psychology, 81(5), 557-574.

Voiers, W. D. (1956). Bombing accuracy as a function of the ground-school proficiency
structure of the B-29 bomb team. (Research Report AF DTRC-TN-56-4) Lackland Air
Force Base, TX: Air Force Personnel and Training Research Center.

Wagner, W. G., Pfeffer, J ., & 0' Reilly, C. A. (1984). Organizational demography and
turnover in top-management groups. Administrative Science Quagrly, 29(1), 74-92.

Wanous, J. P., Stumpf, S. A., & Bedrosian, H. (1979). Job survival of new employees.
Personnel Psychology.

 

Watson, W. E., Kumar, K., & Michaelsen, L. K. (1993). Cultural diversity's impact on
interaction process and performance: Comparing homogeneous and diverse task groups.
Academy of Management Journal, 36(3), 590-602.

Weiner, B. (1979). A theory of motivation for some classroom experiences. Journal of
Educational Psyachologle, 3-25.

Weiner, B. (1985). An attributional theory of motivation and emotion. Pswhological
Review 92 548-573.

 

Weingart, L. R. (1997). How did they do that? The ways and means of studying group
process. In L. L. Cummings & B. M. Staw (Eds.), Resegh in Organizational Behayior.
Volume 19 (pp. 189-240). Greenwich, CT: J AI Press.

Wells, D. L., & Muchinsky, P. M. (1985). Performance antecedents of voluntary and
involuntary managerial turnover. Journal of Applied Psychology, 70(2), 329-336.

 

Werbel, J. D. & Johnson, D. J. (2001). The use of person-group fit for employee
selection: A missing link in person-environment fit. Human resource Management, 40,
227-240.

Werbel, J. D., & Gilliland, S. W. (1999). Person-environment fit in the selection process.

Wesolowski, M. A., & Mossholder, K. W. (1997). Relational demography in supervisor-
subordinate dyads: Impact on subordinate job satisfaction, burnout, and perceived
procedural justice. Journal of Organizatiopal Behavior, 18(4), 351-362.

Westphal, J. D., & Zajac, E. J. (1995). Who shall govern? CEO/board power,
demographic similarity, and new director selection. Administrative Science Quarterly,
19(1), 60-83.

158

Wharton, A. S., Rotolo, T., & Bird, S. R. (2000). Social context at work: A multilevel
analysis of job satisfaction. Sociological Forum, 15(1), 65-90.

Wiersema, M. F., & Bird, A. (1993). Organizational demography in Japanese firms:
Group heterogeneity, individual dissimilarity, and top management team turnover.
Academy of Management Journal, 36(5), 996-1025.

Wilk S. L., Desmarais L. B., Sackett P. R. (1995). Gravitation to jobs commensurate with
ability: Longitudinal and cross-sectional tests. Journal of Applied Psychology, 80, 79-85.

Wilk, S. L., & Sackett, P. R. (1996). Longitudinal analysis of ability-job complexity fit
and job change. Personnel PsychologL49(4), 937-967.

Willet, J. B & Singer, J. D. (1995). It’s Déja vu All Over Again: Using multiple-spell
discrete time survival analysis. Journal of Educatiogrl and Behavioral Statistics, 20, 41-
67.

Willett, J. B. & Sayer, A. G. (1994). Using covariance structure analysis to detect

correlates and predictors of individual change over time. Psycholog'pcal Bulletin, 116,
363-381.

Willett, J. B. & Singer, J. D. (1993). Investigating onset, cessation, relapse, and
recovery: Why you should, and how you can, use discrete-time survival analysis to
examine event occurrence. J ourpal of Conaultingand Clinical Psychology, 61, 952-965.

Williams, K. Y. & O’Reilly, C. A. (1998). Demography and diversity in organizations:
A review of 40 years of research. Research in Organizational Behavior, 20, 77-140.

Williams, W. M., & Stemberg, R. J. (1988). Group intelligence: Why some groups are
better than others. Intelligence, 12, 351-377.

 

Yarnmarino (199x). A conceptual-empirical approach for testing meso and Multi-Level
Theories. In H. L. Tosi (Ed.) Extensions of the anironmenUOrgagization/Person Model
(Volume 2). Greenwich, CT: JAI Press.

Zalesny, M. D., Salas, E., & Prince, C. (1995). Conceptual and measurement issues in
coordination: Implications for team behavior and performance. In G. R. F erris (Ed.).
Research in Personnel and Human Resources Managament. Volume 13 (pp. 81-116).
New York: JAI Press.

Zeidner, J. & Johnson, C. D. (1991). The Economic Benefits of PredictingJob
Performance: Selection Utility (Volume 1). New York: Praeger.

Zenger, T. R., & Lawrence, B. S. (1989). Organizational demography: The differential

effects of age and tenure distributions on technical communication. Academy of
Management Journal, 32(2), 353-376.

159

 

:0:mme:wom :80

A8558 ESE—9::
:0: Ba :o>o::3 88:39: :88 :onwE E 88888
on 2:88 3053868088 E 38:03: 8808:: .NHE

:33 :88,

 

58883:

80:88:88: :88 832 :83 38688 on
23% 8083888028 E .3883: ~38 :88 «8896:: .2:

$25 :83.

 

58383:

88388 05:8 32 8:3 08::

8:3 :85 80928:: 8:68-52 :33 m:88 :8 8:88:88:
98 20>»: O<m¥ 5083 9:28:88 :ow:o::m 8 8 0:05 88
:88 00:88:88: :88 :58 80>»: O<mM :88 :88: 5850:
3:85:88: 8: 88:85:: E3 05:8 :ow888 :83. HE

.054 :88,

 

:o_mmo:wom

.oo:8~::o.tom :88 53: 93388—0: 90985-3 @8825
80:28: 8 855:8 EB 38:08:38 :8818 238 882 .3:

354 :83,

 

:o_mwo:mQM

43:38 o>£:moo :8 w:.:=o:::oo :38 888:8:3
:88 8 830: 53 8053802?“ :88 mo 88/8 882 .mm

654 :83,

 

:o_mmo:wom

3:885:89
:88 8 820: E? mmo:o~nmoo:w< :88 mo ~o>8 802 .v3

65.: :83.

 

:ommmo:wom

5:58 o>u8w8 :88
mo 8888 on: :8 wfizobcoo :38 8:88:88: :88 8 e838
3328: on EB 88: 8838:5838 80:: :88 8g .mm

:33 :88.

 

:meomWeM

.oo8:::o.tom:88 8 @820:
302:8: 8: EB Boom 3085885638 :88: :88 8E. .NI

654 :88,

 

:oamoEom

.oo:8:::o,to@ :88
8 @888: 302880: on 2:» 0:08 3:58 :88: :88 8E, 43

~0>01H amour

 

:oimobwom xoO

88558880250 :8 w::_o:::oo :38 H2 8 comemao 8
:> £55 2 38:: use 8 Ea $8288 :55“ :85 .2:

325 83238:

 

:ommmoaom xoU

PE 8 e828 3028?: on EB 80538856200 63

.83 3:222:

 

:omwmo:wom x00

H> 8 @883: 53383: 3 EB 80:85::8850 .8:

$54 8522?:

 

 

@0582 88283:

 

2350th

 

8850:»: we on»...

 

 

85582 8:885: 858$ 88 838883 :0 8583 A 038:,

160

 

8989:
822:6 :0>0::::::>:

80.82003 8:: 8:8:

3088:0880: 8805:: 8:3: :::B 280: :: m::v::0B 88:3
8 80030 0:08 0: EB 8:30:88: m::.:, .:0>0:::: :0 08:20::
8:3: 8 :::B 00880.88 m: #8:: 8:: :03 :0>0:::: 8:03:05
8:08: EB :88 0:: :::::B 8:00:03 0:: :0 :m: 8:08: 0:.:. .03.:

888880

 

.A8>0:::: 8:8:38 :0: :3: 8>0:::: 28:23

 

 

 

 

 

 

 

 

 

89:89: :0>0: 8:03:05:0 000E805 8B0: 8 :::B 0088088 0:
8>:2:m :0>0::::::>: 030:: :: :80:-:0m:0: 00888083388900 00808:: .2: :0>0:-mm0:0
w::::000:>: :0:8::0m:
8:80:8m 02:0 00:88:88:
::B0:0 :88: :80: 0: 0088: 20288: 0: EB 38820 :0: 00808:: .mmm 88.: :80 .:,
w::::000:>: :0:8::0m:
8:::0:::m 02:0 88: 8>0:::: :88
::B0:0 :88: 00808:: 0: 008:0: 0: EB 38820 :8 :88 00808:: :~:: 88.: :80 H
m::::000:>: :0:8::0m: 5:38 08 80:88:00 28:88: 8:88:
8:80:8m 02:0 0:: :0: w::::0:::00 8:8 :80:>0 0: EB 88:0 ma 00:88:88:
::B0:0 :88: :80: 8 008:0: 20>:8w0: 0: EB 58820 0?: 00808:: .3: 88:: :80 .:.
w::::000:>: :0:8::.m: 5:28 08 802880 28:88: :8>0:0: 0:: :0:
8:::0:::m 02:0 w::=0:::00 8:8 E0030 0: EB 88.8 m::.:. .88: 8>0:::: :88
::B0:0 :88: 8808:: 0: 0088: 0: EB >::m8>:0 0m: :80: 00808:: .w :m: :08: :80 .:r
w::::000:>: :0:8::0m: 52:8 0:: 80:28:00 58:88: :8>0:0:
8:80:25 02:0 0:: :0: w::=0:::00 8:8 E0030 0: EB 80:0 m.::.:. 00:88:88:
::B0:0 :88: :80: 0: 008:0: 20>:8w0: 0: EB 38820 8:08: 00808:: 0:: :08: :80 H
w:::000:>: :0:8:mm: 5:38 0:8 80:8:00 58:88: 8:88: 0:: :0:
8:::0:::m 02:0 w::=0:::00 8:8 ::00:>0 0: EB 80:8 m::.:. .88: 8>0:::: :80:
::B0:0 :88: 00808:: 0: 008:0: 0: EB 58820 8:08: :80: 00808:: .m::: :08: :80 .:r

 

 

6.83 : 2%:

161

 

8:88:85:
:8>:2:m 88:23):

E: :0: 00 0:B 88:03:08 88:08:80 B0: :8:: 888::

0: :80:: 80: 0: EB :m: 3888880800 :0: 00 8:: 88:03:08
88:08:08 :3: 8:: :08 88:08:08 88:03:08 :8 :0
0:0:00 EB :80: 8 :::::B :: 8:880:80 :0 38:8 0:9 03.:

8817880

 

8:80:82
8>:2:m 88:03:):

8:80: :0
:08: :0B0: :8 88:8: 0: 00:8:0: >888 80:: 0: EB :80: 0::
:0 8:: 08 82880800 88:03:08 :8 :00B:0: ::: 0:,:\ .mm::

