x .1 .
=. E: u: ((53
J< 11.3.?! 3.7!.)
(4‘. 2.12.... 3. {HG-.4:

.3 .vrfuztii‘i.

2.2.2
w?t:<{£v..<u_¢~::flt (.35:
.3.) K— : 254

.
35325357 55%;...
a .s. I; r .

I.
:2... 195.431, u...€€£5}é (43
3 53).: a: a. {£243.}.

4

~._7....r€.}-.:. PE.- in!!!)

3...}.15.’ €3..41(¢..lt:}€:1l§1{
4.34.3.1. ~l.-:(£.i§x..l rrlffir

. :(r
1 . 13:45.6:

 

 

12.!) .2? I! I; .II; .3 a 1
1:1, #2,}?! it!!! ‘

5:13,.{1573115

13:531.?) (53.4!!!
14115.! .tIz-lfiwsidllg
.25., .5... A) {:21}. f. r...

 

 

                                                                                                                        

00785 1698

This is to certify that the

dissertation entitled

Reward Contingency and
Functional Turnover

presented by

Charles Robert Williams

has been accepted towards fulfillment
of the requirements for

Ph.D. Business Administration

degree in

 

 

(Z/ /c ”ZZZ/”L

Major professor

 

Date S’///90

MSU is an Affirmative Action/Equal Opportunity Institution 0- 12771

 

LEERARY
Michigan State
‘ ’ 1 «fly

¢..

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.

DATE DUE DATE DUE DATE DUE II

4
fl
VJ

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

=1

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU Is An Aflinnetive Action/Equal Opportunity Institution
czwmmpma-OJ

 

 

 

 

 

 

REWARD CONTINGENCY AND FUNCTIONAL TURNOVER

By

Charles R. Williams

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Management
Graduate School of Business

1990

i@

1 3‘2

veto

ABSTRACT
REWARD CONTINGENCY AND FUNCTIONAL TURNOVER
BY

Charles R. Williams

This field study investigated the relationship between reward-
contingency and functional turnover, which separates turnover into
four categories: poor performing leavers, good performing leavers,
poor performing stayers, and good performing stayers. The study was
conducted across four companies, each of which rewarded its sales
staff according to a different level of objective reward contingency,
from mostly salary, to salary plus commission and bonuses, to straight
commission. It was predicted that perceptions of reward contingency
would be related to pay satisfaction and job satisfaction differences
such that better (poorer) performers would be more (less) satisfied
and more (less) likely to stay under conditions of strong (weak)
reward contingency. Discriminant analysis indicated that objective
reward contingency, education, and tenure accounted for 56% of the
variance across the four functional turnover groups. Subsequent
analyses showed that objective reward contingency accounted for most
(34%) of this variance. Perceived reward contingency, pay
satisfaction, job satisfaction, age and gender did not contribute to
significant differentiation of the four functional turnover groups.
Linear classification functions, derived from the discriminant

analysis, indicated that objective reward contingency alone produced

Linear classification functions, derived from the discriminant
analysis, indicated that objective reward contingency alone produced
an overall classification hit rate of 58.16% when predicting
functional turnover group membership. However, examination of
separate-group hit rates indicated that the significant overall level
of classification resulted from accurate classification of good and
poor performing stayers. Good and poor performing leavers were not
classified accurately, and were most often misclassified as good
performing stayers. These classification results were replicated
using a sample of nonresponders (n = 128) and then cross-validated and
double cross-validated. While there was some shrinkage in
classification accuracy in these additional analyses, overall
classification was still better than chance in each case, and the
pattern of separate-group hit rates was unchanged. Together, these
analyses indicate that objective reward contingency does no more than
separate good performers from poorer performers. Thus, while
objective reward contingency was related to performance, it was not

related to each of the four functional turnover groups.

To Jenny.

iv

 

 

ACKNOWLEDGEMENTS

I wish to express sincere thanks to my dissertation chairman,
.Dr. John Hollenbeck, for his guidance and constructive suggestions in
the preparation of this dissertation. I also would like to thank
committee members Drs. Daniel Ilgen and Neal Schmitt for their
valuable comments and suggestions.

Thanks, also, to my father, Dr. Robert M. Williams, for setting
the example twenty years ago; to my mother, Mary Williams, for her
loving counsel; and to my in-laws, Irvin and Clara Long, for the
frozen care-packages that kept us going at the end of each month.

Finally, I wish to thank my wife, Jenny, for her encouragement,
love, and patience, and my children, Benjamin, Rebecca, and Zachary,

for teaching Daddy more about the practice of management than he could

ever learn in school.

TABLE OF CONTENTS

LIST OF TABLES .................................................... ix
LIST OF FIGURES ................................................... xi
CHAPTER

ONE INTRODUCTION ........................................... l

Turnover vs. Functional turnover:
A performance Difference ........................ 5

Controlling Functional Turnover

Through Employee Attitudes ...................... ll
TWO LITERATURE REVIEW ...................................... 15
Expectancy Theory and Reward Contingency ..... 15

Objective and Perceived
Reward Contingencies ...................... 18

Performance, Satisfaction, and

Reward Contingency ........................ 21
Satisfaction and Functional Turnover ......... 24
The Entire Model ............................. 25
Hypotheses ................................... 28

THREE METHOD ................................ 30
Sample and Procedure ......................... 30
Measures ..................................... 35

Performance ............................. 35

Objective Reward Contingency ............ 38

Perceived Reward Contingency ............ 40

Pay Satisfaction ........................ 43

Job Satisfaction ........................ 43

vi

Functional Turnover ..................... 44

Data Analysis, Power Analysis,

Restatement of Hypotheses .................. 46

Nonequivalent Control Groups ............ 46

Study Design ............................ 47

Hypothesis Testing ...................... 48

Discriminant Analysis ................... 49

FOUR RESULTS ............................... 52
Descriptive Statistics ....................... 52
Nonequivalent Control Groups ............ 52

Intercorrelations Among

Study Variables ....................... 57

Study Design Hypotheses ...................... 57
Null Hypothesis 1 ....................... 57
Null Hypothesis 2 ....................... 61
Null Hypothesis 3 ....................... 63
Null Hypothesis 4 ....................... 73
Standardization ......................... 76

Null Hypothesis 4, Unstandardized
Job Satisfaction and Unstandardized

Pay Satisfaction ...................... 84
Evidence for Study Design ............... 89
Model Testing Hypotheses ..................... 89
Interpreting Discriminant Analysis ...... 9O
Null Hypothesis 5 ....................... 92
Null Hypothesis 6 ....................... 92
Discriminant Functions .................. 101

Analysis of Variance and
Orthogonal Contrasts ................. 101

Classification Analyses ................. lO7

vii

FIVE DISCUSSION ............................

Future Research and Theory ...................

External Equity ..............................

Rewards Other Than Pay .......................
Internal Job Opportunities ..............
Expected Utility ........................
Job Design ..............................

Performance and Turnover .....................

Controllability of Turnover ..................

Field Experiments ............................

REFERENCES ........................................

APPENDIX ..........................................

viii

116

120

120

124

124

125

126

128

130

131

133

143

LIST OF TABLES

TABLE

1 Sales Example for Computing
Functional Turnover ..................... 9

2 Descriptive Statistics by Organization
and for the Final Sample ................ 53

3 Intercorrelations among
study variables ......................... 58

4 Mean Organizational Response
Distributions for Perceived Reward
Contingency by Performance Level
and Pay Level ........................... 64

5 Moderated Regression Analysis on
Standardized Pay Satisfaction ........... 74

6 Moderated Regression Analysis on
Standardized Job Satisfaction ........... 75

7 Moderated Regression Analysis on
Unstandardized Pay Satisfaction ......... 85

8 Moderated Regression Analysis on
Unstandardized Job Satisfaction ......... 88

9 Discriminant Analysis Testing
The Confounded Effects of
Organization and Reward Contingency ..... 93

10 Discriminant Analysis Testing
The Mediated Effects of Pay
Satisfaction and Job Satisfaction
on Functional Turnover .................. 96

11 Discriminant Analysis Testing
The Unique Effects of Perceived
Reward Contingency and Objective
Reward Contingency ...................... 99

12 Group Centroids in the
Discriminant Space ...................... 102

ix

13

14

15

16

17

18

19

20

Structure Coefficients Between
Discriminant Functions and

Discriminating Variables ................

Univariate Analyses of Variance .........

Means, Standard Deviations,
and Orthogonal Contrasts Between

Functional Turnover Groups ..............

Classification Analysis of the
Original Sample (n - 98) Derived
from a Direct Discriminant
Analysis of All Discriminating

Variables ...............................

Classification Analysis of the
Original Sample (n - 98) Derived
From a Discriminant Analysis

Using Objective Reward Contingency .......

Classification Analysis of the
Nonresponders (n = 128) Derived
From a Discriminant Analysis Using

Objective Reward Contingency .............

Cross-Validated Classification
Analysis of Nonresponders (n = 128)
Based on Linear Classification
Functions Derived from the Original
Sample (n - 98) Using Objective

Reward Contingency .......................

Double Cross-Validated
Classification Analysis of the
Original Sample (n = 98) Based
on Linear Classification Functions
Derived from the Sample of
Nonresponders (n = 128) Using

Objective Reward Contingency ..............

103

104

105

108

111

112

113

114

FIGURE

1

10

LIST OF FIGURES

Reward Contingency Model of

Functional Turnover .....................

Perceived Correlations Between
Pay and Performance,

Company A, n - 8 ........................

Fisher's 2’ Transformation of
Perceived Correlations Between
Pay and Performance,

Company A, n = 8 ........................

Perceived Correlations Between
Pay and Performance,

Company B, n - 20 .......................

Fisher’s 2' Transformation of
Perceived Correlations Between
Pay and Performance,

Company B, n - 20 .......................

Perceived Correlations Between
Pay and Performance,

Company C, n a 17 .......................

Fisher’s 2’ Transformation of
Perceived Correlations Between
Pay and Performance,

Company C, n = 17 .......................

Perceived Correlations Between
Pay and Performance,

Company D, n = 53 .......................

Fisher’s 2' Transformation of
Perceived Correlations Between
Pay and Performance,

Company D, n = 53 .......................

Regression of Unstandardized
Pay Satisfaction on Past Performance
by Perceived Reward Contingency

Interaction .............................

xi

17

65

66

67

68

69

70

71

72

87

CHAPTER ONE

INTRODUCTION

The study of turnover by both micro (i e., industrial-personnel
psychologists, and behavioral scientists) and macro (i.e., economists
and sociologists) scholars has been extensive over the last thirty
years. At the micro level of analysis, turnover research is rooted in
the study of job satisfaction. Early reviews of that literature
(Brayfield & Crockett, 1955; Herzberg, Mausner, Peterson, & Capwell,
1957; Vroom, 1964) were consistent in establishing a negative
relationship between job satisfaction and turnover. In their review,
Brayfield and Crockett (1955) stated, "One principal generalization
suffices to set up an expectation that morale should be related to
absenteeism and turnover, namely, that organisms tend to avoid those
situations which are punishing and to seek out situations that are
rewarding" (p. 415). The existence of well over 1000 studies
addressing the topic of turnover (Steers & Mowday, 1981) attests to
this fact. This large interest in turnover by micro researchers is
probably a function of turnover being a relatively clear-cut
individual choice behavior (Mobley, Griffeth, Hand, & Meglino, 1979),
and of the critical consequences it possesses for both individuals and
organizations (Porter & Steers, 1973).

Over time, the definition of turnover has evolved from a singular
notion of quit/stay to the broader notion that, just as there are many

ways to measure job satisfaction or absenteeism, there are also

2
different, yet valid measures of turnover. The first dimension along
which measures of turnover have differed is the level of analysis.
Turnover has been studied at the level of the individual (Porter &
Steers, 1973), occupation or work group (Price, 1977), organization
(Abelson & Baysinger, 1984), industry (Armknecht & Early, 1972; Burton
& Parker, 1969), and at the national level (Eagly, 1965). While the
existence of different measures of turnover at different levels of
analysis demonstrates the multiplicity of views from which turnover
can be studied, it also points to the need for researchers to specify
the kind of turnover measure used in their studies (Muchinsky &
Tuttle, 1979). This must be done so as not to confound study results
across different levels of aggregation (i.e., an ecological fallacy,
Robinson, 1950), a mistake which occurs frequently in relating labor
market data to individual perceptions of job opportunities and
subsequent turnover behavior (Hulin, Roznowski, & Hachiya, 1985). It
is also important to distinguish among different levels of turnover
because our ability to build turnover theory--to predict and
understand turnover--suffers when we treat turnover as a generic
construct (Van der Merwe & Miller, 1971).

The differentiation of voluntary and involuntary turnover was
another major change in the conceptualization of turnover.
Theoretically, this separation became necessary because of the
difference in who initiates the separation, the individual or the
organization (Muchinsky & Tuttle, 1979). Empirical research also
indicates that voluntary and involuntary turnover may have different
antecedents or correlates. Involuntary turnover is better predicted

by aptitudes and performance (MacKinney & Wolins, 1959; McEvoy &

3
Cascio, 1987; Wanous, Stumpf, & Bedrosian, 1979; Wells & Muchinsky,
1985), whereas voluntary turnover is better predicted by individual
attitudes such as job satisfaction, organizational commitment, and
intention to quit (Steel & Ovalle, 1984).

The most recent change in the way turnover is viewed
incorporates the consequences of turnover, that is, the extent to
which voluntary turnover is actually a problem for organizations
(Dalton, Krackhardt, & Porter, 1981; Macy & Mirvis, 1983; McEvoy &
Cascio, 1987). Traditionally, scientists and practitioners have
concerned themselves solely with the rate of turnover. Moreover, it
has been assumed that the consequences of turnover are inherently
negative, that frequent turnover is expensive, and that the rate of
turnover must be reduced. For example, McEvoy and Cascio (1985) have
called for organizational researchers to conduct field experiments
comparing interventions designed to directly reduce the rate of
turnover in organizations.

The alternative perspective is that some kinds and levels of
turnover are actually beneficial for organizations. Several authors
(Abelson & Baysinger, 1984; Dalton, et. a1., 1981; Dalton, Todor, &
Krackhardt, 1982; Mobley, 1982; Muchinsky & Tuttle, 1979; Porter &
Steers, 1973; Shaw, 1980) have argued that turnover is not necessarily
a problem and that treating it as such leads to an overestimation of
the negative consequences of this behavior for organizations.
Instead, turnover is actually desirable in many situations: for
example, (1) when highly paid senior persons leave and are replaced
with lower paid new hires; (2) when organizational contributions to

pensions, which are tied to seniority, revert to the organization if

4.
the employee leaves before they have a vested right to the money; (3)
when poor performers can be replaced by better performers; or (4) when
it creates opportunities for upward mobility (Jeswald, 1974).

Thus, from this perspective, the concern is no longer with the
rate of turnover, but with the nature of turnover. For example,
Dalton and his colleagues (1981) separate turnover into functional
turnover (i.e., an individual who has been evaluated negatively by the
organization leaves) which benefits the organization, and
dysfunctional turnover (i.e., an individual who has been evaluated
positively by the organization leaves) which does not. In a study of
bank teller turnover, Dalton et a1. (1981) found that 42% of all
voluntary turnover was actually beneficial (i.e., poor performers
left) for the organization. Similarly, Hollenbeck & Williams (1986)
found that just over half of voluntary turnover among retail sales
clerks was functional. Thus, in those studies the traditional measure
of turnover (quit/stay) resulted in incorrect assessments of the
negative consequences of turnover.

This dissertation extends previous research on the consequences
of turnover by empirically investigating the relationship between
performance contingent rewards and functional turnover. To start, the
differences between functional turnover and the traditional measure of
turnover will be explicated. It will then be argued that functional
turnover gives a much better account of the extent to which turnover
is costly and, therefore, an organizational problem. Finally, the
argument is made for investigating the relationship between

performance contingent rewards and functional turnover.

5

Turnover vs. Functional Turnover: A Performance Difference

The traditional measure of voluntary turnover simply denotes
whether an employee stayed with or left the organization of their own
volition. According to the human resource accounting literature
(Cascio, 1982; Flamholtz, 1974; Schmidt & Hoffman, 1973; Smith &
Watkins, 1978), turnover is costly to organizations because of the
administrative and replacement costs (i e., separation pay,
unemployment and social security taxes, recruitment, selection,
training, medical exams, etc.; Cascio, 1982) associated with each
employee separation. Published estimates of replacement and
administrative costs vary from $1,000 (Gustafson, 1980), to $2,522
(Mirvis and Lawler, 1977), to $14,609 (Boudreau, 1981) depending on
the job. Reduction in turnover is desirable if it directly reduces
these costs.

Administrative and replacement costs, however, only represent a
portion of the cost of turnover. In many cases, performance should
also be considered a major cost of turnover (Boudreau, 1983; Boudreau
& Berger, 1985; Dalton et. a1, 1981; Dalton et. a1., 1982).

Performance becomes an important cost if we assume that organizations

incur an opportunity performance cost or realize an opportunity

 

performance gaip each time an employee leaves the organization. In
its traditional economic usage, opportunity cost is the difference
between what is earned on an investment and what could have been
earned had the money been invested elsewhere. In terms of turnover or
human capital, opportunity performance costs or gains represent the

difference between what the organization "earned"--the performance

6
contributed by the employee remaining with the organization——and what
the organization "could have earned" through the performance of
another employee.

Presently, there are two ways to measure the opportunity costs
of turnover. The first, functional turnover (Dalton and Todor; 1979;
Dalton et. a1., 1981, 1982), compares the performance of leavers to
the performance of stayers. For example, Hollenbeck & Williams (1986)
compared the performance of leavers to the average performance of
incumbents (i.e., stayers). Turnover was dysfunctional when leavers
were above average performers and functional when they were below
average performers. However, Boudreau (1983) and Boudreau and Berger
(1985) argue that, "Research should focus on relationships between
retainee utility and replacement utility, rather than on stayer and
leaver performance," (p. 610). Thus, turnover flow addresses the
performance difference between leavers and their potential
replacements. Positive turnover flow, an opportunity gain, occurs
when those who stay with the organization are better performers than
their potential replacements, or when leavers are poorer performers
than replacements. On the other hand, negative turnover flow, an
opportunity cost, exists if stayers are poorer performers than
potential replacements, or if leavers are better performers than
replacements.

In many situations, however, the differences between functional
turnover and flow are inconsequential. For example, the mean
performance level of incumbents (i.e., stayers) will be similar to the
mean performance level of replacements if the applicant pool is

constant from year to year, if the organization's recruitment and

7
selection techniques are constant, and if job learning curves are
short (Hollenbeck & Williams, 1986). Indeed, Boudreau and Berger
(1985) concede that incumbent samples are often used as proxies when
estimating the characteristics of applicant populations in utility
research.

Regardless of how they are operationalized, there are two basic
strategies that can be used to minimize the performance-related
opportunity costs associated with turnover. Perhaps the most
frequently used technique is a selection strategy, attempting to
replace those who leave with better performing replacements.

"Even when separators are good performers, it may be best

to allow them to separate because their replacements are

even more valuable. When those retained are only slightly

less valuable than the previous work force, an

organization may find it more cost effective to improve

its recruiting, selection, placement, or training

procedures so that higher quality replacements are

obtained, rather than devoting resources to change the

pattern of turnover," (Boudreau & Berger, 1985, p. 610).

By validating a battery of selection tests against turnover in
addition to performance, a selection strategy might also reduce the
rate of turnover. For example, weighted application and biographical
information blanks, which are highly predictive of both performance
and turnover (Cascio, 1982; Muchinsky & Tuttle, 1979), might be used
to select employees who will be better performers and who are less
likely to quit.

An alternative to the selection approach--and the focus of this
research-~is a retention strategy. Thus, rather than ensuring that
newly hired replacements will be better performers than those who are

leaving the organization, efforts are made to encourage good

performers to stay with the organization. (Hereafter, a description

8
of the "better", "poorer", "good" or "bad" performance of stayers and
leavers represents an explicit comparison to the average performance
of incumbents). In other words, attempts are made to purposively
affect the nature or "pattern of turnover" in the organization. As
noted by Porter & Steers (1973, p. 169), "The loss of an effective
employee may cost far more than the loss of an ineffective one, and
the costs of efforts to retain the latter may well exceed the
benefits."

Table l, which shows the performance and turnover data for a
hypothetical sample of salespersons, illustrates how performance and
turnover can be combined into a measure of functional turnover. The
first step is to compute a performance deviation score for each
individual. This is done by subtracting the average level of
performance exhibited by incumbents from each individual's performance
score. For example, the poorest performer gets a performance
deviation score of -30,000, while the best performer gets a deviation
score of +30,000. Thus, the performance deviation score is analogous
to the opportunity cost or gain associated with individual turnover.

The second step combines each individual’s performance deviation
score with the additional consideration of whether the individual
stayed with or left the organization. For example, the organizational
outcome is dysfunctional when a poor performer stays (-30,000 x +1),
or when a good performer leaves (+30,000 x -1). Similarly, it is
positive when a poor performer leaves(-20,000 x -l = 20,000), or when
a good performer stays, (10,000 x +1 = 10,000).

While the above example (e g. Table 1) demonstrates the

potential performance consequences of functional turnover, it is also

9

Table 1: Sales Example for Computing Functional Turnover

 

 

Widgets Deviation from X Turnover = Functional
Average Performance Turnover
70,000 -30,000 +1 -30,000
80,000 -20,000 -1 +20,000
90,000 -10,000 +1 ~10,000
100,000 -0- -l -0-
110,000 +10,000 +1 +10,000
120,000 +20,000 -1 -20,000
130,000 +30,000 -1 -30,000
700,000 -60,000

 

10
useful in explicating another important point. The standard deviation

of the dollar value of job performance (i.e. SD which is the standard

y.
deviation of performance measured in dollars; Cronbach & Gleser, 1965;
Schmidt, Hunter, McKenzie & Muldrow, 1979) plays a large role in
determining the utility or dollar consequences of functional turnover.
To illustrate, if a widget in the above example sells for 50 cents,
then SDy for this job is $10,000 ($.50 x the s.d. of 20,000), and the
overall performance cost of turnover to the I. M. Widget company is
$30,000 ($.50 x -60,000). If the recruiting, administrative, and
training costs for a new hire approximate $10,000 per person, a total
of $40,000 for four quits, then the I. M. Widget company should be
more concerned with rate of turnover, as opposed to the nature of
turnover (i.e., functional turnover). Therefore, in those jobs where
SDy is small, researchers and practitioners should not be overly
concerned with functional turnover because even if the best employees
are leaving, the small variance in the dollar value of performance
reduces the dollar loss in performance due to turnover. When SDy is
small, efforts should be directed toward a reduction in turnover
because administrative and replacement costs will compose the largest
portion of turnover costs.

But where SDy is large, functional turnover becomes more critical
because the performance costs of turnover can easily exceed
administrative and replacement costs. For example, if a widget sold
for $3, then SDy would be $60,000 ($3 x the s.d. of 20,000), and the
performance costs of turnover would be $180,000 ($3 x -60,000),
greatly exceeding the recruitment, training and administrative costs

of simple turnover.

11

In sum, there are several reasons for researchers to study
functional turnover in addition to the traditional stay/quit measure
of turnover. Most importantly, functional turnover demonstrates both
the positive and negative consequences associated with turnover
because it incorporates performance as an additional, yet major cost
of turnover. Therefore, it gives a truer account of the extent to
which turnover constitutes an organizational problem. Finally, in
cases where SDy is large, functional turnover can obviously have an
impact on organizational effectiveness. Given these considerations,
it is imperative that researchers identify and test potential
interventions which can influence functional turnover.

Controlling Functional Turnover Through Employee Attitudes

In identifying strategies for controlling functional turnover,
it must be remembered that functional turnover is a much more complex
variable than turnover. Functional turnover combines two behaviors,
turnover and performance, which have been addressed separately in
theory and empirical research (March & Simon, 1958). Further, the
goal is not necessarily a reduction in the rate of turnover (though
this is a legitimate goal if the turnover is functional), the goal is
voluntary separation of poor performers and voluntary retention of
better performers. (No doubt some organizations influence functional
turnover through involuntary turnover by firing or threatening to fire
employees whose performance is substandard. However, many
organizations exercise little direct control over involuntary
retention decisions because of labor agreements (Bureau of National

Affairs, 1982); threats of litigation; use of arbitration which often

 

12
results in reinstatement of approximately half of all who appeal
(Jenning & Wolters, 1976); or because of possible negative effects on
organizational culture, recruiting, or employee morale). Since
voluntary turnover is an individual choice behavior (Mobley et a1.,
1979), the general strategy for achieving positive functional turnover
would be to differentially affect the antecedents of voluntary
quit/stay decisions and behavior for better versus poorer performers
so that poorer performers choose to leave and better performers choose
to stay.

