Академический Документы
Профессиональный Документы
Культура Документы
com/
Published by:
http://www.sagepublications.com
On behalf of:
Additional services and information for Psychology of Women Quarterly can be found at: Email Alerts: http://pwq.sagepub.com/cgi/alerts Subscriptions: http://pwq.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav
Psychology of Women Quarterly, 32 (2008), 469482. Wiley Periodicals, Inc. Printed in the USA. Copyright C 2008 Division 35, American Psychological Association. 0361-6843/08
A multivariate model was developed to examine the relative contributions of mastery, stress, interpretive bias, and coping to sex differences in worry. Rumination was incorporated as a second outcome variable to test the specificity of these associations. Participants included two samples of undergraduates totaling 302 men and 379 women. A path analysis of the full sample demonstrated good overall fit and revealed that low mastery, high stress, and high interpretive bias predicted increases in both worry and rumination. There proved to be no interactive effect of sex with the model. Comparisons of mean differences indicated that women reported significantly lower mastery and significantly higher stress, worry, and rumination than men. Results suggest similar processes confer risk for worry and rumination in men and women; however, lower levels of mastery and higher levels of stress in women may contribute to the female preponderance of anxiety and depression.
According to population estimates, anxiety disorders represent one of the most common mental health problems in the United States, affecting approximately 40 million adults in a given year (National Institute of Mental Health, 2006). Considering the enormous cost and impairment associated with anxiety disorders, researchers have begun to call for greater efforts toward prevention (Craske & Zucker, 2002). Such programs are likely to be most feasible and have the greatest impact by utilizing a combination of selected prevention, aimed at individuals who demonstrate greater risk for the disorder, and indicated prevention, aimed at individuals with subclinical symptoms who do not yet meet full diagnostic criteria. Among the clear risk factors for anxiety, sex has been identified as an extremely robust predictor, with women almost twice as likely to develop an anxiety disorder as men (Kessler et al., 1994). With regard to key subclinical symptoms, women have also reliably reported higher levels of worry than men (McCann, Stewin, & Short, 1991; Stavosky & Borkovec, 1987). Worry has been identified as an important cognitive component of many anxiety disorders and is considered
Alyson K. Zalta and Dianne L. Chambless, Department of Psychology, University of Pennsylvania. We thank Kristopher J. Preacher for statistical consultation. Address correspondence and reprint requests to: Alyson Zalta, Department of Psychology, University of Pennsylvania, 3720 Walnut Street, Philadelphia, PA 19104-6241. E-mail: zalta@sas.upenn.edu
to be the hallmark feature of generalized anxiety disorder (GAD). Furthermore, research has suggested that subclinical worry stems from the same vulnerabilities that contribute to anxiety disorders. Results of taxometric analyses have provided evidence for the dimensionality of worry (Ruscio, Borkovec, & Ruscio, 2001), and Gana, Martin, and Canouet (2001) showed that the relationship between worry and anxiety is not bidirectional: Worry had a strong effect on anxiety, but there was no equivalent effect of anxiety on worry. Thus, examining factors associated with sex differences in worry may help to identify mechanisms that contribute to the female preponderance of anxiety disorders. According to contemporary learning theories, anxiety results from both exposure to threatening stimuli and cognitive vulnerability factors that influence the interpretation of these stressors (Mineka & Zinbarg, 2006). Correspondingly, sex differences in anxiety may be due to mens and womens differential perceptions of threatening stimuli. Feminist theories argue that the traditional feminine role, characterized by compliance, dependency, and nurturance (Fodor, 1982), may contribute to womens vulnerability by fostering maladaptive beliefs and responses to stressors (Miranda, 1995). Because girls environments are more circumscribed, they may be more likely to interpret novel stimuli as threatening, less likely to modify stressors using active coping strategies, and more likely to maintain self-perceptions of incompetence and helplessness. Feminist theory also holds that social roles contribute to the diminished social power of women (Matlin, 1987), which may increase the number of stressors they experience in their 469
470 daily lives. Although sex differences in these domains interpretive bias for threat, task-oriented coping, mastery, and stresshave been examined independently, there have been limited attempts to test a cohesive model of anxiety based on these factors. The primary goal of the current study was to develop a multivariate model to examine the relative contributions of mastery, coping, stress, and interpretive bias to worry for men and women. Sex differences in worry could be explained by two alternatives: (a) different processes account for worry in men and women with varying levels of pathogenesis or (b) similar processes account for worry, but men and women demonstrate mean differences in risk. Thus, we sought to test for both interactive and main effects of sex within the model. A secondary goal of this study was to test whether these processes contribute to specific risk for a cognitive style associated with anxiety (i.e., worry) or more general risk as evidenced by comparable prediction of a cognitive style associated with depression (i.e., rumination). In the following paragraphs, we review prior research on the relationship between each of these variables and anxiety, how these variables may account for womens increased risk for worry, and how this evidence was used to construct the hypothesized cognitive model.
