137
724

 

m
/

'1

01:.) r =

5&29 l9 6 */

 

LIBRARY
Michigan State This is to certify that the
University thesis entitled

 

 

 

INDIVIDUAL DIFFERENCES IN WITHIN-PERSON VARIABILITY
OF PERSONALITY: IMPLICATIONS FOR WELL-BEING

presented by

Brendan Michael Baird

has been accepted towards fulfillment
of the requirements for the

MA. degree in Psychology

 

3/67 94

Major Professor’s Signature
December 11, 2003

Date

MSU is an Affirmative Action/Equal Opportunity Institution

o.-.-.-.. .

ui-.-.-49.-.—-.---.-.-A-.-.- —

 

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

 

DATE DUE

DATE DUE

DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 c:/CIRC/DateDue.p65—p.15

 

INDIVIDUAL DIFFERENCES IN WITHIN-PERSON VARIABILITY
OF PERSONALITY: IMPLICATIONS FOR WELL-BEING

By

Brendan Michael Baird

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF ARTS
Department of Psychology

2003

ABSTRACT
INDIVIDUAL DIFFERENCES IN WITHIN-PERSON VARIABILITY
OF PERSONALITY: IMPLICATIONS FOR WELL-BEING
By

Brendan Michael Baird

Although many researchers consider personality variability to be an individual difference,
there is disagreement about the implications of inconsistency for the individual. Some
believe that inconsistency is maladaptive, whereas others believe that variability may
provide a social advantage. In addition, similar measures of consistency often fail to
converge. This paper addresses these issues by measuring variability using two
assessment strategies: a standard self-concept differentiation (SCD) scale across several
social role contexts and an experience sampling measure of self-perceived personality
over actual situations. Across two studies, SCD was related to higher levels of negative
affect and lower levels of positive affect and life satisfaction. However, factor-analytic
SCD scores are confounded by theoretically irrelevant sources of response variance.
Calculating alternative indexes of cross-role variability yielded very different patterns of
relationships with well-being. In addition, only one of those indexes was able to predict
actual changes in self-perceptions from experience sampling methods in Study 2. These
momentary changes in self-perceptions were not Significantly associated with any well-

being measures.

To my parents, Mary Pat and Greg Baird.

iii

ACKNOWLEDGEMENTS

I would like to thank my committee, Isis Settles, Deborah Kashy, Neal Schmitt, and
especially my advisor Rich Lucas, for their hard work and insight. I would also like to thank
Larry Messe for his example of how to be both a great researcher and a great person.
Furthermore, I would like to acknowledge the support and insight of my peers, Eric Sambolec,
Dong-Heon Seok, & Joan Poulsen. In addition, special thanks go to the members of our research
team, especially Kimdy Le and the outstanding undergraduate research assistants.

Finally, most love and gratitude to my family and fiiends, who know me as I am and
bring out the best in me. I must also thank Marisa Sturza for her patience and support through

this difficult process.

iv

TABLE OF CONTENTS

LIST OF TABLES ........................................................................... vi
LIST OF FIGURES .......................................................................... vii
INTRODUCTION ............................................................................ 1
Problems with SCD ..................................................................... 5
Possible solutions ........................................................................ 8
Testing predictive validity .............................................................. 10
Present studies ........................................................................... 12
STUDY 1 ...................................................................................... 14
Method .................................................................................... 14
Results ..................................................................................... 16
Discussion ................................................................................ l9
STUDY 2 ..................................................................................... 21
Method .................................................................................... 22
Results .................................................................................... 24
Discussion ................................................................................ 27
GENERAL DISCUSSION ................................................................. 30
Limitations ............................................................................... 32
Future directions ......................................................................... 35
FOOTNOTES ................................................................................. 39
APPENDICES ................................................................................ 40
REFERENCES ............................................................................... 53

LIST OF TABLES

Table 1. Example of two cases using factor analytic SCD ............................ 39
Table 2. Correlations between indicators of consistency and well-

being (Study 1) ............................................................................... 40
Table 3. Correlations between the Big Five personality factors and indicators

of consistency (Study 1) .................................................................... 41
Table 4. Correlations between the Big Five personality factors and indicators

of well-being (Study 1) ..................................................................... 42
Table 5. Correlation matrix of indicators of consistency (Study 1) .................. 43
Table 6. Correlation matrix of indicators of consistency (Study 2) .................. 44

Table 7. Correlations between indicators of consistency and
well-being (Study 2) ........................................................................ 45

Table 8. Correlations between the Big Five personality factors and
indicators of consistency (Study 2) ....................................................... 46

vi

LIST OF FIGURES

Figure 1. Relationship between repeated-measure mean and

standard deviation ..................................................................................... 47
Figure 2. Distribution of SCD scores for study 1 ................................................ 48
Figure 3. Distribution of SCD scores for study 2 ................................................ 49

vii

INTRODUCTION

Although there has been some evidence that variability in individual differences
may be adaptive (Snyder, 1974, 1979; Paulhus & Martin, 1988), many argue that
personality, by definition, must maintain some degree of stability. Researchers search for
consistent patterns in peoples’ behavior that may be attributable to some stable
underlying disposition. As such, many theorists view inconsistency as abnormal or as a
Sign of maladjustment (Lecky, 1945; Block, 1961; Donahue, Robins, Roberts, & John,
1993; McReynolds, Altrocchi, & House, 2000; Diehl, Hastings, & Stanton, 2001).
Assumptions about the consistent nature of personality predate modern trait theories.
Lecky (1945) argued for a theory of personality that was based on the notion that self-
consistency is the most basic and universal human motive. In direct response to other
motive-based theories of psychosexual drives that were popular at the time, Lecky
suggested that all of an individual’s values are organized into a single system, and that all
behavior is motivated by the need for unification of these values. Finally, Lecky argued
that self-consistency is a subjective experience and therefore cannot be directly inferred
from behavioral observation.

Historically, trait theories of personality have been rooted in assumptions about
stable dispositions or temperaments that guide behavior (McCrae & Costa, 1997). Some
theorists have even argued that these structures originate in heritable physiological
systems within the individual (Eysenck, 1981; Gray, 1981). AS such, many trait
psychologists have tended to look upon behavioral inconsistency as an indication of

pathology (Block, 1961; Donahue, Robins, Roberts, & John, 1993). This

conceptualization has often been supported in research. In perhaps the first empirical
investigation into the nature of personality variability, Block (1961) hypothesized that
people located at either pole of the variability dimension -- those who are either
inconsistent or rigid -- would be socially disadvantaged. He tested this curvilinear
relationship between variability and psychological adjustment by asking participants to
describe their interpersonal behavior across eight relationships. Instead of finding the
relationship he expected, Block found that subjects with extreme variability in their
personality across their relationships possessed higher levels of maladjustment. Those
who exhibited extreme stability, or rigidity, were no less well-adjusted than those with
moderate amounts of variability. This study became part of the foundation for several
subsequent theories of cross-Situational change in personality.

Recently, F leeson (2001) suggested that variability iS actually a component of
personality. He proposed a conceptualization of traits as distributions of state
expressions of those traits. First, he argued that most people regularly express all levels
of all traits, and that Situational factors determine which level will be expressed at a given
moment. Second, although behaviors are difficult to predict for a Single occasion, the
mean of an individual’s distribution of trait expression is stable and predictable. Finally,
Fleeson suggested that other characteristics of these within-person distributions are also
meaningful. In particular, the standard deviation of a distribution may be used as a
measure of cross-Situational consistency. Using hand-held computers to measure
personality across Situations, Fleeson found high within-person variability in trait

expression. Furthermore, he found stable individual differences in personality variability.

Theories of consistency have begun to find application in other disciplines within
personality psychology. Donahue and her colleagues (Donahue, et. al., 1993; also see
Diehl et al., 2001) expanded on the work of Lecky (1945) and Block (1961) by
examining the relationship between consistency in the self-concept and psychological
well-being. According to Donahue et. a1. (1993), the self-concept comprises multiple
components or identities, and individuals differ in the degree to which these identities are
distinct or differentiated from each other. Self-Concept Differentiation (SCD) is an
individual difference that reflects the degree to which ones’ numerous social identities are
differentiated from each other rather than integrated into a unitary self. People who are
differentiated are believed to have chronic problems with self-image and social
relationships, thus predisposing them to greater levels of stress and negative emotions.