In line with theory on individual turnover (March & Simon, 1958;
Mobley, 1982), employee attitudes consistently predict voluntary
turnover, though the average variance explained is typically small, (R2
g .20; Mobley et a1., 1979; Mowday, Porter, & Steers, 1982). Meta-
analytic results confirm that turnover is best predicted by intentions
to leave (corrected mean r = .45), and attitudes towards the
organization and the job (i e., organizational commitment, corrected
mean r = -.36, work satisfaction, corrected mean r = -.28, and overall
satisfaction, corrected mean r = -.25; Steel & Ovalle, 1984). In
accumulating results across studies, McEvoy and Cascio (1985) showed
that attitude-based interventions, such as job enrichment (corrected
mean r = -.l72) and realistic job previews (corrected mean r = —.087),
can be expected to reduce the pgpp of turnover. Hulin (1968) also
found that increases in satisfaction with pay and promotion reduced
the rate of turnover among female clerical workers from 30% to 12%
over one year.

Though capable of reducing turnover, the ability of attitude-

based interventions--or any other intervention--to influence

l3
functional turnover depends on several factors: (1) the relationship
between attitudes and turnover, (2) the relationship between attitudes
and performance, and (3) the relationship between turnover and
performance. Given the well established negative relationship between
turnover and attitudes, such as job satisfaction and organizational
commitment (Steel & Ovalle, 1984), there are several ways that
attitudinal interventions might influence functional turnover.

First, if attitudes were positively related to performance,
functional turnover would result. In this way, individuals with
positive attitudes toward their jobs would be more likely to stay and
would be better performers than those who left. However, reviews of
this literature generally show that the attitudes which predict
turnover do not predict performance. Iaffaldano and Muchinsky (1985)
estimated the true population correlation between performance and
satisfaction to be .17, a value that was not significantly different
from zero. Mowday, Porter, and Steers (1982) concluded that there is
a weak relationship between commitment and job performance, and that a
strong relationship should not be expected as commitment is but one of
many determinants (e.g., ability, situational constraints and role
clarity) of performance. Thus, it seems unlikely that attitude-based
interventions (Mirvis & Lawler, 1977), which attempt to improve
employee attitudes toward their work or organization, would result in,
positive functional turnover.

Secondly, attitudes might be related to functional turnover if
turnover was negatively related to performance. McEvoy and Cascio
(1987) report that the correlation between performance and voluntary

turnover, after weighting for sample size, was - 24, suggesting that

14
voluntary turnover occurs more often among poor performers than better
performers. However, McEvoy and Cascio based this result on only 6 of
the 13 studies which they cited as reporting a correlation between
performance and voluntary turnover, thus creating the possibility of
second order sampling error (Schmidt, Hunter, & Raju, 1988). A re-
analysis (Williams & Livingstone, 1989) including all 13 of McEvoy and
Cascio's studies, plus 13 new performance-turnover correlations,
produced a very similar sample-sized weighted average correlation of -
.22. Because the heterogeneous confidence intervals (i.e., the
credibility intervals ranged from -.60 to .03) did not include zero, -
.17 to -.27, this reanalysis also found that the negative relationship
between performance and turnover was significantly different from
zero.

Since past research indicates that stayers are typically better
performers and more satisfied with their jobs, one way of influencing
functional turnover would be to bring about a positive relationship
between satisfaction and performance. Functional turnover should
result as better (poorer) performers would be more (less) satisfied
and thus even more likely to stay (quit). One means of producing a
positive relationship between these variables, the use of performance

contingent rewards, is described next.

CHAPTER TWO

LITERATURE REVIEW

Expectancy Theory and Reward Contingency

One way to control voluntary functional turnover is to
differentially influence the quit/stay decisions of good and poor
performers so that good performers stay and bad performers leave.
Thus, the first step is to choose a major antecedent of quit/stay
decisions. As discussed above, research has shown that one method of
influencing turnover behavior is to enhance job satisfaction (Hulin,
1968; McEvoy & Cascio, 1985). However, the previous discussion also
makes clear that any attempt to influence functional turnover through
attitude-based interventions will fail unless there is differential
enhancement of satisfaction for high performers versus low performers.
Under the assumption that job satisfaction is negatively correlated
with turnover, better performers must be satisfied with their jobs,
while poorer performers must be dissatisfied, if functional turnover
is to result. The valence model of expectancy theory offers a
theoretical explanation as to how differential satisfaction can occur
across these groups by showing how reward contingency can be linked to
both components of functional turnover, performance and turnover.

Expectancy theory (Porter & Lawler, 1968; Vroom, 1964) is a
theory of individual choice which consists of two models: (1) a

valence mode1--a valence is the attractiveness or anticipated

15

l6
satisfaction of some outcome--that attempts to predict the rank order
or desirability of a set of outcomes, typically job satisfaction,
occupational preference, or the valence of good performance; and (2) a
behavioral choice model, which attempts to predict an individual's
motivational force towards some act (Mitchell, 1974).

With respect to job satisfaction, the valence model of
expectancy theory states that, "the worker's satisfaction with a job
or anticipated satisfaction with an occupation results from the
instrumentality of the job for attaining other outcomes and the
valence of those outcomes," (Mitchell, 1974, p. 1054). Therefore, if
a job offers valued outcomes, and the attainment of those outcomes is
contingent or instrumental on remaining with the organization, then
the choice of staying on the job will be an attractive or potentially
satisfying one. Again, however, the goal is to make the job much more
satisfying for better performers than poorer ones, assuming that
differential satisfaction will then lead to differences in turnover
across these groups. Thus, rather than making the attainment of
valued outcomes contingent on organizational membership, they must be
contingent on performance.

Therefore, according to the valence model of expectancy theory,
reward contingency can be linked to both components of functional
turnover, performance and turnover, as shown in Figure 1. When
rewards are contingent on performance, better performers will be
rewarded more often and will be more satisfied with their jobs than
poorer performers who will be rewarded less often. This hypothesis,
however, must be modified somewhat in light of research concerning an

important boundary condition of the theory; namely, that for

17

Figure l: Reward Contingency Model of Functional Turnover

OBJECTIVE
REWARD CONTINGENCY

PERCEIVE
REWARD CONTINGENCY

PAST 1, ,PAY & JOB .‘FUNCTIONAL
PERFORMANCE ’SATISFACTION ‘rTURNOVER

18

expectancy theory to work, workers must perceive clear performance
reward contingencies or instrumentalities (Dachler & Mobley, 1973;
Graen, 1969; Kopelman, 1976; Kopelman & Thompson, 1976). However, it
is also expected that "objective" reward contingencies existing in the
work place will influence perceived reward contingencies, hence, the
causal arrow from objective reward contingencies to perceived reward
contingencies. Thus, restated, it is hypothesized that perceived
reward contingency will moderate the relationship between individual
performance and job satisfaction such that performance and
satisfaction will have a stronger positive relationship under
conditions of higher perceived reward contingency than in conditions
of perceived lower reward contingency where the relationship will be
weaker. If this hypothesis holds true, then satisfaction will be
related to functional turnover, as actual turnover will be higher
among the less satisfied poor performing workers than among the more
satisfied better performing workers.

In the following sections, the literature pertinent to each link

in the model is reviewed.

Objective and Perceived Reward Contingencies

According to Miner (1980), several studies have been critically
important in establishing that for expectancy theory to "work",
workers must perceive a strong relationship between performance and
valued outcomes.

In the first of these studies, Graen (1969) conducted an
experiment where women were hired on a temporary, part-time basis to

perform clerical tasks. There were two experimental groups and a

19
control group, with the experimental manipulation taking place after
the first and before the second work session. The first experimental
group received verbal praise or recognition contingent on prior
performance, while the second experimental group received a pay
increase "in the hope that you will do much better." Thus, in the
first condition rewards were contingent on performance, while in the
second they were contingent on membership, on being a job incumbent.
Graen summarized the results of his study as follows:

"Only in the achievement feedback treatment was the
contingency between effective performance and the
attainment of a favorable role outcome established in a
concrete manner--presenting the achievement feedback
contingent upon effective performance...Therefore, an
important boundary condition for instrumentality theory is
that contingencies must be established in a concrete
manner between effective job performance and attaining
favorable role outcomes," (Graen, 1969, p. 19).

A field study conducted by Dachler and Mobley (1973) in two
manufacturing organizations also demonstrated this boundary condition

for expectancy theory.

"The fact that the pattern of results obtained from Plant
1 employees supported every prediction derived from the
present VIE model, whereas the pattern of results from
Plant 2 employees in nearly all cases did not bear out
these predictions, provides a strong argument for the
possibility that certain organizational characteristics in
Plant 2 acted as boundary conditions ...Since the piece-
rate system used in Plant 1 specified the contingencies
between levels of performance and obtained salaries, the
perceived relationship between different levels of
performance and money-related outcomes should be more
clearly defined for employees in Plant 1 than for
employees in Plant 2, who were not exposed to any
incentive plan.... the organizational conditions of Plant
1 were more structured, less complex, and more
"assessable" so that it was easier for employees to form
accurate perceptions about the organizational conditions
and contingencies as well as to accurately perceived how
much they could actually achieve in the existing
situation....These interpretations are consistent with the

20

finding that 36 out of 45 performance-outcome

contingencies were stronger in Plant 1 ...," (Dachler &

Mobley, p. 415).

Other multi-company research conducted by Kopelman (1976), and
Kopelman and Thompson (1976) finds similar results. Kopelman's data,
which were obtained from engineers in three different companies, also
indicate that when organizations establish a "strong, concrete
performance-reward contingency," motivation is higher, and motivation
and performance are much more closely related.

schwab (1973) also found that the kind of compensation system
(hourly, piece-rate, or group-incentive) explained significant
variance (A R? a .44) in instrumentality perceptions. Piece-rate
employees perceived significantly stronger instrumentality (4.46 on a
5 point scale) than the group incentive employees (3.95) who in turn
perceived significantly higher instrumentality than the hourly wage
employees (3.13).

Together, the results of these studies suggest the existence and
importance of an expectancy theory boundary condition. In other
words, a strong actual or objective link between performance and
outcomes will not result in increased motivation and performance
unless this objective link is perceived by workers. But while
providing evidence for the existence of this boundary condition, these
studies also indicate that workers are accurately perceiving the
objective relationship between performance and outcomes in their work

settings.

21

Performance, Satisfaction, and Reward Contingency

Most major reviews of job satisfaction (Brayfield & Crockett,
1955; Schwab & Cummings, 1970; Vroom, 1964) have concluded that the
relationship between satisfaction and performance is not significant.
Indeed, Vroom’s (1964) report of a median correlation of .14, from
over 20 years ago, is comparable to a recently made population
estimate of .17, which was derived from a meta-analysis of over 74
studies spanning 40 years of research (Iaffaldano & Muchinsky, 1985).

Despite this small direct relationship, research and theory
indicate that the relationship between satisfaction and performance
increases when rewards are tied to individual performance. "If we
assume, as seems to be reasonable in terms of motivation theory, that
rewards cause satisfaction, and that in some cases performance
produces rewards, then it is possible that the relationship found
between satisfaction and performance comes about through the action of
a third variable-~rewards" (Lawler & Porter, 1967). Similarly,
Cherrington, Reitz and Scott (1971) argue that there is no inherent
relationship between satisfaction and performance, and that "the type
of reward system under which workers perform might strongly influence
the satisfaction-performance relationship" (p. 531).

An experimental study by Cherrington, et a1. (1971) provides the
strongest evidence for the moderating effect of contingent rewards.
In this study, 90 undergraduate students were guaranteed $1 an hour to
complete a test-scoring task where performance was the number of items
correctly scored. They were told that the best performers were to

receive an additional $1. In fact, rewards were randomly distributed

22
after one hour of task performance, thereby creating one group of
appropriately reinforced subjects (i.e., rewarded high performers and
non-rewarded low performers) and one group of inappropriately
reinforced subjects (i e., rewarded low performers and non-rewarded
high performers). Across conditions (random rewards), the correlation
was .10 (n.s.), supporting the hypothesis that there is no inherent
relationship between satisfaction and performance. However, the mean
correlation between performance and satisfaction was .41 (p < .05) for
appropriately rewarded subjects and -.22 (p < .05) for inappropriately
rewarded subjects.

Other experimental studies mirror this result, too. Podsakoff
and Williams (1986) reviewed this literature, noting whether reward
contingency was present, absent, or an unknown in the design of each
study. In samples or conditions where reward contingency was present
(Baird & Hamner, 1979; Cherrington et. a1., 1971; Greenhaus & Badin,
1974; Hamner & Harnett, 1974; Locke, 1965; Pritchard, 1973),
performance was correlated .39 with work satisfaction and .38 with pay
satisfaction. In samples or conditions where reward contingency was
unknown or absent, performance was correlated .12 with work
satisfaction and -.16 with pay satisfaction. Both sets of
correlations were significantly different. Thus, experimental studies
demonstrate that reward contingency moderates the level of the
performance-satisfaction relationship as well as affecting the
direction of the relationship.

Additional support for this relationship is evident in several
field studies. For example, Greene (1973) provided a longitudinal

test of the moderator hypothesis in a field setting, where

 

23
satisfaction, performance, and merit pay data were measured at two
times, approximately one year apart. Measures of satisfaction were
obtained through questionnaires administered to 62 first-line
marketing and finance managers from a large manufacturer of business
and communication equipment. Performance was measured using an
average rating of quality and quantity obtained from two of each
managers' peers. Company salary records were used to obtain measures
of merit pay, which was expressed as the percentage increase (over and
above cost of living increases in salary) over last year’s salary.
Cross-lagged panel analysis indicated that the correlation between
performance at time 1 and job satisfaction at time 2 (p = .49, p <
.05) was significantly greater than the correlation between job
satisfaction at time 1 and performance at time 2 (y = .17, pyp.), or
either of the cross-sectional correlations (p = .20, py§., at time 1,
and p = .23, p < .05, at time 2). These results suggest that reward
contingency acts as a moderator by increasing the relationship between
performance and satisfaction.

Further support for the moderating role of reward contingency
can be found in a study by Jacobs & Soloman (1977) which investigated
the relationship between contingent rewards, performance, and
satisfaction across three levels of employees (salesmen, and first and
second level managers, N = 255) in a commercial chemical company.
Questionnaires were used to obtain measures on the five satisfaction
scales of the Job Description Index (Smith, Kendall & Hulin, 1969) and
perceived reward contingencies. Performance was measured using an
overall rating from each participants' supervisor. Moderated

regression analyses supported the moderator hypothesis as the

24
perceived reward contingency X satisfaction interaction for each JDI
subscale resulted in greater predictability of performance beyond the
simple main effects.

While several studies (Kesselman, Wood and Hagen, 1974;
Motowidlo, 1982) did not support this relationship, field studies in
general supported the contention that reward contingency moderates the
performance-satisfaction relationship. When reward contingency was
present (Bagozzi, 1978, 1980; Baxter, Taaffe, & Hughes, 1955; Bhagat,
1982; Brody, 1945; Greene, 1973; Kesselman et. a1., 1974; Kirchner,
1965; Kopelman, 1976; Matteson, Ivancevich, & Smith, 1984; Motowidlo,
1982; Orpen, 1974; Schneider & Olson, 1970) the correlation between
performance and general satisfaction was .27, which was significantly
different from of the correlation of .17 for those studies where
reward contingency was unknown or absent (Podsakoff & Williams, 1986).
Thus, both laboratory and field studies suggest that when rewards are
perceived as contingent on performance, performance and satisfaction
will be positively related and that better performers will be more

satisfied than poorer performers.
Satisfaction and Functional Turnover

Only one study has examined the relationship between
satisfaction and functional turnover. Using multiple regression
analysis, Hollenbeck & Williams (1986) found that attitudes (e g., the
satisfaction facets of the JDI, organizational commitment, and
intention to turnover) which typically predict turnover (shrunken 32::
.11, p < .01), did pp; predict functional turnover (shrunken RZ== .04,

n.s.). However, bivariate analyses showed that satisfaction with the

25
work itself (; = .21, p < .05) and satisfaction with co-workers (p =
.19, p < .05) were weakly related to functional turnover.
Satisfaction with supervision, job security, and pay, however, were
not related to functional turnover.

Despite these mixed results, it is expected that satisfaction
will be related to functional turnover under conditions of reward
contingency. If perceived reward contingency moderates the
relationship between performance and satisfaction, then differences in
satisfaction between high and low performers should occur. These
differences, in turn, should be related to a pattern of functional
turnover where poor performers are more likely to leave and better

performers are more likely to stay.
The Entirg Modgl

Up to this point, the literature review has addressed single
linkages in the model. The results of several studies, though, are
consistent with predictions of the entire model.

Dreher (1982) found that performance was related to turnover for
a sample of Exxon's managerial, professional, and technical employees.
The turnover in this sample was largely functional as initial
performance ratings (averaged across the first 2 or 3 years with the
company, 3.51 > 3.24, p = 5.23, p < .01), the rate of career
advancement (i.e., the number of promotions per year, .43 > .31, p =
4.63, p < .01), and final performance ratings (i.e., the most recent
performance ratings, 3.78 > 3.15, p = 9.76, p < .01) were
significantly greater for stayers than for leavers. Dreher speculated

that the performance differences between leavers and stayers could be

26
attributed to the company's policy and practice of rewarding superior
performance. Since there is previous evidence (Dreher, 1977)
validating the performance-reward relationship in this company, these
data suggest that performance contingent promotional practices may
contribute to functional turnover.

Similarly, in a study of bank tellers where poorer performers
quit more often than better performers, Stumpf and Dawley (1981) found
that three measures of performance were negatively correlated with
turnover: promotions (i.e., from teller trainee, to junior teller, to
intermediate teller, and then to senior teller; p = -.33, from 1970-
1976; p - -.81 from 1977-1978), difference records (i.e., differences
of over $1 in cash shortages or overages based on a teller's daily
closing tally; ; = -.41, from 1970-1976; p = -.12 from 1977-1978); and
annual supervisor ratings which were compiled into an overall global
index of performance based on the employee's entire tenure (p = -.30,
from 1970-1976; p = -.21 from 1977-1978). Stumpf and Dawley noted
that during the 1970-1976 period, "If no improvement was made over a
few months, the teller would be dismissed. As it became apparent to
new tellers that this was bank policy, those with poor difference
records (a performance measure) may have resigned before their
situation resulted in dismissal" (pp. 160-161). These policies
resulted in an increase in the relationship between performance and
promotions (p = .30 for the 1970-76 group, and p =.46 for the 1977-78
group), suggesting that the employees learned that promotions were
contingent on performance. The possible learning of this performance-
promotion contingency may explain the strong relationship between

promotions and staying on the job during 1977 and 1978. As in

27
Dreher’s (1977) study, these data point to the possibility of a
positive relationship between reward contingency (promotions) and
functional turnover.

Research by Dansereau, Cashman, and Graen (1973) also addresses
the important role that supervisors can have on functional turnover
through tying rewards to performance and communicating this
contingency to their subordinates. In this study, 261 managers from
all departments of a large manufacturing corporation were separated
into two groups on the basis of subordinate ratings of initiating
structure on the Leadership Behavioral Description Questionnaire
(Fleishman & Harris, 1962; Stogdill & Coons, 1957). Evidence from a
previously published study (Graen, Dansereau & Minami, 1972) with this
same data set indicated that subordinates working for managers who
were either high or low on structure perceived a strong relationship
between performance and rewards. However, subordinates who worked
under managers who imposed only a moderate amount of structure (i.e.,
ambiguity) did not perceive rewards as related to performance.
Consistent with the hypotheses presented here, turnover in the
Dansereau et a1., (1973) study was functional under high-low structure
leaders, as the standardized performance mean for leavers was -.34
versus .08 for stayers. The average annual salary was also lower
($455) for leavers than stayers. In contrast, turnover was
dysfunctional under leaders who imposed moderate structure. Here, the
standardized performance means were -.04 for stayers and .25 for
leavers. The salary of the better performing leavers was
significantly higher ($3394) than that of the lower performing

stayers.

28

Finally, in a study of 420 entry level credit managers in a
retail credit institution, Wells and Muchinsky (1985) found that
promoted employees (N-140) performed significantly better on all 12
performance dimensions than did those who voluntarily left (n=l40) the
organization, who in turn were better performers than those fired
(N-140) from the organization (i.e., involuntary turnover). The
authors strongly suggest that performance contingent reward practices
existed in this organization by noting that, "It would be highly
unusual for a successfully performing credit manager to remain at the
entry-level management position for the balance of his or her career,"
(p. 330), "...as it is the "survivors" from this position that often
rise to assume the most critical role for maintaining the
profitability of the company, the job of branch manager," (p. 335).
They concluded that the relationship between performance and turnover
was "probably moderated by organizational practices that reward
performance (or fail to do so)...," (p. 336). Therefore, both the
results of this study and the conclusions drawn by its authors also
tend to support the predictions of the entire model.

Though not conclusive, the results of these four studies
strongly suggest that performance contingent rewards can have a
positive effect on functional turnover.

Hypotheses

Thus, based on expectancy theory, and the results of the

research discussed in the literature review, the following hypotheses

are advanced:

1) Objective reward contingency will have a positive
direct relationship with perceived reward
contingency.

2)

3)

4)

29

The relationship between performance and
satisfaction is moderated by perceived reward
contingency such that performance and satisfaction
will have a stronger positive relationship under
conditions of higher perceived reward contingency
than in conditions of perceived lower reward
contingency where the relationship will be weaker.

The effects of objective reward contingency and
perceived reward contingency on functional turnover
will be mediated through pay satisfaction and job
satisfaction.

If perceived reward contingency moderates the
relationship between performance and satisfaction,
there will be significant pay satisfaction and job
satisfaction differences across the four functional
turnover groups.

.‘fl L

CHAPTER THREE

METHOD

Sample and Procedure

This study used a nonequivalent control group design where four
separate companies represented four different levels of objective
reward contingency, ranging from straight commission and no salary to
mostly salary plus small commissions/bonuses. While not a true field
experiment, the design can be viewed as a "natural experiment" (Cook &
Campbell, 1979, p. 296) because sampling from companies with different
levels of reward contingency approximated the design--though not the
control-~of a field experiment in which levels of reward contingency
were systematically varied.

Indeed, one consequence of this design is that organization is
fully confounded with objective reward contingency. Consequently,
although these organizations were chosen precisely because they paid
their sales representatives different combinations of salary and
commissions, a number of other nonmeasured differences associated with
organization could serve as alternative explanations for the results
of this study.

While this is surely a design limitation, the difficulty of
conducting a controlled field experiment on this topic, rather than a
field study, should also be recognized. To conduct a true field
experiment, one would have to (a) find a company willing to allow the

researcher to change the pay reward contingency for half of its sales

30

31
representatives, or (b) find a company that has two groups of
employees performing identical jobs under identical situations that
then pays those two groups using different levels of reward
contingency, and, furthermore, is willing to participate in research.

Company A is a manufacturer and distributor of different kinds
of filters and filtering processes and sells these products in
industrial, medical and environmental markets. The sales staff (n =
22), which sells products in all three markets, is paid a salary plus
commission. Commission is distributed quarterly and is paid for
achieving a percentage of one’s annual cumulative quota. For example,
at three months the goal is 22% of annual quota, continuing to 46% at
six months, 72% at nine, and 100% of quota at twelve months. For the
full year, sales representatives who achieve 125% of quota can earn a
maximum of $23,000 of incentive pay. Other representative payouts are
as follows: $17,000 at 115%, $11,000 at 105%, $8,000 at 100%, $4,000
at 95%, and $1,500 at 90%, the minimum payout. Overall in Company A,
only 7% of total pay (figured in months, x = $3,253, s.d. = $605) came
from bonuses (x = $227, s.d. = $140), while 93% came from salary (x =
$3,026, s.d. = $711).