ZALTA AND CHAMBLESS et al., 1994). Researchers have argued that men may be more likely to use a task-oriented coping style because boys are often encouraged to deal instrumentally with stress, whereas girls are encouraged to express their distress and seek social support. However, studies that have examined sex differences in task-oriented coping have yielded inconsistent results (e.g., Cohan et al., 2006; Endler & Parker, 1994; Kohn et al., 1994). Examining the role of coping within a larger model of risk may enable us to disentangle contradictory findings by exploring the relative contribution of this factor to sex differences in worry. Because ones coping style is thought to moderate the impact of stressors, both coping and the interaction between stress and coping were incorporated into the model as direct predictors of worry.
Interpretive Bias
In classic presentations of GAD, minor stressors cause worry that is far disproportionate to the actual level of threat. As such, interpretive bias, or the tendency to construe ambiguous stimuli as threatening, is likely to be an important predictor of worry. Research has indicated that clinically and subclinically anxious adults are more likely to select threatening interpretations of ambiguous stimuli compared to normal controls (Eysenck, Mogg, May, Richards, & Mathews, 1991; Mogg et al., 1994). Studies have also shown that inducing interpretive bias in normal participants increases subsequent anxiety (Mathews & Mackintosh, 2000). Based on this experimental evidence, interpretive bias was incorporated in the model as a direct predictor of worry. To our knowledge, sex differences in threat interpretation have not been examined as a source for womens increased vulnerability for anxiety. Feminist theories might argue that women are more likely to demonstrate such cognitive biases. According to such a view, because of limited experiences and constrained environmental exposure, women may be less capable of discerning real threats from false alarms and have a higher tendency to interpret a given stimulus as a potential threat. The current study represents a preliminary attempt to examine this issue.
Stress
Although stress is a universal predictor of anxiety, the specific triggers that signal threat for individuals with GAD are often minor daily concerns such as work, school, finances, family, illness, and interpersonal relationships (American Psychiatric Association, 1994). Feminist arguments speculate that women may experience a greater number of such daily conflicts due to competing social roles (e.g., concerns about workfamily conflicts; Kandel, Davies, & Raveis, 1985; Miranda, 1995) and diminished social power (Matlin, 1987). However, the research testing this hypothesis has shown inconsistent results (e.g., Kohn, Lafreniere, & Gurevich, 1990; Kohn, Hay, & Legere, 1994; Osman, Barrios, Longnecker, & Osman, 1994). Incorporating hassles into the current cognitive model not only allowed us to examine the direct contribution of minor stressors to sex differences in worry, but enabled us to establish the ability of other variables to predict distress beyond this fundamental risk factor. Thus, stress was treated as an exogenous variable in the model and hypothesized to directly predict worry.