In order to measure self-concept differentiation, Donahue et al. (1993) assessed
self-reported personality across five social roles: student, fiiend, romantic partner,
son/daughter, and worker. Participants were first asked to rate themselves on each of 60
attributes for how they behaved “in general”. They were then asked to rate themselves on
these same adjectives for how they behave in each social role. People who have
consistent self—concepts should report expressing the same amount of each characteristic
across every role. For example, a consistent person would rate themselves as a 4 on a 7-
point “assertive” scale each time. If a person is not consistent, they may rate themselves
as being a 2 on assertiveness in some roles and a 6 or 7 in others. Donahue et a1.
computed an SCD score by individually factor analyzing each person’s personality scores
across all five social roles. The first factor that emerged from a principal components

analysis reflects the amount of variance that is shared across roles or, in other words, the

amount of consistency. Therefore, the amount of variance that cannot be accounted for
by this first factor reflects inconsistency, or SCD. Replicating Block’s (1961) findings,
Donahue et. a1. (1993) found that self-concept differentiation was positively related to
maladjustment across several measures of well-being.

The individual difference of SCD continues to be a popular topic of research.
Recently, several studies have found support for the SCD/well-being relation (Sheldon,
Ryan, Rawsthome, & Ilardi, 1997; Diehl et al., 2001; Suh, 2002). Sheldon et a1. (1997)
examined the role of subjective authenticity in self-reports of the Big Five and in ratings
of satisfaction across specific social roles. They found evidence that authenticity may
mediate the SCD/well-being relation. Diehl et a1. (2001) found that the SCD/well-being
relation increases in strength as people age. Sub (2002) argued that the association is due
to norms of individuation that are found in Western cultures. He found support for his
argument by demonstrating that, while there are greater levels of SCD in Korea, the
association with well-being does not exist there. These studies have highlighted several
intervening variables which account, at least partially, for the association between SCD
and well-being. However, there hasn’t been enough empirical research to fully explain
the processes which underlie this association.

Donahue et a1. (1993) argued that individual differences in personality traits that
are associated with SCD might explain why some people are differentiated. In particular,
they found that the traits of Neuroticism, Agreeableness, and Conscientiousness were
moderately correlated with SCD. They interpreted this pattern of associations as
evidence that people high in SCD have problems respecting and following social norms

and, in extreme cases, develop antisocial and delinquent behavior patterns (Donahue et

al., 1993; p.838). These problems then lead to conflicts in the family, and subsequently,
to a failure to integrate their different relationship experiences into a coherent sense of
self (p.844). One question may be whether it is SCD, or just the combination of low
Agreeableness and Conscientiousness and high Neuroticism that would lead to such
outcomes. In fact, issues concerning the measurement properties of SCD have kept these
two alternatives from being fully disentangled.
Problems with SCD

Research has shown that an association between SCD and well-being exists.
However, there are both theoretical and empirical reasons to doubt the validity of SCD.
First, global reports of personality are difficult for respondents to make (Schwarz,
Groves, & Schuman, 1998; Lucas & Baird, 2003). This kind of report may be even more
difficult to make across several social roles. In particular, such a measure may be
severely limited by the respondents’ memory of relevant events or situations within each
role context (Lucas & Baird, 2003). Memory may be biased by selective recall of
positive events, influenced by personal or social expectations and norms for each role, or
misguided by confusion about which events or Situations belong in specific role
categories. In addition, given that there may be roles which a participant enacts less
frequently or with greater irregularity than others (i.e. romantic partner versus family
member), memory effects may not apply equally to each role in the measure.

A second concern about the validity of SCD is that there is little empirical
evidence of its convergence with similar measures. Several recent studies have compared
multiple methods of assessing consistency. McReynolds, Altrocchi, and House (2000)

developed an explicit measure of self-consistency which assessed the degree to which a

person believed that they were the same across situations or relationships. The authors
found a Significant moderate correlation of .41 between their measure of self-pluralism
and SCD. However, Campbell, Assanand, and DiPaula (2003) used a measure Similar to
Donahue et al.’s SCD and found that this measure was only significantly correlated with
one of the four other scales in their study. In yet another study, Diehl et a1. (2001)
compared standard SCD scores with scores on another measure developed by Harter and
Monsour (as cited in Diehl et al., 2001). They found small correlations between these
two measures, and concluded “the two measures did not assess SCD in Similar ways.”
(p.647) In addition, they found that the Harter and Monsour measure was only associated
with one of their 11 well-being indexes. Understanding how and why these Similar
measures tend to disagree is a major goal of this paper.

A final concern, which may explain the problems described in the previous
paragraph, deals with the calculation of SCD. The factor-analytic method, first
developed by Block, is not clearly interpretable and may combine irrelevant sources of
item variance with cross-role changes in personality (Locke, 2003). Donahue and others
(Sub, 2002; Diehl etal., 2001) have interpreted the SCD measure as an index of “the
extent to which an individual’s ordering of the attributes from most to least descriptive
varies fiom role to role” or “the mean intercorrelation among the role identities.”
(Donahue et al., 1993, p.836) While the relationship between this index of variability and
well-being is consistent across studies, the calculation of SCD makes interpretation
difficult. IS SCD really an index of change? How might item content affect the

interpretation of SCD? What can account for the relationships between SCD and other

individual differences? Our first objective is to understand the nature of the variability
that is captured in Donahue’s SCD.

Principal components analysis will extract a large single factor only if there are
strong correlations among variables. In order to have strong correlations, a person must
have a small amount of variance within the same item across each role and have a large
amount of variance across items within the same role. Table 1 contains an example of a
hypothetical case in which two people are equally consistent across roles: both
individuals report being more nervous in Role 3 and more talkative in Role 1, but each
person changes by only one scale point. In fact, within each item, both people only
fluctuate slightly. They have an identical pattern of variance within items across roles
and an average within-item standard deviation of .51. In contrast, within each role,
Person 1 shows more change across items. In Role 1, She reports being a 5 on
“Talkative” and a 1 on “Organized”, while Person 2’s ratings of those same items are 4
and 2, respectively. The standard deviation across items in Role 1 is 1.81 for Person 1
and 0.83 for Person 2.

Using the factor analytic technique, Person 1 would have a very low
differentiation score because the first factor in her data explains almost all of the
variance. Her SCD score would be around 6. Person 2 would have a differentiation
score of around 30 because only 70 percent of the variance in his responses has been
accounted for. According to SCD theory, Person 1’s self-concept is more consistent
across social roles. However, this conclusion is clearly not supported by the data. It is
the large amount of cross-item variability that produces a stronger first factor, not greater

cross-role consistency. Judging by item content, cross-item variance simply reflects the

unique patterning of traits within the individual. This source of variance is irrelevant to
theories of cross-Situational change, and may also be highly sensitive to item scaling.

The fact that factor analytic SCD scores may be determined by people’s trait
patterns is problematic because these traits are also known to be reliably correlated with
well-being. In particular, Extraversion has been linked with positive affectivity and
Neuroticism with negative affectivity (Lucas & Diener, 2000). Since cross-item
variability represents the patterning of traits within a person, including this source of
variance may inflate the association between SCD and well-being. Furthermore,
associations that have been found between SCD and personality traits may result from
including this trait information in the measure of change. The theory of SCD is
fundamentally about response consistency (i.e. cross-role variance), and the way it is
measured should clearly reflect it.

Possible Solutions

Several variants of the factor-analytic SCD are possible. For example, Campbell
et al. (2003) used bivariate correlations to indicate the amount of agreement, or unity, in
the self-concept across roles. Specifically, if a person reported having a similar profile
across each role, then the average correlation among the roles would be high. While this
index may be more straightforward than principal components analysis, it still suffers
from the same confounding effects of cross-item variance. AS I discussed above, factor
analysis begins with the matrix of correlations among roles, and the eigenvalue of the
first principle component is highly related to the average correlation in this matrix (see
Campbell et al., 2003). Therefore, the average correlation among roles does not correct

for confounding item variance, and is not a suitable alternative to factor analysis.

As part of their landmark study, Donahue et a1. (1993) explored 2 indexes of
variability from their self-concept differentiation measure. The first was the factor
analytic score described above. The second was the average standard deviation of
responses on each attribute across roles. The authors suggested that this latter measure
represented “the absolute differences among the (role) identities” (Footnote 1, p.836).
Based on the strong correlations that were found between these indexes, Donahue et a1.
(1993) decided to only report results from the factor-analytic scores in their paper.

F urtherrnore, the authors reported that the two seemed to be similarly related to their
well-being variables. We feel that it is important to understand how these indexes differ
theoretically. In fact, unless item means are all normally distributed, this alternative
approach may not entirely account for the effect of mean scores.