Company B manufactures and sells office furniture products,
ranging from chairs and smaller open office systems to fully
integrated office systems. The sales staff (n = 58) is paid on a
three-tier salary scale ($24,000, $28,000, and $32,000), plus
commissions, which are paid quarterly. Commission based pay starts at
80% of quarterly quota, and is paid as follows: 1.5% for 80-90% of
quota, 3% for 90-100% of quota, 2.3% for 100-200% of quota, 2% for

120-160% of quota and 1.5% for anything over 160% of quota. On

32
average for Company B, 27% of total pay (figured in months, x =
$4,027, s.d. = $1,820) came from bonuses (x = $1,383, s.d. = $1,471),
while 73% came from salary (x = $2,644, s.d. = $647).

Company C manufactures, sells, and distributes conveyor
equipment ranging from gravity rollers to overhead powered chain
conveyors. These products are sold as components, many of which are
smaller and mobile, and as specialized packages for industrial
customers. Sales representatives (n = 42) are compensated on a three-
level salary plus commission system. Under level 1, new recruits are
not assigned a sales territory. Instead, they acquire product
knowledge through in¥house training. Level 1 pays a salary, but no
commission. There were no level 1 sales representatives in the study.
At level 2, salaries range from $19,000 to $30,000. Commissions,
which are paid monthly, are based on 2% of the first $400,000 in
sales, 3% of the next $200,000, and 4% of all sales over $600,000.
Schedule 3 is the same as schedule 2, but adds a year-end volume bonus
and bonus credit. Volume bonuses are $3,000 for sales over $1,000,000
and $3,000 for each additional $500,000 in sales. Finally, above
$2,000,000, $3,000 is awarded for each million in sales. Amounts
between these levels are interpolated. Bonus credit is based on
assigned margin, which is the percentage profit before taxes less all
costs of a sale, in other words, the percentage profit cleared by the
company on sales. Bonus credit for assigned margin bonus starts at
$2,000 for 10% margin, with the addition of $500 for each 1% increase
(for example, from 10% to 11%) in margin. Assigned margin bonuses are
paid at the end of the year on accumulated margin for the year. On

average in Company C, 64% Of total pay (figured in months, x = $4,986,

33
s.d. - $1,991) came from bonuses (x = $3,178, s.d. = $2,041), while
only 36% came from salary (x = $1,808, s.d. = $204).

Company D is a manufacturer and distributor of computerized
automobile engine analyzers which are used to troubleshoot engine
problems and/or evaluate engine exhaust emissions. The sales staff (n
= 334) typically sells its products to independent service stations
and garages. Sales representatives receive no salary and no bonuses.
They are paid on straight commission (figured in months, x = $2,434,
s.d. - $1,394) which, though it varies by product, is typically 10% of
the product price after subtracting product trade allowance. Trade
allowance, which can vary from 10 to 20%, represents the "slack" that
representatives have in negotiating price with customers. Thus, sales
representatives can cut list price from 10 to 20%, depending on the
product and customer demands. However, if sales representatives can
get the customer to pay more for the product by taking a smaller trade
allowance, they then receive a larger commission.

Questionnaire packets containing a cover letter from the
researcher, a cover letter from the company, a questionnaire, and a
stamped, addressed envelope which was to be used to mail the
questionnaire directly to the researcher, were distributed to sales
representatives by the personnel departments of each company in June
1986. A second questionnaire packet was mailed directly to each sales
representative if the initial questionnaire was not returned within a
month. Finally, a third and last questionnaire packet was sent if the
second questionnaire had not been returned within a month of the
second letter. Note that follow-up questionnaire packets were not

sent to non-responders in company D because of the cost of sending

34
additional questionnaires to such a large sales force, and because a
satisfactory number of questionnaires was projected to be returned,
even with a potentially low response rate. Approximately 90% of the
returned questionnaires were received within 6 weeks of the initial
mailing.

The response rates for each company were 9 of 22 (41%) for
company A, 30 of 58 (52%) for company B, 21 of 42 (50%) for company C,
and 75 of 334 (22%) for company D. Thus, the lower overall response
rate (31%, 135 of 435) is primarily attributable to the lower response
rate in company D, which was not unexpected since only one round of
questionnaires was mailed to its sales representatives.

Multivariate analysis of variance, which was used to compare
responders and non-responders, indicated no significant differences in
turnover, tenure, performance, total compensation, reward contingency,
or gender in companies A, B, and C. However, in Company D, there were
differences between responders and nonresponders, multivariate E (5,
242) - 3.68, p g .01. Follow-up univariate analyses indicated no
significant differences in turnover, performance, gender, or reward
contingency, though there were significant differences between
responders and nonresponders in tenure (13 months versus 11 months, F
(1, 246) = 8.46, p < .01), and in average total monthly compensation
($2,364 versus $1803, F (1, 246) = 8.97, p < .01).

Missing data reduced the final, usable sample from 135 to 98. A
total of 45 participants, 8 from Company A, 20 from Company B, and 17
from Company C, were paid according to some salary plus
commission/bonus system. The remaining 53 participants, all from

Company D, were paid straight commission. The average age was 37,

35
while the average tenure was 5 years. Eighty-seven (89%) of the
participants were male, and 24 (24%) of the participants had quit
their jobs. Approximately 6.2% possessed a master’s degree, 14.2% had
some graduate school, 42.9% had graduated college, and 31.6% had some

college, while the remaining 5.1% had only graduated high school.
Measures

Performance. The first step in constructing performance
measures was to standardize performance within each company to remove
level (i.e., mean) and dispersion (i.e., variance) differences in
performance across companies. Performance was standardized by month
within each company, and these standardized monthly scores were then
averaged to create the measures of individual performance. (While
standardization typically produces a mean of 0 and a standard
deviation of l, the performance measures used in this study will vary
somewhat from these levels because the sample used in the study is not
the same as the entire "monthly population" of sale representatives --
which itself varied by month due to hires and turnover-- on which
monthly means and standard deviations were figured).

Standardization was necessary because it is misleading to
compare individual sales performance across companies using raw sales
performance. For example, raw performance data might indicate that
the average sales person for Company C ($164,000 per month) is a
better sales representative than the average sales representative for
Company A ($61,000), when in fact this difference was due to
differences in prices of products or general industry factors (i.e.

)

level of competition, the economic cycle of the industry, and so on).

 

36
Thus, performance scores were standardized to create comparable
measures of performance across companies.

The second step was to partition performance into two distinct
measures, past performance and overall performance. Pa§p_p§pfppmappg,
performance data collected before questionnaire administration, was
used to test the moderating effect of perceived reward contingency on
the relationship between performance and satisfaction. Past
performance was simply the average of standardized, monthly
performance for each sales representative from April to June 1986.
(Thus, sales representatives hired after April 1986 were excluded from
the study since they would be lacking the required three months of
sales data.) Unfortunately, the internal consistency reliability of
this measure was low (a = .44) due to the instability of performance
from month to month.

Companies A, B, and D, however, provided measures of cumulative
sales that could be used to increase the reliability of past
performance. For Companies B and D, this meant an additional 3 months
of data, including January, February, and March of 1986. Company A,
which was on a July to June fiscal year, provided cumulative data back
to July 1985. Cumulative sales were then divided by the appropriate
number of "cumulative months," after adjusting for each employee’s
starting date, to produce a "monthly average" for cumulative sales.
This monthly average was standardized, like other monthly performance
measures, and added to monthly sales for April, May and June,
producing a total of four "months" of sales data, which was divided by
4 resulting in a measure of overall, average, monthly past

performance. The addition of average, monthly cumulative sales

37
increased the average month-to-month correlation to .26 (p g .05), and
internal consistency reliability to .58. However, the a of .58 is
probably an underestimate because treating cumulative sales as a
fourth month leads to a lower reliability than what would have been
estimated had a been computed using six separate, disaggregated months
of data in companies B and D, and 12 separate, disaggregated months of
data in Company A.

Qyppgll_pgg§ppmpppg, which combined standardized past
performance data with standardized monthly sales data obtained after
administration of the questionnaire, was used to compute functional
turnover scores. Unlike cumulative sales which was simply divided by
"cumulative months" to produce an average, monthly figure (i e.,
essentially a fourth month of data on which to figure past
performance), overall performance was computed by adding each extra
month of standardized sales performance (up to the time of
termination, or the end of the study) and dividing by the number of
months of employment during this time. So, in contrast to past
performance, overall performance could be based on as many as 24
months of performance (Company A), 18 months (Companies B and D), or
15 months (Company C), to as few as six months (the minimum),
depending on when the employee quit. Like any summated measure or
scale, the internal consistency reliability of future performance rose
with the addition of each homogeneous item, or in this case, each
additional month of data. The average reliability for future
performance, sample weighting for months of employment (whether the
employee quit and when), was .83. The average month-to-month

correlation was .24 (p g .05).

38

Objective Reward Contingency. "Objective" reward contingency is
the actual relationship between sales performance and pay within a
company. Consistent with expectancy theory's (Vroom, 1964) definition
of instrumentality as the link between levels of performance and
levels of outcome, the observed correlation between individual pay and
performance should be used to measure "objective" reward contingency
in each organization. Further, the theoretically correct correlation
is a within-person correlation (Mitchell, 1974), where the correlation
is computed on an individual sales person's pay and performance over
some time period. This is a more precise way to determine reward
contingency because it can account for individual differences in the
pay-performance relationship within an organization which might arise
from bonuses for meeting one's sales quota or for changes in
commission which arise after achieving a particular sales volume.

Practical considerations, however, prevent the use of the
within-person correlation as the measure of objective reward
contingency. At best, the within person correlation between pay and
overall performance would be based on 20 data points for Company A, 18
for Companies B and D, and 15 for Company C. At worst, it would be
based on a minimum of six data points since the minimum requirement
was six months of performance data. In any event, sampling error and
the effects of outliers (particularly good or bad months of sales
volume) would make the reliability and validity of a within-person
correlation suspect.

It would seem that the next most plausible measure of reward
contingency is an across-person correlation between pay and

performance. Although the across-person correlation cannot reflect

39

individual differences in actual reward contingency within an
organization's compensation system, it would be computed on a
sufficiently large sample in each company. The problem with this
measure is that it is unlikely to reflect differences in reward
contingency across companies. For example, the correlation between
pay and performance will approach 1.00 in companies that pay sales
representatives strictly on a commission basis. The same correlation,
however, is likely to occur in companies that use salary plus
commission pay plans. Because of the small variation in salary, which
is primarily used to adequately cover sales representatives living
costs, salary pay is effectively a constant in these companies. Yet,
there is tremendous variation in commission and bonus pay because of
their direct link to differences in individual performance. The
result, due to covariation between commission/bonuses and performance,
is twofold: a strong correlation between total pay and performance
under salary plus compensation systems, and, an inability to reflect
company differences in reward contingency. After converting the
correlations to Fisher 2’ equivalents (Cohen & Cohen, 1983),
comparisons revealed that the correlations between pay and performance
for Companies B, C, and D were not significantly different from each
other (; = .71 in Company B, p = .79 in Company C, and p = .76 in
company D). Only Company A (p,= .10) was significantly different (p g
.05) from the others.

The measure used here, which does not suffer from these
problems, is the percentage of total pay based on sales commissions
and bonuses. For example, stronger reward contingency is evidenced

when, on average for a sales force, a large percentage of total pay

40
comes from commissions and bonuses, while weaker reward contingency
occurs when most of total pay comes from salary. This measure also
captures individual differences in reward contingency within
companies, which become especially important under salary plus
commission/bonus pay plans as better performers, relative to poorer
ones, get most of their take home pay from commissions and bonuses.

The components used to figure objective reward contingency were
monthly salary, and commissions and bonuses which were converted to
monthly units to parallel salaries. Each of these monthly measures
was then averaged to create the measure of objective reward
contingency. This measure was computed by dividing the total of
average monthly commissions + average monthly bonuses by the total of
average monthly salary + average monthly commissions + average monthly
bonuses. The resulting percentage served as the final measure of
objective reward contingency.

Perceived Reward Contingency. Consistent with Vroom's (1964)
original formulation of expectancy theory as a within-person model,
theoretically correct measurement of instrumentality requires that
participants rate the relationship between multiple levels of
performance and multiple levels of pay.

The measure of instrumentality used here was developed by Ilgen,
Nebeker and Pritchard (1981) in an experimental work simulation
designed to determine the reliability and validity of traditional and
newly developed measures of expectancy theory constructs. Their study
examined both the item format (e g., the wording of items: frequency,
verbal anchors, probabilities, and correlational format) and the

composite index (e.g., the way in which expectancy theory component

41
measures were combined to produce an overall composite measure of
motivation: expected value index or correlation index) of expectancy
theory measures.

The item format selected for use in the present study was the
probabilities format, using the correlation composite. A 2 x 2
analysis of variance for 2 levels of instrumentality (hourly pay
versus contingent pay) and 2 levels of expectancy (low and high,
manipulated by increasing work load) indicated strong validity for
this format (E - 1.27 x 10, p < .001, omega squared = 1.00).
Moreover, Ilgen et. al. found this format to have high test-retest
reliability.

When modified for the sales representatives in this study, this
item read:

"Using the following scale, rate each item by placing a number

between 0 and 100 on each line such that the lines in one row

add to 100."

Chances in 100 of Getting this Annual Pay

No A Slight A Fifty-Fifty A Good Certain This
Chance Chance Chance Chance Will Happen
(0) (25) (50) (75) (100)

0 10 20 30 40 50 60 70 8O 90 100
If your sales performance for the year totalled $ , what

would be your chances of earning an annual income of .......

$00.00 $-2 s.d. $-1 s.d. $X $1 s.d. $2 s.d.
per per per per per per
year year year year year year
+ + + + + = 100"

It should be noted that participants were asked to rate their chances

of receiving six levels pf annual pay, starting with no pay to

42
differentiate sales staffs on straight commission from those on salary
plus commission, and proceeding to five more levels of annual pay
corresponding to the dollar level of annual pay for two standard
deviations below the mean, one standard deviation below the mean,
average pay, one standard deviation above the mean and two standard
deviations above the mean of annual pay for that company. The use of
these normative levels of pay within each company should result in
participants rating levels of pay which are meaningfully equivalent
(low, average or high) across companies.

Participants completed this item (rating their chances of
receiving six levels of annual pay) for pix levels pf annual sales
performance, starting with no sales ($0.00) to differentiate sales
staffs on straight commission from those on salary plus commission,
proceeding to five levels which correspond to the dollar level of
sales performance for two standard deviations below the mean, one
standard deviation below the mean, average performance, one standard
deviation above the mean and two standard deviations above the mean
for annual sales performance in that company. As with pay, the use of
normative levels of raw performance within each company should result
in participants rating levels of performance which are meaningfully
equivalent (low, average or high) across companies.

Participant responses on the six instrumentality items were
combined to form a six by six matrix where the rows represented levels
of pay and the columns represented levels of performance. The entries
in each cell of the matrix were composed of the participant's
perceived probability of receiving that level of pay (row) at that

level of performance (column). The values in this matrix, which

43
essentially constitute a scatterplot of the perceived relationship
between pay and performance, were treated as frequencies in a
bivariate distribution. A within-person correlation coefficient was
then "calculated between the two dimensions in this two-dimensional
space," (Ilgen, Pritchard, Dugoni, & Nebeker, 1980). Finally, the
within person correlations were transformed to Fisher's z' scores to
normalize the distribution of coefficients and satisfy equal variance
assumptions for data analysis (Cohen & Cohen, 1983, p. 271). This
final index, as represented by Fisher’s 2', represented the degree of
perceived linear covariation between pay and performance.

Pay Satisfaction. The pay satisfaction scale of the Job
Descriptive Index (Smith, Kendall, & Hulin, 1969) was used to measure
satisfaction with pay. Evidence of the JDI's reliability, and
convergent and discriminant validity can be found in Johnson, Smith,
and Tucker (1982). Responses were made using the conventional JDI
format: "Y" for yes was coded 3, "N" for no was coded 0, and "?" for
uncertain was coded 1 (Smith, Budzeika, Edwards, Johnson, & Bearse,
1986). Pay satisfaction was standardized within each organization.
Internal consistency reliability for this scale was .81.

Job Satisfaction. Hoppock's (1935) four item scale was used to
measure overall job satisfaction. There are seven alternatives for
each item (scores range from 4 to 28) and participants were instructed
to choose one of the seven responses in each item. Internal
consistency reliabilities for this measure have been acceptable,
ranging from .76 to .89 on a total of 30,000 (McNichols, Stahl, &
Manley, 1978), to as high as .93 (Hoppock, 1935). There is also

support for the construct validity of this measure. McNichols e1.

44

al., also report that the scale is positively correlated (r = .73)
with the work satisfaction scale of the Job Descriptive Inventory, and
a measure of intention to remain in the military (; = .40). Dunham
and Herman (1975) report a correlation of .75 with either the female
faces scale, which Dunham and Herman published, or the original male
faces scale published by Kunin (1955). Job satisfaction was
standardized within each organization. Coefficient alpha was .80.

Functional Turnover. Functional turnover represents the
opportunity performance costs and gains associated with employee
turnover and was computed as follows (Hollenbeck & Williams, 1986).

Functional Turnover = Zi x T.0.

where,

Zi - individual overall sales performance standardized within
each company, and

T.0. = turnover, stays (coded +1) or leaves (coded -1).

Turnover data were obtained in July 1987, one year after
administration of the questionnaire. Participants who were
terminated, retired, or transferred to a different position within the
company were eliminated from analysis.

Computed this way, functional turnover fully captures
opportunity gains and losses because it combines z-scores, which
compare individual performance to average performance within each
company, and employment status (stay/leave). It also correctly treats
opportunity gains and losses as a continuous measure. For example, a
functional turnover score of .50 is clearly a better opportunity gain

than a functional turnover score of .10.

45

Yet despite the ability to measure opportunity gains and losses,
continuous functional turnover scores suffer from two serious
deficiencies at the individual level of analysis where the research
goal is prediction and understanding of the different stay/leave
choices made by good and poor performers. First, from the standpoint
of prediction, most of the variance in the continuous functional
turnover scores will result from variance in performance, rather than
variance in turnover, because performance is multiplied by +1 or -1
when creating the composite score. Thus, it can be argued that
functional turnover simply reflects performance, especially in
situations where there is little turnover or extreme turnover (i.e.,
where all or nearly all cases are simply standardized performance
scores multiplied by +1 or -l).

The second problem is that any continuous functional turnover
score can be interpreted two ways. To illustrate, an opportunity gain
of .50 could result from retention of a good performer (+l)(.50), or
the voluntary loss of a poor performer (-l)(-.50). Therefore,
continuous functional turnover scores must be disaggregated to
determine the kind of opportunity gain or loss, that is, whether
someone leaves or stays and whether they were a good performer or a
poor performer.

As a result, the continuous measure of functional turnover
actually cancels out individual differences in performance and
turnover. This happens because good performers are paired with poor
performers, while stayers are combined with leavers (i.e, opportunity
gains consist of poor performing leavers and good performing stayer;

opportunity losses consist of good performing leavers and poor

46

 

performing stayers). Consequently, there can be no straightforward
interpretation of continuous functional turnover scores at the
individual level of analysis.

Because of these problems, functional turnover was treated as a
categorical variable to be able to differentiate between each of the
four different kinds of functional turnover, and to account for the
different antecedents associated with membership in each of those
groups. Four groups were created by separating stayers and leavers

into above average performers and below average performers. Poor

 

performing leavers (n = 10) were coded 1, good performing leavers (n =
14) were coded 2, poor performing stayers (n = 28) were coded 3, and

good performing stayers (n = 46) were coded 4.

Data Analysis, Power Analysis, Restatement of Hypotheses

Explanation of the data analysis proceeds in four steps:
nonequivalent control groups, study design, hypothesis testing, and
discriminant analysis.

Nonequivalent control groups. One of the threats, among others,
to the internal validity of quasi-experimental designs is the
nonequivalence of groups across treatment conditions. In this study,
preexisting differences between the sales representatives in each
organization could contaminate comparisons across levels of reward
contingency and pose an alternative explanation for study results.
Because of its advantage in handling "unequal cell frequencies" (Cohen
& Cohen, 1983), regression analysis will be used to conduct orthogonal
contrasts for group differences in age, gender, education, and tenure

across organizations (i e., different levels of reward contingency).

47
Since it was impossible to eliminate potential, measured differences a
priori, they will be controlled statistically by entering them as
covariates in all analyses.

Study design. As a "natural experiment," each organization was
to have represented a different level of "naturally" occurring
objective reward contingency, which in turn was to be related to
different levels of perceived reward contingency across organizations.
Additional support for the study design would also be evident if
perceived reward contingency moderated the relationship between past
performance and job satisfaction, and past performance and pay
satisfaction. Stated as null hypotheses:

1) There will be no differences in objective reward
contingency across organizations.

2) There will be no differences in perceived reward
contingency across organizations.

3) Objective reward contingency will not be positively
related to perceived reward contingency.

4) The relationship between past performance and
satisfaction (pay and job) will be invariant (i.e.
not moderate) and not significantly different from
zero across levels of perceived reward contingency.

’

Regression will be used to conduct orthogonal contrasts for mean
differences on objective reward contingency (hypothesis 1) and
perceived reward contingency (hypothesis 2) across companies (Cohen &
Cohen, 1983). A directional t-test will be used to determine if the
partial correlation between objective reward contingency and perceived
reward contingency is significantly different from zero and positive
(hypothesis 3).

Using moderated regression (Saunders, 1956; Stone & Hollenbeck,

1985; Zedeck, 1971), satisfaction (pay and job) will be regressed on

 

48

the past performance (PP) and perceived reward contingency (PRC) main
effects, followed by the PP x PRC interaction (hypothesis 4). If the
addition of the cross-product term at step 2 results in a significant
increase in R2, then the existence of the PP x PRC interaction will be
confirmed. Since this is a directional hypothesis, the interaction,
if significant, will be plotted to assess the predicted nature of the
interaction. Based on already existing research (Cherrington et a1.,
1971; Greene, 1973; Iaffaldano & Muchinsky, 1985; Mitchell, 1974;
Vroom, 1964), it was estimated that the population R2 for satisfaction
would be around .20 and that the double interaction would account for
an incremental R2 of .05. The power for 98 subjects was .68, which is
slightly lower than the recommended level of .80 (Cohen & Cohen,
1983).

Hypothesis testing. Each of the hypotheses related to model
testing is restated as null hypotheses below, followed by the
procedure used to test it, and a power analysis indicating the

probability of rejecting the null hypothesis given a final sample size

of 98.

5) The effects of objective reward contingency and
perceived reward contingency on functional turnover
will not be mediated by pay satisfaction and job
satisfaction.

6) The effect of performance on satisfaction will not

be moderated by perceived reward contingency. That
is, there will be no significant pay satisfaction
and job satisfaction differences across the four
functional turnover groups.

According to James & Brett (1982), mediation models take the

form x-->m-->y where "x has a direct effect on m, m has a direct

effect on y, and x is not related directly to y when m is held

 

49
constant," (p. 308). In the present model, all the antecedents to pay
satisfaction and job satisfaction may be categorized as x, pay
satisfaction and job satisfaction as m the mediators, and functional
turnover as y, the dependent variable. Following James & Brett
(1982), a hierarchical analysis will be used to test this hypothesis.
Pay satisfaction and job satisfaction will be entered in the
hierarchical step before perceived and objective reward contingency.
If pay satisfaction and/or job satisfaction mediate the effects of
these variables on functional turnover, then pay satisfaction and job
satisfaction will account for significant variance in functional
turnover in the preceding hierarchical step (hypothesis 6), and, the
incremental variance from adding the additional variables in the
following hierarchical step will be statistically insignificant
(hypothesis 5). Given a sample of 98, an a =.05, and an expected R2 of
.10 for functional turnover, and with pay satisfaction and job
satisfaction expected to account for almost all of that variance (say
9%) because of their role as mediators, the power to test these
hypotheses is .87.

Discriminant analysis. Testing of hypotheses 5 and 6 is
conducted using hierarchical multivariate discriminant analysis
because 1) the covariates (i.e., age, gender, education, and tenure),
response variables (i.e, pay satisfaction and job satisfaction), and
design variables (objective reward contingency--bonus pay after
partialling total pay-~and perceived reward contingency) are likely to
be intercorrelated, and 2) functional turnover is a categorical

variable.

More specifically, a hierarchical discriminant analysis is used

50
to test hypotheses 5 and 6. The measured covariates (i.e., age,
tenure, education, and gender) will be entered in step 1, followed in
step 2 where job satisfaction and pay satisfaction are entered to test
their effects as mediators, followed by the design variables (i.e.,
total compensation, bonuses/commissions, and perceived reward
contingency) in the last step. The canonical R and Wilks lambda, both
of which are measures of overall level of discriminatory power, as
well as the chi—squared test of significance will be reported for each
hierarchical step.