Mastery
A vast literature on both animals and humans has shown that an organisms history of control over aspects of the environment influences future reactions to novel and frightening stimuli (for reviews, see Chorpita & Barlow, 1998; Mineka & Zinbarg, 2006). According to Chorpita and Barlow (1998), the collective evidence suggests that infants or children given early opportunities to master new or unpredictable situations will be immunized against stressors and buffered against anxiety, whereas individuals repeatedly exposed to uncontrollable stimuli will develop a cognitive vulnerability (i.e., diminished sense of mastery) that increases their risk for developing anxiety. Mastery has been defined
Task-Oriented Coping
Task-oriented coping refers to an active approach in which individuals consciously attempt to alter or reconceptualize stressors in ways to minimize their effect (Parker & Endler, 1992). This style of coping is considered to be an adaptive strategy and demonstrates small to moderate negative correlations with measures of psychological distress (Cohan, Jang, & Stein, 2006; Endler & Parker, 1999; Kohn
471 was specifically associated with symptoms of anxiety. Much like worry, rumination involves repetitive and often intrusive cognitive processes that result in prolonged attention to negative thought. Despite these similarities, research suggests that worry and rumination are distinct constructs (Muris, Roelofs, Meesters, & Boomsma, 2004) that differ in content (Watkins, Moulds, & Mackintosh, 2005) and maintain unique associations with anxiety and depression (Hong, 2007; McLaughlin, Borkovec, & Sibrava, 2007; Watkins et al., 2005). Rumination has been identified as an important predictor of depression and has been shown to mediate sex differences in depressive symptoms (Nolen-Hoeksema et al., 1999). Consequently, both worry and rumination may represent important mechanisms by which men and women incur differential risk for affective disorders.
Current Study
In the current study, we sought to identify the unique pathways by which women and men demonstrate increased worry by integrating a set of cognitive risk factors into a single framework. We proposed a model in which low mastery acts as the primary source of womens increased risk for worry and rumination. However, it could be argued that one develops a diminished sense of mastery as a result of intrusive distressing thoughts. After establishing a well-fitting model roughly consistent with our hypothesis, we sought to test an alternative model in which mastery was specified as an outcome of worry and rumination rather than a precursor. We then sought to test for both interactive and mean differences in men and women, as well as the specificity of these associations in predicting worry and not rumination. The cross-sectional design limited our ability to test causal relationships, but simultaneous assessment of these variables enabled preliminary examination of the proposed pathways in a cohesive model. METHOD
Specicity
Given the high rates of comorbidity between anxiety and depression (Kessler et al., 2003) and the fact that women are also at a twofold greater lifetime risk for depression than men (Kessler, McGonagle, Swartz, Blazer, & Nelson, 1993), it is quite possible that common underlying mechanisms contribute to sex differences in the development of both anxiety and mood disorders. In fact, many of the aforementioned risk factors have been associated with depression, including mastery (Nolen-Hoeksema et al., 1999; Pearlin et al., 1981), stress (DAngelo & Wierzbicki, 2003; Morrison & OConnor, 2005), and task-oriented coping (Cohan et al., 2006; Felsten, 1998; Matheson & Anisman, 2003). Interpretive bias for threat has been identified as a potentially unique contributor to anxiety in that an enhanced sense of danger may be particularly relevant to apprehension about future events (Butler & Mathews, 1983; Hertel, 2002). However, empirical evidence examining the relative contribution of interpretive bias to anxiety versus depression is not conclusive (e.g., Butler & Mathews, 1983; Chan & Lovibond, 1996; MacLeod, Mathews, & Tata, 1986; Mathews & Mackintosh, 2000). Incorporating a correlate of depression into the current model allowed us to examine whether this variable
Participants
University of Pennsylvania undergraduates were recruited from the psychology department study pool. Two independent samples were collected during the fall (n = 377) and spring (n = 304) terms of the 20062007 academic year. The overall sample (N = 681) included 302 men (44.4%) and 379 women (55.6%) ranging in age from 18 to 26 years old (M = 18.3, SD = 1.0, n = 601). The racial composition of the sample was 62.4% White, 19.2% Asian, 6.9% Hispanic or Latino, 5.1% Black or African American, 0.4% American Indian or Alaskan Native, 0.2% Native Hawaiian or Pacific Islander, 4.6% other, and 1.2% unknown. The Depression Anxiety Stress Scales21 (Lovibond & Lovibond, 1993) were used to assess clinical levels of depression and anxiety in the current sample. The majority of the sample reported normal to moderate levels of distress. Based on cutoff scores (Lovibond & Lovibond, 1993),
472 14.6% (n = 97/664) reported severe depression, 15.9% severe anxiety (n = 106 / 666), and 12.8% severe stress (i.e., chronic tension; n = 84/659).