Because SCD is constructed from repeated measures, the standard deviation of an
item across roles iS tied to the mean score for that item. For example, in order for a
person to have an average score of O on the item Trustful, they must rate themselves as a
O on that item in every role. Even if they report being a 1 in a Single role, their average
will be above 0. Therefore, variability can only occur in the middle of the scale. Figure
1 contains an example of a scatter plot of mean scores and standard deviations across
roles for the item WTrustful”. The correlation between the mean and standard deviation is
-.54, reflected in the dashed line. However, the first thing of note is that the overall
distribution of means is skewed, with most people averaging a 2 or above. In addition,
those averages falling between 1 and 3 have the highest standard deviations. These two
conditions create a Situation where there are more observations at the high end of the

“T rustfirl” dimension, and these individuals are also, by definition, more consistent. In

other words, only the individuals lower on trust are variable because they fall in the
middle of the scale. There are very few people who are (consistently) not trustful at all.
Therefore, unless the distribution of item means is approximately normal, the standard
deviation will appear to be a linear effect of the mean, and will contain information about
the trait dimension that is measured by that item.

Since item means are so closely tied to raw standard deviations, the effect of trait
associations between personality and affect may still be driving the relationship between
SCD and well-being. The theoretical quadratic effect of mean scores on standard
deviations can be corrected for to produce another alternative index of variability. This
can be done by computing ordinary least squares regression equations and saving the
residual variance in standard deviations for each item. The average of these residuals
represents the amount of raw variability in responses across roles, after accounting for
non-normality in item distributions. With this index, we can determine whether, at any
given mean score, a person with that score who has a lot of response variance is any
different from someone with that score who has very little response variance. In other
words, we can tell whether a consistent “2” is different fi'om a variable “2”. According to
SCD theory, a consistent “2” should be more well-adjusted.

Testing predictive validity

Most of the research on SCD to date has focused on the pattern of correlations
among individual differences related to SCD (i.e. personality, self-esteem, positive and
negative affect). However, very few studies have tested the ability of SCD to predict
relevant outcomes. In particular, it is important to establish SCD as a measure of

inconsistent self-perceptions in order to accurately interpret the correlations with other

10

scales. One study that tested a similar measure in this way was McReynolds et al.(2000).
They examined the relationship between scores on an explicit measure of self-
consistency and within-person variability of other measures over time. Using a 75-item
annual check-list of life events, they found that self-pluralism (SPS) predicted self-
reported life changes over an 8-year period. In other words, people high on SPS were
also more likely to experience meaningful life changes. In addition, they found that SPS
was related to variability in well-being, depression, and body weight, over that same
period. Therefore, explicitly asking participants how much they change predicted other
relevant measures of change. Because SCD is a construct that is Similar to self-pluralism,
we Should expect it to also predict meaningful changes in a person’s life.

According to SCD theory, people who are highly differentiated change the way
that they perceive themselves based upon changing situational contexts. Donahue et a1.
(1993) argued that SCD may result fi'om “oversensitivity to the expectations and
demands of others” (p.845). This greater susceptibility to situational influences may be
associated with a lack of a true sense of self. Because people with high SCD lack a clear
definition of self, we should expect these people to report a greater variety of self-
perceptions across different Situations. Those who are low in SCD should not see
themselves differently across these Situations because their definition of themselves does
not change based upon the social roles they enact.

AS a test of the predictive validity of SCD, as well as the alternatives discussed
above, I will see how these scores relate to how people report on their behavior as it is
happening. By using diary methods, I can assess self-perceptions over several situations

and participants can provide descriptions of the kinds of roles they are in at those times.

11

This approach may be less dependent upon the specificity of the respondents’ memories
because reports are made in closer temporal proximity to relevant experiences (Bolger,
Davis, & Rafaeli, 2003). In this way, the same process of personality change can be
assessed without relying on participants’ ability to accurately recall events or imagine
what they might be like in a role that they may not fulfill. Using this technique, I will test
whether SCD can predict variability in self-perceptions over on-going experiences. At
the same time, I will be able to evaluate the alternative analyses to find out if prediction
improves as the measure is refined.
Present Studies

High scores on SCD are supposed to indicate psychological “fi'agmentation”, and
be “associated with intrapersonal and interpersonal difficulties marked by emotional
distress, rejection of social norms, and volatile relationships in love and work” (Donahue
et al., 1993; p.859). However, ambiguities in the measure make it difficult to understand
exactly why SCD is related to adjustrrrent. In addition, studies of SCD generally haven’t
addressed Specifically how this construct relates to changes in self-perceptions over time.
As a result, many important issues still require attention. These include: (1) why similar
measures of consistency have different relationships with well-being (Diehl et al., 2001 ),
and (2) how well reports of consistency at the level of the self-concept predict
consistency in self-perceptions of personality across actual Situations. It is clear that
cross-item variance is irrelevant to SCD theory, and that this variance reflects trait
information. It also seems plausible that cross-role variance is closely tied to trait

information as well. However, it is important to find out what effect this information has

12

in the SCD/well-being relation. Furthermore, it is unclear whether this information can
be controlled for without drastically changing the instrument.

The two studies I present here were conducted for several purposes. Study 1 was
designed to replicate the association between SCD and well-being and to explore this
relationship in greater detail. Specifically, I wanted to examine the measurement
properties of SCD to assess the sources of variance that contribute to it. In addition, I
developed an alternative method of computing SCD in order to test the hypothesis that
the relationship between SCD and well-being iS the result of a measurement artifact.
Then, in Study 2, I compared the standard, contextual self-report method of assessing
SCD with an experience sampling measure (ESM) of momentary self-perceptions from
actual situations. Specifically, I was interested to see if self-perceptions of personality
variability, as measured by the standard SCD scale, could predict momentary changes in
self-perceptions. Also, I wanted to explore how these momentary changes are associated
with well-being. Even if the two measures of self-concept differentiation are found to be

related to one another, it is possible that the two are differentially related to well-being.

13

STUDY 1

The main purpose of Study 1 was to test whether the apparent effect of SCD on
well-being is due to reliable associations between personality and affect, rather than to
variability per se. The first step was to replicate the standard SCD/well-being relation
using the factor analytic SCD measure. Next, 1 investigated the properties of this scale to
Show that it includes variance that is irrelevant to self-concept differentiation. I then
investigated whether this irrelevant variance inflates the correlation with well-being by
testing the strength of the association when more appropriate indexes are used. SCD is
supposed to be related to high Neuroticism and low Agreeableness and
Conscientiousness. If this is the case, then the relationship between SCD and well-being
may not be due to an incoherent self-concept, as much as to underlying personality traits
that are known to be related to adjustment.

Method
Participants.

Two-hundred and seventy participants were recruited from introductory
psychology courses offered during the fall semester at a large university in the
Midwestern United States. Each participant received partial course credit for completing
the project. Seventy-seven percent were women and 98% were between 18 and 25 years
of age. Seventy-nine percent identified themselves as White/Caucasian, 8% as Asian, 7%
as Black/Afiican-American, 1% as Native American, and 5% as a race other than those
listed.

Procedure.

14

Participants reported to a large classroom on campus in groups of 15 to 30. First,
participants completed the 50—item International Personality Item Pool (IPIP, Goldberg,
2001), which measures the traits of Extraversion, Neuroticism, Conscientiousness,
Agreeableness, and Openness. This measure consists of a number of brief statements,
and participants are asked to rate how well each item describes their personalities on a 0-
4 scale (i.e., “Am the life of the party”, “Worry about things”, “Use difficult words”,
“Like order”, “Syrnpathize with others’ feelings”). I also assessed each participant’s
level of subjective well-being (SWB). Participants were asked to complete the Intensity
and Time Affect Survey (ITAS; Diener, Smith, & Fujita, 1995) as an assessment of trait-
level affect. This scale consists of 24 moods which are each rated on a 0-4 scale for how
commonly they are felt (i.e., Joy, Sadness, Pride, Anger). The ITAS provides estimates
of trait positive affect (PA) and negative affect (NA). AS a third measure of SWB, I
assessed life satisfaction (LS) using the Satisfaction with Life Scale (SWLS; Diener,
Emmons, Larsen, & Griffen, 1985). This measure consists of 5 statements (“In most
ways my life is close to my ideal.” “The conditions of my life are excellent.” “I am
satisfied with my life.” “So far I have gotten the important things I want in my life.” “If I
could live my life over, I would change almost nothing”) which are rated on a 0-4
agreement scale.

