Momentary Assessment of Interpersonal Process in Psychotherapy
Katherine M. Thomas and Christopher J. Hopwood Michigan State University
Erik Woody and Nicole Ethier University of Waterloo
Pamela Sadler Wilfrid Laurier University
To demonstrate how a novel computer joystick coding method can illuminate the study of interpersonal processes in psychotherapy sessions, we applied it to Shostrom’s (1966) well-known films in which a client, Gloria, had sessions with 3 prominent psychotherapists. The joystick method, which records interpersonal behavior as nearly continuous flows on the plane defined by the interpersonal dimensions of control and affiliation, provides an excellent sampling of variability in each person’s interpersonal behavior across the session. More important, it yields extensive information about the temporal dynamics that interrelate clients’ and therapists’ behaviors. Gloria’s 3 psychotherapy sessions were characterized using time-series statistical indices and graphical representations. Results demonstrated that patterns of within-person variability tended to be markedly asymmetric, with a predominant, set-point-like inter- personal style from which deviations mostly occurred in just 1 direction (e.g., occasional submissive departures from a modal dominant style). In addition, across each session, the therapist and client showed strongly cyclical variations in both control and affiliation, and these oscillations were entrained to different extents depending on the therapist. We interpreted different patterns of moment-to-moment complementarity of interpersonal behavior in terms of different therapeutic goals, such as fostering a positive alliance versus disconfirming the client’s interpersonal expectations. We also showed how this method can be used to provide a more detailed analysis of specific shorter segments from each of the sessions. Finally, we compared our approach to alternative techniques, such as act-to-act lagged relations and dynamic systems and pointed to a variety of possible research and training applications.
Keywords: psychotherapy, process, momentary assessment, spectral analysis, interpersonal circumplex
The purpose of this article is to demonstrate how a novel method for the study of moment-to-moment interpersonal processes can be applied to psychotherapy sessions and to illustrate how this method could enhance understanding of psychotherapy process. To depict the value of this method, we apply it to Shostrom’s (1966) well-known films in which a client, Gloria, met with three prominent psychotherapists with differing theoretical orienta- tions—Albert Ellis (rational– emotive), Frederick Perls (gestalt), and Carl Rogers (client-centered). These filmed therapy sessions are useful for our purpose because they are widely familiar (e.g., Reilly & Jacobus, 2008; Weinrach, 1990) and because we can contrast our novel approach with previous research applying a
more conventional measurement approach to these sessions (Kies- ler & Goldston, 1988).
Assessing Dynamic Aspects of the Therapeutic Relationship
It is virtually a truism that the interpersonal relationship in therapy has a profound impact on therapy outcomes (e.g., Gold- fried, in press; Horvath, Del Re, Flückiger, & Symonds, 2011). The relationship provides the context in which interventions can be successfully implemented, and it may be particularly relevant when interpersonal difficulties are an important aspect of the client’s problems (Anchin & Pincus, 2010). Not only is a positive relationship associated with successful outcomes (Muran & Bar- ber, 2010) but, in addition, strains in the relationship are associated with therapeutic failure (Castonguay, Goldfried, Wiser, Raue, & Hayes, 1996; Henry, Schacht, & Strupp, 1986, 1990). Hence, studying the dynamic aspects of the therapeutic relationship— how it develops, varies, and changes—is important for understanding effective therapy.
However, variation, pattern, and change in interpersonal behavior during an ongoing exchange are subtle and difficult to measure. One previously employed approach has been to segment the stream of behavior into discrete acts and then to examine how each kind of act by one person is related to each subsequent kind of act by the other person. This act-to-act approach has been used successfully to study interpersonal processes in therapy and relate them to therapy out-
This article was published Online First September 2, 2013. Katherine M. Thomas and Christopher J. Hopwood, Department of
Psychology, Michigan State University; Erik Woody and Nicole Ethier, Department of Psychology, University of Waterloo, Waterloo, Ontario, Canada; Pamela Sadler, Department of Psychology, Wilfrid Laurier Uni- versity, Waterloo, Ontario, Canada.
This research was supported by Operating Grant SRG 410-2009-2164 from the Social Sciences and Humanities Research Council of Canada to Pamela Sadler and Erik Woody.
Correspondence concerning this article should be addressed to Katherine M. Thomas, Department of Psychology, Michigan State University, East Lansing, MI 48824. E-mail: [email protected]
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
Journal of Counseling Psychology © 2013 American Psychological Association 2014, Vol. 61, No. 1, 1–14 0022-0167/14/$12.00 DOI: 10.1037/a0034277
1
comes (e.g., Dietzel & Abeles, 1975; Lichtenberg & Heck, 1986; Tracey, 1985; Wampold & Kim, 1989).
The presently proposed method addresses the dynamic aspects of the therapeutic relationship in a different way by capturing ongoing dynamics as a reasonably continuous flow, rather than as a sequence of discrete acts. To some extent, the new method simply imposes a different frame of reference, yielding its own unique insights. Another advantage is that compared with the act-to-act approach, the method described in the present study is more time effective and thus would be more useful in practical circumstances, such as psychotherapy training and supervision (see Pincus et al., in press).
A Theoretical Framework for Assessing Moment-to-Moment Interpersonal Behavior
To effectively measure interpersonal process, a well-validated theoretical and measurement framework is needed. Evidence across several domains of inquiry converges to suggest that two fundamental dimensions, control (dominance to submission) and affiliation (warmth to coldness), account for variability in rela- tional functioning and behavior (Luyten & Blatt, 2013; Wiggins, 1991). These two dimensions can be operationalized using the interpersonal circumplex (IPC; Leary, 1957; Wiggins, 1996; Fig- ure 1), which offers a measurement model for conceptualizing clinically salient features of personality, psychopathology, and social processes (Pincus, Lukowitsky, & Wright, 2010). An ad- vantage of the IPC is that it reflects basic social processes and therefore can be meaningfully applied across theoretical orienta- tions. Indeed, the interpersonal model in general and the IPC in particular have been fruitfully applied to a variety of therapies, including cognitive (Safran, 1984, 1990a, 1990b), cognitive be- havioral (Hayes, 2004), interpersonal (Anchin & Pincus, 2010; Benjamin, 1996), gestalt (Benjamin, 1979), and psychodynamic (Gurtman, 1996; Horowitz, Rosenberg, & Bartholomew, 1993; Strupp & Binder, 1984). For instance, research applying the IPC to psychotherapy has found that patients respond to hostile therapists with self-blame (Henry et al., 1990) and that warmer patients
improve more quickly than colder patients in psychodynamic but not in cognitive behavioral therapy (Puschner, Kraft, & Bauer, 2004).
The IPC also provides a framework for making testable predic- tions about dyadic behavior as it unfolds over time. Empirical and theoretical literature suggests that interactions are most harmoni- ous (i.e., least anxiety provoking and most stable) when individ- uals in a dyad behave in a manner that is similar with respect to affiliation but opposite with respect to control—a pattern referred to as complementarity (Kiesler, 1996; Sadler & Woody, 2003; Sadler, Ethier, Gunn, Duong, & Woody, 2009; Tracey, 2004). Based on this principle, the behaviors of one individual are pre- dicted to invite particular behaviors from the other individual in dyadic interactions (Kiesler, 1996; Leary, 1957). In brief, warmth invites warmth, whereas dominance invites submission.
The principle of complementarity has been used to develop elegant models explaining the persistence of maladaptive interper- sonal behavior and the nature of psychotherapeutic interventions to change such behavior (e.g., Anchin & Pincus, 2010; Andrews, 1989; Carson, 1982; Kiesler, 1996). Work by Tracey (1993; Tracey, Sherry, & Albright, 1999) suggests that alliance-building complementarity early in psychotherapy, followed by change- promoting noncomplementarity once an alliance has been estab- lished, is associated with positive therapeutic outcomes across varied theoretical approaches. Thus, studying interpersonal com- plementarity may provide an important window into client– therapist relationship patterns that play an important role in treat- ment.
A Computer Joystick Method for Coding Momentary Interpersonal Behavior
Sadler and colleagues recently developed a novel joystick method for assessing momentary interpersonal processes in dyadic interactions (Lizdek, Sadler, Woody, Ethier, & Malet, 2012; Sadler et al., 2009). As an observer uses a computer joystick to make observational ratings of recorded interactions, data on interper- sonal communications are captured twice per second and yield time series for each individual’s level of control and level of affiliation throughout an interaction. Data obtained using this method have revealed novel phenomena that occur in interactions, such as cyclical patterns of complementarity (Sadler et al., 2009). Additional research using the joystick method found that female peer dyads with greater complementarity on the warmth dimension liked one another more and performed lab tasks more accurately (Markey, Lowmaster, & Eichler, 2010) and that parallel processes occur between therapy and supervision (Tracey, Bludworth, & Glidden-Tracey, 2012). Each of these studies showed considerable variability in the degree of complementarity observed across dy- ads, indicating that the joystick method is sensitive to dyadic and individual differences that affect interpersonal processes.
The Present Study
Kiesler and Goldston (1988) applied the IPC and the principle of complementarity to Gloria’s sessions with Ellis, Perls, and Rogers by having raters complete the Checklist of Psychotherapy Trans- actions (CLOPT; Kiesler, Goldston, & Schmidt, 1991). This in- strument is a 96-item checklist of interpersonal behaviors that the
C on
tr ol
Affiliation Warm Cold
Dominant
Submissive
Figure 1. The interpersonal circumplex (IPC).
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2 THOMAS, HOPWOOD, WOODY, ETHIER, AND SADLER
rater completes, once for the therapist and again for the client, after having watched a therapy session. Kiesler and Goldston found that in terms of aggregate measures of behavior, Gloria displayed the highest degree of complementarity with Ellis, followed by Rogers, and the least with Perls. Although useful, this approach does not provide any information about the temporal dynamics that un- folded in each session; indeed, it is even insensitive to how long and how often any behavior occurred (each behavior is simply marked as present or absent during a session). Kiesler (1996, p. 91) drew attention to the importance of techniques that might reveal “patterned redundancies occurring over time,” rather than simply a static snapshot of the partners’ overall interpersonal styles.
Accordingly, in the present study, we use the computer joystick method to apply the IPC and the principle of complementarity to the Gloria sessions. There are two main novel implications of this approach.
1. The method provides an excellent sampling of within-person variability in interpersonal behavior for each person in the inter- action. Thus, we asked the following research questions: What patterns of variability for each partner are evident in these psy- chotherapy sessions? How might these patterns of variability illu- minate the nature of the interaction?
2. The method provides a great deal of information about how the streams of behavior by the therapist and client are interrelated. Hence, we asked the following research questions: Do the partners show shifts in their overall levels of control and affiliation, and are these shifts consistent with the principle of complementarity (e.g., linear slopes with diverging levels of control)? Do partners show cyclical or oscillating variations in control and affiliation, and to what extent are these oscillations synchronized and entrained? Finally, what might differing degrees of interpersonal entrainment tell us about the nature of the therapeutic relationship in these sessions?
Method
Procedure
To examine momentary interpersonal behavior throughout Glo- ria’s sessions, raters recorded their impressions of the continuous stream of interpersonal behavior by watching a session, focusing their attention on either Gloria or the therapist, and using a com- puter joystick apparatus to indicate the target person’s momentary standing on the IPC. Subsequently, raters watched the session again and made similar ratings of the other person in the session. The order of these assessments was arranged such that Gloria was never consecutively rated from two different sessions, nor was the same session ever consecutively rated. The joystick was scaled from �1,000 (submissiveness; coldness) to 1,000 (dominance; warmth), and the computer recorded the rater’s joystick placement along both axes twice per second.
Seven undergraduate students underwent careful individual training on the joystick method prior to rating Gloria’s sessions. We used the training protocol outlined by Sadler et al. (2009) to introduce raters to the joystick method. Raters were instructed to make behaviorally anchored ratings by moving the joystick in accord with any of the target person’s statements, nonverbal be- haviors, fluctuations in tone, and so forth, that constituted an increase or decrease in control or affiliation. Thus, raters moved
the joystick in a reasonably continuous way to represent their perceptions of changes in interpersonal behavior. Raters were informed that the joystick position should also represent any times in which the absence of a behavior signified or sustained a mean- ingful interpersonal action (e.g., if an individual remained silent after being asked a question). When no discernible changes in interpersonal behavior were displayed, raters maintained their joy- stick position until the person made a meaningful interpersonal gesture. However, slight gestures, such as eye contact, engage- ment, tone, and so forth, were coded, and thus the joystick was frequently in motion, capturing these behavioral variations. Raters were not told about the concept of complementarity.
As part of their training, raters used the joystick to code the interpersonal behavior in another set of therapy dyads, Shostrom’s (1976) Three Approaches to Psychotherapy, with a client named Kathy. This resulted in six trial assessments of a format identical to the Gloria films. Prior to coding Gloria’s sessions, each rater was required to demonstrate good consistency of his or her ratings with those of previously trained raters (authors Thomas and Hop- wood). All raters consistently demonstrated cross-correlations above .50 with trained raters on the control and affiliation dimen- sions for both individuals in each of the training videos. Sadler et al. (2009) showed that this level of cross-correlation is sufficient to obtain very good reliability of the moment-to-moment ratings, once they are aggregated across the raters.
Once trained, raters coded all three therapists and Gloria with each therapist (i.e., six total coding sessions). At this juncture, further checks were performed on the quality of each rater’s data. Specifically, 2 weeks after initially coding Gloria’s sessions, each rater watched and recoded two individuals (always Gloria from one session and a therapist from a different session). Cross- correlations between initial and follow-up joystick ratings were computed for both axes to assess self-consistency for each rater. Because of relatively low self-consistency (cross-correlations � .50), one rater’s data were discarded from further consideration. In addition, the consistency of each rater’s data with the group average omitting that rater’s data were assessed. All six remaining raters achieved cross-correlations � .50 (M � .55) with the group average across at least 10 of the 12 variable sets (i.e., control and affiliation for each therapist and Gloria with each therapist).
Final Joystick Data
The first 10 data points for each interactant were deleted to allow raters 5 s to orient themselves to the interaction (as in Sadler et al., 2009). Joystick data were then averaged across raters at each time point to obtain the final time series data for each interactant across both IPC dimensions. All subsequent analyses were con- ducted using these data (aggregated across the six raters). These half-second ratings for affiliation and control across the three dyads yielded 12 total bivariate time series. Data collected for each dyad differed based on the amount of time each therapist spent with Gloria. We collected 2,185 data points for Ellis’s session with Gloria (18 min, 12 s); 2,822 data points for Perls’s session (23 min, 31 s); and 3,811 data points for Rogers’s session (31 min, 45 s).
The reliability of the aggregated time series was assessed using an approach that compares the true score (i.e., shared) variance to the total variance for each time series, as described in Sadler et al. (2009). Specifically, the true score variance was estimated as the
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
3MOMENTARY ASSESSMENT OF PROCESS
mean of the cross covariances of the individual raters’ times series, and the total variance was estimated as the variance of the aggre- gated time series. This approach yielded reliabilities of .80 for control and .66 for affiliation, comparable to values obtained in other published work using the joystick method (Markey et al., 2010; Sadler et al., 2009).
In addition to using these data to characterize interpersonal processes over time, we were interested in the global ratings obtained by calculating the mean of each time series (control or affiliation) for each rater and each interactant (i.e., Gloria with Ellis, Ellis, etc.). Past research has demonstrated that these global ratings have strong reliability (Markey et al., 2010; Sadler et al., 2009). The present data are limited for assessing such reliability because of the small number of cases (six targets); however, it is reassuring that Cronbach’s alpha, calculated by treating raters as items, yielded values of .80 (affiliation) and .95 (control).
Calculation of Indices
In addition to the global levels of control and affiliation, calcu- lated as the means across each person’s entire aggregated time series, we derived a variety of other indices, the calculations of which are outlined below.
Indices of within-person variability. For each person in a session, we calculated the standard deviation across the entire time series for control and for affiliation. We also computed the corre- lation between each person’s control and his or her affiliation across the entire time series. These indices provide quantitative information regarding the nature of a person’s variation in inter- personal behavior across a session.
Density plots. As another way to characterize each person’s pattern of interpersonal variability across a session, we used the procedure smoothScatter (R Development Core Team, 2011) in the statistical software package R to derive a bivariate density plot on the interpersonal plane defined by the affiliation and control axes. The procedure parameters used were the following: nbin � 500, bandwidth � 70, transformation � function(x) x ˆ.8. The densest parts of the distribution are colored black, and the less dense parts successively lighter shades of gray. A major advantage of this approach is that it preserves the actual shape of the density distri- bution, which is particularly important if the distribution is not bivariate normal.
Linear trends in levels. For each person in a session, we used ordinary least squares regression to predict the individual’s moment-to-moment interpersonal scores (control or affiliation) using time as the predictor variable. Each regression yielded an intercept, indexing the estimated value at the beginning of the session, and a slope, indexing the rate of linear change over the course of the session. We also calculated the R2, which indicates the proportion of variance explained by the linear trend. The residuals from these regression analyses also provided the data used for spectral and cross-spectral analyses (in which linear trends could otherwise serve as a confound; Warner, 1998).
