You are on page 1of 18

This article was downloaded by: [Florida State University]

On: 21 December 2014, At: 04:38


Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Anxiety, Stress, & Coping: An


International Journal
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/gasc20

The longitudinal relationship between


fear of positive evaluation and fear of
negative evaluation
a b c
Thomas L. Rodebaugh , Justin W. Weeks , Elizabeth A. Gordon
a c
, Julia K. Langer & Richard G. Heimberg
a
Department of Psychology , Washington University , 1 Brookings
Dr, Campus Box 1125, Saint Louis , MO , 63130 , USA
b
249 Porter Hall , Ohio University , Athens , OH , 45701 , USA
c
Psychology Department, 419 Weiss Hall , Temple University ,
1701 N. 13th St, Philadelphia , PA , 19122 , USA
Published online: 12 Apr 2011.

To cite this article: Thomas L. Rodebaugh , Justin W. Weeks , Elizabeth A. Gordon , Julia K. Langer
& Richard G. Heimberg (2012) The longitudinal relationship between fear of positive evaluation and
fear of negative evaluation, Anxiety, Stress, & Coping: An International Journal, 25:2, 167-182, DOI:
10.1080/10615806.2011.569709

To link to this article: http://dx.doi.org/10.1080/10615806.2011.569709

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or
arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Conditions of access and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
Downloaded by [Florida State University] at 04:38 21 December 2014
Anxiety, Stress, & Coping
Vol. 25, No. 2, March 2012, 167182

The longitudinal relationship between fear of positive evaluation


and fear of negative evaluation
Thomas L. Rodebaugha*, Justin W. Weeksb, Elizabeth A. Gordonc, Julia K. Langera
and Richard G. Heimbergc
a
Department of Psychology, Washington University, 1 Brookings Dr, Campus Box 1125, Saint
Louis, MO 63130, USA; b249 Porter Hall, Ohio University, Athens, OH 45701, USA;
c
Psychology Department, 419 Weiss Hall, Temple University, 1701 N. 13th St, Philadelphia,
Downloaded by [Florida State University] at 04:38 21 December 2014

PA 19122, USA
(Received 30 August 2010; final version received 4 March 2011)

Available research suggests that fear of negative evaluation and fear of positive
evaluation are related but distinct constructs that each contribute to social anxiety,
implying a need to focus on these fears in treatment. Yet, this research is almost
entirely based on cross-sectional data. We examined the longitudinal relationship
between fears of positive and negative evaluation over three time points in a sample
of undergraduate students. We tested competing models consistent with two basic
positions regarding these fears: (1) that fear of positive evaluation only appears
to affect social anxiety because it arises from the same, single underlying trait
as fear of negative evaluation, and (2) fears of positive and negative evaluation
are correlated, but clearly distinct, constructs. The best-fitting model was
an autoregressive latent-trajectory model in which each type of fear had a sepa-
rate trait-like component. The correlation between these trait-like components
appeared to fully account for the relationships between these constructs over time.
This investigation adds to the evidence in support of the second position described
above: fear of positive evaluation is best interpreted as a separate construct from
fear of negative evaluation.
Keywords: social anxiety; longitudinal; fear of positive evaluation; fear of
negative evaluation

Fear of negative evaluation (FNE) has long been thought to be central to social
anxiety and social anxiety disorder (Clark & Wells, 1995; Leary & Kowalski, 1995;
Rapee & Heimberg, 1997) and has been a primary focus of treatment (e.g., Heimberg
& Becker, 2002). This focus on FNE would seem to require very little explanation.
Whether experienced acutely or chronically, fear of being negatively evaluated is an
intuitive predecessor to social anxiety in both its normative and problematic forms.
However, more recent findings have suggested that fears of positive evaluation are also
strongly related to social anxiety (Weeks, Heimberg, & Rodebaugh, 2008; Weeks,
Heimberg, Rodebaugh, & Norton, 2008; Weeks, Rodebaugh, Heimberg, Norton, &
Jakatdar, 2009), which may, in turn, imply that fears of positive evaluation could serve
as an additional focus of treatment.

*Corresponding author. Email: rodebaugh@wustl.edu


ISSN 1061-5806 print/1477-2205 online
# 2011 Taylor & Francis
http://dx.doi.org/10.1080/10615806.2011.569709
http://www.tandfonline.com
168 T.L. Rodebaugh et al.

Although perhaps less intuitive, the relationship between fear of positive


evaluation (FPE) and social anxiety has a solid theoretical basis. Gilbert and
colleagues have suggested a psychoevolutionary theory of social anxiety (e.g., Gilbert,
2001). In this theory, social anxiety is a protective mechanism designed to balance
the risks of trying to move up a dominance hierarchy too quickly versus falling out
of the hierarchy completely (i.e., being ostracized). If this theory holds, then social
anxiety should be triggered by concerns about both positive and negative evaluation.
Negative evaluation would serve as a signal of downward movement on the hierar-
chy. Positive evaluation, on the other hand, could serve as a warning signal that
the person may offend important others or be put in a position of having to defend
newly-acquired status.
We find Gilbert’s theory fairly compelling in regard to evolutionary pressures
that could lead to a social anxiety module in human beings. However, it is not
Downloaded by [Florida State University] at 04:38 21 December 2014