Finally, participants were asked to complete an SCD measure that was based on
Donahue et al.’s (1993) scale. First, participants were asked to report their personality in
general, by rating 20 trait adjectives on a 0-4 scale for how well each described them.
These adjectives have been used in a previous study by F leeson (2001) and measured

each of the Big Five Traits. Then, for each of 6 social-role conditions (Friend, Romantic

15

Partner, Family Member, Student, Worker, and Stranger), participants were instructed to
first think about how they see themselves in that role. Then they were asked to rate the
same 20 adjectives for how well they described them in each role. The stranger social
role was included to capture those situations in which the person may be in a social
Situation, but not with someone they know personally.
Results

Scores for SCD were computed for 264 participants (m =33.34, sd = 16.66). Four
participants could not be scored on SCD: two because they provided incomplete data and
two because their responses had no variance which made factor analysis impossible. For
each participant in the final sample, a principle components analysis was conducted using
the social-role conditions as the replications. The percent of variance accounted for by
the first extracted factor was subtracted from 100 to yield the index of differentiation.
The distribution of SCD scores can be found in Figure 1. Scores for PA ranged fi'om
1.00 to 4.00 (m = 2.83, sd = .61), NA ranged from .06 to 3.63 (m = 1.47, sd = .71), and
LS ranged from .00 to 4.00 (m = 2.58, sd = .95).

The first objective was to replicate the relationship between SCD and well-being.
As can be seen in the left column of Table 2, SCD was Significantly related to all three of
the SWB measures. People with high SCD scores reported higher levels of NA and
lower levels of PA and LS. These results replicate previous studies that have supported
the “fragmentation” theory of within-person variability. Also, replicating previous
findings, SCD was negatively related to Extraversion, Agreeableness, and

Conscientiousness, and positively related to Neuroticism (line 1 of Table 3). Openness

l6

was the only trait that was not Significantly associated with SCD. These traits were also
correlated with the 3 SWB variables (Table 4).

Next we wanted to decompose the SCD score into its Specific components. AS
noted above, a strong first factor will only emerge if there are both low cross-role
variance and high cross-item variance in the data. Yet cross-item variance is irrelevant to
SCD. To demonstrate this, I computed both the average cross-item standard deviation (m
= 1.13, sd = .21) and average cross-role standard deviation (m = .70, sd = .25) for each
person. The factor-analytic SCD correlated .79 (p <.01) with cross-role S.d. and .24 (p
<.01) with cross-item S.d. In other words, while the factor analytic SCD score is strongly
related to cross-role variance, it is also significantly associated with theoretically
irrelevant cross-item variance. In addition, cross—item variance was positively correlated
with Positive Affect (r = .22, p < .01) and Life Satisfaction (r = .17, p < .01), and
negatively correlated with Negative Affect (r = -.27, p < .01).

When we examined item standard deviations (thereby removing cross-item
variance fi'om the analyses) cross-role variance was still Significantly associated with
SWB, albeit to a much lesser degree than the overall SCD (see Table 2). It was
positively associated with NA and negatively associated with PA and LS. Whereas the
overall pattern was very Similar to that of the factor analytic index, the magnitude of the
associations decreased considerably. When theoretically irrelevant trait information is
removed fi'om the analysis, the result is a weaker association with well-being. In other
words, this drop illustrates that trait level associations between personality and affect are

confound the factor analytic index of SCD.

l7

Theorists interested in SCD are concerned with variability in the self-concept
across roles. The above analyses Show that factor analysis of SCD confounds variability
with mean levels of traits. Cross-role standard deviations remove some of this
confounding effect. However, for the 20 items in our scale, correlations between item
means and standard deviations ranged in magnitude from .07 to .54, with an average of
.35. Therefore, it is still not clear whether the associations between standard deviations
and well-being are due to individual differences in variability, or to associations between
traits and well-being.

In order to test this alternative explanation, we calculated a quadratic regression
equation for each item predicting cross-role standard deviations fiom mean scores and
squared mean scores. The average effect for the quadratic fimction of the mean was
Significant for every item (,6 range = -.79 to -2.38, average ,8 = -l .59, sd = .43; t range = -
3.89 to -12.60, average t = -7.86, sd = 2.61; all p’s < .01). An inspection of item mean
distributions also confirmed that most of these items were highly skewed.

After calculating the regression equations, we averaged the 20 residual values for
each person (m = .00, sd = .22). AS can be seen in Table 5, this corrected standard
deviation is very strongly related to the raw standard deviations. However, the corrected
standard deviation was not related to any of the SWB measures (NA r = .04, n.s.; PA r =
.03, n.s.; LS r = -.02, n.s). The further drop in correlations with SWB indicates that trait
information may have been driving the association with well-being even when raw
standard deviations were used. Once I controlled for the quadratic effect of mean scores
on standard deviations, there was no evidence for a relationship between SCD and well-

being.

18

The corrected standard deviation seems to account for irrelevant variance that is
inherent in factor analyses of SCD. AS a final test of this, I wanted to see whether
Block’s (1961) hypothesis of an inverted U-Shaped relationship could be supported using
the new index. I computed regression equations predicting the 3 SWB variables from
corrected standard deviations and the square of those values. A positive quadratic effect
emerged for NA (,8 = 1.00, t = 1.80, p <. 10), and negative quadratic effects were found
for both PA (5 = -.90, t = -1.87, p <.10) and Life Satisfaction (15’ = -2.19, t = -2.97, p
<.01). These results suggest that those people with a moderate amount of variance in
their responses across roles tend to report higher levels of subjective well-being. Those
people with extreme rigidity or variability report the lowest levels of subjective well-
being.

Discussion

Previous research has Shown that SCD is reliably related to well-being. We
replicated this finding using the standard procedure for computing SCD. However,
subsequent analyses supported my argument that there are several problems with this
method that may artifactually increase the magnitude of the relation. First the factor
analytic SCD conflates 2 sources of variance: across roles and across items. Of the two,
only cross-role variance is theoretically relevant to SCD because it represents the degree
to which a person’s responses on each item change fiom one role to the next. Cross-item
variance, on the other hand, Simply reflects the unique pattern of traits within the
individual. Because personality traits correlate with well-being, this irrelevant source of
variance inflates the correlation between SCD and well-being. When cross-item variance

was removed from the analysis, the relationship with well-being weakened considerably.

l9

The second issue we addressed relates to the use of raw standard deviation scores
as an index of Situational change. Because of the artifactual association between mean
levels and variance, along with the skewed distributions of item means, this index also
confounds trait information with variance. When I examined the corrected standard
deviation, the association between SCD and well-being dropped to zero. This suggests
that the original relationship between SCD and well-being is entirely due to associations
between mean levels of personality traits and affect. Finally, when response variance
was examined after correcting for these confounds the correlations between consistency
and SWB dropped to near zero. However, these correlations do not represent a lack of a
relationship. Rather, I found that a curvilinear relationship emerged instead (Block,
1961). In Study 2, I compare all three SCD indexes fi'om Study 1 with a measure of

consistency derived fiom reports of momentary self-perceptions.

20

STUDY 2

Previous research on SCD has relied heavily upon global self-reports to assess
response consistency (Donahue et al., 1993, Diehl et al., 2001). Very little research has
been done to examine the broad spectrum of methods that can inform theories of
personality change, yet SCD theory makes very clear predictions about dynamic changes
in the self-concept across situations. If SCD theory is correct, the self-concept of those
with high SCD Should change more dramatically across Situations than it should for those
with low SCD. In this study, I chose to employ an experience sampling technique to test
this possibility (Bolger et al., 2003; Stone, Shiffrnan, & DeVries, 1999). This method is a
newer version of traditional daily-diary methods, where participants provide an on-going
account of their experiences. For each moment that is sampled, a person’s perceptions
of himself or herself can be assessed, as well as aspects of the situation that they are in.
These reports can then be compared to obtain an estimate of the degree to which a
person’s self-perceptions change across roles in actual situations.

Some theorists may argue that SCD involves individual differences in the ability
to organize changing self-perceptions into a consistent self-concept, and that the two
measures should not be expected to associate. However, given the results of Study 1,
these measures may not converge instead because the factor-analytic SCD measure
combines irrelevant sources of variance. Therefore I expect that SCD and situational
changes in personality will not be related. Once these inherent confounds are accounted
for by using the alternative indexes, self-reported variability Should be more strongly

related to situational changes.

21

Finally, I wanted to examine whether a person’s level of change in momentary
self-perceptions is related to their well-being. Even if perceptions of change are able to
predict changes in perception across moments, the two may still be differentially related
to well-being. If momentary changes in self-perceptions are also negatively associated
with well-being, then this may still support the fragmentation view of the self-concept.