Indices of oscillation and entrainment. To derive indices of cyclical processes and entrainment, we conducted spectral and cross-spectral analyses on the detrended data for each session following the procedures detailed in Sadler et al. (2009). The results of these analyses were summarized using three different types of index: rhythmicity, average weighted coherence, and
average weighted phase. Rhythmicity was computed as the propor- tion of variance in a time series that is accounted for by frequen- cies with periods longer than 30 s (the rationale being that, at least in social interactions, frequencies higher than this are likely to represent noise). This range of frequencies was also used in the calculation of the coherence and phase statistics. Rhythmicity values indicate the extent to which variations in control or affili- ation are explained by cyclical patterns.
The average weighted coherence was computed by weighting the coherence value at each frequency band in the cross-spectral analysis by the amounts of variance at the same frequency band in the univariate spectral analyses (Sadler et al., 2009; Warner, 1998). The resulting value is a nondirectional index of the proportion of variance in one time series that can be predicted by the other time series, thereby indicating the attunement of cycles across members of a dyad. Coherence ranges from 0 to 1, with higher values indicating greater entrainment. The average weighted phase was computed by weighting the phase values at each frequency band in the cross-spectral analysis in the same way as described for the coherence. Phase values indicate proportions of a full cycle and range from �.5, through 0, to .5. (Because phase is a circular statistic, the values of �.5 and .5 are logically indistinguishable, both falling half a cycle away from zero.) A phase value of zero indicates that the partners’ behaviors are exactly in phase, with peaks and troughs coinciding exactly. A phase value of .5 or �.5 indicates that the partners’ behaviors are completely out of phase, with peaks for one person coinciding with troughs for the other. Intermediate values can be interpreted as one individual’s variation leading the other person’s variation, as described later in the Results section.
As a final index of entrainment that is not a component of the spectral and cross-spectral analyses, we calculated the cross- correlation of the time series for the two interacting partners for control and for affiliation. This intuitively accessible, directional value indicates how strongly correlated the two partners’ behaviors were throughout the interaction.
Results
Global Levels of Control and Affiliation
The overall means of control and affiliation for Gloria and the corresponding therapist are presented in Table 1. From these means, it is clear that not only did the three therapists have very different interpersonal styles but also that Gloria’s interpersonal style was strongly affected by the therapist with whom she was interacting. The configuration of means is readily appreciated in Figure 2, where a white plus sign denotes each overall interper- sonal style (the centroid, which is the intersection of the person’s control mean and affiliation mean). Among the therapists, Ellis and Perls had dominant styles, whereas Rogers had a submissive style; Rogers had the warmest style and Perls the coldest. Gloria’s overall interpersonal styles show striking complementarity with Ellis and with Rogers. To Ellis’s warm– dominant style, she tended to respond with a warm–submissive style, whereas to Rogers’s warm–submissive style, she responded with a warm– dominant style. In contrast, Gloria’s response to Perls’s cold– dominant style shows the deviation from classical complementarity noted by Orford (1986) and others; overall, she responded with a similarly
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
4 THOMAS, HOPWOOD, WOODY, ETHIER, AND SADLER
cold– dominant style. These findings for overall style are quite similar to the findings of Kiesler and Goldston (1988), even though our method differs considerably from theirs.
Within-Person Variability
Indices of variability of interpersonal behavior for each person are also provided in Table 1. The standard deviations, which index how much each person varied on each interpersonal dimension, show some very large differences in the amounts of variability. For example, Ellis showed almost three times as much variability in control as Rogers, and Perls showed almost three times as much variability in affiliation as Rogers. The amount of variability in Gloria’s interpersonal behavior was generally quite large; for ex- ample, her control behavior in the interaction with Ellis yielded the largest standard deviation in this data set.
Another potentially useful index of the nature of within-person variability is the correlation of control and affiliation (Table 1). The most striking finding is the strong negative correlation be- tween Gloria’s levels of control and affiliation in her interaction with Perls. This finding indicates that as she became more domi- nant with Perls, she also strongly tended to become colder. In contrast, her tendency to be affiliative at times when she was dominant was minimally, but positively, correlated in her sessions with Ellis and Rogers.
These standard deviations and correlations provide valuable quantitative indices of variability; however, they may be somewhat limited in how fully they convey underlying patterns of variability. This is because the actual patterns of variability do not necessarily follow the assumptions of a bivariate normal distribution (e.g., symmetry of the data points around the intersection of the means). Figure 2 shows the density distributions in a manner that preserves their actual shapes. For each person’s interpersonal behavior, the darkest area is, in effect, a bivariate mode, showing what may be regarded as the person’s interpersonal set point in the interaction. The gray areas show the patterns of deviation from this interper- sonal set point.
Note that the actual patterns of variability are often quite asym- metric. Consider, for example, the density distribution for Ellis. His predominant style was strongly dominant, depicted at the top like the head of a comet; however, he tended to diverge strongly from this predominant pattern, switching periodically to a far more submissive style, shown as the tail of the comet. Note that this pattern is not at all bivariate normal. The deviations from the predominant style are mostly in just one direction (downward);
indeed, these asymmetric deviations pull the centroid (denoted by the white plus sign) well below Ellis’s modal style. In response to Ellis, Gloria showed a predominantly warm–submissive style, but the deviations from this interpersonal set point are very extensive, reaching far up into dominant behavior and even straying occa- sionally toward greater warmth. As for Ellis, these asymmetric deviations (upward and to the right) pull Gloria’s centroid outside what is actually her modal interpersonal style in the interaction.
Unlike Ellis, Perls’s pattern of variability around his modal interpersonal style (cold– dominant) is reasonably symmetric. In response to Perls, Gloria’s pattern of variability is very distinctive. Her predominant response is near the origin (neutral in both control and affiliation), but her deviations from this set point extend diagonally very far to the upper left of the IPC (hostile– dominant). The shape of her density distribution is consistent with the strongly negative correlation found between her control and affiliation (in Table 1).
Finally, Rogers and Gloria both show a narrow range of varia- tion on affiliation, but striking variability on control. Note that Rogers’s deviations from his warm–submissive set point are mostly upward, toward greater dominance, whereas Gloria’s de- viations from her warm– dominant set point are mostly downward, toward greater submissiveness. As a result, their density distribu- tions overlap considerably.
Temporal Dynamics That Interrelate the Partners’ Behaviors in the Interaction
Although the foregoing patterns of each partner’s within-person variability are quite interesting, they cannot show important tem- poral dynamic aspects of the interaction that crucially interrelate the partners’ behaviors. To illustrate, consider again the density plots for Ellis and Gloria with Ellis (in Figure 2). We wanted to know whether the occasions during which Ellis’s interpersonal style veers toward submissiveness are associated with the occa- sions during which Gloria’s interpersonal style veers toward greater dominance. Because the density plots collapse across time, they cannot provide us with this kind of information.
One way in which partners’ behaviors may be associated across the time course of the interaction concerns linear trends in each person’s level over time. Information about these linear shifts is provided in the linear regressions portion of Table 2. For example, consider control in the interaction between Ellis and Gloria. For Ellis, the intercept tells us that he began the interaction being somewhat dominant, and the slope tells us that he increased in
Table 1 Means, Standard Deviations, and Within-Person Correlations
Variable
Control Affiliation Correlation of control
and affiliationM SD M SD
Gloria’s behavior Gloria w/Ellis �116.17 283.59 105.36 108.50 .14 Gloria w/Perls 180.78 216.65 �127.56 232.72 �.56 Gloria w/Rogers 123.03 244.38 243.45 80.05 .12
Therapist’s behavior Ellis 471.48 259.56 110.28 58.48 �.32 Perls 295.84 169.94 �11.23 113.56 �.23 Rogers �99.01 89.91 249.75 39.04 .11
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
5MOMENTARY ASSESSMENT OF PROCESS
dominance rather steeply over the course of the interaction. This linear trend explained 24% of his variance in control. For Gloria with Ellis, her intercept tells us that she began the interaction being slightly dominant, and her slope tells us that she decreased in dominance quite steeply over the course of the interaction. This linear trend explained 16% of her variance in control. Thus, consistent with reciprocity of overall shifts in control, Ellis and
Gloria moved apart in control over the course of the interaction, with Ellis becoming more dominant and Gloria more submissive. For Gloria’s sessions with Perls and with Rogers, such linear trends were much less substantial, as indicated by the small R2
values. Another way in which partners’ behaviors may be associated
across the time course of the interaction is probably more impor-
Figure 2. Density plots (centroids, or means on both dimensions, are shown with a white plus sign).
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
6 THOMAS, HOPWOOD, WOODY, ETHIER, AND SADLER
tant for understanding interaction dynamics. As discussed previ- ously, to the extent that the partners show cyclical or oscillating patterns of interpersonal behavior, these oscillations may become synchronized and entrained between the partners. Information about these phenomena is provided in the spectral analysis portion of Table 2. For example, consider again control in the interaction between Ellis and Gloria. The rhythmicity values for Gloria with Ellis and for Ellis tell us that both partners’ behavior in the interaction strongly tended to fall in reasonably regular cycles (specifically, once linear trends are removed, about 90% of the variance in Gloria’s control behavior and about 87% of the vari- ance in Ellis’s control behavior are attributable to cyclical trends). Thus, there is certainly the potential that these cyclical varia- tions could have become entrained between Gloria and Ellis. The value for average weighted coherence indexes the extent of such entrainment. This value is akin to a squared correlation and takes on values between 0 and 1. The obtained value of .75 indicates a high degree of entrainment of cyclical variations in levels of control between Gloria and Ellis. Note that the other average weighted coherence values for control indicate that Gloria and Rogers were similarly highly entrained, whereas there was much more modest entrainment between Gloria and Perls.
The average weighted coherence as an index of entrainment emerges from relatively complex statistical machinations on the data; however, the intuitive meaning of what these values capture is readily conveyed. The top panel of Figure 3 shows Gloria’s and Ellis’s moment-to-moment levels of control over the first 10 min of their interaction. (Showing the entire inter- action makes the time scale too compressed to see patterns clearly.) First, note that as the rhythmicity values indicated, the moment-to-moment variations for both people are reasonably cyclical in nature; indeed, they tend to have a relatively con- sistent period of roughly a minute. Second, note that as the coherence value indicated, these cycles are strongly related, with peaks for one person tending to occur together with troughs for the other person. In contrast, the top panel of Figure
4, depicting moment-to-moment levels of control for Perls and Gloria, shows a less consistently entrained pattern of variation, which is consistent with the lower coherence value obtained from their interaction data. Akin to Figure 3, the top panel of Figure 5, depicting Gloria and Rogers’s moment-to-moment levels of control, shows highly entrained cycles, consistent with relatively high rhythmicity and coherence values.
An index of entrainment that is quite similar to the average weighted coherence but more intuitively approachable is the cross- correlation between the partners. Returning to the top panel of Figure 3, showing Gloria’s and Ellis’s moment-to-moment levels of control, this pattern should yield a strong negative correlation between their moment-to-moment values. Cross-correlations are presented in the last column of Table 2, and we see that the correlation between Gloria’s and Ellis’s levels of control is a whopping �.84. This negative relation is very consistent with the principle of reciprocity of control in interpersonal theory, but here we are applying this principle at the level of moment-to-moment variations in control, not at the level of global interpersonal style (as in, e.g., Kiesler & Goldston, 1988). As shown in Table 2, the cross-correlations for control are clearly negative for the other two therapy interactions, and their relative magnitudes map consis- tently onto the corresponding average weighted coherence values and the time series graphs.
Table 2 also shows the results for entrainment of affiliation. The values for the average weighted coherence indicate strong entrain- ment of moment-to-moment levels of affiliation between Gloria and Ellis, and more modest levels of such entrainment between Gloria and Perls and between Gloria and Rogers. Note that the corresponding cross-correlations for affiliation are positive, rather than negative as for control. This positive relation is very consistent with the principle of correspondence of affiliation in interpersonal theory (e.g., warmth invites warmth; coldness invites coldness). The lower panels of Figures 3–5 show the partners’ levels of affiliation over the first 10 min of each session. It can be seen that with affiliation, the prevailing tendency is for peaks in one person to go together with peaks in the other person, and troughs with troughs.
Table 2 Results From Linear Regressions, Spectral and Cross-Spectral Analyses, and Cross-Correlations
Variable
Linear regressions Spectral analysis
Cross-correlationR2 Intercept Slopea Rhythmicity Avg. weighted
coherence Avg. weighted
phase
Control Gloria w/Ellis .16 84.58 �438.42 .90 .75 .46 �.84 Ellis .24 250.66 482.26 .87 Gloria w/Perls .00 199.94 �32.43 .86 .39 �.46 �.45 Perls .00 313.43 �29.79 .80 Gloria w/Rogers .00 131.14 �10.17 .85 .69 .50 �.77 Rogers .00 �93.53 �6.90 .83
Affiliation Gloria w/Ellis .05 147.00 �90.93 .94 .89 �.01 .61 Ellis .00 116.29 �13.11 .93 Gloria w/Perls .04 �46.05 �137.99 .96 .20 �.07 .28 Perls .04 �50.38 66.27 .93 Gloria w/Rogers .01 253.64 �12.79 .84 .36 .03 .30 Rogers .06 266.04 �20.44 .79
Note. Avg. � average. a Slope is the linear change over a 10-min period.
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
7MOMENTARY ASSESSMENT OF PROCESS
From the cross-spectral analysis, as a supplement to the average weighted coherence, it is possible to calculate the average weighted phase. These phase values for control and for affiliation in each of the three therapy interactions are provided in Table 2. Consider first the phase values for affiliation. If the peaks in one person’s oscillations in affiliation exactly coincided with the peaks in the other person’s oscillations in affiliation, and troughs coin- cided with troughs, then the phase value would equal zero. In these
data, positive phase values indicate that the peaks for Gloria were tending to lead the peaks for the therapist, whereas negative values indicate that peaks for the therapist were tending to lead the peaks for Gloria. The values in Table 2 are quite close to zero, but the �.07 for the session between Gloria and Perls suggests that Perls tended to lead the variations in Gloria’s affiliation levels.
According to the principle of reciprocity on control, we would expect the peaks in each person’s oscillations to occur together
Figure 3. Bivariate time series for Gloria’s (solid) and Ellis’s (dotted) control and affiliation behavior.
Figure 4. Bivariate time series for Gloria’s (solid) and Perls’s (dotted) control and affiliation behavior.
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
8 THOMAS, HOPWOOD, WOODY, ETHIER, AND SADLER
with the troughs in the other person’s oscillations. In this case, the phase value would equal .50 (or –.50, which is logically equiva- lent, as explained earlier). This is the value obtained for the session between Gloria and Rogers, which indicates that neither was leading the other. The phase value of .46 for Gloria with Ellis, because it is .04 less than .50, suggests a slight tendency for Gloria’s peaks in control to precede the troughs for Ellis (specif- ically, by about 4% of a full cycle, or roughly 2.5 s). The phase value of –.46 for Gloria with Perls, because it is .04 away from �.50, suggests a slight tendency for Perls’s peaks to precede Gloria’s troughs. Although these phase values can suggest lead– lag relations between the partners, it is important not to interpret them too simplistically (Warner, 1998).
To summarize the most important findings concerning entrain- ment of interpersonal behavior, Ellis and Gloria showed high levels of entrainment of both control and affiliation, and Rogers and Gloria also showed fairly high levels of entrainment, particu- larly for control. In contrast, Perls and Gloria showed relatively low levels of entrainment for both control and affiliation. In other words, at the level of moment-to-moment variations, their session tended not to follow the interpersonal principles of oppositeness on control and sameness on affiliation. Most likely, this stemmed from an intentional strategy by Perls, in which the client’s inter- personal expectations are deliberately disconfirmed for therapeutic reasons (e.g., Beier & Valens, 1975; Carson, 1982; Kiesler, 1996). Indeed, in the filmed aftermath to the sessions, it became clear that although Gloria found Perls’s behavior frustrating, she also found the session to be intriguing and thought provoking.
Examining Windows of an Interaction
When examining graphs and results from a complete interaction, one might note patterns in the data at particular times that merit further exploration (e.g., showing marked deviations from com-
plementarity). Furthermore, if a clinician were to recall a qualita- tively key moment in an interaction, data from this window could be explored further, both graphically and statistically. In either case, these patterns and key moments can be explored by zooming in to a specified segment of an interaction and examining data from this window of the interaction. Because the joystick method provides so much data (e.g., a 5-min interaction yields 600 data points), the analyses described above, among many others, can generally be applied to brief segments from an interaction. To illustrate this possibility, we examined cross-correlations for no- table segments of Gloria’s interaction with each therapist to test their complementarity during these selected times.
Gloria and Ellis. This dyad exhibited the highest degree of complementarity on the affiliation dimension, with the highest cross-correlations and coherence of the three dyads. They also both tended to behave neutrally to warmly. Therefore, in examining their bivariate time series for affiliation (Figure 3), it was notable that partway through the interaction Gloria became less friendly and was rated on the cold half of the circumplex, even though ratings for Ellis generally remained warmer during this time. To examine this shift in Gloria’s affiliation, we looked at the tran- script1 and data from the 450- to 550-s window:
Ellis: You’re not merely concerned, you’re overconcerned; you’re anxious. Because if you were just concerned, you’d do your best, and you’d be saying to yourself, “If I succeed, great; if I don’t succeed, tough, right now I won’t get what I want.” But you’re overconcerned,
1 From Three Approaches to Psychotherapy, by E. L. Shostrom, 1966, 1976, Santa Ana, CA: Psychological and Educational Films. Copyright 1966, 1976, by E. L. Shostrom. Transcript used with permission. Note that in making behaviorally anchored ratings, tone, gestures, expressions, pos- ture, and so forth were critical. Thus, these transcripts cannot capture all of the important processes occurring in this interaction.