88:-880

 

8:88:85:
8>:>::m 88:23»:

.8583 :o 88:88:: :28:
8 505 0088088 8: #8:: 8:: :08 :0>0:::: 8:03:08 8:08:
53 as: 28 55:3 8:285 28:0 8 833-83 2:: .8:

88:-80:0

 

 

8:80:85:
:8>:2:m 88:03:):

 

.8383 :0 88:3: 088:
8 505 0088088 8: 58:8 8:: :08 888:: 8:03:08 8:08:
:3, as: 2:: 8:83 8:285 2: :o :8 888-088 28 8 E

 

88:-880

 

 

8:88: : 2%,:

162

.mng ... .3836 05 3209 03 bzfimzficn 5:03.800 BEBE

8m “08000000 2833880 was 428me 0:0 96% 0% 05:22.80 3000000080 .65va 05 so 08 0.80850 5:806:00 BEBE

«:90 6205000 000532820300 7: n 2007: ”mm0:05008w< 7: n 03% Maom000>mbxm r: u XE“: 60.0303 mmocozaoouwxx

 

:5 u «-035 0853883. 20 u 956 age-235$ 20 u $5 0858:5628 E0 n 2820 ”862

 

 

 

53 8. 8.- oo. 8. 8. 0w. 3.. m3. 200::
8. 85 Q. G. 0N. Q. C. S. 2.... 9.5:
8.- *3. :3 mm. 8. mm. mo. 3. N3“ fig
8. *3. *8. 63 2.. a. S. 3. m2 méSmU
8. *2. 8. am. 33 R. 5.. mm. mom 035
we. .3. *3 *2. :N. :3 mm. 8. 3m x55
*8. 3. S. 8. 5.- *2. :3 3. 3m 20035
200:: 03.: 03:: «-035 035 0320 75020 am :82

Amruzv

860m 7: 28 EU 05 50303 macaw—0:00 “03000.50 98 ©0000-5005 £80858 55006200 BEBE .mocmcfim 033C039 .N 033-

163

.movnr. 000838606300 7: n 20?: ”0002050000w< r: H 03% €280>mhxm r: H NBA: 6025a 0000030000w<

30 H M-OSMU ”00000300003‘ EU n Ofimo 8300022005 EU n XmEU ”0005326000300 EU N 20020 500% n PM ”00002

 

3w. 8.. 9r mm. t. w_. *wo.
wo.- *mw. 2. *Nv. w". *mm. 2.
3r Om. *3. vm. we. *3. S.-
2. 1m. 2. *mw. *3. mm. 2.-
oo. om. we. *Nn. *ww. mm. M2.-
:. Om. *No. *9”. 3. Rm. 3.
*No. 8.. 2.- mo. 2.- mo. *mw.
Hy:- Hm-o HM-m .530 FM-m Hm-N PM;

207: H

012% .0

Ken: .m

mOSMU .v

OSMU .m

XBMU .N

ZOUHmU A

 

.Aom n Zv 330:: 0003-000 0 00 00800 T: 98 EU 00 macaw—00000 000000-0000. .m 030-.-

164

 

 

 

cm 103 02 nmodw www.3- voow Numb-m mmodm :38-
Ec.mo mo 30.? mum-M: owm.~ meJ wmodm 062
03.3 K Swim mom-E win-m 0mm; 304m 2080.,“

.90-w 00:00 0::->9 08805 00233 gxmngmcfi x005
0£00n<cfim< 802005
03803 05 mo 22:89:00 x0m can 000m . v 053-

165

Table 5. Age composition of the sample (Valid N = 251,355)

 

 

Age Range N %
<20 103,180 54.3%
20'” 42,095 22.1%
”'29 18,996 10.0%
30"” 11,066 5.8%
35‘” 6,658 3.5%
40'“ 3,278 1.7%
“‘49 1,688 0.9%
”‘54 895 0.5%
”'59 435 0.2%
>59 1,025 0.5%

 

Total 1 89,257 100%

 

166

Table 6. Categories of Turnover and Incidence

 

 

Reason N
Voluntary
Other Job 36,948
Moved 6,111
Personal 34,291
Voluntary Other 13,400
Subtotal 90,750
Involuntary Avoidable
Unacceptable Attendance 26,099
Substandard Performance 2,920
Violation of Policy 4,791
Gross Misconduct 2,489
Job Abandonment 726
Involuntary Other 2,510
Resigned in Lieu of Discharge 39,535
Subtotal 26,099
Involuntary Unavoidable
Death 269
No Work Available/ Layoff 17
Unavailable for Scheduled Work Hours 91
Military 1,563
Transfer 262
Other 8,416
Subtotal 1 0,61 8
Total 140,903

%

26.2%
4.3%
24.3%
9.5%
64.4%

18.5%
2.1%
3.4%
1.8%
0.5%
1.8%
28.1%
18.5%

0.2%
0.0%
0.1%
1.1%
0.2%
6.0%
7.5%
100%

 

167

.0820 22000 22.2223 009202...

020 002000 2002 00-2 00203.2 .3. v 2* 0020225 2200000 00-2 momm I comm M Z 20228300 320 200.0 00202.82, 3220000002 02 8mm

1 mow-N M 2 ”00000022020050 0 0000 20022000082- .1- 0002. mm 202-202 n 002 226 202-202 H v02 300082 02002200000 0 20 ”00202

 

 

*00. *0m. *om. No. Lb. mo. 0.260. mm. 0N6 mam 00000200002w< .m
2 .000. *2m. .8.- *02. met 20.. mm. mvd mam .0000 .0
2 *02. *vor *3. 0.26.0.- mor Om. 0N6 mam 02020200032 .0
2 oo. *02. i5. mo. wm2 2N0 QO 52220.0. .m
2 .50.- ...No. *00. 30.2.0. 000.? mam 008- 002mm .26
2 .30.- *0o.- om. mm. mm: 0220.230m 220 .m
2 0.20. mwmdm N20K2m 002 20 .N
2 ommdw 226.02% 2632 .20 .2
n .0 m .26 .m .N .2 Om c.0022

 

0020209 2002-200208000 wcofim 0022200000022 2000 02202220 00322-20000 .2. 0200-2-

168

Table 8. Hierarchical regressions for mean ability levels as a predictor of controllable

profit (N =2,206).

 

 

 

Step and Predictors Statistics for Step Statistics for Predictors
R2 df AR2 Adf AF [3 t

Step 1 .854 3 — -— 4,287.26

Profit PD 4 .64' 39.89

CRI Saturation .04' 4.62

PD 5

# of Sales PDS .32' 19.92
Step2 .855 4 .001 1 13.13

Ability PD 5 .03 ‘ 3.62
Step 3 .855 5 .000 1 .592

CRI Saturation .03 .77

X Ability

 

Notes: ' p < .05. Dependent variable is Controllable Profit for Period 5. PD = Period.

Measures collected at the individual-level are averages calculated within the stores.

169

Table 9. Hierarchical regressions for mean conscientiousness levels as a predictor of

controllable profit (N =2,207).

 

 

 

Step and Predictors Statistics for Step Statistics for Predictors
R2 (if AR2 Adf AF [3 t
Step 1 .854 3 - - 4,290.31
Profit PD 4 .64' 39.89 -
CRI Saturation .04' 4.63
PD 5
# of Sales PDS .32' 19.92
Step2 .854 4 .000 1 2.31
Conscientiousness -.01 -l .52
PD 5
Step 3 .854 5 .000 1 .04
CR1 Saturation X .04 .21
Conscientiousness

 

Notes: ’ p < .05. Dependent variable is Controllable Profit for Period 5. PD = Period.

Measures collected at the individual-level are averages calculated within the stores.

170

Table 10. Hierarchical regressions for mean agreeableness levels as a predictor of

controllable profit (N =2,207).

 

Step and Predictors

Statistics for Step

Statistics for Predictors

 

Step 1

Profit PD 4

CR1 Saturation
PD 5

# of Sales PD5

Step2

Agreeableness
PD 5

Step 3

CR1 Saturation
X
Agreeableness

 

R2 df ARZ Adf AF 0 t
.854 3 - - 4,290.31

.64‘ 39.89

.04' 4.63

.32‘ 19.92
.855 4 .000 1 3.41

-02 -1.85
.855 5 .000 1 .391

-.09 -.62

 

Notes: ' p < .05. Dependent variable is Controllable Profit for Period 5. PD = Period.

Measures collected at the individual-level are averages calculated within the stores.

171

Table 11. Hierarchical regressions for mean extraversion and extraversion2 levels as a

predictor of controllable profit (N =2,205).

 

 

 

Step and Predictors Statistics for Step Statistics for Predictors
R2 df AR2 Adf AF 0 T
Step 1 .854 3 —_ —— 3228.06
Profit PD 4 .64' 39.96
CRI Saturation .04' 4.79
PD 5
# of Sales PDS .32' 19.94
Tenure PD5 .02' 2.78
Step2 .854 4 .00 1 7.33
Extraversion -.02' -2.71
PD 5
Step 3 .854 5 .00 1 .366
Extraversion2 .08 .61

 

Notes: ‘ p < .05. Dependent variable is Controllable Profit for Period 5. PD = Period.

Measures collected at the individual-level are averages calculated within the stores

172

Table 12. Hierarchical regressions for mean conscientiousness X manager tenure levels

as a predictor of controllable profit (N =2,206).

 

 

 

Step and Predictors Statistics for Step Statistics for Predictors
R2 df AR2 Adf AF (3 t
Step 1 .854 3 — — 4,290.31
Profit PD 4 .64' 39.89 '
CRI Saturation .04' 4.63
PD 5
# of Sales PDS .32' 19.92
Step2 .856 5 .001 2 9.86
Manager Tenure .03' 4.17
Conscientiousness -.01 -1 .54
Step 3 .856 6 .000 1 .989
Manager Tenure x .16 1.00
Conscientiousness

 

Notes: ‘ p < .05. Dependent variable is Controllable Profit for Period 5. PD = Period.

Measures collected at the individual-level are averages calculated within the stores.

173

Table 13. Hierarchical regressions for mean agreeableness X manager tenure levels as a

predictor of controllable profit (N =2,206).

 

 

 

Step and Predictors Statistics for Step Statistics for Predictors
R2 df AR2 Adf AF B t
Step 1 .854 3 — —- 4,290.31
Profit PD 4 .64' 39.89
CRI Saturation .04‘ 4.63
PD 5
# of Sales PDS .32' 19.92
Step2 .855 5 .001 2 10.16
Manager Tenure .03' 4.1 1
Agreeableness -.01 -1 .73
Step 3 .855 6 .000 1 .583
Mgr. Tenure x .11 .76
Agreeableness

 

Notes: ' p < .05. Dependent variable is Controllable Profit for Period 5. PD = Period.