Interpretation of the unique contribution of each variable to
group separation, assuming overall significance, will be judged using
structure coefficients (Huberty, 1984; Huberty & Morris, 1989).
Structure coefficients, which are analogous to factor loadings in
factor analysis, indicate the correlation between the original
variable and the linear discriminant function which maximizes group
separation. The greater the greater the structure coefficient, the
more an individual variable contributes to group separation.

If the multivariate analysis indicates significant group
separation, follow up orthogonal mean contrasts between groups will be
conducted on those variables which had significant univariate effects.

Finally, the results of the discriminant analysis will be
corroborated by using the resulting linear classification function
coefficients to assess the accuracy (i.e., hit rates) with which sales
representatives can be correctly classified into each of the four
functional turnover groups. The classification analysis will be
adjusted for differences in group sizes (i e., priors = size) and will

be tested for significance using the proportional chance criterion

(Huberty, 1984).

51

CHAPTER FOUR

RESULTS

Descriptive Statistics

Table 2 presents the means and standard deviations for each
company and for the final sample of 98.

Nonequivalent control groups. Orthogonal contrasts revealed
preexisting differences in gender, education, and tenure, but not age,
between sales representatives in each organization. Companies A and B
had more female sales representatives than companies C and D. The
sales staffs of companies A, B, and C typically had a college degree
or some graduate school, while the staff of company D was more likely
to have had only some college. Company C was characterized by
significantly higher tenure than any other company, followed by
Company B, which had tenure significantly higher than Companies A and
D, who had the lowest tenure.

Additional between company contrasts indicated no differences in
pay satisfaction or job satisfaction, whether standardized or
unstandardized. (Although both the standardized and unstandardized
measures of pay satisfaction and job satisfaction are reported in
Table 2, the standardized form of these measures will be used in
subsequent analyses, as described in the methods section). In terms
of performance, there were two differences. Company A sales

representatives had a significantly lower level of past performance

52

53

 

 

Table 2: Descriptive Statistics by Organization and for the Final
Sample.
Variables Final Company Company Company Company
Sample A B C D
N 98 8 20 17 53
Age
14 37.13 32.25 34.60 36.59 39.00
§Q 8.39 5.75 8.63 8.01 8.37
Gender1
Males 87 4c,d 14¢,d 173.13 528,b
Females 11 4 6 0 1
M .11 .50 .30 0.00 .02
SD .32 .54 .47 0.00 .14
Education2
g 3.85 4.63d 4.40cl 4.06d 3.45%”):0
SD .95 1.06 .75 .75 .87
Tenure
11 5.02 3.131%0 7.4239d 10.46”,cl 2.65b'°
SD 5.25 2.83 5.83 7.52 1.55
Average Salary3
M $1111 $3026b'c'd $2644a'°'d $18083'b'd $ 0000.13,.
SD $1298 $ 711 $ 647 $ 204 $ 00.00
Average
Commission“
M $2169 $ 2270'cl $1383“d $3178” $2434”
§_ $1676 $ 140 $1471 $2041 $1394
Average Total5
Compensation
L4 $3269 $3253c $4027d $4986a'd $24310“
§_ $1840 $ 605 $1820 $1991 $1394

 

Table 2 (cont’d.).

54

 

 

Variables Final Company Company Company Company
Sample A B C D
Objective Reward
Contingency
M .70 .07bvcvd .27“,d .5991"cl 1.003vb'c
SD .37 .05 .21 .16 0.00
Perceived
Reward
Contingency
m 1.65 1.24(1 1.24d 1.48(1 1.93”»
SQ .51 .50 .37 .27 .46
Unstandardized Pay
Satisfaction
M 32.69 37.75 31.40 39.06 30.38
SQ 14.29 5.29 11.93 8.10 16.78
Unstandardized Job
Satisfaction
M 20.53 21.38 20.05 21.71 20.21
SD 3.09 1.85 2.52 2.09 3.60
Standardized Pay
Satisfaction
M .02 .18 -.03 .00 .01
SD .99 .89 1.02 1.05 1.00
Standardized Job
Satisfaction
M -.04 .03 .02 .07 -.10
SD .99 1.07 .93 .78 1.07
Past
Performance
14 .16 -.43d .05 .01 .34a
SD .67 .69 .75 .58 .61

 

55

Table 2 (cont'd.).

 

Variables Final Company Company
Sample A B

Company
C

Company

D

 

Overall
Performance

.16 -.39b-‘='d .14a
.67 .64 .66

IUHZ
C1

Turnover

Quit 74 0d 2d
Stayed 24 8 18

Functional
Turnover

Number of

Poor

Performers

Leavers 10 0 0
Stayers 28 5 10

Number of

Good

Performers
Leavers 14 0 2
Stayers 46 3 8

.11a
.29

17d

0
10

.18a
.40

22a,b,c
31

12
25

 

1O-Male, l-Female.

2l-Some high school, 2=Graduated high school,

3=Some

college, 4-Graduated college, 5=Some graduate school,6=Master’s
degree, 7-M.D., Ph.D., or some other professional degree.
3Monthly salary. Does not include commissions.

l'Average commissions, on a per month basis.
include salaries.

5Average, monthly total compensation, salary
commissions.

aSignificantly different from Company A, p g
bSignificantly different from Company B, p g
°Significantly different from Company C, p g
dSignificantly different from Company D, p 5

Does not
plus

.05.
.05.
.05.
.05.

56
than the sales representatives in Company D. Similarly, Company A
sales representatives' overall performance was significantly lower
than the overall performance of sales representatives in Companies B,
C, and D. It should be noted that these small performance differences
are for the most part irrelevant, since the relevant performance
comparison occurs within companies where individual sales performance
is judged to be above or below average.

In terms of turnover, there was one significant difference
across companies. Company D experienced a much higher rate of
turnover (42%) in the final sample than Company A (0%), Company B
(10%), or Company C (0%). Although this level of turnover was
consistent for Companies B and D which experienced turnover of 4% and
45% the previous year, much more turnover was expected in Companies A
and C, both of which had 20% turnover in the previous year.

Note that while turnover is not the dependent variable, this
configuration of turnover across companies creates a pattern of
functional turnover where all poor performing leavers and almost all
the good performing leavers come from Company D. In contrast, poor
performing stayers, as well as good performing stayers, come from all
4 companies in the sample. Indeed, a discriminant analysis using
organization as the discriminating variables indicated significant
overall differences (p g .0002) in functional turnover across
organizations. These functional turnover differences will be
discussed again in the context of hypothesis testing.

Because almost all the turnover occurred in Company D, objective
reward contingency is effectively a constant of 1.00 for all leavers,

regardless of their performance. As such, objective reward

57
contingency, and the satisfaction differences predicted to be related
to it, should be unable to differentiate poor performing leavers from
good performing leavers. However, reward contingency might still
uncover meaningful differences between each group of leavers and the
good and poor performing stayers.

Intercorrelations Among Study Variables. Table 3 shows the
intercorrelations among study variables. As expected given the study
design, objective reward contingency was negatively related to average
salary (p - -.89, p g .05), and positively related to average
commission (p = .46, p 3.05). Similarly, perceived reward contingency
was negatively related to average salary (p = - 59, p g .05) and
average total compensation (p = -.25, p g .05), and positively related
to average commission (p = .18, p g .05) and objective reward
contingency (; - .56, p g .05).

Study Design Hypotheses

Null Hypothesis 1: There will be no differences in objective
reward contingency across organizations.

Orthogonal contrasts indicated significant differences in
average monthly salary, average commissions (averaged to a monthly
interval), average monthly total compensation, and objective reward
contingency between all companies, resulting in the rejection of null
hypothesis 1. The between company contrasts of objective reward
contingency were particularly important because they served as a check
on the design strategy used in the study: "to gather data from sales
forces working under different levels of reward contingency..".
Company A, with the largest salary and smallest commissions had a

reward contingency percentange of .09, followed by Company B, which

 

58

amopm u" HsfimHOOHHmHmdwonm

manna mn;9<.<mfiwmowmm.

 

 

$8838 H N .6. s m m 4 m m S S
H. mom 111
N. mmsnme 1.0m 111
. N *
u. macnmnwo: I.HN .mo .111
* *

s. .85ch .8 .8 .8 111
w. 504.13% 1.8 .8» .mo» .8» 111

mmHmH<
8. >838» .8» 1.8» 1.8 .8» 1.8 111

ooaawmm.o:
4. 509.6%» .8 .3 .8» .mm» .8» .13» 111

adflmH mm
m. 03.89% .8» 1.3»1.3»1.8»1.8» .8»1.8 111

wmz. noun
6. 868434.84 .8» 1.8» 1.8» 1.8» 1.8» .8» 1.8» .8» 111

wmz. noun. » » » »
40.448.83.665 1.8 .8 .8 .8 .8 .8 .8 1.8 1.5 111

mmdwmmmnnwo: » »
489803.36 .8 .8 1.5 .8 .8 .5 .2 1.8 1.8 .8 111

mmflwmmmOHHomm

 

 

59

Hmowm u Aoo:d_a.v.

 

 

 

 

 

<mHHmUHmm H N u A m m u m 0 Ho HH
* * *

HM. mdmbn. wmk. I.NH .ou I.om .om .OH .Hq .Hm .om I.ow .ow .Nu
mmdwmmmOHMo: » »

Hm. mdmbfl. Q06 .0» .om I.Hw .Hq .0m .0m .Ho l.o< .oo .wm .wq
mmdwmmmOHHo: » » » »

H». wmmfl .om I.Hm I.Hm .oo I.wp .90 .HH .uo .wq .Hq .ou
wMHmoagmbom » » »

Hm. OCQHflHH .HH I.Hm I.Hu .om I.NN .aw .um .um .Hb .Hu I.ow
wmfimo

8. 888% .8 .8 .8» .8» .3 .8» .8» 1.8» 1.8 .3 .8
gmhs.noapm:<

<WHHmUHmm Hm Hu Hp Hm Hm

8. @858 6.8 111
mmflwmmmodwo: »

Hw. mflmDQ. Q06 .NN III
mmflwmmmonwo: »

HA. wmmfl .Hm .ow III
mafimonamdom »

Hm. O<meHH .Hm I.ou .QH III
wmumonsmdnm »

8. m8<§ 1.8 .oo 1.8 1.8 111

sEH:.ooapm:<

 

60

60,310 .4. 0050.9; .

2000.2 H mm. 3800a. em 0501.. wow 050005903800. mama. M0 0501.. won
000303809.

 

»w m .8

Hon—,5? 78:08.

001080 :8: 8:8? 01080688 :8: 8:80 . 01080 8580. 4108888
00500008050 0305000 00:00? 08000006 009100. 83.0.. 05.0.. On. 0050 0.900

udemmmmHOSmH 900000.

A“5.00060 505.95. 0&me 988.0 0508.0 0.89%. comm :00 50500 00358308 .
><0Hmo0 gmmwosm‘ o: m 00H. Bond: gmwm 4 98.50 030810 0.89%. comm :00 8.0850
080300 .

MOW/00000. 30:va dog 0960:8303. 0&3 E100 088M888 4 98% 05.9.5.0 000%.
3. 003,40 H053 0033:0050K 985 0308.0 005%.

Mwmndmnzmn wmswa 00505038?

mamamawmmm $35.5 096034..

olwuncfid‘ Hnmflmwma.

61
had significantly smaller salaries and larger commissions, and an
objective reward contingency percentage of .27. (While it would
appear that the commissions paid in Company B were significantly
larger than those paid in company A, the larger mean and variance in
Company B were due to an exceptional quarterly commission of $31,700
for one sales representative. The median commission of $350 in
Company B is actually very similar to the average commission of $283
paid in Company A). Company C had smaller salaries than Company B,
but had the largest commissions, and an objective reward contingency
percentage of .59, followed finally by Company D, which paid no salary
and 100% commissions.

Null Hypothesis 2: There will be no differences in perceived
reward contingency across organizations.

Null hypothesis 2 was rejected as orthogonal between company
contrasts explained 36% of the variance (2 3 .00001) in perceived
reward contingency. These contrasts revealed a significant difference
in perceived reward contingency between Company D, which paid straight
commission and had the highest levels of perceived reward contingency,
and companies A, B, and C, which paid salary plus commissions and had
a lower level of perceived reward contingency.

Even though null hypothesis 2 was rejected, it should be noted
that the 2-way differences in perceived reward contingency did not
fully complement the 4-way differences in objective reward
contingency. The levels of perceived reward contingency were
generally in the right direction, getting smaller with lower levels of
objective reward contingency (A, 7% commission, PRC = 1.24; B, 27%

commission, PRC = 1.24; C, 59% commission, PRC = 1.48), but were not

62
strong enough to produce more than a 2-way significant difference.
The level of significance for these contrasts was A versus B, p g .98,
A versus C, 2 g .19, and B versus C, p g .09.

Since it is unlikely that the sales representatives in these
companies were ignorant of how much of their pay came from salary or
commissions, it may be that the nonsignificant differences in
perceived reward contingency between Companies A, B, and C were due to
the format of items used to tap these perceptions. Recall that
participants were asked to rate their chances of receiving six levels
of monthly pay ($0.00, $-2§Q, $-l§Q, $X, $+l§Q, $+2§Q) for each of six
levels of monthly sales performance ($0.00, $-2§Q, $-1§Q, $X, $+l§Q,
$+2§Q), such that within levels of performance the "chances of getting
paid" added to 100. A sample item is shown below:

1) If your total annual sales performance was $960,000, what
would be your chance (1 in 100) of getting paid .......

A) $ 000.00 per year ....... +

B) $ 5,000.00 per year ....... +

C) $25,000.00 per year ....... +

D) $45,000.00 per year ....... +

E) $65,000.00 per year ....... +

F) $85,000.00 per year ....... +

TOTAL- 100

However, it was very common for the "chances" within each level of
performance to be distributed among just two or three levels of pay.
For example, distributions of 50/50, 60/40, 70/30, and 80/20, or
50/40/10 and 65/25/10 were common. Often, though, participants simply
shifted similar distributions "down the scale" when rating higher
levels of annual sales performance. If the distribution for the above

item was 0/0/50/50/0/0, many participants would shift the distribution

to 0/0/0/50/50/0 for the next highest level of performance, resulting

63
in a close to linear progression between sales performance and the
chance of getting paid a certain level of compensation. Table 4,
which shows the mean response distributions by performance and pay
level, demonstrates this "down the scale" effect for each company.

Examination of Table 4 also reveals other notable differences.
First, the perceived stability of pay across the first three levels of
performance in Companies A and B, and the first two levels of
performance in Company C, indicates that the salaries in those
companies attenuate the perceived relationship between pay and
performance at lower levels of performance, although not enough to
significantly reduce the size of the perceived relationship between
pay and performance across all six levels of pay and performance. In
contrast at Company D, pay is strictly tied to performance at all
levels of performance, even low performance. For example, if you
don't sell at all in Company D, there is an 89% perceived chance that
you’ll receive no pay, in comparison to a 32% chance in Company A, an
11% chance in Company B, and a 3% chance in Company C.

In sum, this "down the scale" bias may have resulted in a
spuriously high floor effect on the perceived correlation between pay
and performance and thus reduced the chance of obtaining significant
differences in perceived reward contingency across all four
organizations. For comparative purposes, Figures 3 to 9 display the
distribution of perceived reward contingency responses for each
company in terms of correlations, and correlations converted to
Fisher’s g', the latter serving as the measure used in the study.

Null Hypothesis 3: Objective reward contingency will not be
positively related to perceived reward contingency.

Despite the lack of significant differences in perceived reward

64

Table 4: Mean Organizational Response Distributions for Perceived
Reward Contingency by Performance Level and Pay Level

 

 

 

Performance -
Level $$ 0 4% -1§_ X +l§_ +2fl
Chance (1 in 100) of
§§ Compensation for SS Sales Performance:
Pay Level
Company A
1) 0 32 0 0 0 0 0
2) -2_Q 3 13 2 2 1 1
3) -1§Q 47 55 60 45 15 1
4) X 18 19 25 29 20 20
5) +1§Q 0 12 12 22 44 31
6) +2§_ 0 0 1 1 19 47
Company B
1) O 11 7 4 0 0 0
2) -2§Q 3 3 6 4 0 0
3) -1§Q 73 79 64 42 18 4
4) X 13 10 25 44 43 32
5) +1§D 0 0 l 8 32 34
6) +2§_ 0 0 0 3 8 30
Company C
1) 0 3 0 0 0 0 0
2) ~2§Q 4O 12 2 0 0 O
3) -1§Q_ 58 87 70 14 2 1
4) X 0 1 28 61 19 14
5) +1§Q 0 0 0 25 65 38
6) +2§_ 0 0 0 2 14 48
Company D
1) 0 89 50 2 1 0 0
2) -2§Q 5 44 23 1 0 1
3) -1§Q 3 4 67 32 4 1
4) X 2 1 6 44 34 13
5) +1§Q O 0 1 19 41 36
6) +2§_ 0 0 0 3 21 50

 

Frequency

65

Figure 2: Perceived Correlations Between
Pay and Performance, Company A,
n = 8

 

 

 

 

 

Perceived r's

66

Figure 3: Fisher's 2' Transformation of

Perceived Correlations

Between Pay and Performance,

Company A, n

=8

 

 

“33%“me

flgwgflwfififiygfifimx
.9

u
an: :1" cc
A
u u. a

a.€%%%fi§i%%fi§a§

...
Y<xu.u.\u.v...rz .4

”fix.

fig

k!

 

III/IIIIII/

 

agar?

 

>833.”—

o q /IIIIIA

owmmvmmnmmPPvammmmmmwmmwmmmmmm

v-v-v-v—v-v-r-v—v— NNNNNNNNN

OOOOOOOOO

Fisher's 2's

Frequency

A

 

67

Figure 4: Perceived Correlations Between
Pay and Performance, Company B,
n = 20

 

 

 

 

Perceived r's

Frequency

68

Figure 5: Fisher's 2' Transformation of
Perceived Correlations
Between Pay and Performance,
Company B, n = 20

 

 

 

 

A

 

1* I v v v

or-Ncou-mcorxecc)..v-va-anorxcoanv—vamcorxcomm
000000000 Fv-v-v-v-v-v-v-v- NNNNNNNNN

Fisher's 2's

 

Frequency

69

Figure 6: Perceived Correlations Between
Pay and Performance, Company C,

 

 

 

 

 

n=17

8-

6-

4-

2..

O frrrlvt‘fil’v’fv
omv—mwmmmvmmm
QofoquoYou’.
o o o o o o

Perceived r's

 

Frequency

Figure 7: Fisher’s 2' Transformation of
Perceived Correlations
Between Pay and Performance,
Company C, n = 17

 

 

 

 

r frrrrrtvvw

 

O‘rvr.....rr r.......
000000000 v—v—v—v—v-v—v—v-v- NNNNNNNNN

Fisher's z's

 

7]

Figure 8: Perceived Correlations Between
Pay and Performance, Company D,
n = 53

 

50-

Frequency

 

 

 

 

om—mmmmmvmmmwmnmm
‘3 ai0“'7!0“:0“".0‘°.0".0
0 0 0 0 0 0

Perceived r's

Frequency

72

Figure 9: Fisher's 2' Transformation of
Perceived Correlations
Between Pay and Performance,
Company D, n = 53

 

       

 

 

 

Orr r ffrivllvffffivvvr rw—.
ov—vamcotxcova—vaIntonaocan—wamcolxoomm
000000000 v-v—v-v-v-v-v—v-v- NNNNNNNNN

F isher's z's

 

73
contingency across Companies A, B, and C, smaller levels of objective
reward contingency were strongly associated with smaller perceived
levels of reward contingency (partial p = .57, p g .05), leading to
the rejection of null hypothesis 3.

Null hypothesis 4: The relationship between past performance

and satisfaction (pay and job) will be invariant (i.e., not

moderate) and not significantly different from zero across
levels of perceived reward contingency.

Null hypothesis 4 was tested using moderated regression. The
first step in the analysis removed the effects of the measured
covariates, age, tenure, education, and gender. These variables are
also commonly controlled for in analyses of pay satisfaction (Dyer &
Theriault, 1976; Heneman, 1973; Hinrichs, 1969; Lawler, 1966; and
Schwab, 1973). The main effects for past performance and perceived
reward contingency were then entered in steps 2 and 3, with their
interaction entered last.

The results for standardized pay satisfaction appear in Table 5,
while those for standardized job satisfaction are displayed in Table
6. Of the covariates, only age (ARZ== .05, p g .05) explained
significant variance in standardized pay satisfaction. Past
performance explained an additional 4 percent of the variance (p g
.05) in standardized pay satisfaction; however, there was no evidence
for a perceived reward contingency main effect (A32 = .00, pppp). The
past performance (PP) by perceived reward contingency (PRC) was
entered last in step 4, but was not significant.

Parallel analyses regressing the same covariates, main effects,
and interaction on standardized job satisfaction, none of which were

significant, are shown in Table 6.

74

 

 

 

 

 

 

Table 5: Moderated Regression Analysis on Standardized Pay
Satisfaction
Hierarchical
Step Variable R2 AR2 p of AR2
1 Age .05 g .05 .05 p g .05
Tenure .05 g .05 .00 n.s
Education .05 g .05 .00 n.s
Gender .00 g .05 .00 n.s
2 Past .09 g .05 .04 p g .05
Performance
(PP)
3 Perceived .09 g .05 .00 n.s
Reward
Contingency
(PRC)
4 PP x PRC .11 g .05 .02 n.s

 

 

 

75

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 6: Moderated Regression Analysis on Standardized Job
Satisfaction
Hierarchical
Step Variable R2 AR2 p of AR2
1 Age .00 .00 n.s
Tenure .00 .00 n.s
Education .00 .00 n.s
Gender .00 .00 n.s
2 Past .00 .00 n.s
Performance
(PP)
3 Perceived .00 .00 n.s
Reward
Contingency
(PRC)
4 PP x PRC .02 .02 n.s

 

 

 

76

Although null hypothesis 4 could not be rejected for either
satisfaction measure, it is quite possible that standardizing job and
pay satisfaction within companies biased the moderated regression
analyses used to test this hypothesis. Arguments and evidence for
this alternative perspective, that is, for standardizing the
performance variables and for not standardizing the pay satisfaction
and job satisfaction variables, are expressed next.

Standardization. One of the primary uses of standardized scores
is to allow comparison of measures with different metrics (Chiselli,
Campbell, & Zedeck, 1981; Glass& Stanley, 1970). This is accomplished
by subtracting the mean from each individual score and dividing by the
standard deviation, (Xi- X)/S.D., which provides an index, in standard
deviation units, of how far above or below the mean a score is. The
resulting distribution of standard scores, regardless of the original
metric and distribution, will have a mean of 0.00 and a standard
deviation of 1.00. Because of the differences in raw sales
performance measures across companies (i.e., different products,
different prices for the products, and different industry factors),
raw performance scores were standardized by month within each
organization to facilitate comparison of individual sales performance
across companies.

Unlike the performance measures, which were not comparable,
identical pay satisfaction and job satisfaction measures were used in
each organization. Thus, the rationale for standardization of pay
satisfaction and job satisfaction has to be different than that given
for performance. It was argued (during discussion with the committee

at the proposal stage) that standardizing performance without

77
standardizing satisfaction was wrong because between group (i.e.,
across company) differences would be removed for performance, but not
for pay satisfaction and job satisfaction.

No doubt, the basic premise of this argument is correct:
variables involved in the same statistical analysis must be comparable
in terms of between group variance (Sockloff, 1975). Furthermore,
this consideration is not particular to the study of functional
turnover. Indeed, McIntyre (1990) makes the same argument with
respect to combining different data samples for criterion-related test
validation. Likewise, Dreher, Ash, and Hancock (1988) expressed
similar concerns about collapsing data across different interviewers
when determining the validity of the employment interview.

The basic issue in equating between group variance (i.e., or in
deciding whether to standardize) is determining if the differences
across groups are real or unreal Ghiselli et. a1., (1981, pp. 64-73).
Real organization differences, referred to as legitimate mean
differences by McIntyre (1990), arise from actual situational or
dispositional differences in the mean and standard deviation of the

variable being measured. For example:

"Suppose we are measuring heights of school children and
we take a random sample from each school in the city. For
various reasons it may very well be that the average
height of students from the various schools is actually
different. The observed differences among the means of
our sample are not due to sampling error and are not due
to the fact that some schools measure height with a ruler
that has 11 inches per foot and some do not. They are
real: each school is drawing students from genuinely
different subgroups," (Ghiselli et. a1., 1981, p. 73).