Figure 5. Bivariate time series for Gloria’s (solid) and Rogers’s (dotted) control and affiliation behavior.
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
9MOMENTARY ASSESSMENT OF PROCESS
you’re anxious, you’re really saying again what we said a moment ago: “If I don’t get what I want right now, I’ll never get it, and that would be so awful that I’ve got to get it right now.” That causes the anxiety, doesn’t it?
Gloria: Yes, or else work towards it.
Ellis: Yes, but if you . . .
Gloria (interrupts): But if I don’t get it right now, that’s all right, but I want to feel like I’m working toward it.
Ellis: Yeah, but you want a guarantee, I hear. My trained ears hear you saying, “I would like a guarantee of working towards it.” And there are no certainties and guarantees.
Gloria: Well, no, Dr. Ellis, I don’t know why I am coming out that way. What I really mean is “I want a step toward working towards it.” I want . . .
Ellis (interrupts): Well, what’s stopping you?
Gloria: I don’t know, I thought . . . Well, what I was hoping is that whatever this is in me, why I don’t seem to be attracting these kinds of men, why I seem more on the defensive, why I seem more afraid, you could help me [with] what it is I’m afraid of, so I won’t do it so much.
Ellis: Well, my hypothesis is so far that what you’re afraid of is not just failing with this individual man, which is really the only issue when you go out with a new—and we’re talking about eligible males now, we’ll rule out the ineligible ones—you’re not just afraid that you’ll miss this one: You’re afraid that you’ll miss this one, and therefore you’ll miss every other, and therefore you’ll prove that you are really not up to getting what you want and wouldn’t that be awful. You’re bringing in these catastrophes.
Gloria: Well, you sound more strong at it, but that’s similar. I feel like this is silly if I keep this up.
Ellis: If you keep what up?
Gloria: There’s something I’m doing; there’s something I’m doing not to be as real a person with these men that I’m interested in.
Ellis: That’s right. You’re defeating your own ends.
Panel A of Figure 6 shows the bivariate time series for affiliation during this segment of their session. In addition to Gloria’s low levels of friendliness during this time, cross-correlations indicate a very noncomplementary pattern of affiliation (r � –.48). Qualitative anal- ysis of the transcript suggests that Ellis was presuming the extent of Gloria’s problems to be more severe and pervasive than she felt they were. It is also notable that he interrupts her during this time and asserts dominance with phrases such as “my trained ears.” Given that their session was otherwise highly complementary with regard to affiliation, a windowed analysis such as this can illuminate times when their interaction did not flow smoothly and might prove useful for evaluating client–therapist transactions in therapy.
Gloria and Perls. Therapists may employ anticomplementary patterns as a means of increasing anxiety and arousing affect. Among the therapists in this study, Perls’s behavior was the least complementary, and he often challenged Gloria. We examined complementarity during a time when Perls clearly aroused nega- tive affect in Gloria. We chose the following window, which
occurred approximately a third of the way into their interaction (data ranging from 300 to 375 s):
Perls: No, you’re a bluff. You’re a phony.
Gloria: Do you believe, are you meaning that seriously?
Perls: Yeah, if you say you’re afraid, and you laugh, and you giggle, and you squirm, it’s phony. You put on a performance for me.
Gloria: Oh, I resent that, very much.
Perls: Can you express this?
Gloria: Yes sir, I most certainly am not being phony. I will admit this: It’s hard for me to show my embarrassment, and I’m afraid to be embarrassed, but boy, I resent you calling me a “phony.” Just because I smile when I’m embarrassed or when I’m being put in a corner doesn’t mean I’m a phony.
Perls: Wonderful, thank you. You didn’t smile for the last minute.
Gloria: Well, I’m mad at you.
Perls: That’s (Gloria tries to interrupt), that’s right. You didn’t have to cover up your anger with your smile. Now in that moment, in that minute, you were not phony.
Figure 6. Windows into sessions: Gloria (solid) with therapists (dotted).
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
10 THOMAS, HOPWOOD, WOODY, ETHIER, AND SADLER
Gloria: Well, at that minute I was mad though, I wasn’t embarrassed.
Perls: Exactly, when you’re mad, you’re not a phony.
Gloria: I still resent that. I’m not a phony when I’m nervous (hits the couch).
Perls (interrupts): What is that? Again.
Gloria (hits couch): I want to get mad at you. I, I, you know what I would like to do?
Perls (mockingly interrupts): I, I, I.
Gloria: I want you on my level, so I can pick on you just as much as you are picking on me.
Perls: Okay, pick on me.
Panel B of Figure 6 shows the bivariate time series of control during this segment. Cross-correlations suggest that Gloria and Perls’s interaction was not complementary on either dimension during this time (affiliation: r � –.03; control: r � .39). Further, the cross-correlation between Gloria’s own dominance and friendliness was strongly negative (r � –.86) during this win- dow, suggesting that her simultaneous movements toward in- creased dominance and unfriendliness were sometimes even more strongly associated than her overall cross-correlation in Table 2 suggests. These ratings are understandable given that anger projects into the cold– dominant quadrant of the IPC (McCrae & Costa, 1989), and Gloria twice states that she is “mad” during this segment.
Gloria and Rogers. We evaluated data from the following segment of Gloria’s session with Rogers, which has received prior empirical attention (e.g., Reilly & Jacobus, 2008; Weinrach, 1990) and is notably sentimental. Near the end of the session (data ranging from 1,500 to 1,690 s), Gloria and Rogers have the following interchange:
Rogers: You know perfectly within yourself a feeling that occurs when you’re really doing something that’s right for you.
Gloria: I do, I do. And I miss that feeling other times; it’s right away a clue to me.
Rogers: You can really listen to yourself sometimes and realize: “No, no, this isn’t the right feeling; this isn’t the way I would feel if I was doing what I really wanted to do.”
Gloria: But yet, many times I’ll go along and do it anyway. Say, “Oh well, I’m in the situation now, I’ll just remember next time.” Uh, I mention this word a lot in therapy, and most therapists grin at me or giggle or something when I say “utopia.” But when I do follow a feeling, and I feel this good feeling inside me, that’s sort of utopia; that’s what I mean, that’s a way I like to feel whether it’s a bad thing or a good thing. But I feel right about me. This is what I want to accomplish.
Rogers: I sense that in those utopian moments you really feel kind of whole, you feel all in one piece.
Gloria: Yeah, yeah, it gives me a choked-up feeling when you say that because I feel I don’t get that as often as I’d like. I like that whole feeling, that’s real precious to me.
Rogers: I expect none of us get it as often as we’d like, but I really do understand. Mm– hmm, that really does touch you, didn’t it?
Gloria: You know what else I was just thinking? I feel dumb saying it. (Pause) All of a sudden as I’m talking to you I thought, “Gee how nice I can talk to you, and I want you to approve of me, and I respect you, but I miss that my father couldn’t talk to me like you are.” I mean, I’d like to say, “Gee, I’d like to have you for my father.” I don’t even know why that came to me.
Rogers: You look to me like a pretty nice daughter.
Cross-correlations of the control dimension at this time suggest an especially high degree of dominance reciprocity (r � –.84), and examination of the bivariate time series for control (Panel C of Figure 6) indicates the presence of cycling. During this portion of the interview, as Rogers becomes more dominant, Gloria becomes more submissive, and vice versa, with these behaviors occurring in near perfect synchrony. Gloria and Rogers were also especially correspondent (r � .44) in their levels of friendliness during this segment. Therefore, close examination of this portion of their interview provides a window for viewing strong patterns of inter- personal complementarity, with each individual taking control and then stepping back to allow the other to take control, and mutually adjusting to one another as this process unfolds.
Discussion
We used these psychotherapy sessions with Gloria to demon- strate some of the fruitful ways in which the joystick method can be applied to psychotherapy research. Some of the phenomena revealed are very intriguing and would be difficult to capture using other methods.
With regard to our first set of research questions, the patterns of within-person variability in levels of control and affiliation were quite interesting. For example, Ellis showed a very clear predom- inant pattern of moderately friendly, very strongly dominant be- havior, from which he periodically diverged to a markedly more submissive style. What kind of underlying mental process would yield such a pattern? One intriguing possibility is that highly dominant behavior was Ellis’s default, relatively automatic mode, and that the more submissive episodes may have been intentional overrides of this mode, in which, for example, Ellis was trying to be more collaborative. This is the kind of pattern one might see in novice therapists who are highly dominant but whose clinical supervisor has told them to be more collaborative. Effortful pro- cessing would produce the less dominant episodes, and repeated return to the more automatic default would produce the highly dominant set point.
The other density plots also tended to show patterns of variabil- ity in which there was a clear interpersonal set point, with diver- gences away from it in just one direction. In their asymmetry, these patterns are strikingly unlike a bivariate normal distribution. In addition, they are unlike other patterns reported in the interper- sonal literature. Mainly on the basis of the study of multiple- occasion diary data, Moskowitz and Zuroff (2004) have described within-person variability in interpersonal behavior in terms of characteristics like pulse and spin, which are defined in terms of patterns of divergence from a person’s predominant interpersonal style. However, the patterns found here appear to be different from pulse and spin. For pulse, we would expect variation along a radius
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
11MOMENTARY ASSESSMENT OF PROCESS
from the origin through the person’s predominant style; such a direction of variation does not characterize any of the density plots, except perhaps Perls’s. For spin, we would expect angular varia- tion perpendicular to this radius, which also does not characterize any of the density plots well. In short, further study of the partic- ular patterns of variation found in psychotherapy sessions should avoid preconceptions about the form these patterns may take.
The patterns of variability in interpersonal behavior shown by Gloria were remarkably different in response to the three thera- pists. For example, only with Perls did she show a strong inverse relation between her levels of control and affiliation. Her periodic efforts to take control and reset the nature of their interaction tended to be accompanied by striking decreases in her warmth. Indeed, comparing her behavior across the three therapists, her interpersonal variation was mainly in three different directions: with Ellis, toward greater dominance (and sometimes greater warmth); with Perls, toward greater hostile dominance; and with Rogers, toward greater submissiveness. These different patterns of variability provide striking evidence of the powerful effect of a therapist’s style on a client’s interpersonal dynamics.
With regard to our second set of research questions, the dynamic patterns interlinking the partners’ interpersonal behaviors were also quite telling. Of the greatest importance were differences in the degree of entrainment between partners’ oscillating levels of control and affiliation. Relatively high degrees of entrainment, such as those seen in the sessions with Ellis and with Rogers, probably contribute crucially to the sense of a satisfying, predict- able therapeutic alliance, because the entrained variations follow the principles of interpersonal complementarity. In contrast, as mentioned earlier, the relatively low levels of entrainment in the session with Perls probably reflected a deliberate therapeutic strat- egy on his part, in which the client’s typical interpersonal expec- tations are not met or even disconfirmed. This noncomplementary strategy may be an important technique for eliciting reflection about and change in the client’s interpersonal behaviors (e.g., Carson, 1969; Kiesler, 1996; Tracey et al., 1999). In short, inter- personal strategies enacted by the therapist should show up as different patterns of dynamics in the therapy sessions, and the dynamical patterns obtained can be examined to study whether the therapist’s interventions are having the desired effects on the client’s inter- personal patterns.
As we have pointed out elsewhere (Sadler et al., 2009, Sadler, Ethier, & Woody, 2011), measuring variation in interpersonal behavior over time opens up the possibility of distinguishing multiple, and conceptually separate, levels of complementarity. In addition to the degree of complementarity of two partners’ global interpersonal styles, we can examine the complementa- rity of steady shifts in overall levels over the course of an interaction (Sadler & Woody, 2003), conceptualized in the present study as linear trends. For example, we can look for diverging slopes on control, as found here for Gloria and Ellis. Possibly of even greater importance, we can also examine the complementarity of cyclical variations in interpersonal behav- ior. For example, are one person’s oscillations in control at- tuned to the other person’s oscillations in control? As found in the present study, cycles that are opposite in phase (peaks in one person going together with troughs in the other person) are consistent with the interpersonal principle of oppositeness on
control, but the degree of such entrainment varied strikingly across the various psychotherapy interactions.
In addition to allowing us to examine the foregoing kinds of interpersonal dynamics, the computer joystick method has some interesting advantages as a way of measuring global interpersonal style. In particular, it provides a careful and thorough sampling of interpersonal behavior from moment to moment across the entire course of the interaction. Aggregating across this time sampling avoids several possible shortcomings of techniques that rely on retrospection at the conclusion of the interaction. For example, at the conclusion of an interaction, raters may tend to remember better the acts that were consistent with their overall view of the person, or the acts that stood out to them for any reason (e.g., because they were emotional, unexpected, significant, or the like), or the acts that occurred first or last (Kahneman, 2011; Stone & Shiffman, 1994). Thus, an advantage to using the joystick method to measure global interpersonal style is that ratings are made directly and immediately while the interaction is being viewed, limiting error attributable to recall effects.
Comparison With Act-to-Act Relations and Other Statistical Methods
As mentioned earlier, a novel aspect of the computer joystick method is that it captures interaction dynamics as reasonably continuous flows. An approach that is more familiar to many researchers involves segmenting each partner’s stream of behavior into a sequence of discrete acts and then studying time-lagged relations of one person’s behaviors to the other person’s. This method has proven to be very generative in previous research on psychotherapy process (e.g., Lichtenberg & Heck, 1986; Tracey, 1985; Wampold, 1986; Wampold & Kim, 1989). The general assumption underlying much of this work is that a relation found with a time lag supports the hypothesis that one person’s behavior was leading or driving the other person’s behavior. In the cross- spectral analyses of the time-series data from the joystick method, the average weighted phase is analogous to the time lag in the act-to-act approach. Phase indexes the degree of displacement from one person’s peaks to the other person’s peaks, which could readily be expressed as a time lag.
However, it is important to point out that the presence of oscillations in interpersonal behavior (as indicated by the high rhythmicity values in this study) tends to radically transform the possible meaning of a time-lagged relation. To illustrate, consider a case in which two people’s behaviors have exactly the same frequency and are exactly in phase. With a time lag of 0, this would yield a correlation of 1. As we increasingly lag one person behind the other, the correlation drops to 0 and then reverses in sign. When the time lag is long enough that peaks for one person are paired with the lagged troughs for the other person, the corre- lation reaches –1. With further increasing lags, the correlation heads back to 0 and then reaches 1 again. Real data would be more complex, with multiple frequencies superimposed and varying amplitudes (heights of the peaks). Under these conditions, inter- preting the fluctuating values of the various lagged correlations can be very challenging.
Moreover, the presence of oscillations in behavior has important conceptual implications. One is that the future value of a fairly consistently oscillating behavior can be anticipated well ahead of
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
12 THOMAS, HOPWOOD, WOODY, ETHIER, AND SADLER
time. This means that the causes of a later behavior can occur long before the immediately preceding behavior. In addition, oscillations are consistent with circular causality, in which the partners are in a feedback loop and neither can be said to be driving the obtained pattern. In short, with oscillating behaviors such as those evident in these therapy interactions, time-lagged relations need to be interpreted with caution.
In this article, we mostly focused on the spectral and cross-spectral analysis of the time-series data obtained with the joystick method. However, joystick data are very well suited to other promising statis- tical approaches, such as dynamic systems analyses (Boker, 2002; Boker & Wenger, 2007; Salvatore & Tschacher, 2012). These possi- bilities provide a rich vein for further exploration.
Conclusion
This study demonstrates the utility of measuring momentary interpersonal processes in psychotherapy and provides a step to- wards more accurate measurement of interpersonal process. The fine-grained level of analysis outlined in this study has the poten- tial to augment current research on meaningful psychological processes as they occur within therapy sessions. Hill, Nutt, and Jackson (1994) noted the relative infrequency with which the same measure of process is used in multiple studies. One explanation for this may be that measures used to study particular constructs or techniques that are more prominent in particular therapy orienta- tions are less likely to be used by researchers with differing interests and orientations. An advantage of the IPC is that it is reasonably trans-theoretical and hence could allow for increased communication regarding therapy processes across various re- search paradigms (Hopwood, 2010).
As technological advances allowing for the collection and mea- surement of psychological data continue to advance, research that focuses on dynamic processes will play an increasingly important and sophisticated role in psychotherapy research. Some types of disorders, such as borderline personality disorder, may be charac- terized by distinctive cyclical patterns, and these patterns, as well as changes in them due to psychotherapy, could be captured using the methods advanced here (Pincus & Hopwood, 2012). In addi- tion, questions regarding the degree to which interpersonal pro- cesses coalesce and diverge across different therapies are highly amenable to investigation using the methods advanced in this article. Furthermore, using these methods to investigate interper- sonal processes in psychotherapy may be informative for specify- ing the conditions under which particular intervention techniques are most likely to be effective. Finally, the joystick method could also be used in the training and supervision of psychotherapists, as outlined in Pincus et al. (in press).