ZALTA AND CHAMBLESS coping. Respondents were asked to indicate how much you engage in these types of activities when you encounter a difficult, stressful, or upsetting situation on a 5-point Likerttype scale. The short 21-item form has been shown to have a similar factor structure to the original 48-item measure (Cohan et al., 2006). Because the focus in this study was on the problem-solving behaviors that people use in stressful situations, we included only the task-oriented coping subscale (7 items in length) to reduce subject burden. In a large sample of undergraduates, measures of anxiety and depression were negatively correlated with the task-oriented subscale (Cohan et al., 2006). Cronbachs alpha for the current sample was .85. Penn State Worry Questionnaire (PSWQ). The PSWQ (Meyer, Miller, Metzger, & Borkovec, 1990) is a self-report measure designed to assess the extent to which individuals engage in chronic worry. Respondents are asked to rate themselves on 16 items, such as Ive been a worrier all my life, on a scale from 1 (not at all typical of me) to 5 (very typical of me). The PSWQ has strong psychometric properties with particularly good sensitivity and specificity in discriminating individuals with generalized anxiety disorder (Brown, Antony, & Barlow, 1992; Fresco, Mennin, Heimberg, & Turk, 2003; Meyer et al., 1990). Meyer et al. (1990) reported high test-retest reliabilities of .93 over 1 month and .92 over 810 weeks. Cronbachs alpha for the current sample was .93. Ruminative Responses Scale (RRS). The RRS (NolenHoeksema & Morrow, 1991) asks respondents to report what they typically think or do when in a depressed mood. The 22 items are rated on a scale from 1 (almost never) to 4 (almost always) and assess the individuals level of selffocus, focus on symptoms, and focus on the causes and consequences of depression. The measure has demonstrated good convergent validity and reliability, with an alpha coefficient of .90 and 1-year test-retest correlation of .67. Cronbachs alpha for the current sample was .94. Depression Anxiety Stress Scales21 (DASS-21). The DASS-21 (Lovibond & Lovibond, 1993) is a set of three selfreport scales designed to measure the negative emotional states of depression, anxiety, and stress. This measure was used to assess clinical levels of depression and anxiety to permit a description of the sample. Respondents are asked to use 4-point severity/frequency scales to rate the extent to which they have experienced each state over the past week. In both clinical and normal samples, the DASS scales have shown high internal consistency and good convergent and discriminant validity when compared to other measures of anxiety and depression (Brown, Chorpita, Korotitsch, & Barlow, 1997; Clara, Cox, & Enns, 2001). In the current sample, Cronbachs alpha was .88 for the depression scale, .82 for the anxiety scale, and .86 for the stress scale.