Measures collected at the individual-level are averages calculated within the stores.

174

Table 14. Hierarchical regressions for mean ability X manager tenure levels as a

predictor of controllable profit (N =2,206).

 

 

 

Step and Predictors Statistics for Step Statistics for Predictors
R2 (if AR2 Adf AF B t
Step 1 .854 3 —- -— 4,290.31
Profit PD 4 .64' 39.89
CR1 Saturation .04' 4.63
PD 5
# of Sales PD5 .32' 19.92
Step2 .856 5 .002 2 14.45
Manager Tenure .03' 3.96
Ability .03* 3.40
Step 3 .856 6 .000 l -.02
Mgr. Tenure x -.02 -.481
Ability

 

Notes: ' p < .05. Dependent variable is Controllable Profit for Period 5. PD = Period.

Measures collected at the individual-level are averages calculated within the stores.

175

Table 15. Hierarchical regressions for coefficient of variability in conscientiousness as a

predictor of controllable profit (N =2,029).

 

 

 

Step and Predictors Statistics for Step Statistics for Predictors
R2 df AR2 Adf AF B t
Step 1 .853 3 —— —— 3,924.45
Profit PD 4 .64’ 38.15
CRI Saturation .04' 4.13
PD 5
# of Sales PD5 .32' 19.25
Step2 .853 4 .000 1 2.30
Conscientiousness .01 1 .51
COV

 

Notes: ' p < .05. Dependent variable is Controllable Profit for Period 5. PD = Period.

Measures collected at the individual-level are averages calculated within the stores. COV

= Coefficient of variation, the standard deviation divided by the mean for a given store.

176

Table 16. Correlations among 1 1-month aggregates of restaurant level demographic

diversity measures and turnover rate (N = 2496-2498).

 

 

Mean SD 1. 2. 3. 4.
1. Race Diversity 3.76 1.96 1
2. Sex Diversity 4.84 .46 .16 1
3. Age Diversity 26,738 9,934 .14 .17 1
4. Turnover .56 .09 .09 .10 .04 1

 

Notes: All correlations are significant at p < .05. The demographic diversity measures
were aggregated across the time periods by summing the values, whereas the turnover

rate was averaged.

177

Table 17. Multiple regression analyses for the prediction of 1 l-month aggregate of

turnover rates with 1 1-month aggregates of age, sex, and demographic diversity.

 

 

 

Step and Predictors Statistics for Step Statistics for Predictors
R2 df AR2 Adf AF [3 t
Step 1 .015 3 — — 12.81
Age Diversity .02 .78
Race Diversity .07 364'
Sex Diversity .08 4.03'

 

Notes: ' p < .05.

178

.26 .092 3 88:60

.3002 2 w 202.02 .020 202-En2 0822 2000 022 02 ma .3. v 22 00 200002030 000 00220200000 22 520-235 22032 0 GM ”0.0494

 

8.2 00. 2o. 0w. mm. mm. 222. ow. mm. mm. mm. oooom. 560:”. mg
00.2 00. mm. 0w. mm. mm. 222. 92.. R. 3.. mmwom. m2owm. VQM

00.2 00. oo. mm. mm. mm. 2w. 3.. m0. mw0om. mmwhm. mg

00.2 mm. 20. mm. mm. mm. ow. R. w2wom. 03.52. NQM

00.2 00. 20. mm. mm. mm. mm. 030m. New?» 252

00.2 0m. 20. mm. vw. 2w. oooom. voobm. and

8.2 mo. 00. 0w. mm. mmwom. momwm. N25

00.2 mm. mm. 2%. Nwmom. 02632”. 2 2nd

002 va. 0w. Swom. N0me. 0252

cc. 2 mo. N002... mewm. 0%

22.2 hE-ON. Emwm. wad

mam VQM mam Nam 29% m2Q~2 N252 22mm o2QM2 and mad Om 00032 020.2.B>

 

.Awwm.m n 720 002-200 080 22 05 000000 32000220 2202 00-2 0022200000022 000 00222020 0.8200020 .0082 .2 0200.2.

179

.30 .34. B BEE A32 5 w 33 3.6 85a 2%

BE 05 mm wQ< .3. v a an “:35:me can macaw—88o =< .83?“on a 35:» $8 5 emu mo 09 babza om< n Q< ”88 Z

 

co; cm. on. on. 00. 5. mm. av. 9». mm.
co; ow. mm. on. we. mm. E. 9». ov.

0o; ww. mm. :M No. mm. wv. NV.

o0; mm. m». 8. mm. cm. 3.

co; mm. 3.. No. Vm. 9».

0o; mm. on. om. av.

co; mm. It. hm.

co; mm. no.

co; cm.

004

mQ< vQ< mQ< NQ< HQ< QQ< N_Q< ZD< 29¢. oQ<

Mm.
mm.
mm.
Vm.
hm.
ow.
5V.
mm.
mo.
3..
on:
wQ<

Remy
v9:
$.Z
OWNH
cfimg
m5:
3.:
003
2.3
ooNH
$1.2
Om

:SN
mmfim
mmfim
3.3
NBSN
mvsm
EVEN
wvfim
wodm
omdm
Vmwm
:82

mQ<
VQ<
ma
NQ<
_D<
m ~Q<
20¢.
H 2?.
o ~Q<
a2
wQ<

28gb;

.Awwm.m H Zv £55m 0:5 I 05 $88 ~93on own 8m mcouflotoobafi can .32333 @3985 £582 .9 2an

180

.30 .96 E 832$ A32

5 w 85m :9: 85a 05: Eu 05 2 w .6 .8. v a a Edema? 2“ Seasons =< dogmas 233:8 u 8 ”882

 

oo._ 8. mm. cm. 5. on.

9: mm. mm. mm. mm.

00; mm. 3. ow.

0o; mm. mm.

co; cm.

00;

TWO .35 EU NLU EU mEU

mm.

ww.

mm.

3w.

vw.

cm.

00.“

NED

mm.

mm.

mm.

vw.

mm.

cm.

00;

~30

mm.

mm.

vw.

vw.

mm.

mm.

mm.

cm.

00;

QED

fiw.

ow.

on.

on.

El.

mm.

vw.

vw.

mw.

co;

95

2..

mp.

E.

on.

mm.

mm.

on.

em.

mm.

2:

was

mmdwm
mewm
3.15%
3.3%
$.me
oodwm
wm. Em
Nvéwm
om.vwm
ondwm
Smog

Om

mdem
91.0;
50.3;
8.02%
omdm;
$.35
mm.wm$
N53;
3&5;
$.0on

$.35

:82

was

35

THU

98

EU

mEU

N50

:90

030

95

was

29.9;

.meN. H ZV mwotom 2:: Z 05 $83 5on 23:05:00 By mcouflotoobufi cam .mconfifiw 935me £582 .om 038.

181

.20 .39: 96 B 8323 A32 5 m 85m do 85a
2:: we 05 a $9: $0 .8. v a a 58%? em 20:20:00 =< 5.835 38m u mm 52$on 2%:880 u .6 a..o:“.o_.z.-

 

2.

mg.

2.

3.

3.

E.

E.

a.

2.

2.

3.

TWO

3.

.1.

2.

2.

2.

2.

3.

3.

2.

3.

3.

was

3.

2.

E.

3.

E.

3.

3.

2.

2.

2.

3.

EU

2.

2.

S.

E.

C.

8.

2.

Z.

S.

S.

C.

was

2.

3.

2.

3.

E.

E.

2.

E.

C.

C.

2.

EU

.2.

N“.

S.

mg.

Q.

S.

Q.

2.

mg.

2.

ma.

mEU

2.

Z.

Z.

Z.

I.

S.

3.

Z.

S.

a.

Q.

NED

ao.

we.

ao.

ao.

ao.

ao.

ao.

ao.

3.

S.

S.

:55

I.

2.

Q.

I.

Q.

Z.

2.

NH.

3.

2.

NH.

QED

ao.

ao.

ao.

ao.

ao.

ao.

ao.

ao.

ao.

ao.

mo.

amU

S.

o“.

:.

3.

2.

2.

o“.

E.

S.

OH.

2.

M30

mam

VQM

ma

NQM

RE

29%

and

:9

3a

and

ma

2&5,

.Awwmrm n ZV mwotom 2:: Z on“ $88 5683c 308 98 505 052—888 5m 3233:0885 Am 2an

182

.20 .89: 96 B 8325 .3? a w 85m 6.: Baa 2:: Em 2: a $9: $0 .8. v a a :aéfia

8m we. N 80:20:00 .83 $85 a 55E 3% E own mo Qwv 38830 ow< u Q< 62>on 0323580 U mu ”8on

 

we.

be.

he.

ee.

we.

we.

we.

e“.

ee.

ee.

who

be.

we.

we.

we.

we.

we.

we.

we.

we.

be.

be.

$6

me.

he.

we.

we.

ee.

we.

we.

we.

3.

ee.

ee.

was

me.

be.

we.

we.

we.

we.

we.

ee.

me.

we.

was

be.

we.

we.

we.

be.

we.

we.

we.

we.

we.

we.

EU

me.

me.

we.

we.

be.

we.

we.

we.

we.

we.

we.

wine

we.

we.

we.

we.

we.

me.

we.

be.

we.

we.

we.

NED

we.

we.

be.

he.

ee.

we.

we.

we.

we.

be.

we.

:mU

we.

we.

we.

be.

we.

we.

we.

we.

be.

we.

we.

eEU

we.

we.

me.

we.

be.

me.

we.

we.

be.

we.

me.

95

be.

me.

we.

be.

we.

we.

we.

we.

we.

me

mD<

VQ<

mD<

NQ<

~D<

2Q<

N~D<

:Q<

29¢.

QQ<

wD<

032.;

 

.Awwm.m H Ze mwotoa can 2 05 $8an 37¢on awn was Goa 0333980 8w 203288885 .mm 2me

183

Table 23. Example of the period-person data structure used for the multilevel survival

analyses.

 

Store Employee Period Turnover Sex Store Sex Sex X Store Sex

 

l 1 l O 1 .50 .50
1 l 2 O l .50 .50
l 1 3 1 1 .48 .48
1 2 1 0 0 .36 O

l 2 2 l 0 .35 O

1 3 l O 1 .31 .31
1 3 2 O l .32 .32
1 3 35 0 1 45 45

 

184

Table 24. Results of HnLM examining the effect of individual sex X restaurant sex

composition interaction on turnover probabilities.

 

 

 

 

 

 

 

 

Variable Parameter(s) B SE t-ratio Exp(B)
Intercept 700 + 80k -2.066 .021 -100.72' .126
Period 7,0 -0.035 .002 -18.10' .965
Period2 720 0.001 .0001 10.10' 1.001
Female 730 + 63* 0.207 .02 9.84' 1.23
Female Composition 740 .422 .04 11.52' 1.53
Female X Female

750 -.532 .04 -13.62' .588
Composition

 

 

 

 

 

 

 

Notes: *p < .05. SE = Standard error.