References
Anchin, J. C., & Pincus, A. L. (2010). Evidence-based interpersonal psychotherapy with personality disorders: Theory, components, and strategies. In J. J. Magnavita (Ed.), Evidence-based treatment of person- ality dysfunction: Principles, methods, and processes (pp. 113–166). Washington, DC: American Psychological Association. doi:10.1037/ 12130-005
Andrews, J. D. W. (1989). Psychotherapy of depression: A self- confirmation model. Psychological Review, 96, 576 – 607. doi:10.1037/ 0033-295X.96.4.576
Beier, E. G., & Valens, E. G. (1975). People-reading: How we control others, how they control us. Oxford, United Kingdom: Stein & Day.
Benjamin, L. S. (1979). Use of structural analysis of social behavior (SASB) and Markov chains to study dyadic interactions. Journal of Abnormal Psychology, 88, 303–319. doi:10.1037/0021-843X.88.3.303
Benjamin, L. S. (1996). Interpersonal diagnosis and treatment of person- ality disorders (2nd ed.). New York, NY: Guilford Press.
Boker, S. M. (2002). Consequences of continuity: The hunt for intrinsic properties within parameters of dynamics in psychological processes. Multivariate Behavioral Research, 37, 405– 422. doi:10.1207/ S15327906MBR3703_5
Boker, S. M., & Wenger, M. J. (2007). Data analytic techniques for dynamical systems. Mahwah, NJ: Erlbaum.
Carson, R. C. (1969). Interaction concepts of personality. Chicago, IL: Aldine.
Carson, R. C. (1982). Self-fulfilling prophecy, maladaptive behavior, and psychotherapy. In J. Anchin & D. Kiesler (Eds.), Handbook of interper- sonal psychotherapy (pp. 64 –77). New York, NY: Pergamon.
Castonguay, L. G., Goldfried, M. R., Wiser, S. L., Raue, P. J., & Hayes, A. M. (1996). Predicting the effect of cognitive therapy for depression: A study of unique and common factors. Journal of Consulting and Clinical Psychology, 64, 497–504. doi:10.1037/0022-006X.64.3.497
Dietzel, C. S., & Abeles, N. (1975). Client–therapist complementarity and therapeutic outcome. Journal of Counseling Psychology, 22, 264 –272. doi:10.1037/h0076731
Goldfried, M. R. (in press).What should we expect from psychotherapy? Clinical Psychology Review.
Gurtman, M. B. (1996). Interpersonal problems and the psychotherapy context: The construct validity of the Inventory of Interpersonal Prob- lems. Psychological Assessment, 8, 241–255. doi:10.1037/1040-3590.8 .3.241
Hayes, S. C. (2004). Acceptance and commitment therapy, relational frame theory, and the third wave of behavior and cognitive therapies. Behavior Therapy, 35, 639 – 665. doi:10.1016/S0005-7894(04)80013-3
Henry, W. P., Schacht, T. E., & Strupp, H. H. (1986). Structural analysis of social behavior: Application to a study of interpersonal process in differential psychotherapeutic outcome. Journal of Consulting and Clin- ical Psychology, 54, 27–31. doi:10.1037/0022-006X.54.1.27
Henry, W. P., Schacht, T. E., & Strupp, H. H. (1990). Patient and therapist introject, interpersonal process, and differential psychotherapy outcome. Journal of Consulting and Clinical Psychology, 58, 768 –774. doi: 10.1037/0022-006X.58.6.768
Hill, C. E., Nutt, E. A., & Jackson, S. (1994). Trends in psychotherapy process research: Samples, measures, researchers, and classic publica- tions. Journal of Counseling Psychology, 41, 364 –377. doi:10.1037/ 0022-0167.41.3.364
Hopwood, C. J. (2010). An interpersonal perspective on the personality assessment process. Journal of Personality Assessment, 92, 471– 479. doi:10.1080/00223891.2010.513284
Horowitz, L. M., Rosenberg, S. E., & Bartholomew, K. (1993). Interper- sonal problems, attachment styles, and outcome in brief dynamic psy- chotherapy. Journal of Consulting and Clinical Psychology, 61, 549 – 560. doi:10.1037/0022-006X.61.4.549
Horvath, A. O., Del Re, A. C., Flückiger, C., & Symonds, D. (2011). Alliance in individual psychotherapy. Psychotherapy, 48, 9 –16. doi: 10.1037/a0022186
Kahneman, D. (2011). Thinking, fast and slow. Toronto, Canada: Double- day/Random House.
Kiesler, D. J. (1996). Contemporary interpersonal theory and research: Per- sonality, psychopathology, and psychotherapy. New York, NY: Wiley.
Kiesler, D. J., & Goldston, C. S. (1988). Client–therapist complementarity: An analysis of the Gloria films. Journal of Counseling Psychology, 35, 127–133. doi:10.1037/0022-0167.35.2.127
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
13MOMENTARY ASSESSMENT OF PROCESS
Kiesler, D. J., & Goldston, C. S., & Schmidt, J. A. (1991). Manual for the Check List of Interpersonal Transactions–Revised (CLOIT–R) and the Check List of Psychotherapy Transactions–Revised (CLOPT–R). Rich- mond, VA: Virginia Commonwealth University
Leary, T. (1957). Interpersonal diagnosis of personality. New York, NY: Ronald Press.
Lichtenberg, J. W., & Heck, E. (1986). The analysis of pattern and sequence in process research. Journal of Counseling Psychology, 33, 170 –181. doi:10.1037/0022-0167.33.2.170
Lizdek, I., Sadler, P., Woody, E., Ethier, N., & Malet, G. (2012). Capturing the stream of behavior: A computer-joystick method for coding inter- personal behavior continuously over time. Social Science Computer Review, 30, 513–521. doi:10.1177/0894439312436487
Luyten, P., & Blatt, S. J. (2013). Interpersonal relatedness and self-definition in normal and disrupted personality development: Retrospect and prospect. American Psychologist, 68, 172–183. doi:10.1037/a0032243
Markey, P. M., Lowmaster, S., & Eichler, W. (2010). A real-time assess- ment of interpersonal complementarity. Personal Relationships, 17, 13–25. doi:10.1111/j.1475-6811.2010.01249.x
McCrae, R. R., & Costa, P. T. (1989). The structure of interpersonal traits: Wiggins’s circumplex and the five-factor model. Journal of Personality and Social Psychology, 56, 586 –595. doi:10.1037/0022-3514.56.4.586
Moskowitz, D. S., & Zuroff, D. C. (2004). Flux, pulse, and spin: Dynamic additions to the personality lexicon. Journal of Personality and Social Psychology, 86, 880 – 893. doi:10.1037/0022-3514.86.6.880
Muran, J. C., & Barber, J. P. (Eds.). (2010). The therapeutic alliance: An evidence-based guide to practice. New York, NY: Guilford Press.
Orford, J. (1986). The rules of interpersonal complementarity: Does hos- tility beget hostility and dominance, submission? Psychological Review, 93, 365–377. doi:10.1037/0033-295X.93.3.365
Pincus, A. L., & Hopwood, C. J. (2012). A contemporary interpersonal model of personality pathology and personality disorder. In T. A. Wi- diger (Ed.), Oxford handbook of personality disorders (pp. 372–398). Oxford, United Kingdom: Oxford University Press. doi:10.1093/ oxfordhb/9780199735013.013.0018
Pincus, A. L., Lukowitsky, M. R., & Wright, A. G. C. (2010). The interpersonal nexus of personality and psychopathology. In T. Millon R. Krueger & E. Simonsen (Eds.), Contemporary directions in psychopa- thology: Toward DSM–V and ICD-11 (pp. 523–552). New York, NY: Guilford Press.
Pincus, A. L., Sadler, P., Woody, E., Roche, M. J., Thomas, K. M., & Wright, A. G. C. (in press). Assessing interpersonal dynamics. In C. J. Hopwood & R. F. Borstein (Eds.), Multimethod clinical assessment. New York, NY: Guilford Press.
Puschner, B., Kraft, S., & Bauer, S. (2004). Interpersonal problems and outcome in outpatient psychotherapy: Findings from a long-term longi- tudinal study in Germany. Journal of Personality Assessment, 83, 223– 234. doi:10.1207/s15327752jpa8303_06
R Development Core Team. (2011). R: A Language and Environment for Statistical Computing [Computer software]. Retrieved from http://www .R-project.org/
Reilly, J., & Jacobus, V. (2008). Students evaluate Carl Rogers and his relationship with Gloria: A brief report. Journal of Humanistic Psychol- ogy, 48, 32– 41. doi:10.1177/0022167807311944
Sadler, P., Ethier, N., Gunn, G. R., Duong, D., & Woody, E. (2009). Are we on the same wavelength? Interpersonal complementarity as shared cyclical patterns during interactions. Journal of Personality and Social Psychology, 97, 1005–1020. doi:10.1037/a0016232
Sadler, P., Ethier, N., & Woody, E. (2011). Interpersonal complementarity. In L. M. Horowitz & S. N. Strack (Eds.), Handbook of interpersonal psychology: Theory, research, assessment, and therapeutic interventions (pp. 123–142). New York, NY: Wiley.
Sadler, P., & Woody, E. (2003). Is who you are who you’re talking to? Interpersonal style and complementarity in mixed-sex interactions. Jour- nal of Personality and Social Psychology, 84, 80 –96. doi:10.1037/0022- 3514.84.1.80
Safran, J. D. (1984). Assessing the cognitive interpersonal cycle. Cognitive Therapy and Research, 8, 333–347. doi:10.1007/BF01173309
Safran, J. D. (1990a). Towards a refinement of cognitive therapy in light of interpersonal theory: 1. Theory. Clinical Psychology Review, 10, 87– 105. doi:10.1016/0272-7358(90)90108-M
Safran, J. D. (1990b). Towards a refinement of cognitive therapy in light of interpersonal theory: 2. Practice. Clinical Psychology Review, 10, 107– 121. doi:10.1016/0272-7358(90)90109-N
Salvatore, S., & Tschacher, W. (2012). Time dependency of psychotherapeutic exchanges: The contribution of the theory of dynamic systems in analyzing process. Frontiers in Psychology, 3. doi:10.3389/fpsyg.2012.00253
Shostrom, E. L. (Producer & Director). (1966). Three approaches to psychother- apy [Film]. Santa Ana, CA: Psychological and Educational Films.
Shostrom, E. L. (Producer & Director). (1976). Three approaches to psychother- apy [Film]. Santa Ana, CA: Psychological and Educational Films.
Stone, A. A., & Shiffman, S. (1994). Ecological momentary assessment (EMA) in behavioral medicine. Annals of Behavioral Medicine, 16, 199 –202.
Strupp, H. H., & Binder, J. L. (1984). Psychotherapy in a new key: A guide to time-limited dynamic psychotherapy. New York, NY: Basic Books.
Tracey, T. J. (1985). Dominance and outcome: A sequential examination. Journal of Counseling Psychology, 32, 119 –122. doi:10.1037/0022- 0167.32.1.119
Tracey, T. J. G. (1993). An interpersonal model of the therapeutic process. Journal of Counseling Psychology, 40, 396 – 409. doi:10.1037/0022- 0167.40.4.396
Tracey, T. J. G. (2004). Levels of interpersonal complementarity: A sim- plex representation. Personality and Social Psychology Bulletin, 30, 1211–1225. doi:10.1177/0146167204264075
Tracey, T. J. G., Bludworth, J., & Glidden-Tracey, C. E. (2012). Are there parallel processes in psychotherapy supervision? An empirical exami- nation. Psychotherapy, 49, 330 –343. doi:10.1037/a0026246
Tracey, T. J. G., Sherry, P., & Albright, J. M. (1999). The interpersonal process of cognitive– behavioral therapy: An examination of comple- mentarity over the course of treatment. Journal of Counseling Psychol- ogy, 46, 80 –91. doi:10.1037/0022-0167.46.1.80
Wampold, B. E. (1984). Tests of dominance in sequential categorical data. Psychological Bulletin, 96, 424 – 429. doi:10.1037/0033-2909.96.2.424
Wampold, B. E. (1986). State of the art in sequential analysis: Comment on Lichtenberg and Heck. Journal of Counseling Psychology, 33, 182–185.
Wampold, B. E., & Kim, K.-H. (1989). Sequential analysis applied to counseling process and outcome: A case study revisited. Journal of Counseling Psychology, 36, 357–364. doi:10.1037/0022-0167.36.3.357
Warner, R. M. (1998). Spectral analysis of time-series data. New York, NY: Guilford Press.
Weinrach, S. G. (1990). Rogers and Gloria: The controversial film and the enduring relationship. Psychotherapy: Theory, Research, Practice, Training, 27, 282–290. doi:10.1037/0033-3204.27.2.282
Wiggins, J. S. (1991). Agency and communion as conceptual coordinates for the understanding and measurement of interpersonal behavior. In W. Grove & D. Cicchetti (Eds.), Thinking clearly about psychology: Essays in honor of Paul E. Meehl (Vol. 2, pp. 89 –113). Minneapolis: University of Minnesota Press.
Wiggins, J. S. (1996). An informal history of the interpersonal circumplex tradition. Journal of Personality Assessment, 66, 217–233. doi:10.1207/ s15327752jpa6602_2
Received August 15, 2012 Revision received April 16, 2013
Accepted April 18, 2013 �
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
14 THOMAS, HOPWOOD, WOODY, ETHIER, AND SADLER
- Momentary Assessment of Interpersonal Process in Psychotherapy
- Assessing Dynamic Aspects of the Therapeutic Relationship
- A Theoretical Framework for Assessing Moment-to-Moment Interpersonal Behavior
- A Computer Joystick Method for Coding Momentary Interpersonal Behavior
- The Present Study
- Method
- Procedure
- Final Joystick Data
- Calculation of Indices
- Indices of within-person variability
- Density plots
- Linear trends in levels
- Indices of oscillation and entrainment
- Results
- Global Levels of Control and Affiliation
- Within-Person Variability
- Temporal Dynamics That Interrelate the Partners’ Behaviors in the Interaction
- Examining Windows of an Interaction
- Gloria and Ellis
- Gloria and Perls
- Gloria and Rogers
- Discussion
- Comparison With Act-to-Act Relations and Other Statistical Methods
- Conclusion
- References
Journal ol"Personality and Social Psychology 1995. Vol. 68. No. 3. 518-530
Copyright 1995 by the American Psychological Association, Inc. 0O22-3514/95/$3.0O
Self-Esteem as an Interpersonal Monitor: The Sociometer Hypothesis
Mark R. Leary Wake Forest University
Ellen S. Tambor Johns Hopkins University
Sonja K. Terdal Northwestern Michigan College
Deborah L. Downs Ohio State University
Five studies tested hypotheses derived from the sociometer model of self-esteem according to which the self-esteem system monitors others' reactions and alerts the individual to the possibility of social exclusion. Study 1 showed that the effects of events on participants' state self-esteem paralleled their assumptions about whether such events would lead others to accept or reject them. In Study 2, participants' ratings of how included they felt in a real social situation correlated highly with their self-esteem feelings. In Studies 3 and 4, social exclusion caused decreases in self-esteem when re- spondents were excluded from a group for personal reasons, but not when exclusion was random, but this effect was not mediated by self-presentation. Study 5 showed that trait self-esteem correlated highly with the degree to which respondents generally felt included versus excluded by other people. Overall, results provided converging evidence for the sociometer model.
The proposition that people have a fundamental need to maintain their self-esteem has provided the cornerstone for a great deal of work in personality, social, developmental, clinical, and counseling psychology. In the century since William James (1890) first referred to self-esteem as an "elementary endow- ment of human nature," many classic theories of personality have addressed the importance of self-esteem needs, many emotional and behavioral problems have been attributed to un- fulfilled needs for self-esteem, and many psychotherapeutic ap- proaches have focused in one way or another on the client's feel- ings about himself or herself (Adler, 1930; Allport, 1937; Bed- nar. Wells, & Peterson, 1989; Horney, 1937; Maslow, 1968; Rogers, 1959). Among social psychologists, the self-esteem mo- tive has been offered as an explanation of a wide array of phe- nomena, including self-serving attributions (Blaine & Crocker, 1993), reactions to evaluation (S. C. Jones, 1973), self-handi- capping (E. E. Jones & Berglas, 1978), downward social com- parison (Wills, 1981), attitude change (Steele, 1988), and in- group/but-group perceptions (Crocker, Thompson, McGraw, &Ingerman, 1987).
Despite the fact that the self-esteem motive has been invoked to explain so many phenomena, little attention has been paid to the source or functions of the self-esteem motive itself. The field has taken it for granted that people have a motive to protect
Mark R. Leary, Department of Psychology, Wake Forest University; Ellen S. Tambor, Genetics and Public Policy Studies, Johns Hopkins University; Sonja K. Terdal, Department of Sociology, Northwestern Michigan College; Deborah L. Downs, Department of Psychology, Ohio State University.
We thank Robin Kowalski, Cathy Seta, and James Shepperd for their comments on earlier versions of this article.
Correspondence concerning this article should be sent to Mark R. Leary, Department of Psychology, Wake Forest University, Winston-Sa- lem. North Carolina 27109.
their self-esteem without adequately addressing the question of why they should have such a motive or what function it might serve. In five studies we evaluated the hypothesis that the self- esteem system functions as a sociometer that monitors the de- gree to which the individual is being included versus excluded by other people and that motivates the person to behave in ways that minimize the probability of rejection or exclusion.