Measures
Self-Mastery Scale (SMS). The SMS (Pearlin & Schooler, 1978) is a widely used measure that assesses the extent to which people perceive their lives as being under their own control. The scale contains 7 items rated on a scale from 1 (strongly disagree) to 4 (strongly agree). Cronbachs alphas have been reported from 0.75 to 0.78 (DeSocio, Kitzman, & Cole, 2003; Marshall & Lang, 1990; Mizell, 1999), and the 4-year stability coefficient has been reported as 0.44 (Pearlin et al., 1981). The measure has demonstrated the ability to longitudinally predict changes in depression (Pearlin et al., 1981) and treatment outcome for patients with panic disorder and agoraphobia (Bowen, South, Fischer, & Looman, 1994). Cronbachs alpha in the current sample was .78. Inventory of College Students Recent Life Experiences (ICSRLE). The ICSRLE (Kohn et al., 1990) assesses the number of daily hassles and stressors experienced by college students. Students rate how much they have experienced each of 49 items during the past month on a scale from 1 (not at all part of my life) to 4 (very much part of my life). The measure has demonstrated good concurrent validity with other measures of daily hassles (Osman et al., 1994) and an alpha reliability of .89 (Kohn et al., 1990). Cronbachs alpha for the current sample was .93. Modified Interpretive Bias Questionnaire (MIBQ). This measure was modeled after a self-report instrument developed by Butler and Mathews (1983). The original measure asks individuals to rank order three possible interpretations of 14 ambiguous scenarios. For each of the presented scenarios, two of the explanations are neutral and one is negative. Research has demonstrated good criterion validity for this measure, with anxious and depressed patients reporting significantly higher scores compared to matched controls (Butler & Mathews, 1983). Concerned that the ranking format would not produce a good range of scores with a nonclinical sample, we developed a modified questionnaire1 for the purposes of this investigation. This version includes the same scenarios as the original, but asks respondents to rate how likely it is that the negative explanation would pass through your mind if you found yourself in a similar situation on a scale of 1 (not very likely) to 7 (very likely). Cronbachs alpha for the current sample was .88. Coping Inventory for Stressful SituationsShort Form (CISS-SF). The CISS-SF (Endler & Parker, 1990) assesses the extent to which individuals use three different styles of coping: task-oriented, emotion-oriented, and avoidance
473 constraining the path from interpretive bias to worry equal to the path from interpretive bias to rumination. Two types of power are relevant to covariance structural modeling techniques: power to test overall model fit and power to test differences in incremental fit between nested models. Root mean square error of approximation (RMSEA; ) values are commonly used to establish the boundaries to detect between the null (0 ) and alternative (a ) hypotheses in the case of overall fit or between the inclusive model (i ) and the nested model (n ) in the case of incremental fit. For the current study, SAS programs (SAS Institute Inc., 1992) provided by MacCallum and colleagues (MacCallum, Browne, & Cai, 2006; MacCallum, Browne, & Sugawara, 1996) were used to make power determinations for all relevant models using = .05, 0 = .10, a = .05, i = .05, and n = .07.
Procedure
After providing informed consent, students anonymously completed a set of online computer questionnaires for extra credit or in partial fulfillment of course research requirements. Research has indicated that computer versus paper-and-pencil administration has little effect on the psychometric properties of self-report questionnaires (Keel, Heatherton, Dorer, Joiner, & Zalta, 2006) and is often preferred by students (Vispoel, Boo, & Bleiler, 2001). This study was approved by the Human Subjects Committee of the University of Pennsylvanias Institutional Review Board.
RESULTS
Preliminary Analyses
All preliminary analyses were conducted using Stata software version 9.2 (StataCorp, 2005). Of the 681 participants, 645 completed all surveys, and 36 were missing data on at least one of the survey measures. Independent samples t tests were conducted to test for differences between those with and without missing data. Results indicated that individuals with missing data had significantly lower scores on the Inventory of College Students Recent Life Experiences (p < .05) and significantly higher scores on the Penn State Worry Questionnaire (p < .05). Thus, the data did not meet criteria for missing at random (MAR). Although we were unable to determine whether data were MAR or not missing at random (NMAR), researchers