185

Table 25. Results of HnLM examining the effect of individual sex X restaurant sex

composition X time interaction on turnover probabilities.

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable Parameter(s) B SE t-ratio Exp(B)
Intercept 700 '1' 50k -.216 .026 -84.39‘ .1 15
Period 710 -.0189 .003 -6.186' .98]
Period? 7 20 .0008 .00008 10.154. 1.001
Female 730 + 83k .405 .0297 13.63. 1.820
Female 7.0 .599 .0487 12.303‘ 1.499
Composition

Female X Female no .405 .029 13.61' .421
Composition

Female X Period 760 “865 -054 '15-95' .969
Female .
Composition X 770 “'03 -004 ‘6-172 .970
Period

F ] .
Citazgzsfiigga e 780 .053 .006 8.684 1.055
Period

 

 

 

 

Notes: *p < .05. SE = Standard error.

186

 

Table 26. Results of HnLM examining the effect of individual race X restaurant race

composition interaction on turnover probabilities.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable Parameter(s) B SE t-ratio Exp(B)
Intercept 700 + 80,. -1.846 0.022 -91.864 0.158
Period 7.0 -0.043 0.003 -8.431 0.958
Periodz r20 0.001 0.000 9.403 1.001
White M) + 63,. 0.254 0.041 10.217 1.289
Hispanic 740 + 84,. 0.204 0.054 3.757 1.226
White Comosition 750 0.394 0.030 19.335 1.482
Hispanic Composition 760 0.174 0.053 4.082 1.190
White X White 770 -O.678 0.048 -19.195 0.508
Composition

White X Hispanic 780 -0.362 0.063 -8.233 0.696
Composition

Hispanic X White 790 -0.606 0.081 -8.024 0.545
Cormsition

Hispanic X Hispanic 7100 -0.373 0.073 -6.089 0.689
Composition

 

Notes: *p < .05. SE = Standard error.

187

 

Table 27. Predicted turnover hazards as a function of employee race and store racial

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

composition.
Store Racial Composition
80% White 10% White 10% White
10% Hispanic 80% Hispanic 10% Hispanic
10% Black 10% Black 80% Black
Employee Race Emplog Race Employee Race
P White Hisp. Black White Hisp. Black White Hisp. Black
1 13.6% 13.6% 15.9% 15.9% 11.4% 15.5% 17.5% 14.6% 13.7%
2 13.1% 13.1% 15.4% 15.4% 11.0% 14.9% 16.9% 14.1% 13.2%
3 12.7% 12.7% 14.9% 14.9% 10.7% 14.5% 16.4% 13.7% 12.8%
4 12.3% 12.3% 14.5% 14.5% 10.3% 14.0% 15.9% 13.2% 12.4%
5 1 1.9% 12.0% 14.0% 14.0% 10.0% 13.6% 15.4% 12.9% 12.03%
6 11.6% 11.6% 13.7% 13.7% 9.7% 13.2% 15.0% 12.5% 11.7%
7 11.3% 11.3% 13.3% 13.3% 9.5% 12.9% 14.6% 12.2% 11.4%
8 11.0% 11.0% 13.0% 13.0% 9.2% 12.6% 14.3% 11.9% 11.1%
9 10.8% 10.8% 12.7% 12.7% 9.0% 12.3% 14.0% 11.6% 10.8%
10 10.6% 10.6% 12.4% 12.4% 8.8% 12.1% 13.7% 11.4% 10.6%
11 10.3% 10.3% 12.2% 12.2% 8.6% 11.8% 13.4% 11.1% 10.4%
12 10.2% 10.2% 12.0% 12.0% 8.5% 11.6% 13.2% 10.9% 10.2%
13 10.0% 10.0% 11.8% 11.8% 8.3% 11.4% 13.0% 10.8% 10.1%
14 9.9% 9.9% 11.6% 11.6% 8.2% 11.3% 12.8% 10.6% 9.9%
15 9.7% 9.7% 11.5% 11.5% 8.1% 11.1% 12.7% 10.5% 9.8%
16 9.6% 9.6% 11.4% 11.4% 8.0% 11.0% 12.5% 10.4% 9.7%
17 9.5% 9.5% 11.3% 11.3% 8.0% 10.9% 12.4% 10.3% 9.6%
18 9.5% 9.5% 11.2% 11.2% 7.9% 10.8% 12.3% 10.2% 9.5%
19 9.4% 9.4% 11.1% 11.1% 7.9% 10.8% 12.3% 10.2% 9.5%
20 9.4% 9.4% 11.1% 11.1% 7.8% 10.7% 12.2% 10.1% 9.4%
21 9.4% 9.4% 11.1% 11.1% 7.8% 10.7% 12.2% 10.1% 9.4%
22 9.4% 9.4% 11.1% 11.1% 7.8% 10.7% 12.2% 10.1% 9.4%
23 9.4% 9.4% 11.1% 11.1% 7.8% 10.7% 12.2% 10.1% 9.4%
24 9.4% 9.4% 11.1% 11.1% 7.9% 10.8% 12.3% 10.2% 9.5%
25 9.5% 9.5% 11.2% 11.2% 7.9% 10.8% 12.3% 10.2% 9.5%
26 9.5% 9.5% 11.3% 11.3% 8.0% 10.9% 12.4% 10.3% 9.6%
27 9.6% 9.6% 11.4% 11.4% 8.0% 11.0% 12.5% 10.4% 9.7%
28 9.7% 9.7% 11.5% 11.5% 8.1% 11.1% 12.7% 10.5% 9.8%
29 9.9% 9.9% 11.6% 11.6% 8.2% 1 1.3% 12.8% 10.6% 9.9%
30 10.0% 10.0% 11.8% 11.8% 8.3% 11.4% 13.0% 10.8% 10.1%

 

 

 

 

 

 

 

188

 

Table 27 (cont’d)

 

 

 

 

 

 

 

 

 

 

31 10.2% 10.2% 12.0% 12.0% 8.5% 11.6% 13.2% 10.9% 10.2%
32 10.3% 10.3% 12.2% 12.2% 8.6% 11.8% 13.4% 11.1% 10.4%
33 10.6% 10.6% 12.4% 12.4% 8.8% 12.1% 13.7% 11.4% 10.6%
34 10.8% 10.8% 12.7% 12.7% 9.0% 12.3% 14.0% 11.6% 10.8%
35 11.0% 11.0% 13.0% 13.0% 9.2% 12.6% 14.3% 11.9% 11.1%

 

 

 

 

 

 

Notes: P = Period.

189

 

Table 28. Results of HnLM examining the effect of individual race X restaurant race

composition X time interaction on turnover probabilities.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable Parameter(s) B SE t-ratio Exp(B)
Intercept 700 + 80,. -2.019' .022 -91.864 .133
Period 710 -.021' .003 -8.431 .979
Period2 720 .001‘ .000 9.403 1.001
White H, + 83,. .579' .030 19.335 1.784
Hispanic y“, + 64,. .216' .053 4.082 1.241
White Composition 750 .414‘ .041 10.217 1.513
Hisoanic Composition 7.0 .202‘ .054 3.757 1.224
mm X.Wh“e 770 -.926‘ .048 -19.195 .396
Composmon

Wh‘w X.I3‘Spam° 780 -.516‘ .063 -8.233 .597
Composmon

H‘Span“? 3‘ WI“ 790 -.446' .073 -6.089 .640
Composmon

Hispanic X Hispanic .

Composition 7,00 -.651 .081 -8.024 .521
White X Period 7110 -.030' .003 -9.193 .970
Hispanic X Period 7,20 -.006 .005 -1.215 .994
White Composition X .

Period 7,30 -.026 .004 -6.677 .975
Hispanic Composition X

Period 7,40 -.001 .005 -.105 .999
White X White .

Composition X Period 7150 .041 .005 7.830 1.041
White X Hispanic .

Composition X Period 7,50 .025 .006 3.929 1.026
Hispanic X White .

Composition X Period 7160 .012 .007 1.666 1.012
H‘spamc X ”1393“” 7.7., .006 .008 .739 1.006

Composition X Period

 

 

 

 

 

 

Notes: *p < .05. SE = Standard error.

190

 

Table 29. Predicted turnover hazards as a function of employee race and store racial

composition and time.

 

Store Racial Composition

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

80% White 10% White 10% White
10% Hispanic 80% Hispanic 10% Hispanic
10% Black 10% Black 80% Black
Employee Race Employee Race Employee Race
P White Hisp. Black White Hisp. Black White Hisp. Black
1 12.8% 12.8% 15.3% 14.5% 10.1% 13.7% 17.2% 13.3% 12.1%
2 12.4% 12.4% 14.8% 14.2% 9.9% 13.5% 16.6% 13.1% 11.9%
3 12.1% 12.0% 14.3% 13.9% 9.7% 13.2% 16.0% 12.8% 11.7%
4 11.7% 11.6% 13.9% 13.6% 9.5% 13.0% 15.4% 12.6% 11.5%
5 11.4% 11.2% 13.5% 13.3% 9.4% 12.8% 14.9% 12.4% 11.3%
6 11.1% 10.9% 13.1% 13.1% 9.2% 12.6% 14.4% 12.2% 11.1%
7 10.8% 10.6% 12.7% 12.8% 9.1% 12.5% 13.9% 12.1% 11.0%
8 10.6% 10.3% 12.4% 12.6% 9.0% 12.3% 13.5% 11.9% 10.9%
9 10.4% 10.0% 12.0% 12.4% 8.9% 12.2% 13.1% 11.8% 10.8%
10 10.1% 9.8% 11.7% 12.2% 8.8% 12.1% 12.7% 11.7% 10.6%
11 9.9% 9.5% 11.5% 12.1% 8.7% 12.0% 12.4% 11.6% 10.6%
12 9.7% 9.3% 11.2% 11.9% 8.7% 11.9% 12.0% 11.5% 10.5%
13 9.6% 9.1% 10.9% 11.8% 8.6% 11.8% 11.7% 11.5% 10.4%
14 9.4% 8.9% 10.7% 11.7% 8.6% 11.8% 11.4% 11.4% 10.4%
15 9.3% 8.7% 10.5% 11.6% 8.6% 11.7% 11.1% 11.4% 10.3%
16 9.1% 8.6% 10.3% 11.5% 8.5% 11.7% 10.9% 11.3% 10.3%
17 9.0% 8.4% 10.1% 11.5% 8.5% 11.7% 10.7% 11.3% 10.3%
18 8.9% 8.3% 10.0% 11.4% 8.5% 11.7% 10.4% 11.3% 10.3%
19 8.8% 8.2% 9.8% 11.4% 8.6% 11.7% 10.2% 11.4% 10.3%
20 8.8% 8.1% 9.7% 11.3% 8.6% 11.7% 10.0% 11.4% 10.4%
21 8.7% 8.0% 9.6% 11.3% 8.6% 11.8% 9.9% 11.4% 10.4%
22 8.6% 7.9% 9.5% 11.3% 8.7% 11.9% 9.7% 11.5% 10.5%
23 8.6% 7.8% 9.4% 11.4% 8.7% 11.9% 9.6% 11.6% 10.5%
24 8.5% 7.7% 9.3% 11.4% 8.8% 12.0% 9.5% 11.7% 10.6%
25 8.5% 7.7% 9.3% 11.4% 8.9% 12.1% 9.3% 11.8% 10.7%
26 8.5% 7.6% 9.2% 11.5% 9.0% 12.3% 9.2% 11.9% 10.8%
27 8.5% 7.6% 9.2% 11.6% 9.1% 12.4% 9.2% 12.0% 10.9%
28 8.5% 7.6% 9.1% 11.6% 9.2% 12.5% 9.1% 12.2% 11.1%