Explanations of the Self-Esteem Motive
Although few efforts have been made to systematically ad- dress the functions of the self-esteem motive, at least three gen- eral explanations of the motive can be gleaned from the literature.
The most widely acknowledged explanation is that people strive for self-esteem because high self-esteem promotes posi- tive affect by buffering the person against stress and other nega- tive emotions and by enhancing personal adjustment, whereas low self-esteem is associated with depression, anxiety, and mal- adjustment. Research findings attest that people with low self- esteem experience virtually every negative emotion more com- monly than those with high self-esteem (e.g., Cutrona, 1982; Goswick & Jones, 1981; Leary, 1983; Taylor & Brown, 1988; White, 1981). Furthermore, high self-esteem appears to buffer people against feelings of anxiety, enhance coping, and promote physical health (Baumeister, 1993; Greenberg et al., 1992; Tay- lor & Brown, 1988).
Although the link between self-esteem, affect, adjustment, and health is undisputed, it is less clear why self-esteem should produce these effects. One possibility is that, because self-es- teem is associated with confidence and high expectations of suc- cess, high self-esteem is associated with optimism and lowered anxiety. In a variation on this theme, Greenberg, Pyszczynski, and Solomon (1986) suggested that high self-esteem serves as a buffer against the existential anxiety people experience when they contemplate their own fragility and mortality. However, it
518
SELF-ESTEEM FEELINGS 519
is unclear why such a psychological system for buffering people against anxiety and uncertainty should have developed. Indeed, from an evolutionary perspective, we might expect that people who worried about possible misfortunes (including death) would have been more likely to survive and reproduce.
A second set of explanations emphasizes the role of high self- esteem in promoting goal achievement. The motive to seek self- esteem may have developed because high self-esteem enhances people's willingness to strive toward desired goals and to persist in the face of obstacles and setbacks (Bandura, 1977; Greenwald, 1980; Kernis, 1995). In a related vein, Tedeschi and Norman (1985) suggested that people seek self-esteem be- cause self-esteem is associated with feelings of control over one's environment.
In support of explanations that implicate goal accomplish- ment, people with high self-esteem often work harder and per- form better after an initial failure than people with low self- esteem (Perez, 1973; Shrauger & Sorman, 1977). However, it is also true that high self-esteem may lead to nonproductive per- sistence when tasks prove to be insurmountable (McFarlin, Baumeister, & Blascovich, 1984). Although self-efficacious cog- nitions can undoubtedly facilitate achievement, this explana- tion cannot explain why self-esteem is inherently affectively laden (Brown, 1993). Although we concur that it may often be useful for people to think that they possess certain favorable attributes or abilities (i.e., to have high self-efficacy), this does not explain why self-esteem is intimately linked to strong self- relevant emotions.
A third set of explanations of the self-esteem motive involves the possibility that people seek self-esteem for its own sake. Many writers have implicitly assumed the existence of a self- system that maintains a sense of integrity or adequacy (e.g., Epstein, 1973; James, 1890;Steele, 1988). One difficulty with this assumption is that it fails to explain why a motive to behave in ways that promote self-esteem should exist at all. It also does not adequately explain why certain events pose threats to self- integrity and others do not. In fact, to the extent that, over the long run, rewarding encounters and the attainment of one's goals depends on accurate knowledge of oneself, a self-system that functioned purely to elevate one's sense of self may actually be less adaptive than one that sought accurate self-knowledge (Heatherton & Ambady, 1993).
Properties of the Self-Esteem System
Although each of these explanations can explain why people prefer to evaluate themselves positively under certain circum- stances, none clearly and fully explains why people appear gen- erally to need self-esteem and regularly behave in ways to main- tain and enhance it. Before offering an alternative explanation of the self-esteem motive that we believe parsimoniously ex- plains the properties of the self-esteem system, we must clarify precisely what we mean by the term self-esteem.
Self-esteem has often been described as an attitude, specifi- cally an attitude toward oneself (Coopersmith, 1967; Rosen- berg, 1965). Like all attitudes, self-esteem has cognitive and affective components. A distinction can be drawn between the self-concept (beliefs about the self) and self-esteem (evaluation of oneself in light of those beliefs). Although self-esteem is often
based on self-relevant cognitions, not all cognitions about the self, even evaluatively laden ones, are relevant to a person's self- esteem. Each person has many self-beliefs that have no affective quality. People may believe firmly that they are very good or very bad at certain mundane tasks, for example, yet experience no corresponding increase or decrease in their self-esteem.
Self-esteem includes an essential affective quality that "cold" cognitions about the self do not. Brown (1993) persuasively ar- gued that self-esteem is fundamentally based in affective pro- cesses, specifically positive and negative feelings about oneself. People do not simply think favorable or unfavorable self-rele- vant thoughts; they feel good or bad about themselves. Further- more, they fiercely desire to feel good rather than bad. Most previous explanations of the self-esteem motive have difficulty explaining the inherent emotional and motivational qualities of self-esteem. We return to this point shortly.
Although people can be characterized as having some average level of self-esteem over situations and time (trait self-esteem), self-esteem inevitably fluctuates as people move about their daily lives (state self-esteem). To our knowledge, researchers have not previously addressed the question of whether people are motivated to maintain state self-esteem, trait self-esteem, or both. However, we think it is reasonable to assume that people want both to feel good about themselves in the present moment as well as maintain positive self-feelings over time. As will be- come clear, stale self-esteem is of paramount importance in the explanation of the self-esteem system we describe.
Self-Esteem System as a Sociometer
The hypothesis to be considered in this article is that the self- esteem system is a sociometer that is involved in the mainte- nance of interpersonal relations (Leary, 1990; Leary & Downs, 1995). Specifically, a person's feelings of state self-esteem are an internal, subjective index or marker of the degree to which the individual is being included versus excluded by other people (the person's inclusionary status) and the motive to maintain self-esteem functions to protect the person against social rejec- tion and exclusion. We believe that this perspective on self-es- teem more parsimoniously explains the emotional and motiva- tional aspects of self-esteem than other explanations.
Many writers have observed that human beings possess a fun- damental motive to seek inclusion and to avoid exclusion from important social groups and that such a motive to promote gre- gariousness and social bonding may have evolved because of its survival value (Ainsworth, 1989; Barash, 1977; Baumeister & Leary, in press; Baumeister & Tice, 1990; Bowlby, 1969; Hogan, 1982; Hogan, Jones, & Cheek, 1985). Because solitary human beings in a primitive state are unlikely to survive and reproduce, psychological systems evolved that motivated people to develop and maintain some minimum level of inclusion in social rela- tionships and groups.
Successfully maintaining one's connections to other people requires a system for monitoring others' reactions, specifically the degree to which other people are likely to reject or exclude the individual. Such a system must monitor one's inclusionary status more or less continuously for cues that connote disap- proval, rejection, or exclusion (i.e., it must be capable of func- tioning preconsciously), it must alert the individual to changes
520 LEARY, TAMBOR, TERDAL, AND DOWNS
in his or her inclusionary status (particularly decrements in so- cial acceptance), and it must motivate behavior to restore his or her status when threatened. In our view, the self-esteem system serves precisely the functions of such a sociometer.
From this perspective, what have previously been viewed as threats to self-esteem are, at a more basic level, events that make the possibility of social exclusion salient. Events that lower self- esteem appear to be those that the person believes may jeopar- dize his or her social bonds. Ego-threatening events are aversive because they signal a possible deterioration in one's social relationships.
Strictly speaking, then, people do not have a need to maintain self-esteem per se. Self-esteem is simply an indicator of the qual- ity of one's social relations vis-a-vis inclusion and exclusion. To use an analogy, a behavioral researcher from another planet might conclude that Earthlings who drive automobiles are mo- tivated to keep the indicator on their fuel gauges from touching the E (empty); every time the indicator approaches E, Earth- lings behave in ways that move the indicator back toward F (full). However, the fuel gauge is simply a monitor to help peo- ple avoid running out of gas, just as self-esteem is a monitor to help people avoid social exclusion. Thus, our analysis departs sharply from explanations that impute people with an inherent need to maintain self-esteem.
As noted earlier, self-esteem is closely tied to affective pro- cesses (Brown, 1993). Self-esteem involves how people feel about themselves; high self-esteem "feels" good, whereas low self-esteem does not (Scheff, Retzinger, & Ryan, 1989). Most motivational and drive systems elicit aversive affect when poten- tial threats to the organism's well-being are detected (i.e., when needs are not being met). People experience unpleasant affect when they are hungry, thirsty, sleepy, or in physical danger, for example, but feel better when they are well fed, hydrated, rested, and safe. In the case of self-esteem, negative emotions arise when cues that connote disapproval, rejection, or exclu- sion are detected. It is for this reason that "some of the best evidence for changes in self-esteem can be inferred from self- reports of mood" (Heatherton & Polivy, 1991, p. 896).
The affective reactions to changes in self-esteem tend to cen- ter around the "self-relevant" emotions, such as feelings of pride and shame. In particular, losses of self-esteem are associ- ated with feeling foolish, ashamed, inadequate, or awkward, whereas increased self-esteem is associated with pride, self-sat- isfaction, and confidence (Scheffet al., 1989).
Real or potential threats to self-esteem also elicit anxiety. Not only does state self-esteem correlate highly with state anxiety but trait self-esteem and trait anxiety correlate as well (Spivey, 1989). According to the sociometer model, lowered self-esteem and anxiety are coeffects of perceived exclusion (Leary, 1990). Baumeister and Tice (1990) argued that anxiety is a natural consequence of perceived threats to one's social bonds. Viewed from this perspective, social anxiety, which originates from peo- ple's concerns with others' impressions of them (Leary & Ko- walski, in press; Schlenker & Leary, 1982), is also a product of the sociometer. Concerns about other people's impressions raise the specter of disapproval and rejection.
This is not to say that self-esteem is nothing more than mood (see Heatherton & Polivy, 1991). Changes in mood occur for a wide variety of reasons that have nothing to do with social
exclusion or with self-esteem. Even so, decreases in self-esteem are invariably accompanied by negative affect.
Our conceptualization offers a functional perspective on the self-esteem motive and explains several facts about self-esteem. Although space does not permit a full discussion of the implica- tions of this explanation (see Leary & Downs, 1995), we offer a few examples that demonstrate its applicability as a general model of self-esteem. For example, viewing the self-esteem sys- tem as a sociometer explains Cooley's (1902) observation that people's feelings about themselves are highly sensitive to how they think they are being regarded by other people. The more support and approval people receive, the higher their self-es- teem tends to be (Harter, 1993). Such feelings are a direct re- flection of one's inclusionary status, with deflations of self-es- teem alerting the individual to the possibility that their standing in important groups or relationships is in jeopardy.
The sociometer model also explains why people place varying degrees of importance on different domains of the self (e.g., in- tellectual, athletic, social), as well as why the importance people place on these domains correlates highly with the importance they think others place on them. It also explains why self-esteem correlates highly with the individual's performance in domains judged important to others (Harter & Marold, 1991). People strive to excel in domains that will enhance their inclusion by certain other people. As a result, they adopt others' standards, and their self-esteem is affected by performance in domains that others value. People's self-evaluations are also differentially affected when they visualize different significant others (Baldwin & Holmes, 1987), presumably because the socio- meter is sensitive to the idiosyncratic standards of particular people. What may not jeopardize one's image in one person's eyes may lead to rejection by another.
This perspective also helps us understand why people with lower self-esteem are more sensitive to socially relevant cues than those with high self-esteem (Brockner, 1983). People who already feel included, accepted, and socially integrated need not be as concerned with fitting in as people who feel less so (Moreland&Levine, 1989; Snodgrass, 1985).
The sociometer hypothesis also answers the seemingly para- doxical question of why, if people possess a system for main- taining self-esteem, some people have low self-esteem (Baumeister, 1993). The answer is that people do not have a system for maintaining self-esteem per se but a system for avoiding social exclusion. To function properly, the system must lead the person who faces potential exclusion or ostracism to feel badly about it. Over time, people who experience real or imagined rejection repeatedly will have lower trait self-esteem than people who feel warmly included.
Overview of O u r Research
In brief, conceptualizing the self-esteem system as a socio- meter that monitors one's standing with others helps to explain most of its central properties. In addition, it confers an essential function on self-esteem that helps to explain why the need for self-esteem appears to be innate and universal.
Empirical validation of the sociometer conceptualization of state self-esteem requires the evaluation of at least three general theoretical implications. Most fundamentally, the theory stipu-
SELF-ESTEEM FEELINGS 521
lates that changes in self-esteem should be closely associated with changes in the degree to which people perceive they are being included versus excluded by other people (perceived in- clusionary status). Second, research must show that the socio- meter model accounts for such changes better than alternative conceptualizations of self-esteem. Third, research must demon- strate that behavioral reactions to what have been viewed as threats to self-esteem are in fact responses to real, potential, or imagined social exclusion.
Obviously, fully testing these three general implications of the theory will require much research. In this article we report the results of five studies that focused on the first (and most fundamental) theoretical implication just described: that self- esteem is highly sensitive to changes in perceived inclusionary status. Failure to demonstrate an integral link between per- ceived inclusion-exclusion and self-esteem would provide dis- confirmatory evidence for the sociometer perspective. On the other hand, converging support from these five studies would lend credence to the notion that the self-esteem system is in- volved in the monitoring of the quality of one's social bonds vis- a-vis inclusion-exclusion.
Study 1: Self-Feelings and Anticipated Inclusion- Exclusion
At minimum, the sociometer model of self-esteem predicts that the effects of people's own behavior on their self-feelings should closely parallel the degree to which they expect those behaviors may result in rejection or exclusion. Casual observa- tion suggests that this is the case: Events that raise self-esteem (e.g., achievement, recognition, compliments, being helpful, being loved) tend to be associated with improvement in the in- dividual's chances of being accepted and included, whereas events that lower self-esteem (e.g., failure, moral violations, possession of socially undesirable attributes, rejection) are as- sociated with decreased inclusion likelihood.
To test this fundamental prediction, participants in Study 1 indicated how they would expect other people to react to each of several behaviors. In addition, they rated how they would feel about themselves if they had personally performed each of these actions. The sociometer model predicts that these sets of ratings should be positively correlated.
Method
Participants. Seventy-five male and 75 female undergraduate stu- dents served as participants in fulfillment of a course research requirement.
Procedure. Participants completed two questionnaires that were embedded in a much longer instrument. Each questionnaire described 16 behaviors that varied in social desirability (e.g., I lost my temper; I cheated on a final exam; I donated blood; I saved a drowning child). One questionnaire asked respondents to indicate on 5-point scales how they thought others would react toward them if they had performed each behavior (1 = many other people would reject or avoid me, 5 = many other people would accept or include me).
The second questionnaire asked respondents to rate on four 7-point bipolar adjective scales how they personally would feel about themselves if they performed each behavior (i.e., good-bad, proud-ashamed, valu- able-worthless, happy-dejected). Ratings on these adjectives were
summed to provide an index of self-esteem feelings resulting from each behavior; higher ratings indicated more positive self-feelings.
To minimize the extent to which participants' might try to be consis- tent in their responses to the two questionnaires, (a) the items were presented in a different random order on each questionnaire, (b) differ- ent response formats were used (5-point vs. 7-point scales), and (c) the questionnaires were separated by several unrelated instruments that took approximately 30 min to complete. One half of the respondents completed the inclusion-exclusion ratings first, whereas the others com- pleted the self-feelings ratings first.
Results
Each of the four-item self-feelings scales demonstrated an ad- equate degree of interitem reliability (Cronbach's alpha was greater than .70 for all 16 behaviors). Also, considerable agreement was obtained across participants regarding the rela- tive orderings of the 16 situations. Kendall's coefficient of con- cordance was .79 for ratings of inclusion-exclusion and .87 for self-reported self-feelings.
The canonical correlation between respondents' ratings of others' reactions (inclusion-exclusion) and their own feelings of resultant self-esteem across all 16 situations was .70. That is, expectations of the degree to which one's behaviors would result in social inclusion-exclusion correlated highly with the impact of those behaviors on feelings of self-esteem.
On an item-by-item basis, the correlations between expecta- tions of social inclusion-exclusion and state self-esteem across the 16 situations ranged from . 14 to .47, with an average of .32. As can be seen in Table 1, the order of respondents' ratings was virtually identical on the two sets of questions. Thus, the effects of performing the 16 behaviors on self-esteem closely mirrored their effects on others' expected reactions.
Discussion
Participants' ratings of their self-feelings after performing 16 behaviors mirrored their expectations regarding how others would respond to these behaviors. Across all 16 situations, rat- ings of others' expected reactions accounted for nearly 50% of the variance in self-feelings. The correlations for each of the 16 situations ranged from small to moderate, but it must be remembered that they were based on single-item measures of unknown reliability. These data are open to alternative expla- nations but are consistent with the proposition that self-esteem feelings serve as an internal index of the degree to which one's behavior is likely to result in inclusion versus exclusion.