have argued that MAR assumptions are often a reasonable approximation under conditions of NMAR (e.g., Allison, 2003). Simulation studies have shown that full-information maximum likelihood (FIML) estimation is superior to listwise or pairwise deletion approaches under conditions of MAR (Muthen, Kaplan, & Hollis, 1987). Thus, FIML missing data estima & Muthen, tion was implemented using MPlus (Muthen 1998) for all models in the current study. Estimated correlation matrices based on missing data imputation are presented in Table 2 for the full sample and separately for men and women. Measures were tested for univariate normality, and all measures included in the model demonstrated significant skew (p < .01 for all). Consequently, we used an MPlus estimator that produces a maximum likelihood parameter estimate with standard errors and a 2 test statistic that is & Muthen, 1998). robust to nonnormality (MLR; Muthen Because the MLR estimator produces a scaled 2 statistic that does not reflect a normal 2 distribution, the Satorra and Bentler (1999) scaled 2 difference test (TRd) was used to examine the difference between nested models. RMSEA confidence intervals are not reported because procedures
Analysis
Path analysis was conducted to test the models using MPlus & Muthen, 1998). Data colsoftware version 4.2 (Muthen lected in the fall (n = 377) were initially used to test the hypothesized model, and data collected in the spring (n = 304) were used for the purposes of replication. After demonstrating adequate stability in the two independent samples, we combined data from the two terms in order to examine the hypothesized model with maximum power. The hypothesized model was then compared to an alternative model in which mastery was specified as an outcome of worry and rumination. Indices of both absolute and incremental fit are reported, as recommended by Hu and Bentler (1998). After establishing the best fitting model for the overall sample, our primary interest was to evaluate whether the relationships indicated by this model varied as a function of sex. To test this, we conducted two-group path analyses under two conditions: an unconstrained condition in which model parameters were allowed to vary as a function of sex and a constrained condition in which path coefficients, intercepts, correlations, and correlated errors were set to be invariant for men and women. A 2 difference test was then used to examine the comparative fit of these nested models. A secondary goal was to examine whether interpretive bias would demonstrate specificity by having a significantly stronger association with worry than rumination. We computed a Wald chi-square test to examine the effect of
474
ZALTA AND CHAMBLESS Table 1 Estimated Means, Standard Deviations, and Comparisons by Sex for Overall Sample
Overalla Measure SMS CISST MIBQ ICSRLE PSWQ RRS M 21.81 26.49 37.54 103.39 45.76 44.81 SD 3.18 4.83 11.95 21.24 12.58 14.22 M 22.19 26.51 37.59 100.95 43.74 43.05 Menb SD 3.37 4.83 12.37 22.16 12.96 14.04 M 21.50 26.47 37.50 105.33 47.34 46.18 Womenc SD 3.02 4.83 11.60 20.25 11.97 14.17 d (95% CI) .22 (.06, .37) .01 (.14, .16) .01 (.14, .16) .21 (.36, .06) .29 (.44, .14) .22 (.37, .07)
Note. Values represent estimated means and standard deviations based on full-information maximum likelihood missing data imputation using MPlus version 4.2. SMS = Self-Mastery Scale, CISST = Coping Inventory for Stressful Situations Task Orientation, MIBQ = Modified Interpretive Bias Questionnaire, ICSRLE = Inventory of College Students Recent Life Experiences, PSWQ = Penn State Worry Questionnaire, RRS = Ruminative Responses Scale. a N = 681. b n = 302. c n = 379. Significant after Bonferroni correction at p = .01.
have not yet been developed to compute confidence limits for MLR estimation. Participants estimated mean scores and standard deviations based on missing data imputation are presented in Table 1. Table 1 also includes analyses for sex differences on each of the measures. Compared to men, women reported significantly lower levels of mastery and significantly higher levels of stress, worry, and rumination. Effect sizes were small. To assess whether observed sex differences were associated with reporting biases, four multiple regression equations were performed. Mastery, stress, worry, and rumination were entered separately as dependent variables and sex, social desirability, and the interaction between sex and social desirability were entered together as independent variables. Independent variables were centered for analysis as recommended by Aiken and West (1991). Because the interaction term was not a significant predictor of any of the outcome variables (p > .10 for all), social desirability was not included in path analyses.