 

191

 

Table 29 (cont’d)

 

 

 

 

 

 

 

 

 

 

 

 

 

29 8.5% 7.5% 9.1% 11.8% 9.3% 12.7% 9.0% 12.3% 11.2%
30 8.6% 7.5% 9.1% 11.9% 9.5% 12.9% 9.0% 12.5% 11.4%
31 8.6% 7.5% 9.1% 12.0% 9.6% 13.1% 8.9% 12.7% 11.6%
32 8.7% 7.5% 9.1% 12.2% 9.8% 13.3% 8.9% 13.0% 11.8%
33 8.7% 7.6% 9.1% 12.3% 10.0% 13.6% 8.9% 13.2% 12.0%
34 8.8% 7.6% 9.2% 12.5% 10.2% 13.9% 8.9% 13.5% 12.3%
35 8.9% 7.6% 9.2% 12.7% 10.4% 14.2% 8.9% 13.7% 12.5%

 

 

 

 

 

Notes: P = Period.

192

 

Table 30. Results of HnLM examining the effect of individual age X restaurant age

composition interaction on turnover probabilities.

 

 

 

 

 

 

 

 

Variable Parameter(s) 3 SE t-ratio Exp(B)
Intercept 700 + 501 -2.960 .098 -30.l82 .052
Period 7‘0 —.034 .002 -17.788 .966
period2 720 .001 .000 9.796 1.001
Age 730 +8“ .053 .004 12.641 1.054
Age Composition 740 .043 .004 9.869 1.044
Age X Age Composition 750 -.002 .000 -1 1.834 .998

 

 

 

 

 

 

Notes: *p < .05. SE = Standard error.

193

 

Table 31. Predicted turnover hazards as a function of employee age X store age

composition interaction.

 

Store Age Composition

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

(Average Age)
16 19 31
Employee Age
16 19 31 16 19 31 16 19 31
1 11.9% 12.5% 15.1% 12.2% 12.6% 14.3% 13.4% 12.9% 11.2%
2 11.5% 12.1% 14.7% 11.8% 12.2% 13.9% 13.0% 12.6% 10.9%
3 11.2% 11.8% 14.4% 11.5% 11.9% 13.5% 12.7% 12.2% 10.6%
4 10.9% 11.5% 14.0% 11.2% 11.6% 13.2% 12.3% 11.9% 10.4%
5 10.7% 11.2% 13.7% 10.9% 11.3% 12.9% 12.0% 11.6% 10.1%
6 10.4% 11.0% 13.4% 10.7% 11.0% 12.6% 11.8% 11.4% 9.9%
7 10.2% 10.7% 13.1% 10.4% 10.8% 12.3% 11.5% 11.1% 9.6%
8 10.0% 10.5% 12.8% 10.2% 10.6% 12.1% 11.3% 10.9% 9.4%
9 9.8% 10.3% 12.6% 10.0% 10.4% 11.8% 11.0% 10.7% 9.3%
10 9.6% 10.1% 12.3% 9.8% 10.2% 11.6% 10.8% 10.5% 9.1%
11 9.4% 9.9% 12.1% 9.7% 10.0% 11.4% 10.7% 10.3% 8.9%
12 9.3% 9.8% 11.9% 9.5% 9.8% 11.2% 10.5% 10.1% 8.8%
13 9.2% 9.6% 11.8% 9.4% 9.7% 11.1% 10.3% 10.0% 8.7%
14 9.0% 9.5% 11.6% 9.3% 9.6% 10.9% 10.2% 9.9% 8.5%
15 8.9% 9.4% 11.5% 9.1% 9.5% 10.8% 10.1% 9.7% 8.4%
16 8.8% 9.3% 11.4% 9.1% 9.4% 10.7% 10.0% 9.6% 8.3%
17 8.7% 9.2% 11.3% 9.0% 9.3% 10.6% 9.9% 9.5% 8.3%
18 8.7% 9.1% 11.2% 8.9% 9.2% 10.5% 9.8% 9.5% 8.2%
19 8.6% 9.1% 11.1% 8.8% 9.1% 10.5% 9.7% 9.4% 8.1%
20 8.6% 9.0% 11.1% 8.8% 9.1% 10.4% 9.7% 9.4% 8.1%
21 8.5% 9.0% 11.0% 8.8% 9.1% 10.4% 9.7% 9.3% 8.1%
22 8.5% 9.0% 11.0% 8.7% 9.0% 10.3% 9.6% 9.3% 8.0%
23 8.5% 9.0% 11.0% 8.7% 9.0% 10.3% 9.6% 9.3% 8.0%
24 8.5% 9.0% 11.0% 8.7% 9.0% 10.3% 9.6% 9.3% 8.0%
25 8.5% 9.0% 11.0% 8.7% 9.0% 10.3% 9.6% 9.3% 8.0%
26 8.5% 9.0% 11.0% 8.8% 9.1% 10.4% 9.7% 9.3% 8.1%
27 8.6% 9.0% 11.1% 8.8% 9.1% 10.4% 9.7% 9.4% 8.1%
28 8.6% 9.1% 11.1% 8.8% 9.1% 10.5% 9.8% 9.4% 8.1%
29 8.7% 9.1% 11.2% 8.9% 9.2% 10.5% 9.8% 9.5% 8.2%

 

194

 

Table 31 (cont’d)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

30 8.8% 9.2% 11.3% 9.0% 9.3% 10.6% 9.9% 9.5% 8.3%
31 8.8% 9.3% 11.4% 9.1% 9.4% 10.7% 10.0% 9.6% 8.3%
32 8.9% 9.4% 11.5% 9.2% 9.5% 10.8% 10.1% 9.7% 8.4%
33 9.0% 9.5% 11.6% 9.3% 9.6% 10.9% 10.2% 9.9% 8.5%
34 9.2% 9.6% 11.8% 9.4% 9.7% 11.1% 10.3% 10.0% 8.7%
35 9.3% 9.8% 12.0% 9.5% 9.9% 11.3% 10.5% 10.1% 8.8%

 

Notes: P = Period.

195

 

Table 32. Results of HnLM examining the effect of individual age X restaurant age

composition X time interaction on turnover probabilities.

 

 

 

 

 

 

 

 

 

 

 

Variable Parameter(s) B SE t-ratio Exp(B)
Intercept 700 + 5m. -3.63 .152 -23.817 0.0265
Period 710 .071 .012 5.893 1.0731
period? 720 .00076 .000075 10.124 1.0008
Age 730 + 531 .0870 .00030 -11.206 1.0908
Age Composition 740 .066 .0068 12.847 1.0682
Age x Age Composition 750 -.0033 .0069 9.623 0.9967
Age X pen-0d 760 -.005 .0005 -9.921 0.9949
Age Composition X 770 -.0037 .0005 -6.952 0.9963
Period

Age X Age comPOSition 780 .00018 .00002 8.085 1.0002
X Period

 

 

 

 

 

 

Notes: *p < .05. SE = Standard error.

196

 

Table 33. Predicted turnover hazards as a function of employee age X store age

composition X time interaction.

 

Store Age Composition

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

(Average Age)
16 19 31
Employee Age
16 19 31 16 19 31 16 19 31
1 11.3% 12.3% 17.1% 11.7% 12.4% 15.6% 13.3% 12.8% 10.8%
2 11.1% 12.0% 16.3% 11.5% 12.1% 15.0% 12.9% 12.4% 10.6%
3 10.9% 11.7% 15.6% 11.2% 11.8% 14.4% 12.5% 12.1% 10.3%
4 10.7% 11.5% 14.9% 11.0% 11.5% 13.8% 12.2% 11.8% 10.1%
5 10.6% 11.2% 14.2% 10.8% 11.3% 13.3% 11.9% 11.5% 9.9%
6 10.4% 11.0% 13.6% 10.6% 11.0% 12.8% 11.6% 11.2% 9.7%
7 10.3% 10.8% 13.1% 10.5% 10.8% 12.3% 11.3% 11.0% 9.5%
8 10.2% 10.6% 12.6% 10.3% 10.6% 11.9% 11.1% 10.7% 9.4%
9 10.1% 10.4% 12.1% 10.2% 10.5% 11.5% 10.9% 10.5% 9.2%
10 10.0% 10.3% 11.6% 10.1% 10.3% 11.1% 10.6% 10.3% 9.1%
11 9.9% 10.2% 1 1.2% 10.0% 10.2% 10.7% 10.5% 10.1% 9.0%
12 9.8% 10.0% 10.8% 9.9% 10.0% 10.4% 10.3% 10.0% 8.9%
13 9.8% 9.9% 10.4% 9.9% 9.9% 10.1% 10.1% 9.8% 8.8%
14 9.8% 9.8% 10.1 % 9.8% 9.8% 9.8% 10.0% 9.7% 8.7%
15 9.8% 9.8% 9.8% 9.8% 9.7% 9.5% 9.9% 9.6% 8.7%
16 9.7% 9.7% 9.5% 9.7% 9.7% 9.3% 9.7% 9.5% 8.6%
17 9.8% 9.6% 9.2% 9.7% 9.6% 9.1% 9.6% 9.4% 8.6%
18 9.8% 9.6% 8.9% 9.7% 9.6% 8.9% 9.6% 9.4% 8.6%
19 9.8% 9.6% 8.7% 9.7% 9.5% 8.7% 9.5% 9.3% 8.6%
20 9.9% 9.6% 8.5% 9.8% 9.5% 8.5% 9.4% 9.2% 8.5%
21 9.9% 9.6% 8.3% 9.8% 9.5% 8.3% 9.4% 9.2% 8.6%
22 10.0% 9.6% 8.1% 9.9% 9.5% 8.2% 9.4% 9.2% 8.6%
23 10.1% 9.6% 7.9% 9.9% 9.5% 8.0% 9.3% 9.2% 8.6%
24 10.2% 9.7% 7.7% 10.0% 9.6% 7.9% 9.3% 9.2% 8.7%
25 10.3% 9.7% 7.6% 10.1% 9.6% 7.8% 9.3% 9.2% 8.7%
26 10.5% 9.8% 7.5% 10.2% 9.7% 7.7% 9.4% 9.2% 8.8%
27 10.6% 9.9% 7.3% 10.4% 9.8% 7.6% 9.4% 9.3% 8.9%
28 10.8% 10.0% 7.2% 10.5% 9.8% 7.6% 9.4% 9.4% 9.0%

 

197

 

Table 33 (cont’d)

 

 

 

 

 

 

 

 

 

 

 

 

29 11.0% 10.1% 7.1% 10.7% 9.9% 7.5% 9.5% 9.4% 9.1%
30 11.2% 10.2% 7.1% 10.8% 10.1% 7.4% 9.6% 9.5% 9.2%
31 1 1.4% 10.3% 7.0% 11.0% 10.2% 7.4% 9.7% 9.6% 9.4%
32 11.6% 10.5% 6.9% 11.2% 10.3% 7.4% 9.8% 9.7% 9.5%
33 11.9% 10.7% 6.9% 11.5% 10.5% 7.4% 9.9% 9.8% 9.7%
34 12.2% 10.9% 6.8% 11.7% 10.7% 7.3% 10.0% 10.0% 9.9%
35 12.5% 11.1% 6.8% 12.0% 10.9% 7.3% 10.2% 10.2% 10.1%

 

 

 

 

 

 

Notes: P = Period.