Study 2: Personal Experiences Involving Reactions to Exclusion
One weakness inherent in Study 1 is that participants re- sponded to hypothetical target behaviors. As a result, some re- spondents would not have experienced many of these behaviors, either as a participant or as an observer. In the absence of direct experience, respondents may have relied on their personal as- sumptions about how people would react to such behaviors. To the extent that these assumptions may have been based on their own personal evaluations of the actions, we would obtain a spu- rious correlation between respondents' ratings of how others
522 LEARY, TAMBOR, TERDAL, AND DOWNS
Table Study 1: Perceived Effects of the 16 Behaviors on Others' Reactions and Self-Esteem
Rank order of ratings
Inclusion-exclusion Self-feelings Item
1 2 3 4 5 6 7 8 9
10 11 12 13 14
15 16
2 1 3 4 6 5 7 8 9
11 12 10 13 14
16 15
.42
.17
.27
.34
.19
.36
.46
.21
.33
.25
.34
.33
.14
.47
.26
.33
I cheated on a final exam in a course. I carelessly caused a traffic accident in which someone was permanently paralyzed. I dropped out of college. I was unfaithful to my boyfiend or girlfriend. I lost my temper and yelled at someone. I received a negative evaluation on my work performance from my boss. I accidentally sneezed on someone standing in front of me in a checkout line. I took care of a friend's houseplants while she was out of town. I gave a dollar to a begger. I volunteered to donate blood. I was accepted into an honor society. I was voted "best-looking" person in my class. I was a Big Brother or Big Sister to an underpriviledged child. As president of a campus organization, I was responsible for raising $ 15,000 to buy
food and Christmas toys for abandoned children. I saved a drowning child who had fallen into a pool. I donated one of my kidneys to a dying person.
* Correlation between ratings of expected inclusion-exclusion and resultant self-feelings.
would react and their own self-feelings. As a first step in coun- teracting this alternative interpretation of the findings, Study 2 examined the relationship between exclusion and self-esteem in situations that respondents had actually experienced.
To avoid producing demand characteristics that would focus respondents' attention on inclusion and exclusion specifically, we asked the respondents to write essays describing the last time they experienced situations that involved either positive or neg- ative emotions. They then answered questions concerning their reactions to the occasion they described; embedded in these questions were ratings involving perceived exclusion and self- feelings. We predicted that their ratings of how included versus excluded they felt in the situations they described would corre- late highly with their self-esteem feelings in these situations.
Method
Participants. Eighty male and 80 female undergraduates enrolled in introductory psychology courses received credit toward a class re- quirement for their participation.
Procedure. Participants were randomly assigned to write a para- graph about the last occasion on which they experienced one of four negative emotional responses (i.e., social anxiety, loneliness, jealousy, or depression) or the opposite, positive pole of one of these responses— "feeling particularly at ease in a social situation" (nonanxious), "satis- fied about having as many close friends and family as desired" (nonlonely), "particularly secure in a relationship" (nonjealous), or "particularly happy in response to a social situation or relationship" (nondepressed). We felt that writing about interpersonal situations that produced negative and positive affect would lead participants to write about situations that varied in inclusion-exclusion (without cuing par- ticipants into the fact that this was the focus of the study). Each respon- dent wrote an essay about one of these eight types of experiences.
After completing the paragraph, participants rated how they felt on five 7-point scales intended to measure how included or excluded they felt in the situation (i.e., accepted, excluded, welcomed, rejected, included); higher ratings were associated with greater feelings of social
exclusion. A second set of 15 unipolar scales asked respondents how they felt about themselves on the occasion they described (i.e., good, adequate, attractive, inferior, ashamed, bad, socially desirable, popular, likable, proud, worthless, superior, confident, valuable, and competent, each paired with its opposite); higher numbers reflected more positive self-feelings.
Participants were also asked to state in a phrase the primary reason why they felt as they did in the situation they described. This phrase was later coded to determine the extent to which it involved factors relevant to inclusion and exclusion (1 = clear evidence of inclusion, 2 = proba- ble inclusion, 3 = unclear or not relevant to inclusion-exclusion, 4 = probable exclusion, 5 = clear evidence of exclusion).
Respondents also rated their emotional reactions on 7-point scales in- tended to measure anxiety (i.e., relaxed, tense, anxious, calm, nervous), loneliness (i.e., lonely, popular, involved, lonesome, isolated), jealousy (i.e., trusting, jealous, secure, possessive, suspicious), and depression (i.e., cheer- ful, happy, sad, depressed, gloomy).
Results
Cronbach's alpha for the measure of perceived exclusion was .94, and alpha for the index of self-feelings was .95. The Pearson product-moment correlation between ratings of perceived ex- clusion and self-feelings was calculated separately for partici- pants who wrote each of the kinds of essays (i.e., social anxiety, loneliness, jealousy, and depression). As can be seen in Table 2, their ratings of how excluded they felt in the situation correlated highly with how they felt about themselves in that situation (-.68 < rs < -.92, ps < .001). In addition, their attributions for their feelings clearly reflected the effect of exclusion. Attri- butions relevant to exclusion, coded from answers to the open- ended question, correlated as expected with self-feelings (—.38
Cronbach's alphas for the four affective measures each ex- ceeded .84. Respondents' affective ratings {scored so that higher numbers indicated negative affect) also correlated highly with
SELF-ESTEEM FEELINGS 523
Table 2 Study 2: Correlations Between Situational Self-Feelings and Perceived Exclusion
Essay type Perceived exclusion Attribution to exclusion
Social anxiety Loneliness Jealousy Depression
- . 6 8 * * - . 9 2 * * - . 8 3 * * - . 8 0 * *
- . 3 8 * - . 6 8 * * -.64** - . 4 5 *
/?<.00l.
perceived exclusion (.52 < r < .97, ps < .001) and with attribu- tions to exclusion (.23 < r < .76, p% < .08). These data show that changes in inclusionary status were accompanied by affec- tive changes as well.
Discussion
Respondents' retrospective accounts of personal experiences involving positive or negative affect again showed a strong rela- tionship between perceived exclusion and self-feelings. The more excluded respondents reported they felt in each type of situation, the less positively they indicated they felt about them- selves in that setting. Importantly, in each instance, the magni- tude of the correlation approached the reliability of the scales. For all practical purposes, self-feelings were a proxy for per- ceived exclusion.
Furthermore, their reports of why they felt as they did cor- roborated this finding. Attributions that invoked interpersonal exclusion as an explanation correlated with participants' self- feelings in each type of situation.
Study 3: State Self-Esteem in Reaction to Exclusion From a Group
The first two studies showed clearly that self-esteem feelings are strongly tied to perceived social exclusion. However, the cor- relational nature of both of these studies leaves open alternative explanations other than that perceived exclusion causes self-es- teem to decrease. In particular, both studies are open to the ex- planation that people who evaluate themselves positively may assume that others will like and accept them, whereas those with lower self-esteem are primed to perceive others' behaviors as rejecting (Alloy, 1988; Beck, 1967).
To directly examine the causal effects of exclusion on self- esteem, we experimentally manipulated social inclusion-exclu- sion in a third study. In this experiment, respondents were in- formed that they were either included or excluded from a labo- ratory work group. In addition, respondents were told that this selection was based either on a random procedure or on the preferences of other group members. According to the socio- meter hypothesis, when the selection is based on others' prefer- ences, and thus reflects personally on the individual, respon- dents should demonstrate increased self-esteem when they are included and decreased self-esteem when they are excluded, compared with random inclusion and exclusion.
Method Participants. One hundred twelve male and female undergraduates
recruited from introductory psychology courses served as participants and received required credit for their participation.
Procedure. Five respondents participated in each experimental ses- sion, but they reported to separate locations and were brought individ- ually to separate cubicles in the same laboratory to limit interaction among them throughout the study. (Data from 3 respondents were dis- carded to create equal cell sizes.)
Respondents were told that the study involved group and indi- vidual decision making. To begin, participants completed a brief "information exchange questionnaire" that they were told would be shown to the other 4 participants. This questionnaire asked them to rate themselves on thirteen 7-point scales (e.g., open-closed, tense-relaxed, athletic-nonathletic) and to write two short essays on "what it means to be me" and "the kind of person I would most like to be." These questions were designed to provide participants with information that would provide the basis for their subsequent evaluations of one another. After they finished, the researcher circulated each participant's ques- tionnaire to all of the other participants.
After viewing the other participants' responses, each participant com- pleted a form on which he or she rated the other 4 respondents. They also indicated which 2 of the other participants they would most want to work with and which 2 could be most relied on in time of trouble, and they rank ordered the other participants in terms of who they would most want to work with later in the study.
After taking time to ostensibly collate respondents' ratings of one an- other, the researcher distributed sheets that made task assignments. First, respondents were told that 3 of the 5 participants would work together on the decision-making problems as a group and that the other 2 participants would work on the same problems individually. They were then told that (a) they either would work as a member of the 3- person group (included condition) or would work alone (excluded condition) and ( b ) this selection was based either on the other members' preferences (based on the rating sheets that respondents completed earlier) or on a random procedure. (The random procedure was justi- fied by noting that the researcher wanted to assign participants to groups irrespective of their preferences.) Thus, the experimental design was a 2 (included in vs. excluded from the group) X 2 (assignment based on others' preferences or random procedure) randomized factorial.
Respondents then completed a questionnaire that assessed their reac- tions and rated how they currently felt about themselves on twelve 7- point bipolar adjective scales. The adjectives, which were drawn from McFarland and Ross' (1982) low and high self-esteem feelings factors, were as follows: good, competent, proud, adequate, useful, superior, smart, confident, valuable, important, effective, and satisfied, each paired with its opposite. On half of the items, the positive pole was on the left end of the scale, and on half it was on the right. Participants also rated how excluded they felt on three 7-point scales (i.e., included- excluded, rejected-accepted, welcomed-avoided).
Respondents were also asked to rate the other respondents (as a group) on seven 7-point scales (i.e., good, competent, adequate, useful, smart, valuable, and likable, each paired with its opposite) and to indi- cate how much they had wanted to be selected for the 3-person group.
Finally, manipulation check questions asked (a) how the assignments to the experimental tasks were made and (b) whether the participant was assigned to work alone or with the group. Respondents were then fully debriefed, and the rationale for the study and all deceptions were explained.
Results
Manipulation checks. All but 3 of the 112 participants ac- curately reported whether they were assigned to work with the
524 LEARY, TAMBOR, TERDAL, AND DOWNS
3-person group or alone. Thus, the manipulation of task assign- ment was apparently effective.
Twelve participants incorrectly answered the question that asked how assignments to the group were made. Examination of the pattern of errors revealed a roughly equal number of errors across conditions. Analyses conducted with and without these 12 participants did not produce different findings.
Self-feelings. Cronbach's alpha for the self-feelings scale was .84. A 2 X 2 X 2 (Inclusion-Exclusion X Mode of Assign- ment X Gender) analysis of variance (ANOVA) performed on the mean of participants' self-ratings revealed a significant main effect of inclusion-exclusion, F(\, 104) = 9.36, p < .01, that was qualified by a significant Inclusion-Exclusion X Mode of Assignment interaction, F( 1,104) = 9.60, p < .01. Tests of sim- ple main effects showed that inclusion-exclusion affected par- ticipants' self-esteem feelings only if it was attributable to ac- ceptance or rejection by the group, F( 1, 104) = 18.55,p< .05 (see Table 3).
Furthermore, the pattern of data suggests that this effect was attributable to the effects of exclusion rather than inclusion. Whereas participants felt no better about themselves when the group included them than when they were included randomly, F( 1, 104) = 1.42, p > .20, respondents who thought they were excluded on the basis of the group's preferences rated them- selves significantly more negatively than those who believed they had been randomly excluded, F ( l , 104)= 10.00,p< .01.
A main effect of participant sex, F( 1, 104) = 4.01, p < .05, and an Inclusion-Exclusion X Sex interaction were also ob- tained on participants' self-ratings, F( 1, 104) = 6.63, p < .01. Tests of simple effects revealed that, although men (M = 5.6) and women (M = 5.6) rated themselves identically when they were included in the group, women (M = 4.8) rated themselves significantly less positively than did men (M = 5.5) following exclusion (p < .05). Interestingly, this effect was not qualified by mode of assignment. Finally, participants' feelings about themselves correlated strongly with their feelings of being ex- cluded (r = -.75, p < .001).
Ratings of the other participants. Ratings of the other 4 re- spondents also showed a main effect of inclusion-exclusion, F(l, 104) = 14.18, which was qualified by the two-way Inclu- sion-Exclusion X Mode of Assignment interaction, F( 1, 104) = 7.96, ps < .01. As shown in the second line of Table 3, re- spondents rated one another significantly less positively when they believed they had been excluded by the group rather than
when the group had included them, F( 1, 104) = 23.41, p < .001. When the assignment was random, participants' ratings of one another were unaffected by whether they were assigned to work with the group, F(l, 104) = 0.47, p > .40. In addition, participants derogated one another more when exclusion was based on group members' preferences rather than on a random selection procedure, F( 1, 104) = 8.67, p < .01. The more ex- cluded respondents felt, the more negatively they rated the other participants (r = - . 5 2 , p < .001).
Retrospective inclusion motivation. An Inclusion-Exclu- sion X Mode of Assignment interaction was also obtained on the item that asked respondents, "How much did you want to be selected for the 3-person 'central' group?" although the effect just failed to reach the .05 level of significance, F(\, 104) = 3.61, p < .06. As can be seen in the third line of Table 3, a sour- grapes effect was obtained: Participants who thought the group had included them retrospectively indicated a greater desire to be included than those who thought the group had excluded them, F( 1, 104) = 4.74, p < .03. The simple effect of inclusion- exclusion was not significant when selection was ostensibly ran- dom, F( 1, 104) = 0.24, p > .60.
Discussion
The central conclusion to be drawn from the findings of Study 3 is that exclusion that reflects rejection by others pro- duces strong effects on self-feelings and social perceptions. Compared with respondents who thought their peers had se- lected them to participate in the group, those who thought the group had excluded them rated themselves more negatively, more strongly derogated the other group members, and claimed less interest in being a member of the group. By contrast, inclu- sion and exclusion had no discernible effect on respondents' re- sponses when they were based on a random selection procedure. Furthermore, the data showed that exclusion had a stronger effect in lowering respondents' self-feelings than inclusion had in raising them.
These data suggest that exclusion that implies disapproval or rejection results in lowered self-esteem even in contexts in which social inclusion has no identifiable implications for peo- ple's well-being (e.g., survival, assistance, or comfort). In fact, given the experimental situation, we were surprised that respon- dents responded as strongly as they did to being excluded by the group. Participants had neither a history of previous experience
Table 3 Study 3: Effects of Exclusion and Mode of Assignment
Variable
Self-feelings Ratings of other participants Desire to be included
Included
5.5 5.6 5.1
Mode of assignment
Random
Excluded
5.5. 5.5. 5.6
Group choice
Included
5.7b 5.9b 6.5a
Excluded
4.8.b 4.9.b 4.9.
Note. Means in a single row that share a common subscript differ significantly by tests of simple main effects, p<. 05.
SELF-ESTEEM FEELINGS 525
nor expectations of future interaction (beyond this study) with other group members; they did not even know who the other members were. In addition, it was not clear that working on the decision-making problems with the group was in any way more desirable than working alone, and the basis on which respon- dents were excluded was limited and superficial. Yet, those who were rejected by the group suffered a decrease in state self-es- teem. These findings attest to the strength of people's desire to avoid disapproval and rejection in the absence of any tangible benefits of being accepted.
They may also shed light on people's motivation to maintain and foster their connections to seemingly meaningless groups. As research using the minimal in-group paradigm shows, even when people are "members" of a group in name only, they come to identify with the group and its members (e.g., Tajfel, 1981). Given that the motivation toward social inclusion and away from social exclusion is potent, maintained by an inner socio- meter that monitors inclusionary status, even a minimal sense of"belongingness" maybe rewarding.
The effects of exclusion on ratings of the other group mem- bers comes as no surprise, but it deserves attention from the perspective of the sociometer hypothesis. Traditional ap- proaches to self-esteem would explain that, by derogating those who rejected them, individuals could minimize the importance or validity of the others' evaluations and thereby protect their own self-esteem. People's self-esteem is less likely to be dam- aged if they can convince themselves that those who rejected them were socially undesirable people they did not want to as- sociate with anyway. We have no complaints with this explana- tion as far as it goes. Given that perceived exclusion is anxiety producing, once permanent rejection is detected, people may indeed try to reduce their distress through cognitive means, such as by derogating the rejector or minimizing the impor- tance of acceptance.
However, the sociometer model suggests two additional ex- planations. First, the derogation of sources of rejection may re- flect an interpersonal tactic aimed at others who are present. Being rejected calls one's social acceptability into question in the eyes of others who are privy to the rejection. By lambasting the rejector, the individual may lead others to ignore or discount the rejection. (If a person who was recently rejected by a roman- tic partner praises the ex-partner's judgment and social quali- ties, new acquaintances may conclude that such a wonderful person had good reasons for dumping him or her.) Other evi- dence on self-serving reactions to failure and negative evalua- tion shows that such reactions are sometimes for the benefit of others (Leary & Forsyth, 1987; Schlenker, 1980). In Study 3, rejected participants might have derogated the other group members to convince the researcher that they did not deserve to be excluded.