necessary to accept this argument to understand how FPE could be strongly related
to social anxiety. A psychological (vs. psycho-evolutionary) analysis of social anxiety
leads to a similar conclusion. Multiple studies have indicated that people with higher
social anxiety believe they are less acceptable or have more negative characteris-
tics than others (e.g., Moscovitch, Orr, Rowa, Reimer, & Antony, 2009; Rodebaugh,
2009). Positive and negative evaluation are not complete opposites: Both imply that
one is being evaluated and scrutinized. People who believe they are not socially
acceptable would be expected to avoid evaluation of any type, because even positive
evaluation could be the occasion for other people discerning one’s flaws and foibles.
Although there are multiple reasons to believe that FPE should strongly predict
social anxiety, there are several ways that social anxiety could relate to fears of positive
and negative evaluation over time. A basic argument concerns whether fears of
positive and negative evaluation are distinct. The proposition that they are not
distinct has thus far been contradicted by all available research (Fergus et al., 2009;
Weeks, Heimberg, & Rodebaugh, 2008; Weeks, Heimberg, Rodebaugh, & Norton,
2008; Weeks et al., 2009), which suggests that fears of negative and positive evalua-
tion are overlapping, but distinct constructs that each contribute unique variance
to social anxiety. We denote this hypothesis as the related but distinct constructs
(distinct) hypothesis.
An alternative argument suggests that, at some level, the effects of FPE may be
rooted in FNE.1 That is, despite apparent distinguishing features between fears of
positive and negative evaluation, social anxiety’s relationship with FPE could be
entirely due to the negative evaluation that is anticipated as a consequence of positive
evaluation once it is received. For example, a person with higher social anxiety who
is told that she gave a good speech might fear the possibility that the next speech
will not meet the same standard, resulting in even greater negative evaluation than
might have been experienced if the first speech had not been good to begin with.
FPE may thus be motivated by a desire to avoid future negative evaluation. Indeed,
studies by Wallace and Alden (1995, 1997) provide evidence implying that just this
process might occur. In these studies, the authors demonstrated in analogue (1995)
and clinical (1997) samples that, when people with problematic social anxiety
are confronted with apparent social success, they are likely to believe that other people
will now expect more from them. However, Wallace and Alden did not demonstrate
specifically that this perceived change in standard leads to any change in FNE.
It nevertheless seems that such a process is plausible, leading to the hypothesis that
Anxiety, Stress, & Coping 169

FPE represents only fear of delayed negative evaluation. We will refer to this
hypothesis as the delayed hypothesis.
One problem with the delayed hypothesis is that participants in research
supporting the distinctness of FPE were asked to respond regarding their trait levels
of constructs, not their current levels of those constructs. It is difficult to under-
stand how trait FPE could simply reflect future state FNE. However, the previous
research is based entirely on retrospective reporting of traits, and evidence regard-
ing retrospective reporting suggests that people are not skilled at recalling past
psychological experiences (e.g., Henry, Moffitt, Caspi, Langley, & Silva, 1994). If
participants are only truly able to accurately report regarding their current thoughts,
behavior, and mood, it might be that apparent report of traits primarily represents
a report of the state-like aspects of the variables in question. If participant report
of current traits is biased toward current states, then we can begin to see how a
Downloaded by [Florida State University] at 04:38 21 December 2014

plausible version of a delayed model can be constructed. In this model, however,


measures of FPE and FNE both assess the same underlying trait-like variable,
imperfectly assessed by both types of measures due to lack of insight on the part of
respondents. We therefore refer to this model as the single-factor hypothesis. The
single-factor hypothesis represents the only way we have found to express the delayed
hypothesis in a form that (1) is consistent with existing evidence reviewed above and
(2) can be identified with a specific statistical model.
In the single-factor model, responses on FNE and FPE measures are both
expressions of a single, trait-like vulnerability to fear of evaluation. Further, fear
of negative evaluation is assumed to be of primary importance to the development
and maintenance of social anxiety. Although it is logically possible that FPE could
be of primary importance, this possibility seems implausible to us. The assumption
of FNE’s primacy would be borne out in the trait-like vulnerability being more
heavily dependent upon measures of FNE than measures of FPE. At a single time-
point, a measure of FPE could appear to offer a unique contribution to predicting
social anxiety. However, this significant but spurious cross-sectional relationship
would simply be a result of (1) sampling error and limits in reliability in the compa-
rison measure of FNE and (2) limits of participant insight. That is, no single observed
measure could fully assess an underlying latent construct at a single time point
because no single measure has perfect reliability. Further, if participants are biased
to report based on current states, the ability of a single measure of FNE to capture
the underlying trait-like construct across time should be further limited. The unique
variance in social anxiety that a measure of FPE predicts at one time point would
simply be variance more properly accounted for by a single underlying factor, which
draws considerably more variance from FNE measures than FPE measures.
The single-factor model cannot adequately be tested cross-sectionally if, as seems
likely, participants are biased to report based more on their recent experience.
Although models of the underlying structure of FPE and FNE items could be tested
(and indeed have been, as noted above), only multiple measurements across time
would allow estimation of a trait-like latent variable that captures stability across
time. To test whether multiple measurements of FPE and FNE across time are better
accounted for by a single latent variable (defined primarily by FNE) or distinct latent
variables, we turn to longitudinal data in the current study.
It should be noted that the latent structure of FPE and FNE over time has
direct implications for cognitive behavioral therapy of social anxiety disorder. If the
170 T.L. Rodebaugh et al.

single-factor hypothesis is correct, there would be little reason to measure FPE


specifically over the course of treatment. A measure of FPE could be used as an
index of the underlying trait-like variable, and might be useful for full coverage of the
latent construct at a single time point. However, across treatment, there would be
little reason to move beyond measuring FNE at multiple time points. Further,
targeting FNE would be expected to be sufficient to alter FPE if FPE is merely a
state-dependent expression of the underlying level of a single trait variable. If, on the
other hand, the distinct hypothesis is correct, FPE might require separate
measurement and intervention.
The arguments above detail why it is essential to examine these constructs over
time to determine to what extent cross-sectional relationships might merely be
spurious results based on incomplete measurement and lack of participant insight.
Testing the single-factor and distinct hypotheses requires measurement of both
Downloaded by [Florida State University] at 04:38 21 December 2014

types of fears over time and the estimation of latent constructs that account for
stability and change. We collected data from undergraduates over successive weeks to
allow just such tests. We examined models that included latent variables such as slopes
and intercepts (e.g., underlying traits) alongside autoregressive parameters (e.g., the
autoregressive latent trajectory [ALT] model: Bollen & Curran, 2004; Curran &
Bollen, 2001). ALT models, in comparison to other available models (e.g., auto-
regressive models) have the virtue of mapping directly onto our hypotheses to be
tested, by allowing the estimation of latent variables (i.e., underlying traits). In
addition, the ALT modeling framework allowed us to examine the remaining
relationships between variables at different time points (i.e., cross-sectional and
longitudinal relationships), permitting further tests of interest. We also examined
relationships with social anxiety to elucidate the implications of the models. We
expected, based on the distinct hypothesis, that there would be no evidence that
FNE and FPE are best accounted for by a single factor that is defined primarily by
the loadings of FNE measurements; instead, we expected that each variable would
have a distinct underlying latent trait.