Method
Participants

One hundred and five participants were recruited from summer psychology
courses and fiom advertisements posted on campus at a large university in the
Midwestern United States. Our sample consisted of 72% women; 94% were between 17
and 20 years of age. Eighty percent identified themselves as White/Caucasian, 9% as
Asian, 7% as Black/Afiican-American, and 4% as a race other than those listed.
Materials

Questionnaire data for this study were collected using the same measures used in
Study 1: the 50-item IPIP, the 24-item ITAS, the 5-item SWLS, and our 20-item, 6-role
SCD measure. Palm m100 computers, equipped with the Experience Sampling Program
(ESP) were used for momentary assessment of personality. At each sampled moment, a
33-item questionnaire was administered (see Appendix L). Items on this questionnaire
included the same 20 trait adjective items used in the SCD measure from Study 1.
Momentary mood was assessed using 4 affect items: Pleasant, Unpleasant, Sad, and
Happy. These personality and affect items were each rated on a 7-point scale ranging
from 0 (“does not describe me”) to 6 (“describes me very well”). Finally, participants

responded to Six items designed to measure aspects of the situation that they found

22

themselves in at the time the alarm sounded. These items explicitly asked what social
role the person was in just before the alarm sounded. Participants were asked to respond
by indicating yes or no to “Were you being a student?”; “Were you being a fiiend?”;
“Were you being a romantic partner?”; “Were you being a family member?”; “Were you
being a worker?”; or “Were you being a stranger?”. These items were not mutually
exclusive, so participants could report being in more than one role at a time. Each item
was presented one at a time and responses were given by touching numbered boxes on
the screen below each question.

Procedure

Small groups of five to ten participants reported to a small classroom for the first
phase of the study. First, participants completed the 50-item IPIP as a global assessment
of trait personality. Participants were then asked to complete the ITAS and SWLS as
assessments of SWB. For the remainder of the session, participants were trained on the
use of the palrn-top computers.

Phase 2 of data collection began immediately after the end of the first
questionnaire session and consisted of 7 days of experience sampling. During this phase,
data were collected on the hand-held computers. The computers were programmed to
randomly sound an alarm up to 8 times per day during the participants’ waking hours.
Each time the alarm sounded, participants had 10 minutes to respond to the questionnaire.

Phase 3 of data collection took place one day after the completion of phase 2.
Participants returned to our lab to turn in the computers and were asked to complete our
revised version of the SCD measure. Once this was completed, participants were

thanked, debriefed, and paid. Participants received $20 for their completion of all three

23

phases of data collection. An additional $5 was giVen to those participants who
completed at least 5 questionnaires per day during phase 2 (82% completed 35 or more
questionnaires). One participant who decided to terminate their participation in the
middle of the study was compensated for the data that they provided up to that point.
Results

Scores for SCD were computed for 103 participants. Two people who chose to
end their participation early did not provide SCD data. The distribution of SCD scores
can be found in Figure 3. To get a general idea for how much people changed their
responses throughout the week, I computed each person’s overall ESM standard
deviation (m = .99, sd = .20) and also corrected for item skew using the regression
approach from Study 1 to compute the corrected overall standard deviation (m = .00, sd =
.17). These values represent the amount a person changed fiom moment to moment,
regardless of the roles they were in.

Next I wanted to recreate the data structure of the global SCD measure with the
ESM data. Using the six role items, I assigned each moment to the role or roles the
participant reported being in at the time. If a participant reported being in more than one
role at a time, the data for that moment were added to each role category that the person
was in.1 I then computed the average of each item within the roles, resulting in the Item
(20) x Role (6) matrix, with each cell representing the average for that item from multiple
moments. The cross-role ESM standard deviations (m = .34, sd = .28) and corrected
ESM standard deviations (m = 0.00, sd = .16) were then computed. These measures were
based on the roles for which the participant provided data. Most people (90%) provided

data for either 5 or 6 roles. Only 2 people gave reports for 2 or 3 roles, and no one gave

24

reports for 1 role or fewer.2 The most common role for which people had missing data
was romantic partner. When I conducted the analyses separately using only those people
with data for all 6 roles, the overall pattern of results did not change. Therefore, I present
the results fiom analyses on the entire sample.

The intercorrelations among the indexes of variability are presented in Table 6.
AS in Study 1, all of the indexes of variability based on the Donahue et a1. procedure are
Si gnificantly associated with each other. However, these measures are differentially
associated with the ESM indexes. The factor analytic SCD measure does Significantly
predict overall ESM standard deviations. However, this relation drops when the overall
ESM standard deviations are corrected for mean scores. In addition, SCD is not
associated with either the cross-role ESM standard deviations or the corrected ESM
standard deviations. In other words, the factor analytic SCD measure does not predict
actual changes in self-perceptions across situations or roles. The cross-role standard
deviation from the global SCD measure (m = .71 , sd = .20) is also not associated with
changes in self-perceptions across situations or roles in ESM. Only the corrected
standard deviation from the global SCD (m = .00, sd = .16) was significantly correlated
with each index of variability from the ESM data. This suggests that once the effects of
mean scores are controlled for, participants who had more cross-role response variance in
the global measure also had more cross-role response variance during the week that was
sampled.

Scores for PA ranged from 1.63 to 4.00 (m = 2.89, sd = .50), NA ranged from .19
to 2.88 (m = 1.15, sd = .58), and LS ranged from .20 to 4.00 (m = 2.81, sd = .83). As for

the relationship between SCD and SWB, the findings from previous research were

25

replicated as before: factor-analytic SCD was significantly correlated with Positive
Affect, Negative Affect, and Life Satisfaction. However, the pattern of relationships
between the alternative indexes of variability and well being are almost identical to those
found in Study 1 (top 3 lines of Table 7). When I examined cross-role standard
deviations, the relationship with SWB dropped considerably. Furthermore, the corrected
standard deviation was not Significantly associated with any of our SWB measures.

Only one of the indexes of variability from the ESM measure was associated with
SWB. Specifically, the corrected overall standard deviation was positively associated
with PA. This suggests that people whose self-perceptions change more across moments
actually report higher levels of positive affect. Taken alone, this finding contradicts the
fragmentation view of SCD. However, none of other indexes predicted any of the SWB
variables. In particular, the correlations between cross-role standard deviations and well-
being were all nearly zero. The correlations for the corrected standard deviation are
Slightly stronger, but are still not Significant. In fact, they are also in the opposite
direction to that predicted by SCD theory.

As an alternative to the trait measures of PA and NA, I also averaged the affect
items over the ESM measure to asses the amount of positive and negative affect that the
person experienced throughout the week. The global SCD measure was significantly
correlated with PA and NA over the week, but this association again decreased
substantially when confounds were removed. None of the ESM indexes were related to
these measures. In other words, there was no evidence that the amount of PA and NA
experienced during the week was related to the amount of variability in self-perceptions

during that same period.

26

Examining the trait data from the 50-item IPIP scale, I found that several of the
Big Five personality traits were associated with the factor analytic SCD, but that this
association was reduced when I focused only on cross-role standard deviations. As can
be see in Table 8, The corrected standard deviation from the global SCD measure was
only significantly associated with the trait of Extraversion. Neither index of cross-role
change from the ESM data was associated with personality in a way that would be
predicted by SCD theory. The only significant association was between the corrected
ESM standard deviation and the trait of Conscientiousness.

Finally, as in Study 1, I tested Block’s (1961) hypothesis of a curvilinear
association between consistency and SWB by regressing the corrected standard deviation
along with the square of those scores on the three SWB variables. Once again, the
quadratic firnction was negative for both PA ([3 = -2.07, t = -1.43, ns) and LS (,6 = -5.00, t
= -2.08, p <05), and positive for NA (B = 2.84, t = 1.71, p <.10). This suggests that
people who have a moderate amount of response variance on the global SCD scale report
higher levels of SWB.

Discussion

In order to test the predictive validity of the factor-analytic SCD measure, I
partitioned the ESM data in two different ways: across situations, and across roles within
those Situations. I then computed 2 indexes of both the degree to which participants’ self-
perceptions changed across situations and across roles. The SCD measure was
Significantly correlated with changes in self-perceptions across Situations. However, this
relationship was no longer Significant when item skew was accounted for in the ESM

data. The same pattern was true for raw standard deviations from the global SCD

27

measure. Only when item means were completely controlled for in the global data was
SCD able to predict raw and corrected variance across situations. As expected, the
factor-analytic measure also failed to predict changes in self-perceptions across roles.
Controlling for trait confounds in the global data resulted in stronger prediction of cross-
role change. This finding suggests that global reports of changes in personality are related
to actual changes in how a person perceives him or her self across situations, but that
confounds inherent in factor analysis along with deviations from multivariate normality
produce Type II errors in prediction.