In this example, standardization would produce the between groups

variance problem discussed above because it would wipe out real

78

differences that exist between schools. In fact, McIntyre (1990, p.
91) suggests that when there are legitimate mean differences across
subgroups, those differences "..do not represent a problem in the
estimation of the validity coefficient."

In contrast, unreal or spurious differences are caused by
differences in measuring devices, or by rater errors (Dreher et a1.,
1988; McIntyre, 1990). Ghiselli et a1. (1981) offer another example:

"Suppose we learn that Jason received the raw scores of 45
on an arithmetic test and 72 on a history test. 0n the
other hand, Cindy, who is in another class, received a 39
on arithmetic and an 80 on history. What kinds of
between- or within-person comparisons can be made? Not
many unless we convert to z or T scores. We obtain the
means and s.d.’s for Jason's two classes and also for
Cindy's two classes, and conversions to T scores provide
the following: Jason has T scores of 40 and 64 on
arithmetic and history, respectively, whereas Cindy has 50
and 51, respectively. The comparisons we can make are as
follows: First, Jason is doing better in history than in
arithmetic relative to other students’ performance in each
of those classes. He is below average in arithmetic but
above average in history. Likewise, Cindy is above
average, and equal, in her performance in history and
arithmetic. Second, compared to other students in Cindyis
arithmetic class, she is better in arithmetic than is
Jason. The reverse is true for history. None of these
comparisons can be made if we only have raw scores
(italics added)," (p. 68).

Standardization is appropriate in this latter example because the
differences are metric differences, not real ones. In other words,
equivalence in between group variance becomes a concern only if
standardization removes "real" group differences.

Given these guidelines, it would seem that differences in sales
performance across organizations are not real differences but metric
differences produced by differences in the product, the cost of the
product, etc. And, if the performance differences were not real

differences, then standardization would not remove real between group

79
variance. Consequently when performance is standardized, there would
be no need to also standardize pay satisfaction and job satisfaction
to ensure equivalence in between group variance.

The most direct evidence on this issue is the raw sales
performance data in each company. For example, the average monthly
performance and standard deviation were $60,690 and $29,753 for
company A, $126,799 and $214,463 for company B, $82,657 and $164,104
for Company C, and $13,223 and $20,261 for Company D. Any analysis
based on these numbers would be misleading because some of the best
performers in Company D, which has the lowest dollar sales, would be
treated as poor performers when compared with the much higher dollar
sales of representatives in Companies A, B, and C.

To illustrate, recall that standardized past performance was
negatively related (; - -.34) to salary, but positively related (3 =
.40) to commissions, and total compensation (p = .11). These numbers
made sense because when sales performance increased, any increase in
total pay came from commissions, hence their strong positive
relationship with performance. It did not come from salary, which was
generally unchanged from month to month. Indeed, it was the lack of
variability in salary which attenuated the correlation of average
total compensation (salary + bonuses) from that of commissions.

In comparison, the correlations with unstandardized performance
are .33 for bonuses, .52 for total compensation, and .41 for average
monthly salary which, contrary to the correlations with standardized
performance, suggests that total compensation is more strongly related
to performance than bonuses, and that salary actually increases with

performance. In fact, these spurious correlations occur because the

80
sales representatives in Company D, who also have the lowest dollar
sales, were only paid commission and received no salary, whereas the
remaining sales representatives all have larger dollar sales and
received regular monthly salaries. Again, these results seem to
indicate that the differences in performance across organizations are
nothing but scale differences.

A second consideration in determining the threat of "real
differences" is the purpose of each part of the data analysis. If the
purpose was to determine the effect of higher levels of reward
contingency (across companies) on performance, standardization of
performance would be inappropriate because it would remove the effect
of objective reward contingency on performance (i.e., real
differences). However, this is not the purpose as neither the
moderated regression analysis, nor the discriminant analysis, nor the
functional turnover variable itself involves a comparison of absolute
performance across companies.

According to Stone and Hollenbeck (1984), "One variable (e.g.,
z) is said to moderate the relationship between two other variables
(e.g., x and y) if the degree of association between x and y varies as
a function of the value assumed by z . . .," (p. 196). Given that
hypothesis 4 predicts that perceived reward contingency (2) will
moderate the relationship between past performance (x) and
satisfaction (y), standardization of performance within organizations
would produce an inappropriate test of this hypothesis only if it
changed the relationship between performance and satisfaction as it

exists within each level of objective reward contingency. But, since

standard scores are perfectly correlated with raw scores with

81

organizations, changing raw scores to standard scores will not change
an individual's rank order on performance or the shape of the
distribution of performance scores within each organization or level
of reward contingency. And, it is the change in the correlation
between performance and satisfaction across levels of perceived reward
contingency that was being investigated in the moderated regression:
absolute differences in performance were not of interest.

Similarly with the discriminant analysis and functional
turnover, absolute levels of performance were not considered.
Instead, individual performance was relative to average performance
within each company. Thus, the performance comparison is not across
companies (or levels of reward contingency), but within companies.
Because standardization occurs within levels of reward contingency
(i.e., organizations), it should be recognized that reward contingency
is a constant for each individual performer. And, as a constant
within that organization/level of reward contingency, reward
contingency does not affect the relative performance difference
between individual performance and average performance. Note,
however, that the absolute magnitude of this difference across
companies will not be comparable in raw performance scores because of
differences in product prices, etc. All standardization does is
ensure the comparability of functional turnover across organizations
by removing the level and dispersion differences caused by these
factors. Functional turnover scores become comparable across
companies because they are now stated in standard deviations units
which preserve the within company comparison of average performance to

the performance of stayers and leavers.

82

For example, assume that in Company Q, X = $1,000,000
and s.d. = $200,000, and that in Company 2, X = $600,000 and s.d. =
$100,000. Also, assume that two exemplary salespeople (+3 s.d.'s;
$1,600,000 in Company Q, and $900,000 in Company Z) have decided to
leave. Using raw performance, the functional turnover score for the
Company Q sales representative is $-600,000 (-1 x ($1,600,000 -
$1,000,000)) and $—300,000 for the Company Z sales representative (-1
x ($900,000 - $600,000)). Thus, using raw performance it appears that
flow is twice as bad in this instance for Company Q as it is for
Company Z. If we were determining the utility of functional turnover,
this would be the proper conclusion. We know, however, from earlier
discussion, that functional turnover is concerned with relative
performance, not absolute performance, and that the absolute
difference in the Company Q and Company Z functional turnover scores
is due to differences in products they sell. In relative terms, not
the absolute utility of functional turnover, Company Q and Company Z
experience identical consequences when these individuals leave. Both
companies lose salespeople 3 s.d.'s above the mean. Standardization
equalizes these differences by ensuring the comparability of these
deviation scores in terms of level and dispersion. Thus, after
standardization, functional turnover for the sales representative in
Company Q is -3.0 (-1 x +3), the same as it is for the sales
representative in Company Z (-1 x +3).

To summarize, if the performance differences across
organizations are not real differences, and if the moderated

regression analysis and the functional turnover variable rely on

83
within organization relationships and comparisons, then
standardization of performance does not remove real between group
variance. Thus, based on the rationale that "variables involved in
the same statistical analysis must be comparable in terms of between
group variance," there would be no need to standardize pay
satisfaction and job satisfaction to ensure equivalence in between
group variance when it already existed.

In fact, standardization probably accomplishes just the opposite
by removing real organization differences in pay satisfaction and job
satisfaction across companies. While standardization did not remove
any significant level effects, as evidenced by nonsignificant mean
differences in either satisfaction variable across organizations, in
all likelihood it probably removed real differences in the range and
variance of satisfaction across organizations. For example, the
standard deviation of pay satisfaction for Company D (s.d. = 16.78) is

significantly larger than the standard deviations for Company A (s.d.

5.29; Cochran's C (30,2) = .91, p g .000; Bartlett-Box F (1,1282) =

\0

.13, p g .003), Company B (s.d. = 11.93; Cochran’s C (36,2) = .66, p

IA

.044; Bartlett-Box F (1,8088) = 2.83, p g .093), and Company C (s.d.

8 10; Cochran's C (34,2) = .81, p g .000; Bartlett—Box F (1,6010) =
9.58, p g .002).

These variance differences are real differences which generally
can be predicted by the level of reward contingency and the variance
in pay in each company. For example, the standard deviation for pay
satisfaction should be extreme in Company D because the affective
range of pay is extreme. That is, if you sell, you earn commissions,

and you’re more satisfied with your pay. If you don’t sell, you don’t

84
earn any commissions, and ought to be very dissatisfied. The somewhat
smaller standard deviations in Companies B and C can be attributed to
the floor effect of reward contingency as all B and C sales
representatives receive a monthly salary regardless of sales
performance. Similarly, the smallest standard deviation in pay
satisfaction occurs in company A which pays a monthly salary, and has
the lowest level of reward contingency and the smallest range in
bonuses and total pay.

Finally, by removing the variance differences in pay
satisfaction and job satisfaction across organizations,
standardization artifactually creates a restriction in range which
will generally attenuate correlations with other variables. For
example, the unstandardized and standardized correlations for pay
satisfaction shrink, respectively, from .15 to .00 with salary, from
.20 to .15 with bonuses, and from .27 to .14 with total compensation.

In conclusion, the alternative perspective is that it is
appropriate to standardize performance without standardizing pay
satisfaction and job satisfaction because standardizing performance
removes metric differences while standardizing job satisfaction and
pay satisfaction would remove real between group differences.

Null hypothesis 4, unstandardized job satisfaction and

unstandardized pay satisfaction: The relationship between past

performance and unstandardized satisfaction (pay and job) will
be invariant (i.e., not moderate) and not significantly
different from zero across levels of perceived reward
contingency.

The results of the moderated regression analysis using

unstandardized pay satisfaction appear in Table 7. Both age (ARZ==

.05, p g .05) and tenure (ARZ== .06, p g .05) explained 11% of the

 

85

 

 

 

 

 

Table 7: Moderated Regression Analysis on Unstandardized Pay
Satisfaction
Hierarchical
Step Variable R2 AR2 p of AR2
1 Age .05 S .05 .05 p g .05
Tenure .11 g .05 .06 p g .05
Education .11 g .05 .00 n.s
Gender .11 g .05 .00 n.s
2 Past .15 g .05 .04 p g .05
Performance
(PP)
3 Perceived .15 g .05 .00 n.s
Reward
Contingency
(PRC)
4 PP x PRC .19 g .05 .04 p g .05

 

 

86
variance in pay satisfaction, compared to just 5% for age using
standardized pay satisfaction. The main effects were essentially the
same, as past performance explained an additional 5 percent of the
variance (p g .05) in pay satisfaction, while the perceived reward
contingency main effect accounted for none (ARZ=- .00, ppsp). However,
unlike standardized pay satisfaction, the past performance X perceived
reward contingency interaction accounted for significant additional
variance (ARZ=- .04, p g .05) beyond the covariates and main effects
for unstandardized pay satisfaction. The nature of this interaction
is revealed in Figure 10, in which the relationship between past
performance and pay satisfaction is plotted for values il.0 SQ units
of unstandardized past performance. Consistent with previous research
(Cherrington et. a1., 1971; Greene, 1973; and Jacobs & Soloman,
1977), the relationship between performance and unstandardized
satisfaction is positive and stronger (p = .40, p g .05) under
conditions of higher perceived reward contingency than under
conditions of lower perceived reward contingency (L = .03, n s.). In
practical terms, increasing perceived reward contingency one standard
deviation (i.e., from Fisher g' = 1.14 to Fisher ;’ = 2.17; Pearson
correlation equivalents are from p = .815 to p = .975) resulted in
approximately a one standard deviation difference in unstandardized
pay satisfaction between poor performers and high performers.

Parallel analyses, the results of which are shown in Table 8,
were conducted for unstandardized job satisfaction. However, unlike
pay satisfaction, the results for unstandardized job satisfaction were
nearly identical to those for standardized job satisfaction, meaning

they were not significant.

Pay Satisfaction

38

87

Figure 10: Regression of Unstandardized
Pay Satisfaction on Past
Performance by Perceived
Reward Contingency
Interaction

 

   
 
 

 

 

 

- Low Perceived

, Reward Contingency

- r

—‘

- High Perceived

. Reward Contingency
I

 

0

Past Performance

 

88

 

 

Table 8: Moderated Regression Analysis on Unstandardized Job
Satisfaction
Hierarchical
Step Variable R2 p AR2 p of AR2
1 Age .00 pp .00 pp
Tenure .00 pg .00 pp
Education .00 pp .00 pp
Gender .00 pp .00 pp
2 Past .00 pp .00 pp
Performance
(PP)
3 Perceived .00 pp .00 pg
Reward
Contingency
(PRC)
4 PP x PRC .03 pp .03 pp

 

89

Evidence for study design. In summary, the evidence concerning
the study design is positive. First, the primary design consideration
was met since there were significant differences in objective reward
contingency across all organizations. Second, even though the 4-1evel
objective reward contingency design was associated with only a 2—level
perceived reward contingency design (i.e., essentially a comparison
between perceptions of complete contingency and strong partial
contingency), objective reward contingency was strongly correlated (p
= .56) with perceived reward contingency. Thus, the expectancy
theory boundary conditions that workers must perceive clear
performance reward contingencies, and, that objective reward
contingencies should influence perceptions of reward contingencies,
were met. Finally, consistent with previous research on reward
contingency and satisfaction, perceived reward contingency did
moderate the relationship between past performance and unstandardized
pay satisfaction. While this moderated relationship did not hold for
job satisfaction, the absence of that moderated relationship is
understandable since pay satisfaction and job satisfaction are at best
moderately related (see below), and because of evaluative consistency
(Fisher, 1980), meaning that pay-related reward contingency should
moderate the relationship between performance and the specific facet
of pay satisfaction, but not an overall measure of general job

satisfaction.

Model Testing Hypotheses

The statistics used to interpret discriminant and classification

analysis are explained first. Then the results of the analysis,

90
beginning with hierarchical discriminant analysis which is used to
test null hypotheses 5 and 6, proceeding to interpretation of
discriminant functions, univariate analyses, and orthogonal contrasts,
and ending with linear classification analyses, are reported.

Interpreting discriminant analysis. In discriminant analysis,
the significance and contribution of each discriminant function to
group separation is indicated by the canonical correlation and Wilks
lambda. The canonical correlation, RC, associated with each function
represents the degree of relatedness between the groups and the
discriminant function that is literally formed by correlating the
discriminant variate, representing the linear combination of
discriminating variables, with the groups' variate, which consists of
the linear combination of the dummy coded group variables (Klecka,
1980, pp. 36-37). The larger the canonical R for each function, the
more strongly the discriminant function is associated with group
differences.

Wilks lambda is an "inverse" multivariate measure of group
differences. Values near zero indicate strong discrimination between
groups, while values approaching 1 indicate progressively less
discrimination (Klecka, 1980). Lambda parallels R2 in multiple
regression because lambda subtracted from one represents the overall
canonical R2, which is the proportion of variance in between-group
differences accounted for by the discriminant functions (McLaughlin,
1980). Lambda yields both an overall chi-squared test, as well as
providing a chi-squared test for each discriminant function. The
rationale underlying lambda is to keep deriving discriminant functions

until the residual discrimination is nonsignificant, in other words,

91

up to the point where additional discriminant functions fail to
account for significant incremental separation between groups
(Klecka, 1980). Rather than testing the lambda directly associated
with the addition of a discriminant function, "we examine the residual
discrimination in the system prior to deriving that function,"
(Klecka, 1980, p. 38). Thus, the reported lambdas are designated
"Wilks lambda prior to removal of function" in the tables.

Five statistics are used to test the classification analysis.
The overall hit rate, which is the percentage of correctly classified
cases across all groups, is tested by comparing it to the proportional
chance criterion (Huberty, 1984). The proportional chance criterion
indicates the percentage of correct hits that could be expected on the
basis of chance alone (Huberty, 1984). Here, adjusted for differences
in group sizes, that percentage is 33.28%. Huberty (1984, p. 166-167)
provides a formula for a standardized normal statistic, Z. to test the
significance of the overall hit rate compared to the proportional
chance criterion. Last, it is crucial in functional turnover research
to determine the classification accuracy of "separate-group" hit rates
(Huberty, 1984). In this way, it becomes possible to specify which of
the four groups (i.e., poor performing leavers, good performing
leavers, poor performing stayers, or good performing stayers) a set of
variables proves useful in classifying. Huberty (1984, p. 167,
formula 9) also provides a formula for a standardized normal
statistic, Z. to test the significance of separate-group hit rates.

The first test for all discriminant analyses was the Box M test
for homogeneity of variance-covariance matrices across groups

(McLaughlin, 1980). When Box’s M was significant, quadratic

92
classification (classification based on separate group covariance
matrices) was substituted for the default classification based on
pooled covariance matrices. However, no adjustment is typically made
to the discriminant analyses, even though the unequal covariance
matrices will produce a conservative test by creating a bias in favor
of the null hypothesis (McLaughlin, 1980).

Null Hypothesis 5: The effects of objective reward

contingency and perceived reward contingency on functional

turnover will not be mediated by pay satisfaction and job

satisfaction.

Null Hypothesis 6: The effect of performance on

satisfaction will not be moderated by perceived reward

contingency. That is, there will be no significant pay

satisfaction and job satisfaction differences across the

four functional turnover groups.

Before testing these hypotheses, recall that one consequence of
the study design is that organization and objective reward contingency
are confounded. One method of assessing the degree to which these
variables are confounded is to dummy code organization, enter the
three dummy coded organizational variables first in a discriminant
analysis, and then enter objective reward contingency. If these
variables are fully confounded, then the addition of objective reward
contingency should not result in significant increases in variance or
classification accuracy.

As shown in Table 9, organization alone accounts for significant
variance (29%; R? = 1 - Wilks Lambda) in functional turnover, while
allowing classification of the four functional turnover groups (52%)
at a rate significantly better than chance (33.28%). In comparison,

objective reward contingency explained greater variance (34%) by

itself than the organization dummy variables, while also producing

93

HNUHm 8“ cwmnfiwawswsfi vaHKmHm emmdwso awn nonwocsdma mmmmnnm om OHQmSHdewo: mum
wmzde.06588:©m:n<.

 

 

owmofl8gnsmsfi.>sm8<mwm 08mmmwmwomdwo: vaHKmHm
28506 0.85699
<WH8mUHmm zaaUmH om pHHOH Ho
8: mum mwodwmwnmdd Hmao<m8 ocmHmHH

macmnwo: manndwosm mm om mcsndwo: x8 mm 0 $88 wmdm 880 N8

 

 

own 8 .88 .88 88.88 8 .8888 88.888 8.88 It:
888 8 .88 .88 88.88 8 .88888 88.888 8.88» in:
888 8 .88 .88 8.88 8 .88 88.888 8.88» In:
own + 888 8 .88 .88 88.88 88 .88888 88.888 8.88» 8.88»
own + 888 8 .88 .88 88.88 88 .8888 88.888 8.88» 8.8.
888.+ 888 + 8 .88 .88 88.88 88 .88888 88.888 8.88» 8.8.

 

ENG

 

 

94

emowm 8 A8285;

2000. we Hm 9m 0850:8an w mmmonwmdma £89 988888.888:de 950.8808 . 8m gm
UHOpOanosmH 0.3858 08.8de80: 38.8880 . 88.. 88me gm 8:08.885:

0886388888880: 8800ch 958.. 859 8.8895 gm 68.988085 Hump 8: go mumPKmHm .
men: m Hm m mamamawmmm 30558 mgnwmfio 688me o: m moan @8988me UK
8.388883 3.888. m8. 883 . own Hmpfimmgdm 9m 9.828% 00me 0385888380:
58.888888 . own 88.. 03. madman Hmzma 0038833. 88880 “.88 38.88/59 8658.9
nosngg.

w
wmbm

 

95
classification (58.16%) that was higher and significantly better than
the proportional chance criterion. When objective reward contingency
was added to the organization dummy variables, there was a sizable
increase in variance explained, from 29% to 49%. Likewise, the
overall classification hit rate increased from 51.02% to 60.20%, which
was a significant increase in classification accuracy (Zi = 1.82, p g
.05). (For comparison, similar analyses are presented for perceived
reward contingency.)

Because organization and reward contingency are confounded,
organization, along with education, tenure, age, and education (i.e.,
measured pre-existing differences), will be entered first as
covariates in all discriminant analyses. Note, however, that the
results of the previous analysis do indicate that organization and
reward contingency are not fully confounded.

Hierarchical discriminant analysis was used to test null
hypotheses 5 and 6 that pay satisfaction and job satisfaction mediated
the effects of the antecedent variables on functional turnover, and
that pay satisfaction and job satisfaction contributed to significant
differentiation between the four kinds of functional turnover. As
shown in Table 10, the covariates (i.e., organization, age, gender,
education and tenure) entered in the first step of the discriminant
analysis produced significant differentiation (p g .02) of the
functional turnover groups, while also allowing classification
(51.02%) that was significantly better than the proportional chance
criterion

Null hypotheses 5 and 6 were tested in step 2 with the addition

of job satisfaction and pay satisfaction. Neither null hypothesis

 

96

 

 

 

888888 80" 0858898858” 355.888. ammdbd arm 38888.89 mmmmnflm OM 88884 mmdwmmenHo: 85
no.8 88888830880: 0: 88508808888 8858958..
mendgd ESSEHm 08888883088080: 858888888
888me 88:88.88.
”83.8.8888 25869.» 08. 888.808. no
8: gm $038383” 8888988 0.88.8888
58888880: 95088085 we om 9:08.80: M8 88 m $80 888880 NF. 88
mdmm 8 »
OH©8 H mm .00 00.08 NH .00 mH.0Nw 0.80 III!
wow . 550888.80?
0259... embcwm
mdmw N »
mfimU H. H mu m8 80.NH Nu .om m8.omw 0.80 D.m.
883‘. men.
moo mmd
mdmw 8 8 »
mfimfi H‘ H .mm 88 08.0H 00 .OOOH 0N.N8w 0.00 H.0m
mdmp 8 .

888.8888

 

 

 

97

8.88888 88 Sound; .

E.

we Mm gm 850580888 88 mmmoowmnmn £89 358.888.8558.. 9:08.803.
980880588088 0888500 08.882880: 38.8880 . N. #888 gm 8:08.95:

8pm gm

0886330de0: 880053088. om 808.8 8.80. 09088.. m 8m 88 mwmamawuma 508.5888

mamdwmdwn Ummoa o: m 8.0558888 888.9858 8% 888803 A8888. p. 888 .

088w. mmd Mm

883‘ mmnwmmmnnwos. moo mmn 8m no.8 mmdwmmmnnwos. two 8m 0988ng 8653
002853. 038588888880: 8.88.8853 gm 9H8 955% 00me 03858888303888

4888.88.38 .

 

98
null hypothesis could be rejected on the basis of the discriminant
analysis because the overall level of differentiation indicated by
lambda was unchanged. Furthermore, the addition of job and pay
satisfaction did not produce a significant increase in classification
accuracy beyond step 1 of the analysis.

Perceived reward contingency and objective reward contingency
were added to the last step of the analysis, meaning that all
variables were entered in a direct discriminant analysis in step 3.
This full equation resulted in one significant discriminant function
which explained 32% (Canonical 32:: l - .68 = .58) of the variance in
functional turnover group differences. The classification analysis
produce by this equation was also significant when compared to the
proportional chance criterion, and represented a significant
improvement in classification accuracy over step 2 (p S .052).

The improvement in classification accuracy and variance
explained at step 3 stemmed primarily from the addition of objective
reward contingency, not perceived reward contingency. As shown in
Table 11, when perceived reward contingency was added to steps 1 and
2, there was no appreciable difference in either the discriminant or
classification analyses. In contrast, when objective reward
contingency was added to steps 1 and 2, classification accuracy
increased from 51.02% to 62.24% (p g .052), while variance explained
increased from .33 to .54, levels which were similar to the full
equation results in step 3. For comparison, Table 11 also shows the
separate effects of perceived reward contingency and objective reward

contingency.

99

688888 88"

088.088.888.8888 8588888888 8.8888880 56 085588 mmmmonm 08 868088388 888589
0088883888088. 880 0.8. 800.28 8658.88 00888803.