Second, from a practical standpoint, continuing to seek in- clusion by those who have excluded the individual is not an op- timal strategy. When rejection is permanent and irreversible, as it was in this study, the person should turn his or her attention away from the rejectors and toward those who may be more accepting. Focusing on the desirability of being included by a rejecting group may actually impede the person's general suc- cess in establishing and maintaining connections with other people. Thus, derogating the group and minimizing the impor-
tance of its acceptance may be adaptive in terms of facilitating one's ability to move on to other groups and relationships. All three of these processes operate to produce the derogation effect obtained here, and we have no way to choose among them. However, they provide a ripe source of hypotheses for future research.
Overall, male and female respondents reacted similarly to the experimental manipulations. The only gender difference ob- tained showed that women who were excluded rated themselves less positively than did men irrespective of whether they were excluded randomly or because of other respondents' prefer- ences. Although obtained on only a single measure, this finding suggests that women may be more sensitive than men to cues that connote exclusion, possibly because typical patterns of so- cialization in American culture lead them to be more attuned to others' reactions (Snodgrass, 1985) or more motivated to emphasize communal relationships (Eagly & Wood, 1991). If, indeed, this is a general finding, research is needed to explore the sources of gender differences in reactions to exclusion.
Study 4: Interpersonal Exclusion and Self-Esteem Feelings
Study 3 provided concrete evidence that social exclusion re- sults in lowered self-esteem, at least when the exclusion was based on others' personal evaluations and preferences, and this effect occurred even when exclusion has no notable implica- tions for the individual. The purpose of Study 4 was to concep- tually replicate and extend these findings using a somewhat different paradigm and different measures.
In this experiment, participants provided information about themselves via an intercom to an anonymous participant in an- other room. They then received feedback from the other partic- ipant that connoted either inclusion and acceptance or exclu- sion and rejection, or else they received no feedback relevant to inclusion-exclusion. Participants then rated their feelings about themselves on a questionnaire that they believed would be seen by either the same participant who had listened to them previously (and who, in two conditions, had ostensibly provided his or her feedback) or to a new participant.
This latter manipulation was included to examine the possi- bility that the effects of inclusion-exclusion on self-ratings were mediated by self-presentational rather than self-esteem pro- cesses. If respondents know their self-ratings will be seen by someone who is aware of the fact they were previously included or excluded, they may use their self-ratings as a self-presenta- tional strategy in an attempt to support or counteract the prior effects of inclusion or exclusion on their social image. This pos- sibility would be detected if participants' self-ratings after inclu- sion or exclusion differed as a function of who would be seeing their ratings.
Method
Participants. Forty-five male and 45 female undergraduates served as subjects in return for required credit in an introductory psychology course.
Pretesting. As part of a large mass testing procedure conducted early in the semester, all participants rated themselves on 12 evaluatively laden adjectives: cheerful, absent-minded, honest, clear thinking, de-
526 LEARY, TAMBOR, TERDAL, AND DOWNS
ceitful, friendly, forgetful, dependable, arrogant, intelligent, prejudiced, and irresponsible. Ratings were done on 12-point scales with five equally spaced scale labels (not at all, slightly, moderately, very, and extremely). These ratings were used as a pretest measure of self-feelings.
Experimental session. Each session used a mixed-sex pair of partic- ipants who went to different locations to maintain anonymity. They were informed that the study was concerned with how people form im- pressions of others. They would be asked to talk into a microphone about themselves while another respondent of the other sex listened. After the participant signed an informed consent form, the researcher gave each participant a personal information sheet ostensibly com- pleted by the other participant in the session. This information was pro- vided to convince the respondent of the presence of the other respon- dent and involved innocuous demographic information.
Participants then spoke into a microphone for 5 min about topics drawn from a standard list, believing that the other participant was lis- tening. These topics were intended to be moderately disclosing so that the participant would discuss enough personal information for the other person to ostensibly make a personal appraisal. For example, one ques- tion asked participants to describe aspects of themselves they liked best and least.
After the 5-min verbal presentation, participants were randomly as- signed to receive feedback indicating that the other person either liked, accepted, and wanted to interact with them (inclusion condition); to receive feedback indicating that the other participant did not particu- larly like, accept, or want to interact with them (exclusion condition); or to receive no feedback from the other participant (no-feedback condition). In the inclusion and exclusion conditions, the feedback sheet participants received contained ratings on a number of dimen- sions that connoted inclusion and exclusion. For example, one question asked whether the listener would want to continue a conversation with the participant, and another asked whether the listener would want to introduce the participant to a friend. In response to each question, the other participant had ostensibly marked "yes," "no," or "unsure."
It should be noted that although respondents in the inclusion condi- tion received predominately accepting feedback (with a couple of "un- sure" responses marked), those in the exclusion condition received pre- dominately "unsure" responses (with a couple of rejecting answers) to minimize the aversiveness of the manipulation. We felt that uncertain and ambivalent responses would connote sufficient rejection for purposes of the study. To preserve the illusion that the researcher was ignorant of the ratings, this feedback was provided in an envelope, and participants were told not to read the ratings until the researcher left the room.
After reading the feedback, participants were asked to complete a questionnaire about themselves that would ostensibly be shown to ei- ther the participant who had heard and evaluated them or to another participant. Participants rated themselves on the same 12 self-relevant adjectives they had provided during mass testing several weeks earlier using 12-point scales.
On a questionnaire that respondents were told only the researcher would see, they indicated the degree to which the other respondents' perceptions of them were accurate (1 = not at all, 12 = extremely). To assess the effectiveness of the feedback manipulation, they also indi- cated how positively or negatively the other respondent regarded them (1 = extremely negative, 12 = extremely positive). After completing the experimental questionnaires, participants were fully debriefed, with all deceptions explained in detail.
Results
Manipulation check. An ANOVA conducted on partici- pants' ratings of the feedback they received revealed that the manipulation of inclusion-exclusion was highly successful,
F( 1, 86) = 8.76, p < .01. A Tukey's test revealed that partici- pants who received positive feedback reported that they were perceived most positively (M = 11.4), followed by participants receiving no feedback (M = 7.9) and those who received reject- ing feedback (M = 2.4; all ps < .05).
Self-feelings. Examination of the interitem reliability of the 12 self-ratings revealed that one item (intelligent) had an unac- ceptably low item-total correlation. After this item was dropped, Cronbach's alpha coefficient for the remaining 11 items was .79.
A 3 (feedback: acceptance, rejection, none) X 2 (target: same vs. other respondent) ANOVA was conducted on the sum of the 11 self-ratings after reverse scoring the negatively worded items. This revealed only a significant main effect of feedback, F(2, 90) = 4.93, p < .01. Inspection of condition means revealed that respondents receiving accepting feedback subsequently rated themselves more positively (M = 106.5) than did those who received no feedback (M = 103.2) and rejecting feedback (M = 104.4).
To explore the extent to which respondents' self-esteem devi- ated from their "typical" self-feelings following inclusion and exclusion, difference scores were calculated between the sum of the self-ratings obtained during mass testing and the sum of the self-ratings obtained during the experiment itself. A 3 X 2 ANOVA indicated a significant main effect of feedback, F(2, 91) = 5.95, p < .01. A Tukey's test revealed that rejected partic- ipants' self-feelings were significantly more negative relative to the ratings they gave during mass testing (mean difference = —5.9) compared with accepted participants (mean difference = 2.0, p < .05). Participants who received no feedback did not differ from the other two groups (mean difference = -1.4, ps > .05).
The / tests comparing the mean difference score in each feed- back condition with a score of zero (which would reflect "no change" from mass testing) indicated that although partici- pants who were rejected rated themselves significantly more negatively than they rated themselves during mass testing, t( 32) = 3.64, p < .05, the ratings of accepted and no-feedback partic- ipants did not differ from their mass testing ratings (ps > .10). Thus, rejection significantly lowered self-feelings, but accep- tance did not significantly raise them.
Accuracy ratings. Respondents' ratings of the accuracy of the other respondents' perceptions of them were also signifi- cantly affected by the feedback they received, F(2, 86) = 4.60, p < .01. A Tukey's test showed that accepted respondents be- lieved the other respondent to be significantly more accurate (M = 8.2) compared with respondents who were rejected (M = 3.6) and those who received no feedback (M = 5.5, ps < .05). However, the rejected and no-feedback conditions did not differ significantly (p > .05).
Discussion
The results of Study 4 provide a conceptual replication of the primary findings of Study 3 using a different paradigm, cover story, and measures: Those who were accepted on the basis of personal reasons subsequently felt more positively about them- selves than did those who were excluded. Furthermore, as in Study 3, exclusion had a notably stronger effect in lowering par-
SELF-ESTEEM FEELINGS 527
ticipants' self-esteem than inclusion had in enhancing their self- feelings. As we discuss in the General Discussion section, these findings suggest that the self-esteem system may be more sensi- tive to decrements than increments in inclusionary status.
The effects of feedback on self-feelings described earlier were obtained even though participants in the exclusion condition explicitly dismissed the feedback. Respondents who were re- jected rated the other person's perceptions as highly inaccurate, but their feelings about themselves were affected nonetheless. This finding suggests that people need not view exclusion as warranted in order for it to affect self-esteem. Although it stands to reason that self-esteem would suffer when people are rejected because of their actual transgressions and shortcomings, the fact that self-esteem fell even when respondents dismissed the feed- back as inaccurate provides further support that self-esteem is sensitive to others' reactions per se.
The fact that the identity of the target for whom respondents rated themselves did not moderate the effects of exclusion on self-feelings suggests that the findings of Studies 3 and 4 are un- likely to be attributable to participants' attempts to convey cer- tain impressions of themselves to those who had accepted or rejected them. Although it is impossible to prove the null hy- pothesis, the failure to find an effect of target at least renders such an interpretation implausible.
Study 5: Individual Differences in Self-Esteem
In each of the first four studies, we were interested in the effects of social inclusion and exclusion on people's state self- esteem in a given social setting. However, a corollary of the so- ciometer model is that individual differences in trait self-esteem should be related to individual differences in the extent to which people generally feel that they are socially included versus ex- cluded. On one hand, a history of real or perceived exclusion may ultimately result in lowered trait self-esteem (Harter, 1993; Shrauger & Schoeneman, 1979). Furthermore, once formed, self-esteem may color people's perceptions of others' reactions. People with low self-esteem may be more likely to perceive oth- ers' reactions as rejecting than people with high self-esteem.
Method
Participants. Two hundred twenty male and female undergraduates participated in the study to fulfill a requirement for their introductory psychology course.
Perceived inclusionary status. A scale to measure individual differ- ences in perceived inclusionary status was constructed and pilot tested on a sample of 150 respondents. This measure consisted of nine items that assessed the extent to which individuals feel they are generally in- cluded versus excluded by others. Examples included, "People often seek out my company," "I often feel like an outsider in social gather- ings," and "If I want to socialize with my friends, I am generally the one who must seek them out." Cronbach's alpha was .77 in pilot testing. Although unpublished, this scale has demonstrated usefulness in previ- ous research (Miller, in press).
Self-esteem. General self-esteem was measured with two scales. Ro- senberg's (1965) Self-Esteem Scale is a )0-item scale that has high in- ternal consistency ( a = .85) and test-retest reliability (.85) and is per- haps the most widely used measure of dispositional self-esteem.
In a factor analysis of self-relevant mood items, McFarland and Ross (1982) found that the following items loaded on a self-esteem feelings
factor: proud, competent, confident, smart, resourceful, effective, effi- cient, inadequate, incompetent, stupid, worthless, and shameful. Thus, the sum of these items (after reverse scoring negatively worded items) was used as a second measure of self-esteem.
Procedure. Each of the scales just described was administered dur- ing two separate sessions. Regardless of the original response format, participants answered all items on 5-point scales.
Results
Cronbach's alpha was acceptable for all three measures: per- ceived inclusionary status (.80), Rosenberg self-esteem (.88), and McFarland and Ross self-feelings scale (.91). The two mea- sures of self-esteem correlated .75.
The Pearson product-moment correlation between per- ceived exclusionary status and Rosenberg self-esteem scores was -.55 (p < .001). The correlation between perceived exclusion- ary status and the McFarland and Ross measure of self-feelings was-.51(p<.001).
Discussion
As predicted by the sociometer model, the degree to which people think they are generally excluded versus included corre- lated moderately with two different measures of trait self-es- teem. Whereas Rosenberg's (1965) measure asks respondents to indicate their agreement or disagreement with 10 statements about their self-perceived worth, McFarland and Ross's (1982) items consist of self-descriptive adjectives. Although the corre- lational nature of the data does not rule out the possibility that self-esteem feelings mediate perceptions of inclusion rather than vice versa, the data further support the hypothesized link between perceived exclusion and self-esteem.
To extend our analogy of a fuel gauge, trait self-esteem may be conceptualized as the typical or average resting position of the "indicator needle" on the person's sociometer. This position reflects the person's perception of his or her inclusionary status in the absence of explicit cues connoting inclusion or exclusion. As noted, the relationship between perceived exclusion and trait self-esteem is probably reciprocal. A history of exclusion may lead to low trait self-esteem, and having low trait self-esteem predisposes the person to perceive rejection more readily. Be- cause their sociometers are calibrated differently, people with very low trait self-esteem may perceive others as rejecting most of the time, whereas those with higher self-esteem generally feel they are being accepted.
General Discussion
Taken together, these five studies provide converging evidence for the hypothesized relationship between perceived social ex- clusion and self-esteem. In Study 1, we found that participants' self-feelings varied with how they thought others would react to various behaviors vis-a-vis inclusion-exclusion. Study 2 showed that respondents' retrospective reports of how they felt about themselves in recent social encounters correlated highly with how included versus excluded they felt in those situations. Stud- ies 3 and 4 used experimental designs to demonstrate that ex- clusion by other people results in lower state self-esteem than inclusion. Study 5 showed that the degree to which people gen-
528 LEARY, TAMBOR, TERDAL, AND DOWNS
erally believe that others include versus exclude them correlated negatively with two different measures of trait self-esteem. Al- though alternative explanations may be offered for some of these findings, we believe that the consistency of our findings across widely disparate studies, using different paradigms and mea- sures of self-feelings, provides converging support for the hy- pothesized link between perceived social exclusion and self- esteem.
Of course, an empirical demonstration of this relationship does not necessarily indicate that the function of the self-esteem system is to monitor social exclusion; such a functional expla- nation is difficult to test directly. Even so, we believe that the sociometer model provides a parsimonious explanation of both our results and previous findings. State self-esteem appears to function as a subjective marker that reflects, in summary fash- ion, the individual's social standing in a particular social setting and thus serves to apprise the individual of changes in his or her inclusionary status (Leary, 1990; Leary & Downs, 1995). In essence, one function of self-esteem may be to provide a rela- tively fast and automatic assessment of others' reactions vis-a- vis inclusion and exclusion. Such an ongoing inclusion assess- ment mechanism would enhance the individual's likelihood of establishing and maintaining supportive social relationships and of avoiding social exclusion (Baumeister & Tice, 1990).
From the earliest days of psychology and sociology, theorists interested in the self have suggested that people's self-images, as well as their self-esteem, are based heavily on their perceptions of the evaluative reactions of other people. In particular, sym- bolic interactionists have long maintained that one's self-per- ceptions reflect others' perceptions of and reactions to the indi- vidual (Cooley, 1902; Mead, 1932; see Shrauger & Schoene- man, 1979). The sociometer perspective shows clearly why this is the case. The self-esteem system serves its primary function only if it is sensitive to others' reactions.
In everyday life, inclusion-exclusion and interpersonal eval- uations are highly confounded. We tend to associate with those we regard positively while avoiding those we regard negatively. To the extent that social evaluations are closely related to inclu- sion and exclusion (Baumeister & Tice, 1990), people's self- esteem is often affected by evaluative feedback. Yet, we believe that the degree to which others appear to include versus exclude the individual, rather than the nature of others' evaluations per se, is the most important determinant of self-esteem. Further- more, we believe that, in behaving in ways that promote self- esteem, people are striving to enhance their inclusionary status rather than to be evaluated positively per se. We find it easy to understand why a potent mechanism to increase inclusion would have evolved among humans but more difficult to un- derstand why a motive to be perceived positively would have developed.
Throughout this article we have alternated between referring to self-esteem as a means of enhancing inclusion and as a means of avoiding exclusion. Although the data on this point are only suggestive, we believe that the sociometer system responds pri- marily, if not exclusively, to exclusion rather than to inclusion. First, at a conceptual level, most motivation and drive systems, both physiological and psychological, respond to deprivation states rather than to less-than-complete satiation. For example, people are far more motivated to avoid being hungry than they
are to remain full. Just as there is little merit in a system that constantly motivated a person to maintain a full stomach, there would be little reason for a psychological system to evolve that pushed a person toward greater and greater inclusion by in- creasing numbers of people. In fact, the excessive social respon- sibilities associated with multiple group memberships may be disadvantageous. The sociometer system would serve its pur- pose if it simply assured that the person maintained sufficient social connections with a relatively small set of personally sig- nificant people (Baumeister & Leary, in press).
Second, in both Study 3 and Study 4, respondents who thought they were excluded showed a decrement in self-esteem, but those who were included showed no corresponding incre- ment. Similarly, previous writers have discussed the asymmetry of positive and negative feedback. Although receiving positive reactions may be mildly pleasant, negative reactions carry far more weight. Not only does a slightly negative reaction have a much greater impact on most people than even a strongly posi- tive one, but a single negative reaction can counteract and undo a plethora of accolades. Although other explanations are possi- ble (e.g., because most interactions range from neutral to posi- tive, positive responses from others lack the saliency and diag- nosticity of negative ones), the sociometer hypothesis provides a parsimonious explanation of this pattern. Specifically, our psychological systems are designed to detect and place greater emphasis on reactions that connote exclusion than reactions that connote inclusion.