Method
Participants
A total of 77 participants were available who met all of the criteria described in
the analytic procedure section. Participant completion of the three time points
under consideration was: Time One: n 77, Time Two: n56, Time Three: n 46.
Participants had a mean age of 20.32 (SD 3.56) and most (n 58, 75%) were
women. Although most were white (n 40, 52%), there was considerable ethnic
diversity in the sample, with participants self-identifying as African-American (n15,
20%), Asian or Asian American (n9, 12%), Mixed Ethnicity (n5, 7%), Latino
(n4, 5%), and Native American (n1, 1%). An additional three participants (4%)
indicated that a category for their ethnicity was not provided.

Measures
Participants were administered the following measures, with instructions to rate each
item based on the past week (instead of the typical instructions for each measure).
Anxiety, Stress, & Coping 171

(1) Fear of Positive Evaluation Scale (FPES; Weeks, Heimberg, & Rodebaugh,
2008). The 10-item FPES uses a 10-point Likert-type rating scale, ranging from
zero (not at all true) to nine (very true). The 10 items in the scale include two
reverse-scored items that are not utilized in the calculation of the FPES total
score. The FPES has demonstrated strong internal consistency (all as.80) in
both undergraduate (Weeks, Heimberg, & Rodebaugh, 2008; Weeks, Heimberg,
Rodebaugh, & Norton, 2008) and clinical (Fergus et al., 2009) samples and
five-week retest reliability (intraclass correlation coefficient .70) in under-
graduate samples (Weeks, Heimberg, Rodebaugh, & Norton, 2008). Further-
more, the FPES has demonstrated strong convergent and discriminant
validity (Fergus et al., 2009; Weeks, Heimberg, & Rodebaugh, 2008; Weeks,
Heimberg, Rodebaugh, & Norton, 2008), as well as factorial validity, in a series
of confirmatory factor analyses in both undergraduate (Weeks, Heimberg,
Downloaded by [Florida State University] at 04:38 21 December 2014

& Rodebaugh, 2008; Weeks, Jakatdar, & Heimberg, 2010) and clinical (Fergus
et al., 2009) samples. The FPES demonstrated good internal consistency at
all 3 time points in the present study (all as.82).
(2) Brief Fear of Negative Evaluation Scale-Straightforward Items (BFNE-S;
Rodebaugh et al., 2004; Weeks et al., 2005). The BFNE (Leary, 1983) is a
12-item self-report measure of fear and distress related to negative evaluation
from others. Items are rated on a five-point Likert-type scale, ranging from
one (Not at all characteristic of me) to five (Extremely characteristic of me).
Rodebaugh et al. (2004) and Weeks et al. (2005) have reported that the eight
straightforwardly-worded items are more reliable and valid indicators of fear
of negative evaluation than the reverse-scored items in undergraduate and
clinical samples, respectively. Consequently, Rodebaugh et al. and Weeks et al.
have suggested utilizing only the eight straightforward BFNE items to calcu-
late the total score. The BFNE-S has demonstrated excellent internal
consistency (all as.92), factorial validity, and construct validity in under-
graduate (Rodebaugh et al., 2004) and clinical (Weeks et al., 2005) samples. The
12-item BFNE was administered; however, only the BFNE-S items were
utilized in the present analyses. The BFNE-S demonstrated good internal
consistency at all three time points in the present study (all as .90).
(3) Social Interaction Anxiety Scale-Straightforward items (SIAS-S; Rodebaugh,
Woods, & Heimberg, 2007; Rodebaugh, Woods, Heimberg, Liebowitz, &
Schneier, 2006). The Social Interaction Anxiety Scale (SIAS; Mattick & Clarke,
1998) is a 20-item measure of anxiety in dyads and groups. Rodebaugh and
colleagues (2006, 2007) have reported that the 17 straightforwardly-worded
items of the SIAS are more valid indicators of social interaction anxiety
than the reverse-scored items in both undergraduate and clinical samples. This
17-item score, hereafter referred to as the SIAS-Straightforward (SIAS-S)
score, has demonstrated excellent internal consistency (a .93) and factorial
validity in undergraduate samples and has demonstrated strong construct
validity in both undergraduate and clinical samples (Rodebaugh et al., 2007).
The 20-item SIAS was administered; however, only the SIAS-S items were
utilized here. The SIAS-S demonstrated good internal consistency at all three
time points in the present study (all as.93).
(4) Social Phobia Scale (SPS; Mattick & Clarke, 1998). The SPS measures fear of
public scrutiny and consists of 20 items which are scored on a five-point
172 T.L. Rodebaugh et al.

Likert-type rating scale ranging from zero (Not at all characteristic or true of
me) to four (Extremely characteristic or true of me). The SPS has demonstrated
strong internal consistency in clinical, community, and undergraduate samples
(as range from .89 to .94) (Mattick & Clarke, 1998), and adequate retest
reliability over periods of 1 to 2 weeks (r .66) in a sample of undergraduates
(Heimberg, Mueller, Holt, Hope, & Liebowitz, 1992). Furthermore, Mattick
and Clarke (1998) reported that the SPS adequately discriminated among
patients with anxiety disorders (social anxiety disorder, agoraphobia, specific
phobia) and between individuals with social anxiety disorder and controls
(undergraduate and community samples), providing evidence of construct
validity. The SPS demonstrated good internal consistency at all three time
points in the present study (all as.91).
(5) Social anxiety index. The SIAS-S and SPS were highly correlated at each time
Downloaded by [Florida State University] at 04:38 21 December 2014

point (all rs.74) and were thus standardized and averaged as an overall index
of social anxiety symptoms during the week.