Path Analyses
The hypothesized model (Model A, Figure 1) was first tested using the fall term data (n = 377). This model predicted that mastery mediated by coping and interpretive bias would be associated with both worry and rumination. Stress and the interaction between stress and coping were predicted to have direct effects on worry and rumination. Stress and coping were centered for these analyses (see Aiken & West, 1991). As required in path analyses, firstorder factors (e.g., stress and coping) were permitted to correlate with the second-order factor (e.g., the interaction between stress and coping) and with each other. Correlations were also permitted for all exogenous variables. A correlated disturbance was specified for the PSWQ and RRS due to the close relationship between anxiety and depression. Power to test the overall fit of this model was
.62. The hypothesized model demonstrated poor overall fit based on all four fit statistics used, 2 (5, N = 377) = 89.34, p < .0001, comparative fit index (CFI) = 0.81, RMSEA = 0.21, standardized root mean squared residual (SRMR) = 0.08. Modification indices that would provide a significant reduction in the chi-square were supplied by MPlus and used to generate an alternative, post hoc model that would provide more adequate fit to the data. Following recommendations from Bollen (1989), we sought to include a minimum number of modifications that were theoretically consistent with the a priori model. A large number of significant modification indices were produced for the hypothesized model. Of these, three additional paths were deemed theoretically justifiable and were added to a revised model: direct paths from mastery to worry and rumination, and a direct path from stress to interpretive bias. Power to test overall fit was .34 for the revised model (Model B). This model provided excellent fit to the data according to all four fit statistics, 2 (2, N = 377) = 1.09, p = 0.58, CFI = 1.00, RMSEA = 0.00, SRMR = 0.01. Given the post hoc nature of the revised model, we sought to replicate findings in an independent sample. Data collected in the spring term (n = 304) were used to test the stability of the revised model (Model B). Power to test for overall fit was .37. The revised model demonstrated good fit to the spring term data based on all four fit indices, 2 (2, N = 304) = 0.14, p = 0.93, CFI = 1.00, RMSEA = 0.00, SRMR = 0.00. An examination of the individual parameter estimates for the revised model in the fall and spring samples indicated that only two paths changed in significance.2 To test whether this represented a significant difference in the samples, two-group path analyses were conducted under an unconstrained condition in which parameters were allowed to vary as a function of the sample and a constrained condition in which path coefficients, intercepts, correlations, and correlated errors were set to be invariant
475
(.93) .413
(.94)
Correlation matrix by sex SMS SMS CISST MIBQ ICSRLE PSWQ RRS .376 CISST .369 .176 .088 .097 .041 MIBQ .327 .091 .479 .412 .440 ICSRLE .225 .409 .380 .518 .576 PSWQ .399 .107 .383 .379 .465 RRS .439 .196 .333 .515 .353
Note. In the full sample matrix (N = 681), correlations along the diagonal represent internal reliability coefficients (Cronbachs ). In the matrix by sex, correlations for men (n = 302) are below the diagonal, and correlations for women (n = 379) are above the diagonal. Values represent estimated correlations based on full-information maximum likelihood missing data imputation using MPlus version 4.2. SMS = Self-Mastery Scale, CISST = Coping Inventory for Stressful Situations Task Orientation, MIBQ = Modified Interpretive Bias Questionnaire, ICSRLE = Inventory of College Students Recent Life Experiences, PSWQ = Penn State Worry Questionnaire, RRS = Ruminative Responses Scale. p < .05. p < .01. p < .001.
Fig. 1. Hypothesized model. Stress Coping = cross product of stress and coping.
for the fall and spring samples. There proved to be no significant difference in 2 between these models, TRd(17) = 17.20, p = 0.44, indicating equivalence of the model parameters for the two samples. Thus, the spring sample replication indicated good stability of the revised model in an independent sample. As such, the fall and spring data
were combined to test the fit of the hypothesized and revised models and examine sex differences with maximum power. For the combined sample (N = 681), power to test overall fit was .92 for Model A and .55 for Model B. The original hypothesized model (Model A) demonstrated poor overall
476
Fig. 2. Revised model for overall sample, 2 (2, N = 681) = 1.19, p = 0.55, Comparative Fit Index = 1.00, Root Mean Square Error of Approximation = 0.00, Standardized Root Mean Squared Residual = 0.01. All paths are labeled with standardized coefficients, and solid lines are significant at p < .05. Circles represent residual variances. Stress Coping = cross product of stress and coping.