Study 2 replicated the general SCD effect that factor-analytic scores are
negatively related to well-being. However, as in Study 1, this effect was due to the
confounding effect of mean trait scores on well-being. When cross-item variance was
removed, and the association between item means and standard deviations were
accounted for, the association between consistency and well-being dropped to near-zero.
As firrther evidence of this confound, I found Si gnificant moderate associations between
SCD and the Big Five personality traits, as well as associations between those traits and
SWB. The correlations between the factor analytic SCD measure and personality should
not be interpreted as an indication of some Shared dispositional qualities underlying both.
Instead, theses associations reflect that fact that trait information is imbedded in the item
content of the SCD scale.

The fragmentation theory of self-concept differentiation was also not supported
using ESM methods. Overall, variability in self-perceptions over time was not associated
with trait SWB. In addition, many of the correlations were in the opposite direction as

would be predicted by SCD theory. Finally, the amount of change in the ESM data was

28

also not related the amount of positive or negative affect experienced throughout the
week. Therefore, having changing momentary self-perceptions does not seem to have
immediate repercussions for the individual, nor is it reflected in their general level of

adjustment.

29

GENERAL DISCUSSION

Across two studies, the negative relationship between Donahue et al.’S (1993)
measure of Self-Concept Differentiation and well-being was Shown to be an artifact of
measurement. In particular, using factor analysis to compute an estimate of SCD is
inappropriate because it conflates cross-item variance with cross-role variance. Focusing
only on the theoretically meaningful source of variance resulted in substantially
weakened correlations between differentiation and well-being. In addition, as Study 1
illustrated, due to the nature of repeated measures designs, cross—role standard deviations
are also confounded with mean scores. Once this dependency was accounted for, self-
concept differentiation and well-being were no longer related. Finally, Study 2 provided
evidence that an estimate of cross-Situational consistency derived from ESM data was
also unrelated to well-being. These results fail to support the fragmentation view of self-
concept differentiation.

Traditional methods of computing SCD also lack predictive validity. That scores
from factor analysis do not predict changes in self-perceptions is not surprising, given
that this index has been found to be weakly related to other self-reports of self-
consistency (Campbell et al., 2003; Diehl et al., 2001). This seems to be due to the
imprecision of factor analysis as a measure of change. Conflating multiple sources of
variance makes factor-analysis difficult to interpret, and may be an imprecise method of
assessment in person-centered research (Locke, 2003).

There are more direct ways to assess self-consistency. Several measures of the

construct tend to rely on explicit self-reports, where people are asked outright how

30

consistent they believe they are. These kinds of questions do not depend upon ratings of
trait descriptors and may be less susceptible to the influence of trait-level associations.
One drawback to this approach is that these items may be prone to social desirability
biases if norms of consistency are strong in one’s culture (Sub, 2001). An advantage of
contextual reports is that they may be more easily specialized for particular research
questions. For instance, if a researcher is interested in differences in self-perceptions in
the family versus at work, contextual reports can focus Specifically on those two
domains. The use of contextual self-reports in the study of change may be strengthened
by systematically removing the pre-existing effects of personality on the outcome
variable of interest.

The results of these studies do not support the idea that self-consistency is a
universal human motive (Lecky, 1945). In fact, the distributions of the several indexes of
cross-role change indicate that moderate levels of variability are common (see also
Fleeson, 2001). Furthermore, there is evidence that a curvilinear trend exists in the
association between these indexes and subjective well-being when analyses isolate
appropriate response variance (Block, 1961). These results are by no means conclusive.
Lecky’s theory may be right in some instances and partitioning variance across roles is
not the only way to test for the consistency motive. For instance, consistency within
roles may be more important for adjustment than consistency between them. Researchers
may find that reports of changing self-perceptions within roles (i.e., worker or family)
have a stronger association with SWB.

It has been argued that the individual difference of SCD is strongly related

to personality traits (i.e. Neuroticism, Agreeableness, and Conscientiousness; Donahue et

31

al., 1993), such that people high on certain traits may be predisposed to experience SCD.
Our analyses provide a different account. The high correlations between SCD and
personality may reflect the redundancy of correlating factor analytic SCD scores with
personality traits. When cross-item variance was removed fiom the calculation of SCD, a
very different pattern of associations with personality began to emerge. In fact, neither
response consistency nor situational consistency seem to be reliably predicted by any of
the big five personality traits that are discussed in SCD theory.

Finally, modern trait theorists are beginning to focus on individual differences in
state consistency (Fleeson, 2001). Generally, this kind of research utilizes repeated
measures data, often in the form of experience sampling. In F leeson’s conceptualization
of traits as density distributions of states, the mean of the distribution of state expressions
of approximates a person’s trait level of that characteristic. He argues that mean scores
can be considered independent of state variance. However, as I have Shown, this may
only be true if trait scores are normally distributed. If the population distribution of a
trait is skewed, then the within-person variance of states may be related to the mean of an
individual’s density distribution.

Limitations

One concern with the approach that I am proposing is that, by way of statistical
manipulation, it controls for too much. If there are traits that underlie SCD which are
captured in Specific items in the scale, those traits should be allowed to influence the
consistency with which a person responds to all of the other items. For example, if a
person has high SCD because he or She is neurotic, then we should not control for

neuroticism when examining the association between SCD and well-being. However, it

32

is important to point out that none of the procedures used in these studies controlled for
global trait ratings. Instead, variance associated with mean levels for each item was
removed only fiom the same item’s SD. For instance, mean levels of irritability were
removed only when calculating the standard deviation for irritability. It would still be
possible for highly irritable people to report high levels of variance on other items. Thus,
it is theoretically possible for the corrected SD measure to be strongly correlated with
traits such as extraversion and neuroticism.

Using regression equations to derive an individual difference measure may also
be somewhat problematic. When using the residual value as a corrected standard
deviation, the equation from which this value is derived is sample-specific. One
advantage of the factor analytic approach is that a person’s SCD score will remain the
same regardless of what sample they are in. Using the quadratic residual score will lead
to different scores for the same person if they are put into a different sample. Therefore,
to ensure a reliable estimate of relative response consistency, data Should be collected
from larger samples. Alternatively, other multivariate statistics may be explored which
can control for both item content and non-normal data.

Another issue that Should be considered is whether one ought to expect our
different measures of consistency to converge at all. It could be argued that contextual
self-reports and ESM self-reports assess very different processes, and each method has its
own set of advantages and disadvantages. One advantage of using diary methods is that
they allow individuals to report on the roles that are relevant to them. Contextual self-
reports, on the other hand, impose a role structure on the self-concept by forcing

participants to rate themselves for roles that the may or may not fulfill in their actual

33

lives. On the other hand, one week of ESM may not give us a reliable estimate of what a
typical week is like in the person’s life. This subtle difference may be part of the reason
why I did not find strong agreement between the different methods of assessment.

A criticism of the experience sampling measure is that it may not be broad enough
to be compared with SCD, which is based upon a larger collection of experiences. In
other words, comparing self-perceptions based on years of experiences with self-
perceptions during such a small window of time doesn’t tell us much. In order to get a
representative sample of experiences to measure consistency in self-perceptions across
Situations I may need to collect data for months. The issue may be whether the week that
I assessed can be assumed to be typical or normal for the participant. Maybe it was more
stressful for a number of people, or less social for some. However, assuming that each
person was affected by a unique combination of experiences, the “typicality” of the week
may be random across the sample. Barring any major event or set of circumstances that
would affect everyone (e.g., finals week, natural disaster, or political turmoil), the extent
to which a week is normal for everyone Should be normally distributed throughout our
sample.