198

 

Table 34. Results of HnLM examining the effect of individual age X restaurant age

composition X extraversion interaction on turnover probabilities.

 

 

 

 

 

 

 

 

 

 

 

 

Variable Parameter(s) B SE t-ratio Exp(B)
Intercept 700 + 50;. -1.916’ .065 -29.659 .147
Period 710 -.051' .003 -14.774 .950
Periodz 720 .002' .000 10.370 1.002
Age 730 + 531 .050' .018 2.751 1.051
Age Composition 740 .026 .017 1.531 1.027
Age X Age Composition 750 "001 ~00] '1 ~446 -999
Extraversion 760 + 561 .093 .152 .613 1.098
Extraversion X Age 770 -.010 .007 -1.477 .990
Extraversion X Age -.005 .006 -.810 .995
Composition 730

Extraversion X Age X .000 .000 1.200 1000
Age Composition 790

 

 

 

 

 

 

Notes: *p < .05. SE = Standard error.

199

 

Table 35. Results of HnLM examining the effect of individual race X restaurant race

composition X extraversion interaction on turnover probabilities.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable Parameter(s) B SE t-ratio Exp(B)
Intercept 700 + 80,. -1.566 .137 -11.462' .209
Period 7,0 -.053 .003 -15.937' .948
Periodz 720 .002 .000 1 1.248. 1.002
White 730 + 83,. -.574 .291 -1.970' .563
Hispanic 740 + a... -1.604 .523 -3.068' .201
White Composition 3'50 -.559 .322 -1.736‘ .572
Hispanic Composition 7 60 -.579 .532 -1.089 .561
mgofiim‘l‘m 770 .353 .441 .801 1.423
gggofiggzpmc y... .236 .696 .339 1.266
8818;213:118 3mm ygo 2.090 .744 2.808‘ 8.082
8818:1686: :‘SPamc 7.00 1.282 .804 1.594 3.603
Extraversion 7110 + am -.051 .034 -1.500 .950
Extraversion x White 7,20 + 3,20,. .241 .074 3.234‘ 1.272
Extraversion x Hispanic ym + 5,30. .443 .131 3.380' 1.557
88:85:88: X Wh‘te y... .205 .081 2.541‘ 1.228
22:23:33: x H‘Spamc 7,50 .168 .132 1.275 1.183
Extraversion X White X .

White Composition 7.50 -.238 .112 -2.134 .788
Extraversion X Hispanic .

X White Composition 7'60 7610 '187 ‘3'264 .544
Extraversion X White X

Hispanic Composition 7'70 7154 '175 7884 '857
Extraversmn X Hispanic 7130 _.427 .201 2.124. .652

X Hispanic Composition

 

 

 

 

 

 

Notes: *p < .05. SE = Stande error.

200

 

Table 36. Results of HnLM examining the effect of individual sex X restaurant sex

composition X extraversion interaction on turnover probabilities.

 

 

 

 

 

 

 

 

 

 

 

 

Variable Parameter(s) B SE t-ratio Exp(B)
Intercept 700 + 80,. -2.367 .224 -10.564' .094
Period 710 -.055 .003 ~15.965' .946
Period? 7 20 .002 .000 1 1.219' 1.002
Female 730 +831; .287 .341 .841 1332
Female Composition 740 '37] A47 329 1.449
Female X Female 750 -.490 .619 -.791 .613
Composition .
Extraversion 7 60 + 5 6k .155 .057 2.705' 1.168
Female X Extraversion 77° "067 '086 '-779 -935
F omalo Composition X 780 -.053 .1 14 -.466 .948
Extraversion

F 1 l

Czfiigsfiiiga e 790 .092 .157 .584 1.096
Extraversion

 

 

 

 

 

 

Notes: *p < .05. SE = Standard error.

201

 

Table 37. Predicted turnover hazards as a function of employee age X store age

composition X extraversion interaction.

 

Store Racial Compositions

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

10% White 80% White
80% Hispanic 10% Hispanic
Employee Race
Hispanic Black Hispanic Black

EX=3 EX=5 EX=3 EX=5 EX=3 EX=5 EX=3 EX=5
1 11.3% 12.3% 17.1% 11.7% 12.4% 15.6% 13.3% 12.8%
2 11.1% 12.0% 16.3% 11.5% 12.1% 15.0% 12.9% 12.4%
3 10.9% 11.7% 15.6% 11.2% 11.8% 14.4% 12.5% 12.1%
4 10.7% 11.5% 14.9% 11.0% 11.5% 13.8% 12.2% 11.8%
5 10.6% 11.2% 14.2% 10.8% 11.3% 13.3% 11.9% 11.5%
6 10.4% 11.0% 13.6% 10.6% 11.0% 12.8% 11.6% 11.2%
7 10.3% 10.8% 13.1% 10.5% 10.8% 12.3% 11.3% 11.0%
8 10.2% 10.6% 12.6% 10.3% 10.6% 11.9% 11.1% 10.7%
9 10.1% 10.4% 12.1% 10.2% 10.5% 11.5% 10.9% 10.5%
10 10.0% 10.3% 11.6% 10.1% 10.3% 11.1% 10.6% 10.3%
11 9.9% 10.2% 11.2% 10.0% 10.2% 10.7% 10.5% 10.1%
12 9.8% 10.0% 10.8% 9.9% 10.0% 10.4% 10.3% 10.0%
13 9.8% 9.9% 10.4% 9.9% 9.9% 10.1% 10.1% 9.8%
14 9.8% 9.8% 10.1% 9.8% 9.8% 9.8% 10.0% 9.7%
15 9.8% 9.8% 9.8% 9.8% 9.7% 9.5% 9.9% 9.6%
16 9.7% 9.7% 9.5% 9.7% 9.7% 9.3% 9.7% 9.5%
17 9.8% 9.6% 9.2% 9.7% 9.6% 9.1% 9.6% 9.4%
18 9.8% 9.6% 8.9% 9.7% 9.6% 8.9% 9.6% 9.4%
19 9.8% 9.6% 8.7% 9.7% 9.5% 8.7% 9.5% 9.3%
20 9.9% 9.6% 8.5% 9.8% 9.5% 8.5% 9.4% 9.2%
21 9.9% 9.6% 8.3% 9.8% 9.5% 8.3% 9.4% 9.2%
22 10.0% 9.6% 8.1% 9.9% 9.5% 8.2% 9.4% 9.2%
23 10.1% 9.6% 7.9% 9.9% 9.5% 8.0% 9.3% 9.2%
24 10.2% 9.7% 7.7% 10.0% 9.6% 7.9% 9.3% 9.2%
25 10.3% 9.7% 7.6% 10.1% 9.6% 7.8% 9.3% 9.2%
26 10.5% 9.8% 7.5% 10.2% 9.7% 7.7% 9.4% 9.2%
27 10.6% 9.9% 7.3% 10.4% 9.8% 7.6% 9.4% 9.3%

 

202

 

Table 37 (cont’d)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

28 10.8% 10.0% 7.2% 10.5% 9.8% 7.6% 9.4% 9.4%
29 1 1.0% 10.1% 7.1% 10.7% 9.9% 7.5% 9.5% 9.4%
30 11.2% 10.2% 7.1% 10.8% 10.1% 7.4% 9.6% 9.5%
31 11.4% 10.3% 7.0% 11.0% 10.2% 7.4% 9.7% 9.6%
32 11.6% 10.5% 6.9% 11.2% 10.3% 7.4% 9.8% 9.7%
33 11.9% 10.7% 6.9% 11.5% 10.5% 7.4% 9.9% 9.8%
34 12.2% 10.9% 6.8% 11.7% 10.7% 7.3% 10.0% 10.0%
35 12.5% 11.1% 6.8% 12.0% 10.9% 7.3% 10.2% 10.2%

 

Notes: P = Period

203

 

 

 

000080303 808 0B2 .EB 000300000 00

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0 Z 2005 00003030060000 5 560030 _0>0_ E03 3000.85 A 2.. _0>0q 800,—.
.gwsgfi 0500“ B2 FEB 0.6.05
EB :05 80w0008 0.508-:3: .EB 0800“ 00.“ 0000808000
93 m~0>0_ O<mv~ 00050; 030009300 Swaobm 0 mm 205 «05
:03 000080803 :80“ Ba m_0>0_ O<m¥ 8000 0008 0002500

0 2 030020200 05 880008 EB 0300“ Bwflfifi 800.“. HE 034 800,—.
.00§§0.twm :82 FEB 920003300 30923-3 3:025

02 0005300 0 «3.300 EB 005006.05 E08 («0 £0>0_ 0002 0E ~0>0A 8009.
5:50 030300 00% wan—00000 0am 0000.58004—

0 Z :80“ 00 030“ EB 0000030000?“ 0.88 m0 _0>0_ 0002 .mm ~0>0A 800,—.
00580800.“

0 Z 808 9 0220“ EB 0000030000904 802 m0 ~0>0_ 0002 .vm _0>0q 803.
5:30 03:38 800“
m0 $00E0 05 00m mchbcoo Baa 00:08.8.t0a 8000 0» E00200

0 Z 303500 03 EB 0008 30000030080000 0008 803 2:. .m: ~0>0A 800%
00008003.“ 800“ 8 c0500

02 3035.00 00 EB 0008 3000:030060000 0008 800“ 2:. .mm 004 803.
00008080.“ :80“

00> 0H E00200 3032000 00 EB 00000 3:50 0008 808 05. :I _0>0A 800,—.
00050050060000 00m $530000 baa H2 8 Bmommo 00

oz 5 03238 9 082 868 B 53 866386 05% £05 .2: 08: 36505

0 Z H2 00 E00200 3020030 09 EB 3000003006300 .03 _0>0.._ 3:03:05

02 .H> 8 @0500 303330 00 EB 00000005006300 .w: #004 30232:

Eamon tommsm $005093.” mmm050mwm m0 09C.