 

0880888888888” 2.58%me

0888888388880: 255.888

 

 

 

280.86 88889.8
8868886808 288808. om 888.808. «0
83 gm mwodwmwomsd 5% 958.8888
€30: 5.5988088 we on 88.580880: M8 mm m 380 wmdm Nmn N8
8888 8 8 8. 8 .88 .88 88.88 88 .88 88.888 8.88» 8.8
8.888
8.888 8 8 8. 8 .88 .88 88.88 88 .8888 88.888 8.88» 8.88»
088
8.88 8 .88 .88 8.88 8 .88 88.888 8.88» in
own 8 .88 .88 88.88 8 .88888 88.888 8.88» l:

 

100

 

208m. ma 8m 9m 08503888 x mmmoo8mnmg £8.85 8885838888888 8859803. 8m gm
888.038.889.58 03% 088888880: 38. 88¢. N8 889.. gm 8:08.95:
0838888880: 80:88.88 989.. Ed 8595 .8..an 8 ma mdmv 8 A88. 23. 88.0: 8 8m
88 385888.08me 50888888 mamd8md8o 68me o: 8 8.0588888 68.98509 .88. magma A888
88. 883. mdmv 8 868mg Om mom 09808580: 03.858. #88388 858 9m 95
000mm 08.08888888805 <mH8mU8mm. mdmv 8 838mg 08 8898. 888.858.880.880: 850 80.8

mmd8mmm0n8oa. E0 8m 86880888850 88938.9 00588850908. own 8m 08d. mon8<m 88288.0
00:88:82.8?

.888 m .88

 

lOl

Discriminant functions. Returning to step 3 of the main
analysis where all variables were entered simultaneously: the group
centroids, shown in Table 12, indicated that the discriminant function
was able to differentiate stayers from leavers, and poor performing
stayers from good performing stayers. However, as evidenced by their
similar group centroids, good performing leavers were not readily
differentiated from poor performing leavers.

Examination of the structure coefficients in Table 13 indicates
that most of the variance in the discriminant function can be
attributed to objective reward contingency because of the large
structure coefficient (r = .77) associated with commissions and the
more modest structure coefficients associated with the remaining
discriminating variables. This coefficient for objective reward
contingency suggests, in conjunction with the group centroids, that
poor performing leavers and good performing leavers receive most of
their total pay from commissions, followed by good performing stayers,
and then poor performing stayers.

Analysis of variance and orthogonal contrasts. Follow-up
univariate analyses of variance, reported in Table 14, indicated that
objective reward contingency, tenure, and education accounted for
significant differences in functional turnover groups. The means and
standard deviations for each functional turnover group, as well as the
orthogonal contrasts on variables which had significant univariate
effects are shown for each functional turnover group in Table 15.

Orthogonal contrasts for education showed that poor performing
stayers had more education than either good performing leavers or good

performing stayers. Contrasts for tenure revealed that stayers had

 

 

 

102

Table 12: Group Centroids in the Discriminant Space

 

Group Centroids

 

 

Group Function 1
1) Poor Performing Leavers -l.12
2) Good Performing Leavers -.92
3) Poor Performing Stayers .96

4) Good Performing Stayers -.O6

 

 

103

Table 13: Structure Coefficients Between Discriminant
Functions and Discriminating Variables

 

 

Variable Function 1*

Age .11
Gender -.24
Education -.29
Tenure -.23
Pay Satisfaction .02
Job Satisfaction -.08
Objective Reward .77
Contingency

Perceived Reward .14
Contingency

Organization 1 -.26
Organization 2 -.29
Organization 3 -.l9

 

 

Note. Organization 1, 2, and 3 are dummy coded variables
for organization. Structure coefficients represent
the correlation between individual variables and the
linear discriminant functions which maximize group
separation.

*2 .05

|/\

104

 

 

Table 14: Univariate Analyses of Variance.
Variable Univariate F p of F
Organization 1 1.97 .12
Organization 2 2.48 .07
Organization 3 2.34 .08
Age .76 .52
Gender 1.59 .20
Education 2.90 .04
Tenure 2.79 .05
Pay Satisfaction 1.04 .38
Job Satisfaction .31 .82
Objective Reward 16.38 .00
Contingency
Perceived Reward .73 .53
Contingency

 

Note. Analyses were conducted with 3 and 94 degrees of freedom.

105

 

 

Table 15: Means, Standard Deviations, and Orthogonal Contrasts
Between Functional Turnover Groups.
Poor Good Poor Good
Performing Performing Performing Performing
Variables Leavers Leavers Stayers Stayers
Org 1 0.00 0.00 0.18 0.07
( .00) ( .00) ( .39) ( .25)
Org 2 0.00 0.14 0.36 0.17
( .00) ( .36) ( .49) ( .38)
Org 3 0.00 0.00 0.25 0.22
( .00) ( .00) ( .44) ( .42)
Age 39.00 35.79 35.64 38.04
(9.97) (7.50) (7.82) (8.68)
Gender 0.00 0.07 0.21 0.09
( .00) ( .27) ( .42) ( .28)
Education 3.30 3 . 363 a . 212“ 3. 783
( .79) ( .74) ( .92) ( .99)
Tenure 2.233“ 2.673“ 6.3812 5.5112
(2.04) (1.06) (5.90) (5.71)
Pay 26.00 35.57 31.86 33.78
Satisfaction (19.91) (12.94) (12.41) (14.34)
Job 19.90 20.71 20.89 20.39
Satisfaction (2.99) (3.54) (2.48) (3.36)
Perceived 1.80 1.75 1.57 1.64
Reward ( .25) ( .57) ( .50) ( .55)
Contingency
Objective 1.0034 .913 .39124 ‘7713
Reward (.00) ( 25) ( 38) (.29)
Contingency
Total 31,2552“ $3,1841 $3,057“ $3,86213
Compensation (S 801) ($2,100) ($1,437) ($1,835)
Commissions $1,255“ $2,80713 $1,067“ $2,844”
(S 801) ($2,014) ($1,182) ($1,544)

 

106
Table 15 (cont'd.).
Note. Group standard deviations are presented in parentheses below

group means. Tenure contrasts between 1 and 4, and 2 and 4,
are marginally significant at p g .07.

1Significantly different from poor performing leavers, p g .05.
2Significantly different from good performing leavers, p g .05.
3Significantly different from poor performing stayers, p g .05.
“Significantly different from good performing stayers, p g .05.

107
significantly higher tenure than leavers, regardless of performance.

The most important contrasts, however, were for objective reward
contingency. Poor (1.00) and good performing leavers (.91) received
all or most of their pay from commissions, which was significantly
larger than the percentage earned by good performing stayers (.77)
which, in turn, was significantly higher than the percentage of
commissions relative to total salary (.39) received by poor performing
stayers.

Additional contrasts indicated that good performing stayers had
significantly higher total pay than poor performing stayers and
leavers. However, the same contrast with good performing stayers was
nonsignificant (p g .19). There was also no difference in total pay
between good performing leavers and poor performing stayers, but both
of those groups' total pay was significantly larger than the pay
received by poor performing leavers. Contrasts for commissions showed
that good performing stayers and leavers earned equivalent
commissions, which were significantly larger than the commissions
earned by poor performing stayers and leavers.

Thus, when total pay and commissions are considered together,
it's evident that poor performing leavers, good performing leavers,
and good performing stayers, received most of their total compensation
from commissions, while poor performing stayers received most of their
pay from salary. In fact, the average monthly salary received by poor
performing stayers ($1,990) was larger than average commission
received by poor performing leavers ($1,255).

Classification Analyses. The full classification matrix derived

from direct entry of all variables (step 3) is presented in Table 16.

108

emU8m 88”

08mmm8m8omw8o: >dm8<m8m 08 «cm 0H8o8sm8 mmat8m A: H 888 Umw8<mm 8803 m
U8Hmow D8mofi8a8smsfl >:m8<m8m 08.888 O8mow8a8smd8so <mH8mU8mm

 

08mmm888mm 8:80 QHQCU

 

 

wodzm8 mHocU.zmaUmHm:8p owmmm 8 8 8
8
8V woow mowmona8so.wmm<mHm 8o 8 8 o 8
8888 88888 888 2888
88 0000 mMHmons8so.Hmm<mHm 88 8 8 8 8
, 3.8 888.888 3.888 888.888
88 moon mowmona8so mwmkmfim 88 o o 88* 88
888 888 888.888 888.888
88 noon mmHmona85© mnmkmwm 88 o 8 8 88»
8888 88.888 888.888 888.888

 

zono. o<mHm88 08mmm8m8omn8o: moocnmo%.zmm 88.888.

Hmdm 8m m805888omsfi8<.8mnfimn #58: 038508 Am M .088.

>mdmfi8mw 8:98omdmm numd ofiocw 38d

109
As mentioned above, the overall classification rate of 62.24% was
significant when compared to the chance criterion. Based on a formula
presented by Huberty (1984), this level of classification represented
a 43% improvement over chance classification.

In addition to assessing the overall level of classification,
functional turnover research must also examine separate-group hit
rates to determine which of the four functional turnover groups is
accurately classified. In actuality, the significant level of overall

classification demonstrated in Table 16 results only from accurate

 

prediction of good and poor performing stayers. Neither good nor poor
performing leavers were accurately classified. In other words, the
large amounts of explained variance reported for the discriminant
analysis results from the prediction of performance, from separating
poor performing stayers from good performing stayers.

Because of the upward bias in classification accuracy that
occurs when "validation is based on the same cases used to derive the
classification functions" (Klecka, 1980, p. 51), this classification
analysis was double cross-validated using questionnaire nonresponders
(n - 128, after attrition due to missing data) from each organization.
However, since most of the discriminating variables were obtained by
questionnaire, only a limited double cross-validation could be
attempted. The data common to responders and nonresponders were
overall performance and turnover, which permitted categorization of
functional turnover groups, and total compensation and commissions,
which served as the discriminating variables.

A discriminant analysis of the original sample using only

objective reward contingency (see Table 9) yielded one significant

110
discriminant function and an overall classification hit rate of 58.16%
which was significant against the proportional chance criterion. The
full classification matrix, shown in Table 17, again shows that the
significant overall level of classification is attributable to
prediction of good and poor performing stayers.

A replication of that discriminant analysis using the sample of
nonresponders yielded one significant discriminant function (Canonical
R = .35, Wilks lambda - .88, chi-squared = 15.88, g: = 3, p g .0012)
and an overall classification hit rate of 46.88%, which was

significant against the proportional chance criterion (g = 2.89, p g

 

.0019). That nonresponder classification matrix is shown in Table 18.
Once more, the significant overall classification rate is attributable
to prediction of good and poor performing stayers.

The classification coefficients from the original sample were
then used to classify the sample of nonresponders into functional
turnover groups. The cross-validation classification matrix is shown
in Table 19. There was some deterioration in predictive power because
the cross-validation sample of nonresponders was predicted with 42.97%
accuracy (g = 1.96, p 5 .0250), compared to 62.24% in the original
sample. However, in contrast to previous analyses, only good
performing stayers were classified at a rate better than chance.

The linear classification functions from the sample of
nonresponders were then used to double cross-validate cases from the
original sample into functional turnover groups. The double cross-
validation classification matrix is shown in Table 19. This time,

there was much less deterioration in predictive power because the

ll]

 

 

8.888888 88" 08888888888880: >388<m8m 08 go 08.808858 82888888 A: u 888 88.843 88.08 m
0888583858 85888888.. 88mg 98“.. 888.3 8658.8 003888.8de

 

0885888883 8580 08.086

 

 

88988888 08686 335888 ommmm 8 8 8
8
88 88008. 868.808.5850 8898on 80 o o 8 80
8888 8888 8888 888888
88 0009 888.808.5850 88358.8 88 o o 8 88
8888 88088 3.888 888.888
88 888. 8888.08.83.88 88888.8 88 8 8 88... 88
8888 8888 888.888 888.888
88 888 89.88.5888 88888 88 8 8 8 88s
8888 8888 888.888 888.888

 

208. 958.888 0888388888880: 8809888088 Sum 88 . 888 . $88.88." 8388888.. gun 88,0588 8888..
8.88am 8m 303888082888. vmflflmfi .985 08.858 6 M .088 .

112

 

em68m 8m" 08mmm8m8omn8os.>:m8<m8m om firm zosHmmUosamHm A: n 8wmv DmH8<mQ 8H0; m
U8mow8585m5fl vbm8<m8m cw8so Oquon8<m wmzmHa.ndsn8:om:n<

 

08mmm8MHmQ 83Ho muosw

 

 

>nd5m8 chasm zmaUmHmU8w nmmmm 8 m u
A
88 MUCH mMHMOHBHDQ Hmm<mHm No 0 o M 8m
Aowv Aowv A8owv Aoowv
NV 0009 umwm0H58:©.Hmm<mHm m o o o m
Aowv Aowv Aowv AHOowv
88 moow wMHHOHBHSQ mnmwmfim 8m 0 8 8m mm
80$ 8on 88.8.3 88.808
3 8008 868.808.2880 magma 88 8 8 88 3»
80$ 8on 380$ 380$

 

zowm. o<mHm88 08mmm8m8omd8o: mQQCHmo<.swm pm.mmw. >mnmfi8mw 8sa8omdmm nsmn owoaw 38¢
Hmdm 8m m8od8M8omsn8<VUmdde dam: armsom Aw M .omv.

113

 

 

amgm 8m" . $1588..ng 08mmm8m8omfl8o: 258835 On zonnmmvoama A: H 883 wmmma
0: 888g 08mmm8w8omd8o: 559803 09.8859 8.8.05 gm 083808.858 mm868m
A: n 88 88mg 03. mon8<m wmzwa 003888.853

 

08mmm8m8mn 8380 0.8.086

 

 

>958 08.086 396055888 ommmm 8 N u
A
88 8.008. 38.808.58.30 8mm<m8m No 8 o M 8m
803 808 A888 3803
NV 0009 E05888 8mm<m8m m o o o m
803 303 803 388on
3 88008. 85,8»ng mamvxmwm 88 o o 88 mm
80$ 808 38.8on 38.808
88 0009 388,055 mdm<m8m mm o 0 MN 88.»
80$ 83 88.808 88.808

 

20.5. 05388 08mmm8m8omn8o: woncfimow 5m. 3.8%?
3.8m 8m m8od8m8nmfid8w Umdflmw 9m: 25:8 6

~6me5? gownmm GEN @358 Ed
.A. .888 .

111+

 

.8885 No" coagm Onomml<m889mdma 08mmm8m8omd8o: §<m8m OH. 9m 03085888 mmBEm
A: u mg 88.8me 0: 8.8588. Bmmflflomfio: 350888085 Own/69 5.08 gm
$8680 08. 2038880395 A: n 8N3 88mg 058% 03.888885 wmzma

oosngg

 

08mmm8m8m9 8380 0896

 

 

880.958 0386 3988808.ng nmmmm 8 w u
8
88 8608. £053.50 8mm<mum 8o 8 o 0 80
888 888 888 38888
3 0009 £08.33 Hmm<mwm 88 o o 8 88
888 888 3.888 $8.888
88 888. 38.8.0383 83888.8 88 8 8 88* 88
8888 8888 8.88.888 38.888
88 0000 865808583 mdmwmnm 8m 0 o m 88»
888 8888 888.888 888.888

 

zonm. 95388 08mmm8m8omw8o: 88005308. £88m 3.8888.

>mnmfl8mw 859808188 $881“ 08.086 EH

3mm 8m m8058m80mbd8< 598an gm: 08558 E M .03 .

115
double cross-validated classification accuracy of 57.14% was
significant when compared to the proportional chance criterion (; =
5.01, p 5 .00001), and was almost the same as the original
classification accuracy of 58.16%. Again, the significant overall
classification hit rate was due to classification of good and poor

performing stayers.

 

CHAPTER FIVE

DISCUSSION

The purpose of this field study was to investigate the
relationship between reward contingency and functional turnover. The
study was conducted across four companies, each of which rewarded its
sales staff according to a different level of objective reward
contingency, from mostly salary, to salary plus commission and
bonuses, to straight commission. The results bearing on each of the
substantive hypotheses, starting with hypothesis 6 and then hypothesis
5, are discussed below. (The results pertaining to the design
hypotheses (1 -4) were discussed in Chapter 4).

Stated in the affirmative, hypothesis 6 predicted that if
perceived reward contingency moderated the relationship between
performance and satisfaction, pay satisfaction and job satisfaction
would be related to functional turnover (i e., lead to accurate
discrimination between functional turnover groups). Yet, despite this
moderated relationship for pay satisfaction, null hypothesis 6 could
not be rejected because neither pay satisfaction nor job satisfaction
contributed to significant separation of the four functional turnover
groups. The only sizable difference in satisfaction, though not
statistically significant, was that poor performing stayers, all of
whom were paid on commission and had the lowest level of total

compensation, reported lower pay satisfaction than the good performing

116

 

 

ll7
leavers and either group of stayers.

In explaining why work attitudes (i.e. facets of job
satisfaction, job involvement, organizational commitment, and
motivation to turnover) that typically predicted turnover did not
predict functional turnover, Hollenbeck and Williams (1986, p. 609)
concluded, "The failure of job attitudes to predict turnover
functionality can largely be attributed to the fact that, whereas
related to one component of turnover functionality (i.e., turnover
frequency), attitudes were unrelated to the second component (i e.,
performance)." So even though reward contingency was present and
perceived in each organization, and even though perceived reward
contingency moderated the relationship between performance and
satisfaction, neither pay satisfaction nor job satisfaction was
related to either component of functional turnover in the present
study. However, objective reward contingency was related to both
turnover and performance.

One possible post hoc explanation for the absence of significant
differences in satisfaction across functional turnover groups,
especially pay satisfaction, is procedural justice (Thibaut & Walker,
1975; Greenberg, 1987). In other words, the presence of a commission
system, regardless of the magnitude of differences in objective reward
contingency across companies, may be sufficient to develop the belief
that one's total compensation was justly earned, even if one's pay was
low. This explanation was tested by conducting orthogonal contrasts
between functional turnover groups on two items from the JDI pay
satisfaction scale that were related to the issue of procedural

justice. Some support was found for this hypothesis because there

 

118

were no differences across functional turnover groups on either item,
"Less than I deserve" (E = 1.07, p g .37), and "Underpaid" (E = .44, p
g .72). This suggests that poor performing leavers, good performing
leavers, and good performing stayers, all of whom received most of
their pay from commissions, felt no differently about the fairness of
their pay than did poor performing stayers, who received most of their
pay from salary rather than commissions.

Stated in the affirmative, hypothesis 5 predicted that pay
satisfaction and job satisfaction would mediate the effects of

objective and perceived reward contingency. However, null hypothesis

 

5 also could not be rejected because objective reward contingency had
a significant direct effect on functional turnover after controlling
for pay satisfaction and job satisfaction. Even though pay
satisfaction and job satisfaction did not account for differences in
functional turnover, the full discriminant analysis still accounted
for 56% of the variance in functional turnover groups, as well as an
overall classification accuracy of 62.24%, which represented a 43%
improvement over chance levels of prediction and was significant when
judged against the proportional chance criterion.

Examination of discriminant function structure coefficients and
group centroids, in conjunction with follow-up univariate analyses and
orthogonal contrasts, indicated that objective reward contingency
accounted for most of that discriminatory power and classification
accuracy. Indeed, objective reward contingency accounted for 34% of
the variance in functional turnover groups and classification accuracy
of 58.16%, which was significant when judged against the chance

criterion.

 

119

Importantly, however, examination of separate-group hit rates
indicated that the significant overall level of classification
resulted from accurate classification of good and poor performing
stayers. Good and poor performing leavers were not classified
accurately, and were most often misclassified as good performing
stayers. This pattern of classification results was replicated using
a sample of nonresponders (n = 128) and then cross-validated and
double cross—validated. While there was some shrinkage in
classification accuracy in these additional analyses, overall
classification was still better than chance in each case, and the
pattern of separate-group hit rates was unchanged. Together, these
analyses indicate that objective reward contingency does no more than
separate good performers from poorer performers. In other words,
objective reward contingency is strongly related to performance, but
not to each of the four functional turnover groups.

These results are at odds with meta-analytic data which suggest
that reward contingency should be related to functional turnover.
Williams and Livingstone (1990) found that reward contingency
moderated the relationship between performance and voluntary turnover.
When reward contingency was present, the sample-sized weighted average
; was -.30 with confidence intervals from -.27 to -.34. But when
reward contingency was absent, the sample-sized weighted average g was
-.18 with confidence intervals from -.15 to -.2l. As such, these
meta-analytic data suggest that performance contingent pay is
associated with a positive pattern of functional turnover where better
performers are more likely to stay and poorer performers are more

likely to leave.

120

In summary, the results of this study indicate a relationship
between objective reward contingency and functional turnover, but that
relationship occurs only because of accurate separation of good and
poor performing stayers. Objective reward contingency did not lead to
accurate classification of either good or poor performing leavers.
And despite differences in objective reward contingency across
companies, there were not significant pay satisfaction and job
satisfaction differences across the four functional turnover groups.
Finally, the effects of objective reward contingency were not mediated

by differences in pay satisfaction and job satisfaction.

 

Future Research and Theory

There are several theoretical and methodological issues that
should be considered when conducting functional turnover research in

the future.

External Eguity

Perhaps the greatest theoretical weakness of the general model
predicting positive functional turnover under conditions of high
reward contingency is that it focuses on internal equity, how rewards
are tied to performance within one company, without explicitly
considering external equity, how much one could make by quitting and
accepting a job with another company. Bergmann, Hills, and Priefert
(1983) note that, "Externally, the organization must match the going
rate' of the relevant labor market to remain competitive. Internally,
employees must perceive equity both within and across jobs," (p. 17).

External equity is especially important given that workers

121

usually improve their pay when they make an external job change. In
reviewing a number of studies, Parnes (1954) argued that roughly 50-
60% of all quitters received better pay in their new jobs, while Rees
and Schultz (1970) found that 57% of quitters took better paying new
jobs. This effect, however, is even more pronounced for prearrangers,
those who line up a new job prior to quitting their old one. Reynolds
(1951) data indicated that 60% of prearrangers improved their wages
compared to only 25% of "non-arrangers." In another study, Lansing

and Mueller (1967) found that when quitting involved a geographic

 

move, 72% of prearrangers increased their pay in contrast to 55% of
those who did not prearrange a new job in their new location. That
workers usually leave their present jobs for better paying jobs
suggests that making rewards contingent on performance may be
necessary, but not sufficient to produce positive functional turnover.
Total compensation, even if it's performance based, must also be
competitive in the external labor market.

Several authors have suggested that external job opportunities
influence the withdrawal process because they have a direct effect on
job satisfaction (Hulin et. a1., 1985; Michaels & Spector, 1982). For
example, if alternative better paying jobs were plentiful, the
availability of those jobs might act to directly lower satisfaction
with one's present job (Hulin et. a1., 1985). "The more abundant and
desirable the alternatives, and the greater the expected utility of
these other activities to a worker, the less the satisfaction
experienced with the present job," (p. 242). A meta-analysis by

Carsten and Spector (1987) offers some support for this View because

the negative correlation between job satisfaction and turnover, as

122

well as the positive correlation between behavioral intentions and
turnover, both become stronger when unemployment is down and weaker
when unemployment is up. As they described it, "when there are few
alternatives, dissatisfied employees who wish to quit cannot do so and
there will be a small correlation observed between satisfaction and
turnover. When there are many alternatives, dissatisfied employees
who wish to quit can, and the observed relationship will become
stronger," (p. 378).

Although pay satisfaction and job satisfaction were unrelated to

functional turnover in this study, better paying external job

 

opportunities, according to this reasoning, could counteract the job
satisfaction of better performers, even when rewards are linked to
performance. In turn, lowered job satisfaction coupled with an offer
for a better paying job might then lead to turnover. In contrast, if
there were few alternative jobs, or if those jobs paid no better than
other jobs, high reward contingency systems ought to lead to greater
retention of better performers (Dreher, 1982; Wells & Muchinsky,
1985).