In order to test the precision of experience sampling, 1 analyzed data from another
study in which participants completed 2 different weeks of ESM, separated by 8 months
(Lucas, Baird, Le, & Eng, 2003). First of all, I found moderate to strong correlations
between the two waves of data for the percentage of time that respondents Spent in each
role. That is, the amount of time that participants reported being in each role during
week 1 was significantly correlated with the amount of time they reported being in the

same roles in week 2 (Friend r_= .65; Student 1' = .44; Romantic Partner 3 = .41; Family

34

Member r = .63; Worker 1' = .50; Stranger r = .60; all p_’S < .01). Trait Extraversion was
measured on a 3-item scale (“affectionate”, “assertive”, and “sociable”). These items
were rated, using a 7-point scale, for how much the participant felt that way at the
moment. The standard deviation of extraversion across moments, irrespective of role,
was moderately correlated across the two weeks (r = .60, p < .01). Similar role items as
were used in Study 2 were used to assign each moment to the appropriate social roles. I
computed the corrected cross-role standard deviation for both waves of data and found
that these measures were not Significantly associated with each other (r = .12, us.) This
suggests that, while the amount of overall situational change in extraversion during one
week is reliable, the amount of change in that trait across social roles is not. It is unclear
whether this result can be generalized to other personality traits, or whether it is limited to
self-perceptions of extraversion. However, based on this small amount of evidence, one
might conclude that subjective feelings of inconsistency wax and wane. Certain periods
of a person’s life, and possibly even certain weeks, may be characterized by greater levels
of consistency in self-perceptions, but these fluctuations should be probabilistically
related to his or her trait level of SCD.
Future Directions

There are numerous other ways to operationalize within-person variability. For
example, it may be interesting to know whether a person who is variable at the level of
the self-concept is also more likely to be perceived as variable by other people.
Informant reports may be one way to get a better picture of how an individual changes

across his or her relationships. According to fiagmentation theory, a person with a highly

differentiated self-concept might also show a greater amount of variance across observer

35

reports of personality given by informants. For example, comparisons could be made
between ratings given by a persons’ parent and those given by his or her boss or best
fiiend. If the contextual self-report measures of SCD are valid, then they should predict
the extent to which these informants agree on their perceptions of the individual. In
addition, if SCD is characterized by difficulties in social relationships, then informant
ratings of relationship quality should be lower for people with higher levels of self-
reported SCD.

The composition and intrapersonal correlates of the self-concept may also be
influenced by social norms or internal and external expectations. There are several ways
that this may be tested. Currently, I am conducting a series of studies in which I examine
the judgmental processes required by contextual self-reports. In particular,
conversational norms against providing redundant information may operate to inflate
variability in self-reports. Also, it may be important to know how role-related
expectations affect self-perceptions within each role. What others expect of us, or what
someone expects of him or her self, may Shape the self-concept. The conflict that Lecky
referred to may be Similar to feelings of discrepancies between an actual and ought self
(Higgins, 1987).

Future research Should focus on potential moderators of the relationship between
consistency and well-being. One previous study examined the role that feelings of
authenticity, or genuineness, play in the relationship between consistency and adjustment
(Sheldon et al., 1997). These authors found that subjective authenticity and SCD were

independent predictors of well-being. Therefore, inconsistent people might still be well

36

adjusted as long as they feel like they are in control of their own actions. Experience
sampling designs may be well-suited for similar studies of subjective authenticity.
Finally, variability in self-perceptions may be a legitimate interpersonal strategy
in which changes in self-expression are in the service of relationship development and
maintenance. Laboratory research using social dilemmas might be one way to test
whether the self-consistency dimension can be conceptualized as an approach to
interpersonal interaction. In these paradigms, choices must be made between 2 or more
decisions, at least one of which would result in a temporary gain for oneself, and another
would result in a temporary loss. Individuals could be paired with a series of “partners”
with whom they would take turns choosing between whether to benefit themselves or the
other person. Based upon the characteristics and behavior of partners, we might expect
that consistency would have an effect on how a person approaches the task. Specifically,
consistent people may exhibit the same strategy in social dilemma tasks regardless of the
characteristics of their interaction partner. On the other hand, variable people might be
more likely to change their approach to the task, as they come into contact with different
interaction partners. In addition, both strategies could result in positive outcomes.

Some people prefer to express themselves in the same way to all people. Other
individuals prefer to express different aspects of themselves, depending upon what they
think iS most appropriate in given situation. In this way, variable people might see this as
an opportunity for them to capitalize on the rewards of being able to relate to a variety of
people. This strategy might become increasingly popular in a world where a greater
variety of people regularly come into close contact with each other. Taking the same

interpersonal approach in every situation might actually become a barrier to the

37

development of healthy and productive social relationships. Therefore, it may be more

important to Shape self-expressions to the goal of harmonious interpersonal interaction.

38

FOOTNOTES

1. By assigning a single moment to multiple roles we may be attenuating the
differences between roles. However, we feel that this may be a more appropriate
assessment of actual within-person change. In particular, traditional SCD measures do
not account for the possibility that a person can enact multiple roles at once and that these
roles may have differential overlap with each other. For example, for most people the
romantic partner and friend roles might overlap more than the family and worker roles.
Other research is currently looking at how asking for judgments for each role separately
might actually inflate differences among roles and prime self-discrepancies.

2. Because the social role items in our ESM scale consisted of 6 dichotomous

(yes/no) items, it is possible for a person to complete the questionnaire and not report
being in any roles at the moment.

39

APPENDICES

4o

APPENDIX A

Table 1

Example of two cases using factor analytic SCD.

 

 

 

 

 

 

Person 1 Person 2
Item Role 1 Role 2 Role 3 _S._d_._ Role 1 Role 2 Role 3 3.9..
Talkative 5 4 4 .58 4 3 3 .58
Nervous 1 1 2 .58 2 2 3 .58
Assertive 4 5 4 .58 3 4 3 .58
Organized 1 1 l .00 2 2 2 .00
Insightful 5 5 4 .58 4 4 3 .58
Hardworking 2 1 1 .58 3 2 2 .58
Caring 4 5 5 .58 3 4 4 .58
Irritable 1 2 1 .58 2 3 2 .58
S.d. 1.81 1.93 1.67 0.83 0.93 0.71
% Variance 93.84 69.80
Explained

 

Note. These two hypothetical cases have identical amounts of cross-situational variance
in their reports. The difference in the percentage of variance explained by the first factor
can be attributed to the difference in the cross-item variance.

41

APPENDIX B

Table 2

Correlations between indicators of consistency and well-being (Study 1).

 

Consistency Measure

 

S D S.D. Corrected S.D.
fi -.44** -.25** .03
Well-Being M .41** .19** .04
Measure
LifeSat -.35** -.21** -.02

 

Note. SCD = factor-analytic measure of self-concept differentiation; SD. = average
cross-role standard deviation for all 20 items; Corrected SD. = the average quadratic
residual from regressing mean levels on standard deviations per item; PA = positive
affect; NA = negative affect; LifeSat = satisfaction with life. *p<.05; **p<.01.

42

APPENDD( C

Table 3

Correlations between the Big Five personality factors and indicators of consistency

 

 

 

(Study 1).
Consistency Personalig Trait
Measure
Extraversion Agreeableness Conscientiousness Neuroticism Openness
S D -.22** -.l9** -.27** .35" -.01
SD. -.11 -.07 -.26** .16** -.05
Corrected SD. .08 .12 -.13* .06 .04

 

 

Note. *p<.05, **p<.01.

43

APPENDIX D

Table 4

Correlations between the Big Five personality factors and indicators of well-being (Study

 

 

 

 

1).
Well-Being Personality Trafi
Measure
Extraversion Agreeableness Conscientiousness Neuroticim Qpenness
BA .40** .39" .07 -.30** .16"
NA -.33** -.12* -.05 .71** .03
LifeSat .27** .1 7* .14* -.39** .05

 

Note. *p<.05, **p<.01.

APPENDIX E

Table 5

Correlation matrix of indicators of consistency (Study 1).

 

Consistency Measure

 

L12 SJ; Corrected S.D.
SC_D 1.00
Q .79** 1.00
Corrected SD. .45** .85" 1.00

 

 

Note. SCD = factor-analytic measure of self-concept differentiation; SD. = average
cross-role standard deviation for all 20 items; Corrected SD. = the average quadratic
residual from regressing mean levels on standard deviations per item. *p<.05; **p<.01.

45

APPENDIX F

Table 6

Correlation matrix of indicators of consistency (Study 2).

 

Consistency Meme

 

1 2 3 4 5 6 7
1. SCD 1.00
2. Cross-role SD .79** 1.00
3. Corrected SD .42** .82** 1.00
4. Overall ESM SD .20M .35** .35** 1.00

5. Corrected Overall ESM SD-.10 .17 .39“ .81** 1.00
6. Cross-role ESM SD -.04 .14 .24* .40** .47" 1.00

7. Corrected ESM SD -.13 .06 .24* .30** .49“ .75“ 1.00

 

Note. SCD = factor-analytic measure of self-concept differentiation; Cross-role SD =
average cross-role standard deviation for all 20 items; Corrected SD = the average
quadratic residual from regressing mean levels on standard deviations per item; Overall
ESM SD = the total amount of variance in responses over the week; Corrected Overall
ESM SD = the average quadratic residual from the total variance over the week; Cross-
role ESM SD = cross-role standard deviation from the moment data; Corrected ESM SD
= the average quadratic residual from moment data across roles;. *p<.05; **p<.01.

46

APPENDIX G

Table 7

Correlations between indicators of consistency and well-being (Study 2).