Procedure
Data collection for the present study was managed using the Sona-Systems soft-
ware package (www.sona-systems.com) and was restricted to Temple University
undergraduate students currently enrolled in a psychology course. All participants
received partial academic credit for their participation. Participants provided
informed consent prior to participation. Participants were asked to complete the
FPES, BFNE, SIAS, and SPS once each week for five weeks, among other measures
included for a related study. As part of the related study, participants completed
13 measures at the original Times One, Three, and Five; and nine measures at the
original Time Two and Time Four (see Selection of participants from overall dataset
for additional details on this issue). Carlbring and colleagues (2007) have reported
that anxiety measures completed using online versus paper-and-pencil adminis-
tration formats demonstrate similar psychometric properties and are very strongly
correlated. Similarly, Hedman and colleagues (2010) concluded that social anxiety
measures, such as the SIAS and SPS, show similar psychometric properties when
administered over the internet and when completed in paper-and-pencil format.
Participants were invited to participate in the present study via an online posting
on the Sona-Systems website portal for Temple University. A full description of
all study procedures was provided in the posting, and potential participants who
remained interested after reading the study description provided informed consent
and completed the study measures for Time One. Participants were informed during
the consent process that that they would receive weekly requests/reminders via email
to complete the additional study measures over the next four weeks.
One of the authors (EAG) monitored Time One survey completions and followed
up with participants weekly via email to: (1) remind participants to complete
the next round of questionnaires within the next 24 hours; and (2) to provide
instructions for logging in and completing the next questionnaire series. All survey
completions were time-stamped, and the timestamps were reviewed to ensure that all
participants retained in the study analyses completed the surveys within the
instructed timeframe (i.e., within five to 10 days of completing the previous survey)
Anxiety, Stress, & Coping 173

(see Selection of participants from overall dataset section below for additional details
on data management).

Data analytic procedure


Selection of participants from overall dataset
A total of 299 participants completed some portion of the study from which the
current data were drawn, but most of these participants provided partial data, and
most provided data at only one time point. Participants provided data at up to
five time points, but because missing data were extensive at the later time points
(Time Four and Time Five), we included responses from only the first three time
points in the present analyses (with the exception of multiple imputation; see below).
Downloaded by [Florida State University] at 04:38 21 December 2014

The participants for the current study were selected based on the following rules,
above and beyond those noted above: (1) participants completed at least two of the
initial three time points (n 96) and (2) participants spent at least 15 minutes
completing each longer questionnaire battery (i.e., Time One and Time Three) and
at least eight minutes completing the shorter questionnaire battery (i.e., Time Two).
In regard to the time selection rules, the lower bound for time was considered the
minimal amount of time necessary to give sufficient attention to the items for
responses to be deemed viable. Regarding the latter rule, participants who failed to
meet these standards were entirely removed from the study (rather than individual
participant time points being deleted) because failure to spend sufficient time
providing responses at one time point was considered to indicate lack of participant
investment in the study overall.

Analyses
For the sake of consistency, all data analyses were performed using Mplus version
4.2 (Muthén & Muthén, 19982006) using the robust maximum likelihood esti-
mator (referred to as MLM in Mplus). This estimator is robust to violations of
multivariate normality.
Missing data were handled by multiple imputation performed in Amelia II
(Honaker, King, & Blackwell, 20062008). Multiple imputation was conducted on a
dataset that included BFNE-S and FPES scores and the social anxiety index
composite (from the SIAS-S and SPS) from Time One to Time Five. We included
measurements made at Times Four and Five to improve estimation of missing data.
For example, some participants might have completed measures at Time Four or
Time Five but not one of the previous time points. The addition of these two time
points should therefore improve prediction of missing data despite the fact that Time
4 and Time 5 data were not analyzed here. Of note, we found no significant
correlation between the variables analyzed and missingness at any of the three viable
time points or an overall tendency toward missing data (ps .10). We also examined
all demographic variables in the dataset to determine whether any were associated
with missing data. Because we found that ethnicity was associated with missing
data at Time Two (x2[6] 12.97, p .044), we also included that variable in the
multiple imputation dataset.
174 T.L. Rodebaugh et al.

We had no reason to believe that missing data were dependent upon the missing
values themselves and thus considered the data missing at random. A ridge prior
value of four was included, as recommended by Honaker et al. (20062008), because
some of the initial invariance matrices were not invertible. Honaker et al. note that
non-invertible matrices may be caused by matrices being singular, which may, in
turn, be caused by highly correlated variables in the dataset. Because the current
dataset consists of repeated measures of related constructs, a singular matrix due to
highly correlated variables appeared quite likely. A ridge prior allows resolution of
this problem. The value of four was based on Honaker et al.’s recommendations.
We investigated the ability of the Amelia program to appropriately produce multiply
imputed datasets by examining three diagnostics it provides (for each variable to
be used in analyses, when diagnostics were available by individual variables): (1) over
imputation; (2) observed vs. imputed distributions; (3) over dispersed start values.
Downloaded by [Florida State University] at 04:38 21 December 2014

Please see Honaker et al. for details regarding these diagnostic procedures, each of
which examine a potential way that imputation might fall short of the goal of
properly representing the likely values of the missing data. All diagnostics indicated
that the program was able to provide appropriate multiple imputation given the
data that we provided it.2
We tested ALT models, which include trait and slope latent factors as well as
autoregressive parameters.3 These models were fitted in almost the same manner
described by Rodebaugh, Curran, and Chambless (2002), who in turn based their
strategy on the original descriptions of the models (Bollen & Curran, 2004; Curran
& Bollen, 2001). The exception was that we included Time One variables in the
slope and intercept because they were not estimable without doing so. We first
examined each construct separately, testing in succession for: (1) a significant trait-
like component, and (2) a significant slope component that improved fit. The trait-
like component was specified as a latent variable that contributed equally to each
measurement point of the variable in question. That is, although the mean and
variance of the trait was estimated, each instance of measurement was fixed to have
a loading of one on the underlying trait. The slope variable was also permitted to
have a varying slope and intercept, but the loadings of the variables were fixed to
a linear progression (0, 1, and 2). We then tested whether adding autoregres-
sive parameters improved fit. Further steps were available in principle (e.g., fixing
autoregressive parameters to equality), but no models supported further steps.
Further, we conducted tests of two models specific to the single FNE factor
hypothesis: A multivariate model in which both FPE and FNE had the same
underlying trait-like variable upon which each variable loaded to the same extent
and a multivariate model in which FPE and FNE measurements were permitted
to vary in their loadings on an underlying latent factor.
In evaluating global fit, we consulted the following fit indices: (1) Tucker-Lewis
incremental fit index (TLI; Tucker & Lewis, 1973), (2) comparative fit index (CFI,
Bentler, 1990), (3) root mean square error of approximation (RMSEA; Steiger &
Lind, 1980), and (4) standardized root mean square residual (SRMR; Bentler, 1995;
Jöreskog & Sörbom, 1981). The magnitudes of these indices were evaluated with the
aid of recommendations by Hu and Bentler (1999). Essentially, for the TLI and CFI,
values of .90 and above were considered adequate, whereas values of .95 or above
were considered very good; for the RMSEA and SRMR, values of .08 and below
were considered adequate and .05 or less very good.
Anxiety, Stress, & Coping 175