The presently available evidence, however, does not support the
notion that alternative employment opportunities are associated with
negative functional turnover. Meta-analysis results (Williams &
Livingstone, 1989) suggest that poorer performers, not better
performers, are more likely to leave when more employment
opportunities exist because there is a stronger negative relationship
between performance and turnover when unemployment is low (for
national, sample-size weighted average r = -.25 with confidence

intervals from -.22 to - 28; for industry, -.22 with confidence

123

intervals from -.19 to -.26; for occupation, —.29 with confidence
intervals from -.26 to -.32; for state, -.28 with confidence intervals
from -.24 to -.33; and for city, -.27 with no confidence intervals

because sampling error accounted for 100% of the variance across
studies) and a weaker negative relationship when unemployment is high
(for national, sample-size weighted average r = -.15 with confidence
intervals from -.11 to - 19; for industry, -.18 with confidence
intervals from -.11 to -.25; for occupation, -.14 with confidence
intervals from -.09 to -.l9; for state, -.11 with confidence intervals

from -.06 to -.l6; and for city, -.25 with confidence intervals from -

 

.03 to -.27). Since reward contingency also leads to stronger
negative relationships between performance and turnover, the use of
reward contingency may be even more successful when alternative
employment opportunities are frequent because poorer performers, who
are less satisfied when rewards are contingent on performance, may
become even more dissatisfied when alternative external jobs are
widely available, too.

Unfortunately, the effects of alternative employment
opportunities were not estimated for this study. Therefore, we do not
know the extent to which pay was perceived as competitive compared to
alternative employment opportunities, the extent to which sales
representatives actually had better or poorer alternative employment
opportunities, or the extent to which other job alternatives actually
affected pay satisfaction and job satisfaction. Undoubtedly, further
research is needed to clarify the combined effects of satisfaction,
reward contingency, and alternative employment opportunities on

functional turnover.

124
Rewards Other ThagiBay

Although the results of this study suggest that increases in
reward contingency are associated with positive functional turnover,
it is unlikely that the commission-based method of reward contingency
used in this study can be applied to most jobs.

Internal job opportunities. One readily available reward in
many companies that can potentially counteract the influence of
external job opportunities is the availability of internal job
opportunities (March & Simon, 1958; Mobley, 1982). Preliminary
research (Jackofsky and Peters, 1983; Todor, 1980) indicates that the
processes for both organizational and job turnover are similar, and
that internal mobility can greatly decrease employee turnover (Dalton
& Todor, 1987). Moreover, the point of measuring and attempting to
influence functional turnover is to retain better employees
throughout the entire organization. Therefore, research that ignores
internal movement, that excludes workers who have quit one job to take
another within the same organization, misrepresents the effects of
retention strategies. Just because a good worker takes another job in
the same company doesn't indicate that the retention strategy was a
failure. Rather, it indicates success.

For example, several studies (Dreher, 1982; Stumpf and Dawley,
1981; and Wells and Muchinsky, 1985) already suggest that judicious
use of promotions may help organizations improve functional turnover.
In Dreher's study of managerial, professional, and technical employees
at Exxon, leavers, who were significantly poorer performers, were

promoted at a substantially slower rate than stayers. Stumpf and

 

125

Dawley (1981) also found that rate of promotion (i.e., indexed by the
speed of promotion within the first 2 years; for example, from teller
trainee, to junior teller, to intermediate teller, and then to senior
teller) was negatively correlated with turnover (3 = -.33, from 1970-
1976; r = -.81 from 1977-1978) for bank tellers. And in a study of
entry level credit managers, Wells and Muchinsky (1985) reported that
promoted employees performed significantly better on all 12
performance dimensions than did those who left voluntarily.

Expected Utility. The uniformity of results across these
promotion studies, however, does not imply that all an organization
must do is make promotion contingent on performance. Rather, Wells
and Muchinsky (1985) speculate that there is also an important
feedback component at work. Based on Ashford and Cummings (1983) who
state "it is clear that individuals often use such feedback
(performance appraisal information) in making turnover decisions"
(1983, p. 375), Wells and Muchinsky argue that early performance
appraisal information sets up a self-fulfilling prophecy that leads
employees to "infer" their potential for future success with a company
which, in turn, affects quit/stay decisions. For example, the credit
managers in Wells and Muchinsky's study received performance appraisal
information relatively early, after 2, 3, and 4 months on the job. In
Dreher's (1982) study, early career variables, such as initial
performance appraisal, initial job level, and initial rating of
potential (which estimated the probable, final job level that the
individual would achieve with the organization), were significantly
better for stayers who were promoted more often than leavers. And in

the Stumpf and Dawley study (1981), the authors noted:

126

"After 3 to 4 months, average and above average tellers were
promoted to junior tellers, and on to intermediate tellers after
9 to 12 months service. The promotion to senior teller one or
more years later required one to be able to perform all aspects
of the job with little or no supervision.

During the initial period of employment, promotional increases
provide direct feedback on teller assimilation of training.
Early promotion increases also serve to notify tellers of
management’s view of the future potential" (1981, p. 152).
Together, these studies suggest that the absence or presence of
early rewards and promotions may shape expectations concerning future

potential with the company, independent of job satisfaction. In fact,

Mobley (1982) and his colleagues (1979) expressly separate the two in

 

their model of employee turnover, viewing satisfaction as an affective
reaction to one's present job, and the "expected utility" of one's
present job as the expectancy "that the job will lead to future
attainment of various positively and negatively valued outcomes"
(1979, p. 518) within the same company.
"It is important to recognize that just as the dissatisfied
employee may not quit, given positive expectations about future
roles in the organization, the currently satisfied employee who
has negative expectations about the future in the organization
may quit. For example, expected negative changes in the job,
the lack of perceived desirable promotional opportunities, and
expected negative changes in policy, practices, or conditions

may lead currently satisfied employees to seek external jobs,"
(Mobley, 1982, pp. 128-129).

Job Design. Dunham (1977), Schneider, Reichers, and Mitchell
(1982), and Campion (1989) have investigated the relationship between
job design and ability because changes in the design of jobs may
affect the selection, placement, training, and compensation of
workers. One of their primary considerations has been that
"redesigned jobs may not be compatible with the abilities of job

incumbents," (Schneider et. a1., 1982, p. 568). This concern arises

127
because increases in job complexity are associated with increases in
the abilities required to perform redesigned jobs.

For example, Dunham (1977) found positive relationships (from
.21 to .49) between measures of job characteristics and nine General
Aptitude Test Battery (GATB) scores derived from the Position Analysis
Questionnaire (McCormick, Jeanneret & Meacham, 1972). Schneider et.
a1. (1982) directly coded GATB scores for 140 jobs and found them to
be positively correlated with job characteristics (i.e., identity,
variety, required interdependence, predictability, autonomy, and
feedback). Campion (1989) estimated job ability requirements from the
Dictionary of Occupational Titles as well as the GATB scores for 200
jobs across two samples. He then used factor analysis to reduce the
31 different ability measures to a smaller and more interpretable
number of common factors. The first factor from that analysis,
substantive complexity (which reflected intelligence, verbal,
numerical, spatial, form, clerical, data, people, and talk ability
scales), correlated .55 with the motivational components of jobs
(i.e., autonomy, responsibility, extrinsic feedback, task variety,
task identity, task significance, learning, and participation) in his
first sample and .46 in the second.

While this research forewarns that job redesign may
unintentionally create a mismatch between job requirements and worker
abilities, it also seems to suggest that workers with greater ability,
who on average will be better performers, will respond positively to
job enrichment and enlargement. For example, Schneider et. al. (1982)
found that two task characteristics, variety and autonomy (from .46 to

.52), were as strongly related to work satisfaction as were the

128
intelligence, verbal, and numerical scales of the GATB (from .39 to
.51).

Therefore, it is predicted that better performing employees will
value enriched and enlarged jobs. According to Hackman and Oldham
(1980):

"For jobs high in motivating potential, then, people who

have sufficient knowledge and skill to perform well will

experience substantially positive feelings as a result of
their work activities. But people who are not competent
enough to perform well will experience a good deal of
unhappiness and frustration at work, precisely because the

job ’counts' for them and they do poorly at it," (p. 84).

And if better performers experience positive feelings as a result of

enriched and enlarged jobs, perhaps they may be more likely to stay.

Performance and Turnover

Despite knowing that performance and turnover are inversely
related, that low unemployment and reward contingency moderate this
relationship such that poor performers are more likely to leave and
good performers are more likely to stay when external jobs are
plentiful and when rewards are contingent on performance, and, that
the relationship between performance and turnover is also u-shaped
such that both good and poor performers, relative to average
performers, are more likely to quit (Williams & Livingstone, 1989),
there is little else that we know about the relationship between
performance and turnover.

For example, we don't know if good and bad performers frame
turnover decisions differently. When poor performers quit, is it
because they're withdrawing from a negative situation? Or,

conversely, when better performers, is it because they are attracted

129
to a better job rather withdrawing from their present job? In other
words, we don't even know if good and bad performers leave for the
same or different reasons.

Similarly, despite predictions of a positive relationship, "that
the greater the job performance, the greater number of actual or
perceived alternative job offers an individual receives," Jackofsky
(1984, p. 77), we don't know if high performers actually have or
perceive more external employment alternatives, and if they do, how
that influences turnover decisions. Jackofsky and Peters (1980),

using ability measured by selection tests (e.g., cognitive, clerical,

 

and mathematical ability) as a surrogate for performance, found that
even though employees with more ability had stronger expectations of
finding alternative employment (; = .45, p < .05), they were not any
more likely to quit than low ability employees. Ekpo-Ufot (1976)
found that the higher one’s self-perceptions of ability, the more
likely one was to stay on the job (; = -.34, p g .05), just the
opposite of Jackofsky's prediction. Likewise, we know little about
the relationship between actual employment alternatives and
performance. Even Jackofsky (1984), who has made the argument for a
positive relationship between performance and actual employment
alternatives, admits that the evidence is anecdotal: "Support for this
link can be found, not in empirical literature, but from statements
made by personnel administrators," (p. 77).

And, although some preliminary hypotheses have been made here
with respect to job design, we also don't know if better performers
value different rewards than do poorer performers. Nor do we know how

strong reward contingency must be before positive functional turnover

130
results.
Efforts to successfully influence functional turnover are likely

to be unsuccessful unless we address these basic questions.

Controllability of Turnover

The impact that an organization can have on functional turnover
through any variable or intervention, not just reward contingency, is
limited by the portion of turnover that can be controlled by
organizations. Dalton et. al. (1981, 1982) and Abelson (1987) argue

that much voluntary turnover is unavoidable due to reasons such as

 

returning to school for continued education, transfer of a spouse,
death in the family, retirement, health, and so on. Empirical
research suggests that a substantial percentage of turnover may occur
for these reasons and, therefore, cannot be directly controlled by
organizations. For example, Dalton, et. a1., (1981) indicated that
for bank tellers fully 46% of voluntary turnover was unavoidable and
could not be controlled. Abelson (1987) found that 35% of nursing
personnel turnover in nursing homes was also unavoidable.

The practical implication is that interventions designed to
produce positive functional turnover can only be expected to influence
avoidable separations, individuals who leave for better pay, working
.conditions, leadership, etc. In Abelson’s study, 65% of voluntary
turnover occurred for such reasons, while in Dalton et. al.'s study
only 54% of voluntary turnover was controllable. The scientific
implication is that judgments of the success or failure of
organizational programs designed to influence the pattern of

functional turnover may be invalid to the extent that voluntary

131
turnover is not separated into avoidable and unavoidable quits.

Thus, controllability or avoidability represents another
important refinement in the measurement of turnover. Failure to make
this refinement could lead to a loss of prediction, understanding, and
parsimony when testing turnover models. In the present study, the
extent of unavoidable turnover and its possible effect on the results
if unavoidable voluntary quits were removed from the analysis cannot
be discerned. Yet, given the extent of unavoidable turnover in the

Dalton et. a1. (1981) and Abelson (1987) studies, future functional

 

turnover research should attempt to distinguish between avoidable and

unavoidable quits.

Field Experiments

Finally, perhaps the most important methodological change that
must be made in measuring and determining the antecedents of
functional turnover is a move away from predictive designs which
predominate in turnover research toward the use of basic field

experiments.

"Although thousands of studies have been done, only a few field
experiments have addressed the crucial problem that is of most
interest to managers--how to reduce turnover among those

employees an organization wishes to retain," (McEvoy and Cascio,
1985, p. 351).

While researchers may encounter difficulties in obtaining research
sights when conducting field experiments, and while it may be
difficult to randomly assign subjects to treatments (treatments will
more likely have to be assigned to groups of subjects to accommodate
host organizations), field experiments can help control a number of

threats to internal validity, such as history, maturation, selection,

 

132
mortality, interactions of selection-maturation and selection-history,
all of which may provide alternative explanations when comparing
results across studies or, as in this study, when trying to make sense
of functional turnover results which have been compiled across

different organizations.

 

REFERENCES

REFERENCES

Abelson, M.A. (1987). Examination of avoidable and unavoidable
turnover. Journal pf Applied Psychology 72 382-386.

’ _’

Abelson, M.A., & Baysinger, B.D. (1984). Optimal and dysfunctional
turnover: Toward an organizational level model. Academy pf
Management Review, 2, 331-341.

Armknecht, P.A., & Early, J.F. (1972). Quits in manufacturing: A
study of their causes. Monthly Labor Review, 95 31-37.

—’

Ashford, S.J., & Cummings, L.L. (1983). Feedback as an individual
resource: Personal strategies of creating information.

Bagozzi, R.P. (1978). Salesforce performance and satisfaction as a
function of individual difference, interpersonal, and

situational factors. Journal pf Marketing Research, 15, 517-
531.

Bagozzi, R.P. (1980). Performance and satisfaction in an industrial
sales force: An examination of their antecedents and their
simultaneity. Journal pf Marketing, 44, 65-77.

Baird, L.S., & Hamner, W.C. (1979). Individual versus system
rewards: Who's dissatisfied, why, and what is their likely
response? Academy pf Management Journal, 22,783-792.

Baxter,B., Taaffe, A.A., & Hughes, J.F. (1955). Unpublished study,
cited in Brayfield & Crockett, 1955.

Bergmann, T.J., Hills, F.S., & Priefert, L. (1983). Pay compression:

Causes, results and possible solutions. Compensation Review,
15, 17-26.

Bhagat, R.S. (1982). Conditions under which stronger job performance-
job satisfaction relationships may be observed: A closer look at

two situational contingencies. Academy pf Management Journal,
25, 772-789.

Boudreau, J.W. (1981). A conceptual model of employee movement
utility. Unpublished doctoral dissertation, Purdue University.

Boudreau, J.W. (1983). Effects of employee flows on utility analysis

of human resource productivity improvement programs. Journal pf
Applied Psychology, 68, 396-406.

133

134

Boudreau, J.W., & Berger, C.J. (1985). Decision-theoretic utility
analysis applied to employee separations and acquisitions
[Monograph]. Journal pf Applied Psychology, 10, 581—612.

Brayfield, A.H., & Crockett, W.H. (1955). Employee attitudes and
employee performance. Psychological Bulletin 52 396-324.

Brody, M. (1945). The relationship between efficiency and job
satisfaction. Unpublished M.A. thesis, New York University.

Bureau of National Affairs. (1982). Collective bargaining
negotiations and contracts. Daily Labor Report, 206A, 8.

 

Burton, J.F., & Parker, J.E. (1969). Interindustry variation in
voluntary labor mobility. Industrial and Labor Relations
Review, 22, 199-216.

Campion, M.A. (1989). Ability requirement implications of job
design: An interdisciplinary perspective. Personnel Psychology,
42, 1-24.

 

Carsten, J.M., & Spector, P.E. (1987). Unemployment, job
satisfaction, and employee turnover: A meta-analytic test of the
Muchinsky model. Journal pf Applied Psychology, 1;, 374-381.

Cascio, W.F. (1982). Costing human resources: The financial impact
pf behavior ip organizations. Boston: Kent.

 

Cherrington, D.J., Reitz, H.J., & Scott, W.E., Jr.. (1971). Effects
of contingent and noncontingent reward on the relationship
between satisfaction and task performance. Journal pf Applied
Psychology, 52, 531-536.

Cohen, J., & Cohen, P. (1983). Applied multiple
regression/correlation f9; phg behavioral sciences, (2nd ed.).
Hillsdale, N.J.: Erlbaum.

Cook, T.D., & Campbell, D.T. (1979). Quasi-experimentation: Design
and analysis issues for field settings. Chicago: Rand McNally.

 

Cronbach, L.J., & Gleser, G.C. (1965). Psychological tests and
personnel decisions. Urbana, Ill.: University of Illinois
Press.

 

Dachler, H.P., & Mobley, W.H. (1973). Construct validation of an
instrumentality-expectancy-task-goal model of work motivation:
Some theoretical boundary conditions [Monograph]. Journal pf
Applied Psychology, 58, 397-418.

Dalton, D.R., Krackhardt, D.M., & Porter, L.W. (1981). Functional
turnover: An empirical assessment. Journal pf Applied
Psychology, 66, 716-721.

135

Dalton, D.R., Todor, W.D. (1979). Turnover turned over: An expanded
and positive perspective. Academy pf Management Review, 4, 225-
235.

Dalton, D.R., Todor, W.D. (1987). The attenuating effects of
internal mobility on employee turnover: Multiple field
assessments: Journal pf Management, 13, 705-711.

Dalton, D.R., Todor, W.D., & Krackhardt, D.M. (1982). Turnover
overstated: The functional taxonomy. Academy pf Management
Review, 1, 117-123.

Dansereau, F., Jr., Cashman, J., & Graen, G. (1973). Instrumentality
theory and equity theory as complementary approaches in
predicting the relationship of leadership and turnover among
managers. Organizational Behavior app Human Performance, 10,
184-200.

Dreher, G.F. (1977). Nonrespondent characteristics and respondent
accuracy in salary research. Journal pf Applied Psychology, 62,
773-776.

Dreher, G.F. (1982). The role of performance in the turnover
process. Academy pf Management Journal, 25, 137-147.

Dreher, G.F., Ash, R.A., Hancock, P. (1988). The role of the
traditional research design in underestimating the validity of
the employment interview. Personnel Psychology, 41, 315-327.

Dunham, R.B. (1977). Relationship of perceived job design
characteristics to job ability requirements and job values.
Journal pf Applied Psychology, ,62, 760-763.

Dunham, R.B., & Herman, J.B. (1975). Development of a female Faces
Scale for measuring job satisfaction. Journal pf Applied
Psychology, pp, 629-631.

Dyer, L., & Theriault, R. (1976). The determinants of pay
satisfaction. Journal pf Applied Psychology, 61, 596-604.

Eagly, R.V. (1965). Market power as an intervening mechanism in
Phillips Curve analysis. Economics, 32, 48-64.

Ekpo-Ufot, A. (1976). Self-perceived abilities-—relevant in the task
(SPART): A potential predictor of labor turnover in an
industrial setting. Personnel Psychology, 22, 405-416.

Fisher, C.D. (1980). On the dubious wisdom of expecting job
satisfaction to correlate with performance. Academy pf
Management Review, 5, 607-612.

Flamholtz, E. (1974). Human resources accounting. Encino, CA:
Dickenson Publishing.

 

136

Fleishman, E.A., & Harris, E.F. (1962). Patterns of leadership
behavior related to employee grievances and turnover. Personnel

Psychology, 25, 43-56.

Ghiselli, E.E., Campbell, J.P., Zedeck, S. (1981). Measurement theory
for the behavioral sciences. San Francisco: W.H. Freeman and
Company.

Glass, G.V., & Stanley, J.C. (1970). Statistical methods 53 education
and psychology. Englewood Cliffs, N.J.: Prentice-Hall.

Graen, G.B. (1969). Instrumentality theory of work motivation: Some
experimental results and suggested modifications [Monograph].

Journal pf Applied Psychology, 55.

 

Graen, G.B., Dansereau, F., & Minami, T. (1972). Dysfunctional
leadership styles. Organizational Behavior and Human
Performance, 1, 216-236.

 

Greenberg, J. (1987). Reactions to procedural injustice in payment
distributions: Do the means justify the ends? Journal pf

Applied Psychology, 12, 55-61.

Greene, C. N. (1973). Causal connections among managers' merit pay,
job satisfaction, and performance. Journal pf Applied

Psychology, 52, 95-100.

Greenhaus, J.H., & Badin, I.J. (1974). Self-esteem, performance, and
satisfaction: Some tests of a theory. Journal pf Applied
Psychology, 52, 722-726.

Gustafson, H.W. (1980). Force-loss cost analysis. New York:
American Telephone and Telegraph Co.

 

Hackman, J.R., & Oldham, G.R. (1980). Work redesign. Reading, Mass:
Addison-Wesley.

Hamner, W.C., & Harnett, D.L. (1974). Coal setting, performance, and

satisfaction in an interdependent task. Organizational Behavior
and Human Performance, 12, 217-230.

 

Heneman, H.G. (1973). Impact of performance on managerial pay levels
and pay changes. Journal pf Applied Psychology, 52, 128-130.

Herzberg, F., Mausner, B., Peterson, R.O., & Capwell, R.F. (1957).
Job attitudes: Review of research and opinions. Pittsburgh,

PA.: Pittsburgh Psychological Services.

Hinrichs, J.R. (1969). Correlates of employee evaluations of pay
increases. Journal pf Applied Psychology, 55, 481-489.

Hollenbeck, J.R., & Williams, G.R. (1986). Turnover functionality
versus turnover frequency: A note on work attitudes and

organizational effectiveness. Journal pf Applied Psychology,
1;, 606-611.

137

Hoppock, R. (1935). Job Satisfaction. Harper and Brothers, New
York. Reprint Edition by Arno Press, New York, 1977.

Huberty, C.J. (1984). Issues in the use and interpretation of
discriminant analysis. Psychological Bulletin, 25, 156-171.

Huberty, C.J., & Morris, J.D. (1989). Multivariate analysis versus
multiple univariate analysis. Psychological Bulletin, 105, 302-
308.

Hulin, C.L. (1968). Effects of changes in job-satisfaction levels on
employee turnover. Journal pf Applied Psychology, 52, 122-126.

Hulin, C.L., Roznowski, M., & Hachiya, D. (1985). Alternative
opportunities and withdrawal decisions: Empirical and

theoretical discrepancies and an integration. Psychological
Bulletin, 21, 233-250.

Iaffaldano, M.T., & Muchinsky, P.M. (1985). Job satisfaction and job

performance: A meta-analysis. Psychological Bulletin, 21, 251-
273.

 

Ilgen, D.R., Nebeker, D.M., Pritchard, R.D. (1981). Expectancy theory
measures: An empirical comparison in an experimental

simulation. Organizational Behavior and Human Performance, 22,
189-223.

Ilgen, D.R., Pritchard, R.D., Dugoni, B.L., & Nebeker, D.M. (1980).
models fog th; predictions pf perfgrmance lg ap experimental
work simulation, (Technical Report). San Diego: Navy Personnel
Research and Development Center.

Jackofsky, E.F. (1984). Turnover and job performance: An integrated
process model. Academy pf Management Review, 2, 74-83.

Jackofsky, E.F., & Peters, L.H. (1980). Task-relevant ability and
turnover: Test of a model. Southwest Academy 95 Management
Proceedings, 161-165.

Jackofsky, E.F., & Peters, L.H. (1983). Job turnover versus company
turnover: Reassessment of the March and Simon participation
hypothesis. Journal pf Applied Psychology 68 490-495.

’ ..._.7

Jacobs, R., & Solomon, T. (1977). Strategies for enhancing the
prediction of job performance from job satisfaction. Journal 95
Applied Psychology, 52, 417-421.

James, L.R., & Brett, J.M. (1984). Mediators, moderators, and tests
for mediation. Journal pf Applied Psychology, 52, 307-321.

Jenning, K., & Wolters, R. (1976). Discharge cases reconsidered.
Arbitration Journal 31 164-180.

138

Jeswald, T.A., (1974). The cost of absenteeism and turnover in a
large organization. In W.C. Hamner & F.L. Schmidt (Eds.),

Contemporary problems 1p personnel. Chicago: St. Clair Press. .

 

Johnson, S.M., Smith, P.C., Tucker, S.M. (1982). Response format of
the Job Descriptive Index: Assessment of reliability and
validity by the multitrait-multimethod matrix. Journal pi

Applied Psychology, 21, 500-505.

Keller, R.T. (1984). The role of performance and absenteeism in the
prediction of turnover. Academy 2: Management Journal, 21, 176-
183.

Kesselman, G.A., Wood, M.T., & Hagen, E.L. (1974). Relationships
between performance and satisfaction under contingent and
noncontingent reward systems. Journal pf Applied Psychology,
52, 374-376.

 

Kirchner, W.K. (1965). Relationships between general and specific
attitudes toward work and objective job performance for outdoor
advertising salesmen. Journal pf Applied Psychology, 22, 455-
457.