 

 

 

Consistency Measure P_A N_A Q ESMPA ESMNA
1. SCD -.28** .31** -.30** -.24* .49“
2. Cross-role SD -.l9 .19* -.22* -.23* .32**
3. Corrected SD .01 .09 -.09 -.09 .11

4. Overall ESM SD .12 .00 .01 -.02 .17

5. Corrected Overall ESM SD .27** -.10 .11 .08 -.18

6. Cross-role ESM SD .07 -.06 .05 -.11 -.09

7. Corrected ESM SD .14 -.10 .11 -.06 -.15

 

Note. SCD = factor-analytic measure of self-concept differentiation; Cross-role SD =
average cross-role standard deviation for all 20 items; Corrected SD = the average
quadratic residual from regressing mean levels on standard deviations per item; Overall
ESM SD = the total amount of variance in responses over the week; Corrected Overall
ESM SD = the average quadratic residual from the total variance over the week; Cross-
role ESM SD = cross-role standard deviation from the moment data; Corrected ESM SD
= the average quadratic residual across roles from the moment data; PA = positive affect;
NA = negative affect; LS = satisfaction with life; ESMPA = average positive affect
during the week; ESMNA = average negative affect during the week. *p<.05; **p<.01.

47

APPENDIX H

Table 8

Correlations between the Big Five personality factors and indicators of consistency

 

 

 

 

 

 

(Study 2).
Consistency Personglity Tray

Measure

Extgversion Agreeableness Conscientiousness Neuroticism Openness
SCD -.42** -.26** -.18 .36" -.23*
SD. -.38** -.19 -.02 .34** -.21*
Corrected SD. -.20* .02 .15 .21 -.15
S.D.Mom -.09 .08 .18 -.03 -.17
Corrected SD. -.14 .12 .27" -.14 -.09
Mom

 

Note. SCD = factor-analytic measure of self-concept differentiation; SD. = average
cross-role standard deviation for all 20 items; Corrected SD. = the average quadratic
residual fi'om regressing mean levels on standard deviations per item; S.D.Mom = cross-
role standard deviation from moment data; Corrected S.D. Mom = the average quadratic
residual from moment data. *p<.05, **p<.01.

48

APPENDIX I

Figure 1

Relationship between repeated-measure mean and standard deviation.

 

 

2.0

1.5‘

1.0!

Standard deviation

0.0 r

 

 

 

 

‘-

1 5 3

I
.s
OI
bl
0’

Average across roles

 

Note. The item represented is “Trustful”. The dashed line indicates the linear
relationship, and the solid line represents the quadratic relationship between these
variables. Data points that are occupied by more than one case are indicated by multiple
dashes.

49

APPENDIX I

Figure 2

Distribution of SCD scores for study I.

 

 

 

    

5.0 15.0 25.0 35.0 45.0 55.0 85.0 75.0 85.0 95.0

SCD

 

 

Note. N = 266, Mean = 33.34, SD = 16.66.

50

APPENDIX K

Figure 3
Distribution of SCD scores for study 2.

 

30

20. ' ——-—

 

 

 

 

 

 

 

 

 

5.0 15.0 25.0 35.0 45.0 75.0 85.0 95.0

SCD

 

Note. N = 103, Mean = 32.07, SD = 14.89.

51

APPENDIX L

For the following adjectives, please indicate how well each describes you at the moment
before the alarm sounded.

M

4 — Describes me very well

3 — Describes me well

2 — Describes me moderately

1 — Describes me a little

0 — Does not describe me

1. Talkative 11. Assertive
2. Cooperative 12. Rude
3. Organized 13. Hardworking
4. Irritable 14. Optimistic
5. Intelligent 15. Inquisitive
6. Energetic 16. Adventurous
7. Trustful 17. Warm
8. Dependable 18. Responsible
9. Insecure 19. Vulnerable
10. Philosophical 20. Creative

SCALE:

4 — Very strongly

3 — Strongly

2 - Moderately

1— Slightly

0 — Not at all

22. How PLEASANT are you feeling?
23. How UNPLEASANT are you feeling?
24. How SAD are you feeling?
25. How HAPPY are you feeling?
26. Were you engaged in a recreation activity before the alarm sounded? NolYes
27. Please describe the situation you were in before the alarm sounded. Was the
situation: 1 = Alone
2 = Semi-social (with others but not interacting with them)
3 = Social (interacting with others)
28. Were you being a student? YES|NO
29. Were you being a fiiend? YES|NO
30. Were you being a romantic partner? YESlNO
31. Were you being a family member? YES|NO
32. Were you being a worker? YESINO
33. Were you being stranger? YES INC

52

REFERENCES

53

REFERENCES

Block, J. (1961). Ego identity, role variability, and adjustment. Journal of Consulting
Psychology, 25, 392-397.

Bolger, N., Davis, A., & Rafaeli, E. (2003). Diary methods: Capturing life as it is lived.
Annual Review of Psychology, 54, 579-616.

Campbell, J .D., Assanand, S., & DiPaula, A. (2003). The structure of the self-concept
and its relation to psychological adjustment. Journal of Personality, 71, 115-140.

Diehl, M., Hastings, C.T., & Stanton, J .M. (2001). Self-concept differentiation across the
adult life Span. Psychology and Aging, 16, 643-654.

Diener, E., Emmons, R.A., Larsen, R.J., & Griffen, S. (1985). The satisfaction with life
scale. Journal of Personality Assessment, 49, 71-75.

Diener, E., Smith, H., & Fujita, F. (1995). The personality structure of affect. Journal of
Personality and Social Psychology, 69, 130-141.

Donahue, E.M., Robins, R.W., Roberts, B.W., & John, OR (1993). The divided self:
Concurrent and longitudinal effects of psychological adjustment and social roles
on self-concept differentiation. Journal of Personality and Social Psychology, 64,
834-846.

Eysenck, H.J. (1981). Introduction and general features of the model. In H.J. Eysenck
(Ed.) A Model for Personality (pp. 1— 37). New York: Springer-Verlag.

F leeson, W. (2001). Toward a structure- and process-integrated view of personality:
Traits as density distributions of states. Journal of Personality and Social
Psychology, 80, 101 1-1027.

Gray, J .A. (1981). A critique of Eysenck’s theory of personality. In H.J. Eysenck (Rd).
A Model for Personality (pp.246-276). New York: Springer-Verlag.

Higgins, ET. (1987). Self-discrepancy: A theory relating self and affect. Psychological
Review, 94, 319-340.

Lecky, P. (1945). Self-consistency: A theory of personality. New York: Anchor Books.

Locke, K.D. (2003). H as a measure of complexity of social information processing.
Personality and Social Psychology Review, 7, 268-280.

54

Lucas, R.E., & Baird (2003). Global self-assessment. Manuscript under review.

Lucas, R.E., Baird, B.M., Le, K., & Eng (2003). Extraversion and emotional reactivity.
Unpublished data.

Lucas, R.E., & Diener, E. (2000). Personality and subjective well-being across the life
Span. In V.J. Molfese & D.L. Molfese (Eds.). Temperament and personality
development across the life span (pp.211-234). Mahwah, NJ: Lawrence Erlbaum
Associates.

McCrae, R.R., & Costa, RT. (1997). Personality trait structure as a human universal.
American Psychologist, 52, 509-516.

McReynolds, P., Altrocchi, J ., & House, C. (2000). Self-pluralism: Assessment and
relations to adjustment, life changes, and age. Journal of Personality, 68, 347-
381.

Paulhus, D.L. & Martin, CL. (1988). Functional flexibility: A new conception of

interpersonal flexibility. Journal of Personality and Social Psychology, 55, 88-
1 01 .

Schwarz, N., Groves, R.M., & Schuman, H. (1998). Survey methods. In D.T. Glibert &
S.T. FiSke (Eds.), The handbook of social psychology: Vol. I (pp.143-l79). New
York: McGraw-Hill.

Sheldon, K.M., Ryan, R.M., Rawsthome, L.J., & Ilardi, B. (1997). Trait self and true
self: Cross role variation in the Big Five personality traits and its relation with

psychological authenticity and subjective well-being. Journal of Personality and
Social Psychology, 73, 1380-1393.

Snyder, M. (1974). Self-monitoring of expressive behavior. Journal of Personality and
Social Psychology, 30, 526-537.

Snyder, M. (1979). Self-monitoring processes. In L. Berkowitz (Ed.), Advances in
experimental and social psychology (V 01. 12, pp. 86-128). New York: Academic
Press.

Stone, A.A., Shiffrnan, S.S, & DeVries, M.W. (1999). Ecological momentary
assessment. Well-Being.

Suh, EM. (2002). Culture, identity consistency, and subjective well-being. Journal of
Personality and Social Psychology, 113, 151-159.

55