Chi-square difference tests were used to test whether adding parameters to


the ALT models resulted in significantly improved model fit. Because we used
multiple imputation and the robust chi-square statistic, it was not possible to
conduct the chi-square difference test directly (because Mplus only supplies the
average of the chi-square statistics across datasets, which is not sufficient for
conducting nested tests). We therefore conducted difference tests within each dataset;
in most cases, the tests resulted in the same finding across all datasets. We report
exceptions to complete agreement for the difference tests across datasets below.
Further, it was not always possible to use nested chi-square tests for the comparisons
of most interest (i.e., the single-factor model with unrestricted loadings was not
nested within the most strongly supported distinct model). We therefore evaluated
comparative model fit using the Information Criterion proposed by Akaike (AIC;
Akaike, 1974) and the Bayesian Information Criterion (BIC; Schwarz, 1978). These
Downloaded by [Florida State University] at 04:38 21 December 2014

statistics have different theoretical bases, but both aim to quantify how well a model
approximates the theoretically best model for the data (Kuha, 2004). As recom-
mended by Kuha (2004), we used both indices together with the rationale that it
is unlikely that both indices are incorrect when they agree on which model should
be selected. Lower values of both indices are preferred, and there is no absolute
metric to determine good fit with either index (i.e., they are to be used comparatively
rather than for a single model in isolation).

Results
Distributions of observed variables
The observed variables, prior to imputation, were approximately normally distrib-
uted, with no skewness or kurtosis estimates having an absolute value greater than
one. We nevertheless used MLM estimation because univariate normality does not
assure multivariate normality. Table 1 displays the univariate statistics of the obser-
ved (not imputed) data. Interpretation is complicated by the fact that each

Table 1. Observed univariate statistics for variables in models.

Scale Week n M SD Range Possible range

FPES 1 77 23.14 13.38 058 090


2 56 23.98 13.67 061 090
3 46 23.41 14.43 059 090
BFNE-S 1 76 20.50 6.79 840 840
2 56 20.83 7.21 839 840
3 46 19.00 7.40 837 840
SIAS-S 1 77 20.39 12.32 156 068
2 56 20.66 13.17 055 068
3 46 20.11 13.72 049 068
SPS 1 77 19.14 12.30 056 080
2 56 21.42 14.19 053 080
3 46 19.67 14.57 060 080
Note: FPES, Fear of Positive Evaluation Scale; BFNE-S, Brief Fear of Negative Evaluation Scale,
Straightforward Version; SIAS-S, Social Interaction Anxiety Scale, Straightforward Version; SPS, Social
Phobia Scale. In each case participants responded regarding the past week (e.g., not in general).
176 T.L. Rodebaugh et al.

measurement is of the related construct during the past week. However, it should
be evident from the ranges of the variables that participants with both very high
and very low social anxiety were represented. In nearly all cases, the value at the
bottom of the scale was represented; the highest values observed were consistent
with scores that would be expected of participants with social anxiety disorder. For
example, the highest observed SIAS-S value at each time point was well above a
suggested cut-off for probable social anxiety disorder of 34 (despite the fact that
the cut-off was derived from a version of the SIAS that included three additional
items; Brown et al., 1997).

ALT Models
FPE univariate model
Downloaded by [Florida State University] at 04:38 21 December 2014

FPE had a trait-like component with a significant mean (M 23.46, z 16.31,


pB.001) and variance (variance 154.55, z5.84, p B.001). Model fit was good to
excellent with the trait-like component representing the only relationship between
the measurements of FPE (CFI .99, TLI.99, RMSEA .06, SRMR .06).
Adding additional parameters did not improve fit according to nested chi-square
tests across all five datasets (adding a slope parameter: Dx2s[3] B 6.98, ps.070;
adding unconstrained autoregressive parameters: Dx2s[2] B3.66, ps.160). There-
fore, only the intercept was retained in the model.

Social anxiety univariate model


Much like the FPE model, the social anxiety model consisted of a trait-like
component only. As might be expected given that the social anxiety variables were a
composite of two standardized scores, the trait-like component did not have a
significant mean (M 0.00, z 0.00, p 1.00); however, it did have a significant
variance (variance  0.74, z 6.42, pB.001). The model had excellent fit
(CFI 1.00, TLI 1.00, RMSEA.03, SRMR .05). Across all datasets, neither
the addition of a slope nor unconstrained autoregressive parameters could improve fit
(slope: Dx2s[3] B5.30, ps.151; autoregressive: Dx2s[2] B 5.13, ps .076). Therefore,
only the intercept was retained in the model.

FNE univariate model


Fitting of the FNE model proved more complex. A slope factor significantly impro-
ved fit according to four of the five datasets (Dx2s[3] 10.32, psB.016). With the
slope factor retained, only one degree of freedom was left with which to test
autoregressive parameters. However, it was also mathematically impossible for an
autoregressive parameter to improve fit, because model chi-square was so low in each
dataset that it could not be significantly improved (largest x2[1] 2.63, p .105). The
trait-like factor had a significant mean (M 20.55, z26.86, pB.001) and variance
(variance 41.13, z 10.17, pB.001), whereas the slope factor had a marginally
significant mean (M 0.58, z1.95, p .051) and variance (variance 5.48,
z1.72, p.085). Thus, participant report of FNE tended to decrease over time,
although this tendency also varied among participants.
Anxiety, Stress, & Coping 177

FPE and FNE multivariate model


The initial multivariate model consisted of the FPE and FNE univariate models along
with the following parameters between the FNE and FPE measurements: Each
variable was permitted to prospectively predict the other, and each measure was
allowed to correlate with the other at the same time point (i.e., the error terms of the
variables were allowed to correlate at each time point). It was noted that the slope
parameter for FNE no longer had even a marginally significant mean or variance in
this model (ps.25). We therefore dropped the slope term for the sake of parsimony.
The final model is displayed in Figure 1 and fit acceptably to excellently (CFI.99,
TLI.98, RMSEA.07, SRMR.08). Notably, FNE and FPE showed no tendencies
to prospectively predict each other; their temporal relationships were entirely accounted
for by their strongly correlated trait-like components (r.59, pB.001).
Downloaded by [Florida State University] at 04:38 21 December 2014