Kernis and his colleagues have recently shown that people's reactions to esteem-threatening events are moderated not only by their level of self-esteem but by its stability. Some people show little variation in self-esteem across situations and time, whereas other people's self-esteem is exceptionally labile. Peo- ple with unstable self-esteem, whether low or high, show more extreme emotional and behavioral reactions to events involving negative evaluations by other people and other threats to self- esteem (for a review, see Kernis, 1993). In our view, people with unstable self-esteem essentially have an unstable sociometer that overresponds to cues that connote acceptance and rejec- tion. For such people, minor changes in inclusion or exclusion result in large changes in the sociometer (and self-esteem). In extreme cases of unstable self-esteem, the fluctuations of the sociometer may be only minimally tied to real changes in in- clusionary status, much like a faulty gas gauge that registers "full" one minute, "half full" a few minutes later, then "three- quarters full" after that.
Several theorists have conceptualized the self as having at least two distinct facets, which are commonly labeled public and private^Baumeister, 1986; Buss, 1980; Carver &Scheier, 1981; Fenigstein, 1987; Greenwald & Breckler, 1985; Schlenker, 1985). For example, ego-task analysis theory (Greenwald, 1982; Greenwald & Breckler, 1985) suggests that different as- pects of the self are sensitive to different aspects of self-evalua- tion and perform different "tasks" in the service of protecting the ego. Whereas the public self "is sensitive to the evaluation of others and seeks to win the approval of significant outer audi- ences" (Greenwald & Breckler, 1985,pp. 132-133), the private self evaluates oneself on the basis of the individual's internalized standards. We concur that, once the self develops in childhood,
SELF-ESTEEM FEELINGS 529
people are able to evaluate themselves from the perspectives of both themselves and various other people.
However, we propose that this ability to perceive and evaluate oneself from varying perspectives does not necessarily involve the sociometer mechanism that we have described in this arti- cle. After all, people are able to evaluate all manner of stimuli, including themselves, and some of their self-judgments are ir- relevant to their feelings of self-esteem. As we have argued, the self-esteem system appears specifically designed to detect real or potential changes in the individual's inclusionary status and to elicit emotional and motivational processes in response to threats to one's connections with other people. Some events to which the sociometer responds are "private" ones involving thoughts and feelings that, if known by others, might jeopardize their inclusion in important groups and relationships, whereas other such events are "public" ones that others may easily ob- serve. Even so, the sociometer responds to both private and pub- lic events in terms of their potential effects on inclusion-exclu- sion. Thus, the sociometer and the self-esteem feelings it medi- ates are responsive to both private and public self-relevant events.
In summary, the self-esteem system appears to function as a sociometer designed to detect possible deleterious changes in people's inclusionary status. Furthermore, rather than serving primarily to maintain one's inner sense of self, the self-esteem motive prompts people to behave in ways that maintain their connections with other people.
References
Adler, A. (1930). Understanding human nature. New York: Greenberg. Ainsworth, M. D. S. (1989). Attachments beyond infancy. American
Psychologist, 44, 709-716. Alloy, L. (1988). Cognitive processes in depression. New York: Guilford
Press. Allport, G. W. (1937). Personality: A psychological interpretation. New
York: Holt. Baldwin, M. W., & Holmes, J. G. (1987). Salient private audiences and
awareness of self. Journal of Personality and Social Psychology. 52, 1087-1098.
Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behav- ioral change. Psychological Review. 84, 191-215.
Barash, D. P. (1977). Sociobiology and behavior. New York: Elsevier. Baumeister, R. F. (1986). Public self and private self. New York:
Springer-Verlag. Baumeister, R. F. (Ed.). (1993). Self-esteem: The puzzle of low self-
regard. New York: Plenum. Baumeister, R. F., & Leary, M. R. (in press). The need to belong: Desire
for interpersonal attachments as a fundamental human motivation. Psychological Bulletin.
Baumeister, R. F., & Tice, D. M. (1990). Anxiety and social exclusion. Journal of Social and Clinical Psychology. 9, 165-195.
Beck, A. (1967). Depression: Clinical, experimental, and theoretical aspects. Philadelphia: University of Pennsylvania Press.
Bednar, R. L., Wells, M. G., & Peterson, S. R. (1989). Self-esteem: Par- adoxes and innovations in clinical theory and practice. Washington, DC: American Psychological Association.
Blaine, B., & Crocker, J. (1993). Self-esteem and self-serving biases in reactions to positive and negative events: An integrative review. In R. F. Baumeister (Ed.), Self-esteem: The puzzle of low self-regard (pp. 55-85). New York: Plenum.
Bowlby, J. (1969). Attachment and loss: Vol. 1. Attachment. New York: Basic Books.
Brockner, J. (1983). Low self-esteem and behavioral plasticity. Some implications. Review of Personality and Social Psychology, 4, 2 3 7 - 271.
Brown, J. D. (1993). Self-esteem and self-evaluations: Feeling is believ- ing. In J. Suls (Ed.), Psychological perspectives on the self'(Vol. 4, pp. 2 7 - 5 8 ) . Hillsdale, NJ: Erlbaum.
Buss, A. H. (1980). Self-consciousness and social anxiety. San Fran- cisco: Freeman.
Carver, C. S., & Scheier, M. F. (1981). Attention and self-regulation: A control-theory approach to human behavior. New York: Springer- Verlag.
Cooley, C. H. (1902). Human nature and the social order. New York: Scribner.
Coopersmith, S. (1967). The antecedents of self-esteem. San Francisco: Freeman.
Crocker, J., Thompson, L., McGraw, K.., & Ingerman, C. (1987). Downward comparison, prejudice, and evaluations of others: Effects of self-esteem and threat. Journal of Personality and Social Psychol- ogy, 52, 907-916.
Cutrona, C. E. (1982). Transition to college: Loneliness and the process of social adjustment. In L. A. Peplau & D. Perlman (Eds.), Loneli- ness: A sourcebook of current theory, research, and therapy (pp. 2 9 1 - 309). New York: Wiley.
Eagly, A. H., & Wood, W. (1991). Explaining sex differences in social behavior: A meta-analytic perspective. Personality and Social Psy- chology Bulletin, 17, 306-315.
Epstein, S. (1973). The self-concept revisited: Or a theory of a theory. American Psychologist, 28, 404-416.
Fenigstein, A. (1987). On the nature of public and private self-con- sciousness. Journal of Personality, 55, 543-554.
Goswick, R. A., & Jones, W. H. (1981). Loneliness, self-concept, and adjustment. Journal of Psychology, 107, 237-240.
Greenberg, J., Pyszczynski, T., & Solomon, S. (1986). The causes and consequences of a need for self-esteem: A terror management theory. I n R . F. Baumeister (Ed.), Public selfand private self (pp. 189-207). New "York: Springer-Verlag.
Greenberg, J., Solomon, S., Pyszczynski, X, Rosenblatt, A., Burling, J., Lyon, D., Simon, L., & Pinel, E. (1992). Why do people need self- esteem? Converging evidence that self-esteem serves an anxiety- buffering function. Journal of Personality and Social Psychology, 63, 913-922.
Greenwald, A. G. (1980). The totalitarian ego: Fabrication and revi- sion of personal history. American Psychologist, 35, 603-613.
Greenwald, A. G. (1982). Ego task analysis: An integration of research on ego-involvement and awareness. In A. H. Hastorf & A. M. Isen (Eds.), Cognitive social psychology (pp. 109-147). New York: Elsevier.
Greenwald, A. G., & Breckler, S. (1985). To whom is the self presented? In B. R. Schlenker(Ed.), The self and social life (pp. 126-145). New York: McGraw-Hill.
Harter, S. (1993). Causes and consequences of low self-esteem in chil- dren and adolescents. In R. F. Baumeister (Ed.), Self-esteem: The puzzle of low self-regard (pp. 8 7 - 1 1 6 ) . New York: Plenum.
Harter, S., & Marold, D. B. (1991). A model of the determinants and mediational role of self-worth: Implications for adolescent depression and suicidal ideation. In G.Goethals& J.Strauss (Eds.), TheselfAn interdisciplinary approach. New York: Springer-Verlag.
Heatherton, T. F., & Ambady, N. (1993). Self-esteem, self-prediction, and living up to commitments. In R. F. Baumeister (Ed.), Self-es- teem: The puzzle of low self-regard (pp. 131-145). New \brk: Plenum.
Heatherton, T. F., & Polivy, J. (1991). Development and validation of a
530 LEARY, TAMBOR, TERDAL, AND DOWNS
scale for measuring state self-esteem. Journal of Personality and So- cial Psychology. 60, 895-910.
Hogan, R. (1982). A socioanalytic theory of personality. In M. Page (Ed.), Nebraska Symposium on Motivation (pp. 55-89). Lincoln: University of Nebraska Press.
Hogan, R., Jones, W. H., & Cheek, J. M. (1985). Socioanalytic theory: An alternative to armadillo psychology. In B. R. Schlenker (Ed.), The self and social life (pp. 175-198). New York: McGraw-Hill.
Homey, K. (1937). The neurotic personality of our time. New York: Norton.
James, W. (1890). Principles of psychology. New York: Dover. Jones, E. E., & Berglas, S. (1978). Control of attributions about the self
through self-handicapping strategies: The appeal of alcohol and the role of underachievement. Personality and Social Psychology Bulle- tin. 4.200-206.
Jones, S. C. (1973). Self- and interpersonal evaluations: Esteem theories versus consistency theories. Psychological Bulletin. 79, 185-199.
Kernis, M. (1993). The roles of stability and level of self-esteem in psy- chological functioning. In R. F. Baumeister (Ed.), Self-esteem: The puzzle of low • self regard (pp. 167-182). New York: Plenum.
Kernis, M. (1995). Efficacy, agency, and self-esteem. New York: Plenum.
Leary, M. R. (1983). Understanding social anxiety: Social, personality and clinical perspectives. Newbury Park, CA: Sage.
Leary, M. R. (1990). Responses to social exclusion: Social anxiety, jeal- ousy, loneliness, depression, and low self-esteem. Journal of Social and Clinical Psychology, 9. 221-229.
Leary. M. R., & Downs, D. (1995). Interpersonal functions of the self- esteem motive: The self-esteem system as a sociometer. In M. Kernis (Ed.), Efficacy, agency, and self-esteem. New York: Plenum.
Leary, M. R., & Forsyth, D. R. (1987). Attributions of responsibility for collective endeavors. In C. Hendrick (Ed.), Group processes (pp. 167-188). Newbury Park, CA: Sage.
Leary, M. R., & Kowalski, R. M. (in press). Butterflies, blushes, and bashfulness: Social anxiety and interpersonal behavior. New York: Guilford.
Maslow, A. H.( 1968). Motivation and personality. New York: Harper & Row.
McFarland, C , & Ross, M. (1982). Impact of causal attributions on affective reactions to success and failure. Journal of Personality and Social Psychology, 43, 937-946.
McFarlin. D. B., Baumeister, R. F., & Blascovich, J. (1984). On know- ing when to quit: Task failure, self-esteem, advice, and nonproductive persistence. Journal of Personality. 52, 138-155.
Mead, G. H. (1932). Mind, self, and society. Chicago: University of Chicago Press.
Miller, R. S. (in press). On the nature of embarrassability: Shyness, social-evaluation, and social skill. Journal of Personality.
Moreland, R., & Levine, J. (1989). Newcomers and oldtimers in small groups. In P. Paulus(Ed.), Psychology ofgroup influence (Vol. 2, pp. 143-186). Hillsdale, NJ: Erlbaum.
Perez, R. C. (1973). The effect of experimentally-induced failure, self-
esteem, and sex on cognitive differentiation. Journal of Abnormal Psychology, 81, 74-79.
Rogers, C. (1959). A theory of therapy, personality, and interpersonal relationships, as developed in the client-centered framework. In S. Koch (Ed.), Psychology: A study of a science (Vol. 3, pp. 184-256). New York: McGraw-Hill.
Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University Press.
Scheff, T. J., Retzinger, S. M., & Ryan, M. T. (1989). Crime, violence, and self-esteem: Review and proposals. In A. M. Mecca, N. J. Smelser, & J. Vasconcellos (Eds.), The social importance of self-esteem (pp. 165-199). Berkeley: University of California Press.
Schlenker, B. R. (1980). Impression management. Monterey, CA: Brooks/Cole.
Schlenker, B. R. (1985). Identity and self-identification. In B. R. Schlenker (Ed.), The self and social life (pp. 6 5 - 9 9 ) . New York: McGraw-Hill.
Schlenker, B. R., & Leary, M. R. (1982). Social anxiety and self-presen- tation: A conceptualization and model. Psychological Bulletin. 92, 641-669.
Shrauger, J. S., & Schoeneman, T. J. (1979). Symbolic interactionist view of self-concept: Through the looking glass darkly. Psychological Bulletin, 86, 549-573.
Shrauger, J. S., & Sorman, P. B. (1977). Self-evaluations, initial success and failure, and improvement as determinants of persistence. Journal of Consulting and Clinical Psychology, 45, 784-795.
Snodgrass, S. E. (1985). Women's intuition: The effect of subordinate role on interpersonal sensitivity. Journal of Personality and Social Psychology, 49, 146-155.
Spivey, E. (1989). Social exclusion as a common factor in social anxiety, loneliness, jealousy, and social depression: Testing an integralive model Unpublished master's thesis, Wake Forest University, Win- ston-Salem, NC.
Steele, C. M. (1988). The psychology of self-affirmation: Sustaining the integrity of the self. Advances in Experimental Social Psychology, 21, 261-302.
Tajfel, H. ( 1 9 8 1 ) . Human groups and social categories. Cambridge, En- gland: Cambridge University Press.
Taylor, S. E., & Brown, J. D. (1988). Illusion and well-being: A social psychological perspective on mental health. Psychological Bulletin, 103, 193-210.
Tedeschi, J. T , & Norman, N. (1985). Social power, self-presentation, and the self. In B. R. Schlenker (Ed.), The self and social life (pp. 293-322). New York: McGraw-Hill.
White, G. L. (1981). Some correlates of romantic jealousy. Journal of Personality, 49, 129-147.
Wills, T. A. (1981). Downward comparison principles in social psy- chology. Psychological Bulletin, 90, 245-271.
Received October 13, 1993 Revision received August 24, 1994
Accepted August 28, 1994 •
Guidelines for Article Reaction Papers
Over the course of the semester, you’ll be assigned three articles that relate to particular course topics. For each article reaction assignment, you are to read and carefully evaluate the assigned article and develop something to say about it during class discussion. Bring a typewritten question, with your attempt to answer the question, or a reaction to the article (1 page in length, double-spaced, 12-point font). The question/reaction should be analytic (miniature essay questions) rather than questions of fact or trivia. In a reaction paragraph, you might propose an experiment, or relate the article to another article you’ve read. During class, aim to share your ideas freely and support your arguments thoughtfully. The size of our class makes it unlikely that everyone will be able to share something during class; however, the intellectual excitement of a course depends on the active participation of all students, so we’ll do the best that we can. Article reaction papers must be turned in on their specified due dates. Late article reaction papers will not be accepted without an institutional (Instructor approved) excuse.
Guidelines for Article Reaction Papers
Over the course of the semester,
you’ll be assigned three articles that relate to particular
course topics.
For each article reaction assignment, you are
to read and
carefully evaluate the
assigned
article
and develop something to say about
it
during class discussion. Bring a
typewritten
question, with your attempt to answer the question, or a reaction to the article (1
page in length, double
-
spaced, 12
-
point font). The question/reaction should be analytic
(miniature essay questions) rather than questions of fact or trivia. In a reaction paragraph, you
might propose an experiment, or relate the article to another article you’ve read. During class,
aim to share your ideas freely and support your arguments tho
ughtfully. The size of our class
makes it unlikely that everyone will be able to share something during class; however, the
intellectual excitement of a course depends on the active participation of all students, so we’ll
do the best that we can.
Article r
eac
tion papers must be turned in
on
their specified due dates.
Late article reaction papers will not be accepted without an institutional (Instructor approved)
excuse.
Guidelines for Article Reaction Papers
Over the course of the semester, you’ll be assigned three articles that relate to particular
course topics. For each article reaction assignment, you are to read and carefully evaluate the
assigned article and develop something to say about it during class discussion. Bring a
typewritten question, with your attempt to answer the question, or a reaction to the article (1
page in length, double-spaced, 12-point font). The question/reaction should be analytic
(miniature essay questions) rather than questions of fact or trivia. In a reaction paragraph, you
might propose an experiment, or relate the article to another article you’ve read. During class,
aim to share your ideas freely and support your arguments thoughtfully. The size of our class
makes it unlikely that everyone will be able to share something during class; however, the
intellectual excitement of a course depends on the active participation of all students, so we’ll
do the best that we can. Article reaction papers must be turned in on their specified due dates.
Late article reaction papers will not be accepted without an institutional (Instructor approved)
excuse.
Get help from top-rated tutors in any subject.
Efficiently complete your homework and academic assignments by getting help from the experts at homeworkarchive.com