 

Klecka, W.R. (1980). Discriminant analysis. Beverly Hills, CA:
Sage.

Kopelman, R.E. (1976). Organizational control system responsiveness,
expectancy theory constructs, and work motivation: Some

interrelations and causal connections. Personnel Psychology,
22, 205-220.

Kopelman, R.E., & P.H. Thompson. (1976). Boundary conditions for
expectancy theory predictions of work motivation and job
performance. Academy pf Management Journal 19 237-258.

’ —’

l
|
Kulik, J.A., Kulik, C.-L.C., & Cohen, P.A. (1979). A meta-analysis of i
outcome studies of Keller’s personalized system of instruction.
American Psychologist, 22, 307-318.
1

Kunin, T. (1955). The construction of a new type of attitude measure.
Personnel Psychology, 2, 65-78.

Lansing, J.B., & Mueller, E. (1967). The geographic mobility pi
labor. Ann Arbor, Mi.: University of Michigan Press.

Lawler, E.E. (1966). Manager’s attitudes toward how their pay is and

should be determined. Journal pf Applied Psychology, 52, 273-
279.

Lawler, E.E., & Porter, L.T. (1967). The effect of performance on
job satisfaction. Industrial Relations, 1, 20-28.

Locke, E.A. (1965). The relationship of task success to task liking
and satisfaction. Journal 92 Applied Psychology, 22, 379-385.

139

 

MacKinney, A.C., & Wolins, L. (1959). Validity information exchange.
Personnel Psychology, 12, 482-483.

Macy, B.A., & Mirvis, P.H. (1983). Assessing rates and costs of
individual work behaviors. In S.E. Seashore, E.E. Lawler, P.H.
Mirvis, & C. Cammann (Eds.), Assessing organizational change.
New York: Wiley.

March, J.C., & Simon, H.A. (1958). Organizations. New York: Wiley.

Matteson, M.T., Ivancevich, J.M., & Smith, S.V. (1984). Relation of
type A behavior to performance and satisfaction among sales
personnel. Journal 92 Vocational Behavior, 25, 203-214.

McCormick, E.J., Jeanneret, P.R., & Mecham, R.C. (1972). A study of
job characteristics and job dimensions as based on the Position

Analysis Questionnaire (PAQ). Journal 9: Applied Psychology,
52, 347-368.

 

McEvoy, G.M., & Cascio, W.F. (1985). Strategies for reducing
employee turnover: A meta-analysis. Journal 22 Applied
Psychology, 12, 342-363.

McEvoy, G.M., & Cascio, W.F. (1987). Do good or poor performers
leave? A meta-analysis of the relationship between performance
and turnover. Academy pi Management Journal, 22, 744-762.

McIntyre, R.M. (1990). Spurious estimation of validity coefficients
in composite samples: Some methodological considerations.
Journal 2: Applied Psychology, 15, 91-94.

McLaughlin, M.L. (1980). Discriminant analysis in communication
research. In P.R. Monge & J.N. Cappella (Eds.), Multivariate
Techniques 1p Human Communication Research. New York, NY:
Academic Press.

McNichols, C.W., Stahl, M.J., & Manley, T.R. (1978). A validation of

Hoppock's job satisfaction measure. Academy pf Management
Journal, 21, 737-742.

Michaels, C.E., & Spector, P.E. (1982). Causes of employee turnover:
A test of the Mobley, Griffeth, Hand, and Meglino model.
Journal 91 Applied Psychology, 21, 53-59.

Miner, J.B. (1980). Theories pf organizational behavior. Hinsdale,
ILL.: The Dryden Press.

Mirvis, P.H., & Lawler, E.E. (1977). Measuring the financial impact
of employee attitudes. Journal 91 Applied Psychology, 22, 1-8.

Mitchell, T.R. (1974). Expectancy models of job satisfaction,
occupational preference and effort: A theoretical,
methodological, and empirical appraisal. Psychological
Bulletin, 21, 1053-1077.

 

 

140

Mobley, W.H. (1982). Employee turnover: Causes conseguences, and
control. Reading, Mass.: Addison-Wesley.

 

Mobley, W.H., Griffeth, R.W., Hand, H.H., Meglino, B.M. (1979).

Review and conceptual analysis of the employee turnover process.

’ -’

Psychological Bulletin 86 493-522.

Motowidlo, S.J. (1982). Relationship between self-rated performance
and pay satisfaction among sales representatives. Journal pi

Applied Psychology, 21, 209-213.

Mowday, R.T., Porter, L.T., & Steers, R.M. (1982). Employee-

organization linkages: The psychology pi commitment,
absenteeism and turnover. New York: Academic Press.

 

Muchinsky, P.M., & Tuttle, M.L. (1979). Employee turnover: An
empirical and methodological assessment. Journal pi Vocational
Behavior, 12, 43-77.

Orpen, C. (1974). The effect of reward contingencies on the job
satisfaction-task performance relationship: An industrial
experiment. Psychology, 11, 9-14.

Parnes, H.S. (1954). Research pp labor mobility. New York: Social
Science Research Council.

 

Podsakoff, P.M., & Williams, L.J. (1986). The relationship between
job performance and job satisfaction. In E.A. Locke (Ed.),

Generalizing from laboratory ip field settings (pp. 205-253).
Lexington, Mass: Lexington Books.

 

Porter, L.W., & Lawler, E.E. (1968). Managerial attitudes and
performance. Homewood, Ill: Irwin.

Porter, L.W., & Steers, R.M. (1973). Organizational, work, and
personal factors in employee turnover and absenteeism.
Psychological Bulletin, 22, 151-176.

Price, J.L. (1977). The study pi turnover. Ames, Iowa: Iowa State
University Press.

Pritchard, R.D. (1973). Effects of varying performance-pay
instrumentalities on the relationship between performance and
satisfaction.: A test of the Lawler and Porter model. Journal

pi Applied Psychology, 52, 122-125.

Rees, A., & Shultz, G.P. (1970). Workers and wages ip ap urban labor

market. Chicago, 111.: University of Chicago Press.

Reynolds, L.G. (1951). The structure pi labor markets. New York:
Harper.

Robinson, W.S. (1950). Ecological correlations and the behavior of
individuals. American Sociological Review 15 351-357.

I _ ’

 

141

Saunders, D.R. (1956). Moderator variables in prediction.
Educational and Psychological Measurement, 12, 209-222.

Schmidt, F.L., & Hoffman, B. (1973). Empirical comparison of three
methods of assessing utility of a selection device. Journal pi
Industrial and Organizational Psychology, 13-22.

Schmidt, F.L., Hunter, J.E., McKenzie, R.C., & Muldrow, T.W. (1979).
Impact of valid selection procedures on work-force productivity.
Journal pi Applied Psychology, 22, 609-626.

Schmidt, F.L., & Hunter, J.E., & Raju, N.S. (1988). Validity
generalization and situational specificity: A second look at the
75% rule and Fisher's z transformation. Journal pi Applied

Psychology, 12, 665-672.

Schneider, B., & Olson, L.K. (1970). Effort as a correlate of
organizational reward system and individual values. Personnel
Psychology, 22, 313-326.

Schneider, B., Reichers, A.E., & Mitchell, T.M. (1982). A note on
some relationships between the aptitude requirements and reward

attributes of tasks. Academy pi Management Journal, 25, 567-
564.

Schwab, D.P. (1973). Impact of alternative compensation systems on
pay valence and instrumentality perceptions. Journal pi Applied
Psychology, 52, 308-312.

Schwab, D.P., & Cummings, L.L. (1970). Theories of performance and
satisfaction: A review. Industrial Relations, 2, 408-430.

Smith, H.L., & Watkins, L.E. (1978). Managing manpower turnover
costs. The Personnel Administrator, 22, 46-50.

Smith, P.C., Budzeika, K.A., Edwards, N.A., Johnson, S.M., & Bearse,
L.H. (1986). Guidelines for clean data: Detection of common
mistakes. Journal pi Applied Psychology, 11, 457-460.

Smith, P.C., Kendall, L., & Hulin, C.L. (1969). The measurement pi
satisfaction 1p work and retirement. Chicago: Rand McNally.

Sockloff, A.L. (1975). Behavior of the product-moment correlation
coefficient when two heterogeneous subgroups are pooled.
Educational and Psychological Measurement, 25, 267-276.

Steel, R.P., & Ovalle, N.K. (1984). A review and meta-analysis of
research on the relationship between behavioral intentions and
employee turnover. Journal pi Applied Psychology, 22, 673-686.

Steers, R.M., Mowday, R.T. (1981). Employee turnover and post-
decision accommodation processes. In L.L. Cummings & B.M. Staw

(Eds.), Research 1p Organizational Behavior (Vol. 3).
Greenwich, Conn.: JAI Press.

 

142

Stogdill, R.M., & Coons, A.E. (1957). Leader behavior' Its
description and measurement. Columbus, Ohio: Bureau of
Business Research.

 

 

Stone, E.F., & Hollenbeck, J.R. (1984). Some issues associated with
the use of moderated regression. Organizational Behavior and
Human Performance, 22, 195-213.

Stumpf, S.A., & Dawley, P.K. (1981). Predicting voluntary and
involuntary turnover using absenteeism and performance indices.
Academy pi Management Journal, 22, 148-163.

Thibaut, J., & Walker, L. (1975). Procedural justice: A
psychological analysis. Hillsdale, NJ: Erlbaum.

Todor, W.D. (1980). A movement paradox: Turnovail Transfer?. Paper
presented at the annual meeting of the Academy of Management.

Van der Merwe, R., & Miller, S. (1971). The measurement of labor
turnover: A critical appraisal and a suggested new approach.
Human Relations, 22, 233-253.

 

Vroom, V.H. (1964). Work and motivation. New York: Wiley.

Wanous, J.P., Stumpf, S.A. & Bedrosian, H. (1979). Job survival of
new employees. Personnel Psychology, 22, 651-662.

Wells, D.L., & Muchinsky, P.M. (1985). Performance antecedents of
voluntary and involuntary turnover. Journal pi Applied
Psychology, 12, 329-336.

Williams, G.R., & Livingstone, L.P. (1989). A second look at the
relationship between performance and voluntary turnover.
Manuscript submitted for review.

Zedeck, S. (1971). Problems with the use of moderator variables.
Psychological Bulletin 76 295-310.

’ _’

 

APPENDIX

 

 

APPENDIX

Study Questionnaire

Michigan State Studies pp Attitudes and Organizational Memberahip

This questionnaire is part of an ongoing program of scientific
research at Michigan State University's Graduate School of Business.
Specifically, this research is concerned with attitudes and how they
affect 1) individual job performance, and 2) employee turnover. In
order to comply with the ethical standards established by the American
Psychological Association and the Academy of Management, it is
essential that all participants in this research be made aware of:

1) what is reguired of each participant,
2) the rights of each participant, and
3) the responsibilities of each participant.

1) Participant Requirementa
* if you choose to participate in this research, we will
need no more than 30 minutes of your time.

In addition to the questionnaire, we also need to know
your performance with the company. Without access to
these records, we will be unable to determine how your
attitudes toward your job are related to your job
performance and organizational membership.

Complete confidentiality for all participants is
guaranteed. See below for more.

2) Participant Rights
* Participation in this research is voluntary. further,
should you choose to participate, you may end your
participation at any time.

Each participant is guaranteed complete confidentiality.
No one in your organization will have access to or ever
see your individual questionnaire responses. Thus, you
can be assured that your participation in this research
will in no way influence your job with your company.

143

 

 

144

Your confidentiality will be protected when you seal the
questionnaire in the pre-addressed, pre-stamped envelope
and mail it directly to Chuck Williams at the Graduate
School of Business, Michigan State University.

3) Participant Responsibilities

* Please provide information that is honest, accurate and

complete. By completing the entire questionnaire as
honestly as possible, you can help increase our
understanding of the relationships between employee
attitudes, performance and turnover.

Print and sign your name to the Informed Consent
Declaration shown below.

Also print your name on the first page of the
questionnaire.

Your name must be on the questionnaire in order to link
your attitudes as measured by the questionnaire to your
performance and organizational membership.

Remember, your confidentiality is guaranteed.

**********************************************************************
Informed Consent Declaration

I certify that I have read and understand the requirements, rights and
responsibilities incurred by both the participants and the researcher
in this study. I voluntarily agree to participate in this research
and give the researcher permission to collect questionnaire data from
me and review my sales performance records.

(PRINT YOUR NAME)

 

(SIGN YOUR NAME)

 

**********************************************************************
THANK YOU FOR PARTICIPATING IN THIS RESEARCH STUDY

Charles R. Williams
232 Eppley Center
Department of Management
Michigan State University
East Lansing, MI 48824
(517) 353-5415
PLEASE GO ON TO THE NEXT PAGE

145
SECTION 1

----------------------- INSTRUCTIONS-----------------------—-----—-—---
Please provide general information about yourself in the following
spaces. Fill in the blanks or choose ( J ) the option which describes
you. Please respond to a11 categories.

Name: Sex: Male Female

City/State (Territory) Where you Work or Where you
are Based:
City State

 

How long have you worked for this company?

years months

What is your job title?

 

Associate Sales Representative
Showroom Representative

Sales Representative

Senior Sales Representative
other

 

How long have you been doing this job for this company?

 

years months

How long have you been in sales for this company?

years months

How much education have you had? Place a check mark in the
appropriate space.

Some High-School

Graduated High-School

Some College

Graduated College

Some Graduate School
Master's Degree

M.D., Ph.D., or some other
professional degree

PLEASE GO ON TO THE NEXT PAGE AND SECTION

146
SECTION 2

----------------------- INSTRUCTIONS-----------------------------------
In the following sections, place a "Y" (for Yes) next to the item if
it accurately describes your job. Place an "N" (for No) next to the
item if it does not accurately describe your job. If you cannot
decide whether the item accurately describes your job or not, place a
"?" (for Uncertain) beside the item.

 

 

PAY

Income adequate for Insecure

normal expenses

Satisfactory profit Less than I deserve
sharing
Barely live on Highly paid

income
Income provides Underpaid

luxuries

Bad

PLEASE GO ON TO THE NEXT PAGE AND SECTION

147
SECTION 3

----------------------- INSTRUCTIONS------—----------------------------
For each of the 4 items below, check the one item which best
represents your beliefs and feelings.

1) Choose the one of the following statements which best tells how
well you like your job. Place a check mark next to that
statement.

I hate it.

I dislike it.
I don't like it. '

I am indifferent to it.

I like it.

 

I am enthusiastic about it.
I love it.
2) Check one of the following to show how much of the time your
feel satisfied with your job.
All of the time
Most of the time.
A good deal of the time.
About half of the time.
Occasionally.
Seldom.

Never.

PLEASE GO ON TO THE NEXT PAGE

3)

4)

148

Check the one of the following which best tells how you feel
about changing your job.

I would quit this job at once if I could get
anything else to do.

I would take almost any other job in which I
could earn as much as I am earning now.

I would like to change both my job and my
occupation.

I would like to exchange my present job for
another job in the same kind of work.

I am not eager to change my job but I would do
so if I could get a better job.

I cannot think of any jobs for which I would
exchange mine.

I would not exchange my job for any other.
Check one of the following to show how much you think you
compare with other people:

No one likes his job better than I like mine.

I like my job much better than most people
like theirs.

I like my job better than most people like
theirs.

I like my job about as well as most people
like theirs.

I dislike my job more than most people dislike
theirs.

I dislike my job much more than most people
dislike theirs.

No one dislikes his job more than I dislike
mine.

PLEASE GO ON TO THE NEXT PAGE

 

 

149
SECTION 4

----------------------- INSTRUCTIONS-----------------------------------
The following questions are concerned with how long it takes a new
sales representative in your company to learn how to do his/her sales
job well. Circle the time on the scale below which you believe
indicates how long it takes a new sales representative in your company
to learn how to do that part of their sales job. for example, if it
takes 1 year and 3 months, you would circle .....

4
1 2 3 OR MORE
YEAR YEARS YEARS YEARS
0--/--/--/--l--/--/--/--2--/--/--/--3--/--/--/--l+
3 6 9 3 6 9 3 6 9 3 6 9
MONTHS MONTHS MONTHS MONTHS

How long does it take a new sales representative in your company to...

1) learn the design and specifications of the company products?
A
1 2 3 OR MORE
YEAR YEARS YEARS YEARS
1 -2--/--/--/--3--/--/--/--4

0--/--/--/-- --/--/--
6 9 3 6

3 9 3 6 9 3 6 9
S

MONTHS MONTH MONTHS MONTHS
2) learn the applications and functions of company products?
4
1 2 3 OR MORE
YEAR YEARS YEARS YEARS
1 -2--/--/--/--3--/--/--/--4

0--/--/--/-- --/-—/--
3 6 9 3 6

9 3 6 9 3 6 9
MONTHS MONTHS

MONTHS MONTHS

3) establish a base of steady customers?
4
l 2 3 OR MORE
YEAR YEARS YEARS YEARS
O--/--/--/--1--/-- - 2--/--/--/--3--/--/--/--4
3 6 3 9 3 6 9 3 6 9
MONTHS MO S MONTHS MONTHS

/
6
NTH

PLEASE GO ON TO THE NEXT PAGE

 

 

How long does it take a new sales representative in your company to...

4)

5)

6)

7)

8)

150

reach a sales level achieved by most experienced sales
representatives in your company?

4
l 2 3 OR MORE
YEAR YEARS YEARS YEARS

--2--/--/--/--3--/--/--/--4
9 3 6 9 3 6 9
S MONTHS MONTHS

0--/--/--/--1--/--/--
3 6 9 3 6
MONTHS MONTH

use established customers to make new sales contacts?

4
1 2 3 OR MORE
YEAR YEARS YEARS YEARS
0--/--/--/--1--/--/--/--2--/--/--/--3--/--/--/--4
3 6 9 3 6 9 3 6 9 3 6 9
MONTHS MONTHS MONTHS MONTHS

learn to communicate product information clearly?

4
l 2 3 OR MORE
YEAR YEARS YEARS YEARS

--2--/--/--/--3--/--/--/--4
9 3 6 9 3 6 9
S MONTHS MONTHS

0--/--/--/--1--/--/--
3 6 9 3 6
MONTHS MONTH

learn to communicate a sales presentation clearly?

4
1 2 3 OR MORE
YEAR YEARS YEARS YEARS
1 -2--/--/—-/--3--/--/--/—-4

0--/--/--/-- --/--/--
3 6 9 3 6

9 3 6 9 3 6 9
MONTHS MONTHS

MONTHS MONTHS

to exceed all sales targets and objectives?

4
l 2 3 OR MORE
YEAR YEARS YEARS YEARS
0--/--/--/--1--/--/--/--2--/--/--/--3--/--/--/--4
3 6 9 3 6 9 3 6 9 3 6 9
MONTHS MONTHS MONTHS MONTHS

PLEASE GO ON TO THE NEXT PAGE

151

How long does it take a new sales representative in your company to...

9) What is your OVERALL estimate of how long it takes a new sales
representative in your company to learn how to do their sales
job well?

4
l 2 3 OR MORE
YEAR YEARS YEARS YEARS

0--/--/--/--1--/--/--/--2--/--/--/--3--/--/--/--4
3 6 9 3 6 9 3 6 9 3 6 9
MONTHS MONTHS MONTHS MONTHS

PLEASE GO ON TO THE NEXT PAGE

152
SECTION 5

----------------------- INSTRUCTIONS---------------------------------—-
The six questions below also try to get at what you feel is the
relationship between pay and performance for sales people in this
company. The questions are in a somewhat unusual format. We will be
asking you to indicate the probability, or chance in 100, that you
will receive XXX dollars of pay annually if you sell XX,XXX dollars of
product in a years time. For example, if your company guaranteed you
a salary of $15,000 a year plus 10% commission on your sales volume,
you would earn $65,000 ($15,000 + $50,000) on sales of $500,000 and
would make the following response:

If your total annual sales performance was $500,000, what would
be your chances (l in 100) of getting paid .....

 

 

 

 

 

 

A) $000.00 per year ........ +
B) $05,000 per year ........ +
C) $25,000 per year ........ +
D) $45,000 per year ........ +
E) $65,000 per year ........ +
F) $85,000 per year ........ +

TOTAL - 100

On the other hand, if you were paid on straight commission and did not
receive a guaranteed salary, you would respond differently.

If your total annual sales performance was $500,000,
what would be your chances (1 in 100) of getting

 

 

 

 

 

 

paid ........
A) $000.00 per year ........ +
B) $05,000 per year ........ +
C) $25,000 per year ........ +
D) $45,000 per year ........ +
E) $65,000 per year ........ +
F) $85,000 per year ........ +
TOTAL = 100

Don't worry if the pay and performance figures don‘t match up
perfectly. for example, the sales representative on straight
commission would have earned $50,000 for the year. Since $50,000 was
not listed, the two closest amounts of pay were used: 90 (for a 90%
chance) was written next to $45,000 and 10 (for a 10% change) was
written next to 65,000. JUST MAKE SURE THAT ALL THE RESPONSES 1N EACH
QUESTION 222 12 122.

 

Finally, we have obtained dollar estimates of average, low and
high pay, and average, low and high performance for sales
representatives in your company. Therefore, the pay and performance
numbers used in this part of the questionnaire represent average, low
and high pay and performance levels for sales representatives in your
company.

PLEASE GO ON TO THE NEXT PAGE

153

Chances in 100 of Getting this Annual Pay

No A Slight A Fifty-Fifty A good Certain This
Chance Chance Chance Chance Will Happen
(0) (25 (50) (75) (100)

O 10 20 30 4O 50 60 70 80 90 100

Complete each question by placing a number between 0 and 100 on each
line. Make sure that the total for each question adds to 100.

1) If your total annual sales performance was $000,000, what would
be your chance (1 in 100) of getting paid ........

 

 

 

 

 

 

A) $000.00 per year ........ +
B) $05,000 per year ........ +
C) $25,000 per year ........ +
D) $45,000 per year ........ +
E) $65,000 per year ........ +
F) $85,000 per year ........ +

TOTAL - 100

2) If your total annual sales performance was $220,000, what would

be your chance (1 in 100) of getting paid ......

 

 

 

 

 

 

A) $000.00 per year ........ +
B) $05,000 per year ........ +
C) $25,000 per year ........ +
D) $45,000 per year ........ +
E) $65,000 per year ........ +
F) $85,000 per year ........ +

TOTAL = 100

3) If your total annual sales performance was $960,000, what would

be your chance (1 in 100) of getting paid ........

 

 

 

 

 

 

A) $000.00 per year ........ +
B) $05,000 per year ........ +
C) $25,000 per year ........ +
D) $45,000 per year ........ +
E) $65,000 per year ........ +
F) $85,000 per year ........ +

TOTAL = 100

PLEASE GO ON TO THE NEXT PAGE

154

Chances in 100 of Getting this Annual Pay

No A Slight A Fifty-Fifty A good Certain This
Chance Chance Chance Chance Will Happen
(0) (25 (50) (75) (100)

0 10 2O 30 40 50 60 70 80 90 100

Complete each question by placing a number between 0 and 100 on each
line. Make sure that the total for each question adds to 100.

4) If your total annual sales performance was $1,700,000 what would
be your chance (1 in 100) of getting paid ........

 

 

 

 

 

 

A) $000.00 per year ........ +
B) $05,000 per year ........ +
C) $25,000 per year ........ +
D) $45,000 per year ........ +
E) $65,000 per year ........ +
F) $85,000 per year ........ +

TOTAL - 100

5) If your total annual sales performance was $2,440.000, what

would be your chance (1 in 100) of getting paid ........

 

 

 

 

 

 

A) $000.00 per year ........ +
B) $05,000 per year ........ +
C) $25,000 per year ........ +
D) $45,000 per year ........ +
E) $65,000 per year ........ +
F) $85,000 per year ........ +

TOTAL - 100

6) If your total annual sales performance was $3,180,000, what

would be your chance (1 in 100) of getting paid ......

 

 

 

 

 

 

A) $000.00 per year ........ +
B) $05,000 per year ........ +
C) $25,000 per year ........ +
D) $45,000 per year ........ +
E) $65,000 per year ........ +
F) $85,000 per year ........ +

TOTAL = 100

THANK YOU FOR COMPLETING THIS QUESTIONNAIRE.

 

    

IIIIIIIIIIIIIIIIIIIIIIIII

11111111111111|11111111111111111111 111911111 11