Single-intercept and single-factor FPE and FNE models


Might the strong correlations between the FPE and FNE intercepts indicate that
the two constructs shared a single underlying trait-like intercept? A single-factor

FPE
Intercept
Factor

1 1 1

FPE 1 FPE 2 FPE 3

.59*
.02 .04

.01 -.03

FNE 1 FNE 2 FNE 3

1 1 1

FNE
Intercept
Factor

Figure 1. Fear of positive evaluation (FPE) and fear of negative evaluation (FNE)
multivariate autoregressive latent-trajectory (ALT) model. Unstandardized parameters are
shown. Not shown, for the sake of simplicity, are covariances between the measures at each
time point (specifically, FNE and FPE error variances were permitted to correlate at each time
point). *p B.001; for other parameters, ps .21.
178 T.L. Rodebaugh et al.

hypothesis would suggest that was the case. Prior to fitting these models, we
re-imputed the datasets using standardized variables for the FPE and FNE measures
so that (1) the model fit would not be penalized due to the different metrics of
the two measures and (2) parameters would be more interpretable across measures.
However, substantive results were very similar without standardization. A model
identical to the above except that FPE and FNE scores shared a single underlying
intercept fit poorly (CFI .78, TLI .72, RMSEA .31, SRMR .29). Thus, there
was no indication that FPE and FNE shared a single underlying trait component.
However, a single-factor model in which FNE was primary would suggest that
FNE should load more strongly on the underlying latent variable, whereas this model
required that the trait contributed equally to each measure. Therefore, we fitted
a model in which a single factor was underlying both FNE and FPE measure-
ments, with loadings permitted to vary. This model did not fit uniformly well, with
Downloaded by [Florida State University] at 04:38 21 December 2014

two indices indicating good fit and two indicating poor fit (CFI .96, TLI.69,
RMSEA .32, SRMR .03). Although this model is not nested within the ALT
model that reflects the distinct hypothesis, it can be compared using its AIC and
BIC values (913.74 and 971.77, respectively).
Lower values indicate a superior model for both indices. When the multivariate
model reflecting the distinct hypothesis was fitted again using the standardized
data (to permit comparison of AIC and BIC), its AIC and BIC values were both
lower (913.23 and 967.13, respectively) than the unrestricted single-factor model’s
values. It should also be noted that the single-factor unrestricted model produced
unexpected longitudinal relationships: Each prospective relationship between FPE
and FNE was significant but negative. Finally, FNE loadings on this underlying
factor (1.00, 1.72, 1.97) were not uniformly stronger than the FPE loadings (.99,
1.76, 1.52).

Addition of social anxiety to the model


The social anxiety univariate model was added to the FNE and FPE multivariate
model, because it proved to be the best fitting model of those tested. The measure
of social anxiety was permitted to correlate with both fears of evaluation within
each time point in this model, and each construct was permitted to predict
the other constructs at the next time point. We were thereby able to test the
prospective relationships between the two types of evaluation fears and social anxiety.
In these tests, no variable ever displayed a tendency to predict prospectively (all
ps.10). The relationships among these variables appeared to be completely
accounted for by the correlations between their underlying trait-like components
(FNE with social anxiety: r.78, pB.001, FPE with social anxiety: r.75, p B.001).
The model showed generally good fit (CFI .96, TLI .91, RMSEA.14,
SRMR.08), especially when it is remembered that the RMSEA can tend to be
inflated in samples of 200 or less (Curran, Bollen, Chen, Paxton, & Kirby, 2003).

Discussion
We conducted this study to assess the degree to which FPE and FNE exist as
separate constructs that hold separate relationships with social anxiety over time.
There was no support for a model in which FNE and FPE were found to arise from a
Anxiety, Stress, & Coping 179

single underlying factor. Rather, the strongest support was for a model in which FPE
and FNE at each time point were explained by underlying, moderately correlated
trait-like components. The best-fitting model suggested that, at each time point,
these constructs had no meaningful prospective relationship over the course of three
weeks. All possible prospective relationships among FPE, FNE, and social anxiety
appeared to be accounted for by underlying trait components of these constructs.
Thus, this evidence suggests that participants in this study had a well-established
level of each construct prior to the beginning of the study and that it was this trait-
like level of each construct that most strongly determined their responses at each
time point. Attempts to fit a single-factor model of the two constructs resulted in
inferior fit and nonsensical longitudinal relationships (FPE significantly predicting
lower future FNE). Further, even if the best-fitting single-factor model were
accepted, it did not provide support for the contention that FNE was more strongly
Downloaded by [Florida State University] at 04:38 21 December 2014

related to that single factor than FPE.


In previous studies of FPE (Weeks, Heimberg, & Rodebaugh, 2008; Weeks,
Heimberg, Rodebaugh, & Norton, 2008; Weeks et al., 2009; Weeks et al., 2010),
authors have acknowledged that the precise relationship between FPE and FNE is
unclear, despite the fact that the available evidence suggested that they were not
isomorphic. Hypotheses in which FPE was an expression of the same underlying
factor as FNE across time remained plausible. In this study, the first to use longitu-
dinal data to evaluate the relationship between FPE and FNE, all evidence pointed
toward the distinct hypothesis. Although we cannot rule out the possibility that
fears of positive evaluation sometimes represent delayed FNE, it seems clear that
FPE is at least not always merely delayed FNE or merely an artifact of cross-
sectional measurement and lack of participant insight.
Our findings may provide some guidance for future studies regarding the
constructs of FNE and FPE. The best fitting models suggest caution in interpreting
cross-sectional relationships among these variables as representing causal relation-
ships. Although we do not doubt that, at some level, fears of evaluation cause social
anxiety and are likely to have reciprocal relationships among each other, such
apparent relationships were better accounted for by underlying traits. Cross-sectional
relationships found in other studies might be based on causal relationships, but they
do not appear to be causal relationships that are expressed over the course of a
few weeks.
It must be emphasized that our findings are based on weekly ratings and thus do
not necessarily speak to much wider time frames or much narrower ones. We believed
that a week-to-week design was a good starting place for assessing the temporal
relationships among these variables, but we encourage other researchers to assess
these variables (and others) over a wider range of times. Notably, the evidence that
seems to imply the possibility that fears of positive evaluation merely represent
concerns about later negative evaluation concerns changes in affect and expectations
regarding standards over the course of minutes and hours, not weeks (Wallace
& Alden, 1995, 1997). Our results also were reliant upon naturally-occurring changes
in FPE and FNE, triggered by naturally-occurring evaluation experiences encoun-
tered by our participants. It might be that assessment of the nature of these
evaluation experiences or assignment to certain types of evaluation experiences
would be necessary to detect prospective relationships between FNE and FPE that
go beyond underlying trait-like aspects of these variables.
180 T.L. Rodebaugh et al.