 

 

£30m Ea m0m050§3 («0 ng .mm 030,—.

204

 

00>

000007003 005 00500

00000000000000 33000300 000mm: 00B 08000 03 w0§00B 0000.05
00 000030 00000 00 EB 0300000200 00:. .00>00000 .00 000.500:
00030 0 000B 0000300000 03 00000 0000 0000 00>00000 300003003
0030000 EB 00000 05 0003B 30023003 05 .00 0c 003000 00F .33

3000400000

 

OZ

.3583 500—903 000 003 00>0003 b00023
~0>0_ 300330050 00002002 00B0_ 0 HEB 0000300000 00
00.500 00.0 00000000000 00000-0000000000300000 000000003 .33

6000300000

 

02

800000.000m
00000 00 0000—00 30300w00 00 EB 3300030 0000 00000005 .NNE

_0>0q 0000B

 

02

mDHMH HD>OEB END“
Emacuodm Cu Bawfiu DD =¢5 0Aummuo>m© KOm F80“ Vogue: fiNI

.00/04 800,—.

 

02

5:50 000 0000000000 320000000 0005—00
05 00.0 w0Eo00000 0000 000030 00 EB 000.0000 002. 000000000000
00000 00 000300 30300m00 00 EB 3300030 0m0 000000003 .0 _ m

_0>0q 0000.0.

 

02

5:50 000 08000000 320000000 000>0_00 000 00.0
w0E000000 0000 000030 00 EB 000000 00:. .00000 00>00000 00000
000000000 00 0000—00 00 EB 3300030 0m0 00000 00000005 .w 0:

_0>0q 0000...

 

00>

5:50 000 0000000000 3200000000 00000200
000 00.0 w0E000000 0000 000030 00 EB 0000000 00:. 00008000000
00000 00 000—00 30300000 00 EB 3300030 303000 0000000003 0:.—

~0>0A 0000.3.

 

02

5:50 000 0000000000 300000000 000>0_00 05 00.0
w0Eo00000 0000 000030 00 EB 000.000 00C. .0000 00>0008 0000
000000003 00 000300 00 EB 5000030 3000 8000 000000000 .mfim

00000.3 800,—.

 

 

02

 

A00>00000 b08235
000 008 00>00000 b00033 .0000 000mm: 00 0000300000
00 03000 00000000006008 00 5300030 000000000 .QE

 

_0>0A 800,—.

 

 

3.0003 mm 030%

205

 

 

02

.E 8.: ow 0:B 23232: 568235 32 SE: 8353

3 385 mm“: on EB E 33039“:onva Ho: ow “m5 £3239:
56.5235 :wE $5 :95 56.5233 333232: 5 co
32:2: EB :88 a 22:3 E oEQEwocsc («o muooto 05H .33

334-380

 

8%

65:2 :0
232 $32 8 .5588 8 @238 bwcobm 208 on E3 Emu: 05
mo 85 ES mofimfiwofion @363me cm 5053 5 2F .82

_o>oq-mm80

 

8%

355% 82:3: 5&2
m :23 88683 2 Emma: 35 :03 .5383 3323?: 868::
:E :32 2: 5:? 1.5222353: 83:-me 0:: .8:

654-380

 

 

mo>

 

.5883 :o 82:3: 5:3:
a :23 ©8283: mm Emma 85 :03 $553 3323?: 8605
:E :32 2: 555 3.232: 2: :o a 88%? 0:: s : :

 

_0>anmmOuU

 

 

€283 mm 2an

206

 

 

Individual Level Factors

(e.g., pattern of member skills.
attitudes. personality
charachteristics)

 

 

 

Performance Outcomes

(e.g., performance quality.
speed to solution, number of

 

 

 

 

 

 

 

 

 

 

 

 

 

 

errors)
Group Level Factors
Group Interaction
(e.g., structure, level of Process
cohesiveness. group size)
Other Outcomes

(e.g., member satisfaction.
group cohesiveness. attitude
change. sociometric structure)

 

 

 

 

 

Environment Level Factors

(e.g., group task charachten'stics,
reward structure, level of
environmental support)

 

 

Figure 1. Hackman’s IPO approach to teams

207

 

.oocegotom :88 mo 5:2on 2: warn—.898 $608 Boaouoofir .N eBwE

flogmcoo _o>oi_
3:229:

985
_m>o._-mwo.o

£02680
.93 Ewe...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  

 

<

mocmctotod H cozmzuoz + wddflm
" L
n o
“ 35850
W 9:85
m m

0 coszLou—mcmfi W m ._._n_ o
A a
qllocmELoton. 3895 i j; 28 E o

 

 

 

 

 

 

 

 

208

mwozbwcoo _o>o._
.3222.

6th
65.7320

 

 

LM>GCH2 _

.5383 .«o dong—ooh: 05 waives: 3on 30:885. .m onE

 

 

 

 

 

 

380330
9:85

 

 

 

 

 

'III"|"

'l'l'lll'lll'lllll'll'l'l'lAIII'|||II|"l|"||llll'l|'l'l|ll'|lll' I"IIIIII|||

flozbwcoo
_m>m._ Ewe...

 

 

 

 

 

 

 

 

 

 

 

 

 

209

 

.1358 22:33 #83033 ciao 538w Ecum— oEEfi a mo coufinomoaom .v 053":

wv

(‘

w:

v o::_H

 

 

ctr

m o::_k

 

 

 

 

mcr

 

 

+_ n

N o::_H

 

 

Ncr

 

 

mm

+o

m:

:Qoo:ouc_

The:

 

 

 

 

210

 

/>_
‘4?
'A\
\V

 

 

 

 

 

 

 

 

 

 

 

8. F3. 8. 8. 9r. 8 8.

aaaaaaaaaaaa

. 'n racial diversity for a sample of 15 restaurants.

 

 

 

3:68— :88.“ one? 36:38 308 :0 88:58 .3355 BNEBwSfim we BE I \t 8:me

 

_ 3995 sawmil—

 

Datum

v m N P m: N: F: or m w

 

 

 

 

 

 

 

 

 

06.

m6.

N¢.

:6.

To

Burpecn Jones 133m manepueis

213

 

 

 

 

 

 

 

Figure 8 - Trends in age diversity

\ /\4;2v\
v/ ‘ ‘

 

 

 

_o\/’\

OOOOO

for a sample of 15 restaurants.

 

 

 

 

 

mwfiwaofi 88mm ego—m basic own mo 83:58 Jami @3832me we BE I m 8:me

 

T5520 wm<lol_
3:3.

 

 

 

 

\\

 

 

 

 

 

 

 

 

 

 

 

md-

md-

5.0-

06-

md-

v.0-

v.0

Nd

6019901 macs 1aasn pertinent-Ins

215

 

 

 

 

 

 

 

 

 

 

\/

 

 

 

 

217

O O O C

O

0. 8. 8 8. §_

0 O 8 In C

N ‘- (fi 1-

93 9} U9 ' Q
”Old I’llo

ntrollable profit for a sample of 15 restaurants

Figure 11 - Trends in co

 

$532 88mm one? ESQ 0323.580 Mo 83838 ‘5de 3&9“?me mo SE I 2 oSmE

 

3.3..
2 N?

3 or

 

 

 

 

 

 

 

 

 

 

 

 

 

Tc-

mmd-

mud.

Nd.

mfio.

.mo.

mod.

mod

v.0

5"!P9°'l Jones 1aasn penpmpuets

219

 

36520 308 .00 £26— EBMEE 3.85m no women 580 032—888 no.“ mocouoomwb 0206on I 2 onE

 

_omow$.l . omowm.+...-... omowolil 9.80-1 Bom.-lol_

totem
v m N F 2 NF 5 o. a m

 

 

000.0»

 

 

000.05

 

000.me

word a; afiuqu |enuuv

 

 

 

08.0mm

 

 

 

0006mm

221

ESQ 0352868 no bmfioZv 368 .«o “can: :39 $.80» 082080 I E 2&5

3.9.020 .203. ton—28:80
omé 00... 0nd 00.0 end. 00.7 8.? 00d.

F P b

 

00.00; /

 

 

mafia w»

 

Nn.n0uu

 

 

no.0 «N»

 

 

REV

 

 

mm;

003

mm;

coma

now»

05»

9N»

omww

(spunnom 0!) 1001c! Imuuv

222

Chart 1: Voorty Total Prom = $172,240 (2RD = -.02)

Chart 2: Yurty Total Profit I $165,610 (2., I .52)
Group 0

Group C

 

Chart 3: Yearly Total Profit I $157,718 (2.,- 1.15)

Group D

    

Group C

Group B

Group A

Chan 5: Vlafly Total Profit I $178,419 (2.9 I -.51)

Group B

 

Chat 4: Youty Total Profit I 8153.705 (ZmI 1.48)

Group D

    
 
 

GrowA

   

GroupB

Chart 6: anty Total Profit I $188,625 (ZmI 4.34)
Group B

  

Figure 17 — Predicted yearly controllable profit for several hypothetical racial
compositions.

 

.888.“ 080 28.830
9.8 30:: 53> 8380 0059:0950 00 39:28 :23»:on 25?: no woman. 228:8 3338 038383 082020 I 2 Semi

 

_ .925 253581 _

 

carom
888%.monammmaowmmamwmw.mommrmptorvaFmFNFFFSm o N o m w n «F

p h b b h P b b h P h b p b b h h b L L b h h h b p h b p p p F _ _ o

I/ to

0.0

/

0.0

 

 

 

 

 

 

Ienwns wemwna

 

0.0

 

5.0

/

0.0

 

 

 

 

 

225

.mcoummomfioo x8 20% ~58h00 92 05 xom mo 800:3 a mm 0.353 02683 08200.5 I 0N Paw—m

 

«£23m .2 u $21 02635 .n. n. .271 mgoaem. .s. u. #8:? 833m .n. n. nEarl:

carom

 

 

 

 

 

 

 

u\

 

 

 

 

{am

#0?

$3

o\o.v.—.

$9.

e\o®_.

o\oom

o\on

obvw

maze“ JaAomnl

226

.080 0% 98060088 80% 80.8950 95 05 x8 mo 000050 a 8 08$: 5683 0320on I 3 053m

 

Tgoaem. .2 .u. $21 £23m .n. .n. $21 ogoaem. .s. .u. Ascalar magaEm .n. .u. soar-L

 

 

 

 

 

 

 

 

 

canon
00 Co 8 7v z Z 7y 7y Ir Ir Ir It It
9 8 Ir 6 I. 9 8 Ir 6 I. 9 8 Ir 6 I. C. 8 Ir
p n p b b h p L _ h L b n — h p p l— b P _ Ll 1— PIP _ p p p h P h L 51 °\°©
$0
I.
I -I $9 .n.
I U
lll/ M
m
\ H
\ m
\ 1| I I I ' O
\‘1 I! \oNr m.
\
\
I 1“. $3
4 $0—

 

 

 

227

80208088 38 “5.5280 05.80.00 850 05 38 mo 02850 a 8 088: .8583 08200.5 I mm 0.53»

 

I$2.>>$2dsmmeo§ I$2.>>$2dsmIsoE I$2.>>$2.aEm>> -I;
I $8 .2, $2 .95 m I61 I $8 .>> $2 .88 I l I $8 .>> $2 .95 Bill
I $2 .2, $8 .95 mlol I $2 .2, $8 .95 Ill I $2 .2, $8 .95 3+

0Otom
mm «a an an Fm on aw mu km on mm ¢~ mm mm 2N cm a. or $2 or m2 v2 m2 N. 22 o. m m N o m w m N F

I- » b —

 

 

 

b t— bl bl L P L F lP r n p b b p P n I— . Ll LI lp P F P P P p b F

 

 

 

 

 

 

 

 

 

 

r
a
I

 

 

 

 

 

$\.

$0

$5

$0 P

pJezeH renown-L

$0 P

o\o\. w

228

.080 08 800600800 008 80.88000 80000.00 0080 .008 00 000080 0 8 0.88: 00888 0002005 I mm 0530

 

 

I$2.>>$2dsmmsoi I$2.>>$2.aEmI...¢... I$2.>>$2.aEm>>-l;
I$8.>>$2dem_mlol I$8.>>$2.aEmIl I$8.>>$2.aEm>>III
I$2.>>$8.aEmmlol I$2.>>$8.quII.I I$2.>>$8.aEm>>lll

 

 

mmvmmmum5000~0N5NONON¢NONNNRouge:ore:mFNF : 0F 0 0 5 0 m v m N

00:00

.P L - IF 5 h p 5 b _ 5 _ h h

n

P

P

P

$5

 

 

 

$0

 

 

 

 

 

 

o\o E.

- $2

I {one

$5 5

 

 

 

 

pJezeH JaAOUJnl

229

0520388 0mm E05302 E000¢c=0 025 28 0mm mo 58.59 0 00 Emma: 00353 0000605 - vm 05me

 

0?. .050 5 0? 220 5 :0. .. 0? .95 2 02 220 5 5o. ..
00< .95 F, 02 220 2 Idl 0? .95 2 02 290 2 lol

 

0? .95 5 00< 0.90 2+ 0? .95 2 03 0.20 Slol 03 .95 0. 00< 220 2+

0? .050 2 0?. 220 5 - u! -
0m< .aEm 0F 0m< 08$ 9 III

 

 

Boron
mm 3” mm mm 5 on 8 on R 0a mu 0m 8 mm & ca 2 0. t 0. 8

n b h h h p p - h P b F -

F r

vrmrmwpwowmmnomvmww

h u u b p h h h °\°N

 

 

 

$a

 

- $3

- $9

 

 

- $2

 

$5

 

 

 

 

pJezeH JGAOUJnl

230

.08: 0:0 0223:0888 03 05.58002 Heat—0 00.32 .0? mo couoqa 0 00 0850: .5553 0000:00E - mm 05me

 

00< .200 .0 00< 0.20 5...... 0m< .020 2 00< 0.20 Bio... 00< .020 0. 00< 0.20 20 . - I :
0m< dEm 5 mm< 0.06 3 IO! 0m< dEm 9 0m< 0.2w 9 IT mm< .95 or mm< 0.05 m? I‘...
00< .020 20 00< 0.20 02+ 00< .020 2 00< 0.20 Bull 00< .020 02 00< 0.20 02+

 

 

 

catam

mmvmmmmmPmomawmmnwommwvmmwmmRowen—0:are: 2w... 3 or a m x. m w v m NF

I b - p b b n L p

 

 

 

 

 

 

 

 

 

 

 

 

o\oN

$m

. o\o—.F

$m _.

o\om _‘

. $2

pJBZBH JGAOUJnl

231

2000003000000
.00 0_0>0_ 030 0:0 00000000090000 30000 0000000000 8000,0008 030 .0000 .00 0000005.: 0 00 @0000: 00.6.0000 0000000000 I om 000wE

 

10.00 32.2 0&0 m 0&0 :6. .. I080 258. 02m m 008 - -l- - $0.00 >38. 2.0 I muxmicl 10803080020 I naval...

 

I$om >>$ow 02m m mnxw till: I$ow >>$ow 05w m mnXmIIOI I$ow >>$o—. 9cm I mnXmloll I$ow >>$ow 08m I muxm+

 

 

00:00
000000000000000000000000000000000000000000:0000:00 0 0 0 0 0 0 0 00

b \P P F b b P b h p b b b p P * F LIL L L p h Ly h b b P P h b r b °\°o

 

1$m

 

 

$9

 

 

- $3

 

pJBZBH JSAOUJ n1

 

- $0N

$m~

 

 

 

 

$0M

 

232

003000 .00 b08080 0:080:0m I R 00:30

 

 

.0>0c.:k

 

 

_0>04
032205

II'II'I"|IIII"IU

Umtm
_0>0..-000.o

_0>04
E000

H

A05; 00
:0.

 

 

005000800
030050800

 

 

 

 

 

   

 

x0w.0m<.000m

 

 

 

 

 

3:30.

 

 

l||'||l|

 

 

030m
.0>0c.:k

 

 

 

 

 

30.030
0200000800

 

 

$0.020
.0000

'II'I'III l|'l|

 

 

 

 

000.08.00.00

 

 

 

233

Appendices

234

Appendix A - IPIP items used in construct validity study.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Item Scale Reverse
Coded?
Am the life of the party. Extraversion No
Feel little concern for others. Ageeableness Yes
Am always prepared. Conscientiousness No
Don’t talk a lot. Extraversion Yes
Am interested igeople. Agreeableness 'No
Leave my belongings around. Conscientiouesness Yes
Feel comfortable aroundJ3eople. Extraversion No
Insult people. Agreeableness Yes
Pay attention to details. Conscientiouesness No
Keep in the background. Extraversion Yes
Sympathize with others’ feelings Agreeableness No
Make a mess of things. Conscientiouesness Yes
Start conversations. Extraversion No
Am not interested in other people’s problems. Agreeableness Yes
Get chores done right away. Conscientiouesness No
Have little to say. Extraversion Yes
Have a soft heart. Agreeableness No
Often forLet to put things back in their proper place. Conscientiouesness Yes
Talk to a lot of different people at parties. Extraversion No
Am not really interested in others. Agreeableness Yes
Like order. Conscientiouesness No
Don’t like to draw attention to myself. Extraversion Yes
Take time out for others. Agreeableness No
Shirk my duties. Conscientiouesness Yes
Don’t mind being the center of attention. Extraversion No
Fee] others’ emotions. Agreeableness No
Follow a schedule. Conscientiouesness No
Am quiet around strangers. Extraversion Yes
Make people feel at ease Agreeableness No
Am exacting in my work. Conscientiouesness No

 

235

 

Appendix B — Factor analysis results for job performance and skill area ratings

 

Factor Titles and Loadin s

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Working Team and Customer Attendance,
Conscien- Interper- Focus Tardiness, and
tiously & sonal Dependability
Taking Skills
Initiative
Takes initiative/is self .76 .28 .05 .13
directed
Requires Close Supervision .73 .20 .18 .13
Completes Assigned Tasks .68 .07 .16 .26
Goes Beyond Required . 67 .32 .17 .17
Duties
Ability to Follow . 61 .47 .09 .14
Procedures
Follows Polic1es & .60 .11 .28 .20
Procedures
Makes Careless Mistakes .60 .08 .21 .10
Communication Skills .19 .71 .25 .05
Flexibility .04 .64 .09 .39
Leadership Characteristics .48 .63 .08 .03
Team Skills .42 .62 .16 .27
Ability to Learn Procedures .54 .55 .03 .06
Customer Perspective .33 .54 .49 .03
Agreeableness .15 .51 .41 .3 8
Stability .39 .40 .31 .34
Pleasant to Customers
Is Rude or Impatient with .23 .16 .78 .04
Customers
Does Not Get along With .14 .00 .77 .13
Coworkers
Meets Customers Needs .39 .24 .63 .05
Does not Get Along With .01 .19 .51 .30
Coworkers
Misses Work .15 .07 .11 .81

 

 

 

 

236

 

 

Appendix C — Intercorrelations and internal consistencies for the performance composites

 

 

 

Variable

Performance Measure 1 2 3 4 _5_ _6_
1. Working. Consc1entiously and (.91) .47 .70 .3 9 .77 . 91
Taking Initiative
2. Customer Focus (.82) .48 .26 .44 .61
3. Team and Interpersonal Skills (.86) .39 .78 .89
4. Attendance, Tardiness, and
Dependability ('86) '45 '57
5. Overall Job Performance Rating -- .83
6. Performance Composite Average (.95)

 

Notes: Internal consistency estimates are on the diagonol. N = 843 — 867. All
correlations are significant at p < .05.

237

.Amo. - S. .3 0800000000

00000 _0000 0005 0003000 0002:0000 00.0 9000000003 03303 00000020000000 w:_0: 00003200 003 00000 _0000 ED 00:. 0500:0800

0055000000 000.0 05 .00 8000:3800 00:: 0005003005. 0 00 0:000:80 00:0:0000000 .3. v 0 * .0: I vow M Z ”00002

 

 

 

 

 

 

 

 

 

2. *3, S. *3. *2. *2. 00020000000050

*mm. *2. *2. *NH. *om. 1m. 05:00.

* 0.0.. *2. *oo. .32. *2. *2. 000000000000 E000.

*0 0. *N _. mo. .52. *3. *0 _. 8000:0000 00300m

*3. .00... *3. *2. * 0m. *3. 0000000500 20005
00000005 EU
*wm. im. #2. 0mm. 0.0.0.. 1“. 008m 0000. EU

3503:0000 0:000 23% 02005:: $200 0%
0000000000000 00000000000 0% 00050000. 008000000 080000005 3032050000000 000< :2m 02. 0:0
0:000:50 :000>O 000000000000. 000005000000 02.
00:0:0000000 0:0 E000. 00:00:03
0:205:05

 

 

 

 

 

 

 

 

02000 EU 00.0 00000005000 bmcz0> .00 b08800 I Q 5000000.

238

          

NSTATE UNIV VERS SITY LIBR

III I I I MIIII III I

3 1293 02327 0915