Our results must also be interpreted within the limitations of our data. We
examined undergraduate students in a sample of moderate size that showed a fair
amount of diversity. A group of participants with a wide range of FPE and FNE
was ideal given our intent to examine the relationships of these constructs over
time. However, a community sample would have been arguably more ideal for our
purposes; use of an undergraduate sample was based on convenience alone. Viable
data were obtained across three weeks; more time points would have been preferable.
For example, more time points might have clarified the unstable slope parameter
found for FNE in the univariate model but not in the multivariate models. Data were
obtained via the internet, which brings up the possibility that paper-and-pencil
measures would have resulted in different findings. We have no reason to believe
this would be the case given that preliminary evidence suggests that measures related
to social anxiety result in similar responses across these administration methods
Downloaded by [Florida State University] at 04:38 21 December 2014

(Hedman et al., 2010). However, the literature regarding equivalence of internet


and paper-and-pencil methods of administration is sparse, making firm conclusions
on this matter impossible.
In conclusion, our results provide no support for the contention that FPE is
simply reflective of delayed FNE. FPE and FNE appear to be moderately correlated,
but clearly separable constructs that have trait-like components that explain much of
their inter-correlation over time. It seems clear given this evidence as well as past
studies (Weeks, Heimberg, & Rodebaugh, 2008; Weeks, Heimberg, Rodebaugh, &
Norton, 2008; Weeks et al., 2009; Weeks et al., 2010) that FNE and FPE are not
isomorphic. We suggest that it will be more fruitful to turn to questions of how
each of these constructs contributes to aspects of social anxiety, with a particular
focus on social anxiety that has become problematic and reduction of social anxiety
disorder symptoms.

Acknowledgements
The authors wish to thank anonymous reviewers for helpful comments on a previous draft of
this manuscript.

Notes
1. The inverse of this account, that FNE is primarily rooted in FPE, is implausible to us and
seems difficult to defend even as an intellectual exercise; we therefore do not attempt to
defend it here.
2. The figures resulting from these diagnostic tests are available from the first author.
3. We also tested autoregressive models, which are computationally (if not theoretically) more
parsimonious. However, we were unable to find a model that showed uniformly acceptable
fit. Because such models provide little information regarding the hypotheses at hand (i.e.,
because they do not include underlying latent variables), we do not present the results in
detail, although they are available from the first author.

References
Akaike, H. (1974). A new look at the statistical model identification. IEEE Transactions on
Automatic Control, 19, 716723.
Bentler, P.M. (1990). Comparative fit indexes in structural models. Psychological Bulletin, 107,
238246.
Anxiety, Stress, & Coping 181

Bentler, P.M. (1995). EQS: Structural equations program manual, Version 5.0. Los Angeles:
BMDP Statistical Software.
Bollen, K.A., & Curran, P.J. (2004). Autoregressive Latent Trajectory (ALT) models: A
synthesis of two traditions. Sociological Methods & Research, 32, 336383. doi: 10.1177/
0049124103260222
Brown, E.J., Turovsky, J., Heimberg, R.G., Juster, H.R., Brown, T.A., & Barlow, D.H. (1997).
Validation of the Social Interaction Anxiety Scale and the Social Phobia Scale across the
anxiety disorders. Psychological Assessment, 9, 2127.
Carlbring, P., Brunt, S., Bohman, S., Austin, D., Richards, J., Ost, L.G., & Andersson, G.
(2007). Internet vs. paper and pencil administration of questionnaires commonly used in
panic/agoraphobia research. Computers in Human Behavior, 23, 14211434.
Clark, D.M., & Wells, A. (1995). A cognitive model of social phobia. In R.G. Heimberg, M.R.
Liebowitz, D.A. Hope, & F.R. Schneier (Eds.), Social phobia: Diagnosis, assessment, and
treatment (pp. 6993). New York: Guilford Press.
Curran, P.J., & Bollen, K.A. (2001). The best of both worlds: Combining autoregressive and
Downloaded by [Florida State University] at 04:38 21 December 2014

latent curve models. In L.M. Collins & A.G. Sayer (Eds.), New methods for the analysis of
change (pp. 107135). Washington, DC: American Psychological Association.
Curran, P.J., Bollen, K.A., Chen, F., Paxton, P., & Kirby, J.B. (2003). Finite sampling
properties of the point estimates and confidence intervals of the RMSEA. Sociological
Methods & Research, 32, 208252.
Fergus, T.A., Valentiner, D.P., McGrath, P.B., Stephenson, K., Gier, S., & Jencius, S. (2009).
The Fear of Positive Evaluation Scale: Psychometric properties in a clinical sample. Journal
of Anxiety Disorders, 23, 11771183. doi: 10.1016/j.janxdis.2009.07.024
Gilbert, P. (2001). Evolution and social anxiety: The role of attraction, social competition, and
social hierarchies. Psychiatric Clinics of North America, 24, 723751.
Hedman, E., Ljotsson, B., Ruck, C., Furmark, T., Carlbring, P., Lindefors, N., &
Andersson, G. (2010). Internet administration of self-report measures commonly used in
research on social anxiety disorder: A psychometric evaluation. Computers in Human
Behavior, 26, 736740. doi: 10.1016/j.chb.2010.01.010
Heimberg, R.G., & Becker, R.E. (2002). Cognitive-behavioral group therapy for social phobia:
Basic mechanisms and clinical strategies. New York: Guilford Press.
Heimberg, R.G., Mueller, G.P., Holt, C.S., Hope, D.A., & Liebowitz, M.R. (1992). Assessment
of anxiety in social interaction and being observed by others: The Social Interaction
Anxiety Scale and the Social Phobia Scale. Behavior Therapy, 23, 5373.
Henry, B., Moffitt, T.E., Caspi, A., Langley, J., & Silva, P.A. (1994). On the ‘remembrance of
things past’: A longitudinal evaluation of the retrospective method. Psychological
Assessment, 6, 92101. doi: 10.1037/1040-3590.6.2.92
Honaker, J., King, G., & Blackwell, M. (20062008). Amelia II: A program for missing data
(Version 1.2-17). Authors.
Hu, L., & Bentler, P.M. (1999). Cutoff criteria for fit indices in covariance structure analysis:
Conventional criteria versus new alternatives. Structural Equation Modeling, 6, 155.
Jöreskog, K.G., & Sörbom, D. (1981). LISREL V: Analysis of linear structural relationships by
the method of maximum likelihood. Chicago: National Educational Resources.
Kuha, J. (2004). AIC and BIC: Comparisons of assumptions and performance. Sociological
Methods & Research, 33, 188229. doi: 10.1177/0049124103262065
Leary, M.R. (1983). A brief version of the Fear of Negative Evaluation Scale. Personality and
Social Psychology Bulletin, 9, 371375.
Leary, M.R., & Kowalski, R.M. (1995). The self-presentation model of social phobia. In R.G.
Heimberg, M.R. Liebowitz, D.A. Hope, & F.R. Schneier (Eds.), Social phobia: Diagnosis,
assessment, and treatment (pp. 94112). New York, NY: Guilford Press.
Mattick, R.P., & Clarke, J.C. (1998). Development and validation of measures of social phobia
scrutiny fear and social interaction anxiety. Behaviour Research and Therapy, 36, 455470.
Moscovitch, D.A., Orr, E., Rowa, K., Reimer, S.G. & Antony, M.M. (2009). In the absence of
rose-colored glasses: Ratings of self-attributes and their differential certainty and
importance across multiple dimensions in social phobia. Behaviour Research and Therapy,
47, 6670. doi: 10.1016/j.brat.2008.10.007
182 T.L. Rodebaugh et al.

Muthén, L.K., & Muthén, B.O. (19982006). Mplus user’s guide (4th ed.). Los Angeles, CA:
Muthén & Muthén.
Rapee, R.M., & Heimberg, R.G. (1997). A cognitive-behavioral model of anxiety in social
phobia. Behaviour Research and Therapy, 35, 741756.
Rodebaugh, T.L. (2009). Hiding the self and social anxiety: The Core Extrusion Schema
Measure. Cognitive Therapy and Research, 33, 90109. doi: 10.1007/s10608-007-9143-0
Rodebaugh, T.L., Curran, P.J., & Chambless, D.L. (2002). Expectancy of panic in the
maintenance of daily anxiety in panic disorder with agoraphobia: A longitudinal test of
competing models. Behavior Therapy, 33, 315336.
Rodebaugh, T.L., Woods, C.M., & Heimberg, R.G. (2007). The reverse of social anxiety is not
always the opposite: The reverse-scored items of the Social Interaction Anxiety Scale do not
belong. Behavior Therapy, 38, 192206.
Rodebaugh, T.L., Woods, C.M., Heimberg, R.G., Liebowitz, M.R., & Schneier, F.R. (2006).
The factor structure and screening utility of the Social Interaction Anxiety Scale.
Psychological Assessment, 18, 231237.
Downloaded by [Florida State University] at 04:38 21 December 2014

Rodebaugh, T.L., Woods, C.M., Thissen, D.M., Heimberg, R.G., Chambless, D.L., & Rapee,
R.M. (2004). More information from fewer questions: The factor structure and item
properties of the original and Brief Fear of Negative Evaluation scale. Psychological
Assessment, 16, 169181.
Schwarz, G. (1978). Estimating the dimension of a model. The Annals of Statistics, 6, 461464.
Steiger, J.H., & Lind, J.C. (1980, May). Statistically-based tests for the number of factors. Paper
presented at the Annual Spring Meeting of the Psychometric Society, Iowa City, Iowa.
Tucker, L.R., & Lewis, C. (1973). A reliability coefficient for maximum likelihood factor
analysis. Psychometrika, 38, 110.
Wallace, S.T., & Alden, L.E. (1995). Social anxiety and standard setting following social
success or failure. Cognitive Therapy & Research, 19, 613631.
Wallace, S.T., & Alden, L.E. (1997). Social phobia and positive social events: The price of
success. Journal of Abnormal Psychology, 106, 416424. doi: 10.1037/0021-843x.106.3.416
Weeks, J.W., Heimberg, R.G., Fresco, D.M., Hart, T.A., Turk, C.L., Schneier, F.R., &
Liebowitz, M.R. (2005). Empirical validation and psychometric evaluation of the Brief Fear
of Negative Evaluation Scale in patients with social anxiety disorder. Psychological
Assessment, 17, 179190.
Weeks, J.W., Heimberg, R.G., & Rodebaugh, T.L. (2008). The fear of positive evaluation scale:
Assessing a proposed cognitive component of social anxiety. Journal of Anxiety Disorders,
22, 4455.
Weeks, J.W., Heimberg, R.G., Rodebaugh, T.L., & Norton, P.J. (2008). Exploring the
relationship between fear of positive evaluation and social anxiety. Journal of Anxiety
Disorders, 22, 386400.
Weeks, J.W., Jakatdar, T.A., & Heimberg, R.G. (2010). Comparing and contrasting fears of
positive and negative evaluation as facets of social anxiety. Journal of Social and Clinical
Psychology, 29, 6894. doi: 10.1521/jscp.2010.29.1.68
Weeks, J.W., Rodebaugh, T.L., Heimberg, R.G., Norton, P.J., & Jakatdar, T.A. (2009). ‘‘To
avoid evaluation, withdraw’’: Fears of evaluation and depressive cognitions lead to social
anxiety and submissive withdrawal. Cognitive Therapy and Research, 33, 375389. doi:
10.1007/s10608-008-9203-0

You might also like