fit based on all four fit statistics, 2 (6, N = 681) = 149.45, p < .0001, CFI = 0.77, RMSEA = 0.19, SRMR = 0.08. By contrast, the revised model (Model B) demonstrated good fit based on all four fit statistics, 2 (2, N = 681) = 1.19, p = 0.55, CFI = 1.00, RMSEA = 0.00, SRMR = 0.01. The overall fit of this model (Model B) was maintained when analyses were limited to those with complete data (n = 645), providing further confidence in these findings. The revised model (Model B) was used to generate an alternative model (Model C, Figure 3) in which mastery was specified as a direct outcome of worry and rumination, rather than a predictor of interpretive bias, coping, worry, and rumination. All other paths remained the same for Model B and Model C. Power to test the overall fit of Model C using the full sample (N = 681) was .92. Although the SRMR (0.06) was in an acceptable range, the alternative model (Model C) demonstrated poor fit based on all other fit statistics, 2 (6, N = 681) = 109.60, p < .0001, CFI = 0.88, RMSEA = 0.16. As such, the revised model (Model B) appeared to provide superior fit to the data and was used in all subsequent analyses of sex differences and specificity. Examination of the individual paths for Model B indicated that high levels of stress and low levels of mastery were significantly associated with increases in interpretive bias, and all three variables were significantly associated with increases in worry and rumination. This model accounted for 30% of the variance in worry and 37% of the variance in rumination. Parameter estimates for this model
are presented in Table 3 and standardized path coefficients in Figure 2. The revised model (Model B) was used to conduct twogroup path analyses under an unconstrained condition in which model parameters were allowed to vary as a function of sex and a constrained condition in which path coefficients, intercepts, correlations, and correlated errors were set to be invariant for men and women. There proved to be no significant difference in 2 between these models, TRd(17) = 24.48, p = 0.11, despite adequate power (.997) to test for incremental fit. These findings indicated the absence of sex interactions with the predictors of worry and rumination. Thus, parameter estimates and path diagrams are presented only for the full sample. The revised model (Model B) was also used to test the effect of constraining the path from interpretive bias to worry, equal to the path from interpretive bias to rumination. The Wald chi-square test was not significant, 2 (1, N = 681) = .95, p = .33, indicating that interpretive bias did not have a stronger association with worry than with rumination. DISCUSSION Despite the female preponderance of anxiety disorders, surprisingly few studies have examined the mechanisms associated with womens increased risk for anxiety (Lewinsohn, Gotlib, Lewinsohn, Seeley, & Allen, 1998). We constructed a multivariate model to examine sex differences in variables associated with worry, a cognitive
477
Fig. 3. Alternative model (Model C) for overall sample, 2 (6, N = 681) = 109.60, p < .0001, Comparative Fit Index = 0.88, Root Mean Square Error of Approximation = 0.16, Standardized Root Mean Squared Residual = 0.06. All paths are labeled with standardized coefficients, and solid lines are significant at p < .05. Circles represent residual variances. Stress Coping = cross product of stress and coping.
precursor of anxiety, using rumination, a cognitive precursor of depression, to test the specificity of these associations. Path analyses indicated that high levels of stress and low levels of mastery were associated with increases in interpretive bias, and all three variables were associated with increases in worry and rumination. This model appeared to account equally well for the prediction of outcomes in both sexes, suggesting that that similar cognitive processes were associated with worry and rumination for men and women. However, mean differences between men and women indicated that womens lower mastery and higher stress may contribute to their increased vulnerability for worry and rumination. Our findings are congruous with the conceptual framework outlined by Chorpita and Barlow (1998), and they support their theory regarding sex differences in worry and rumination. According to their proposal, repeated exposure to uncontrollable events may foster psychological vulnerability that increases the likelihood of interpreting events as threatening and experiencing distress when faced with novel stimuli. Our final model demonstrated good fit and proved to be superior to a model in which a diminished sense of mastery was specified as a consequence of uncontrollable and intrusive thoughts. Drawing from feminist theories, we argue that women might be more likely to develop stable beliefs of helplessness through chronic exposure to situations in which they lack control. According to one author, mastery exploration can alleviate fear, but girls are not reinforced for this type of mastery. . . there is a differential sex role message about mastery of fears presented to males and females (Hoffman, 1972, p. 182). Research has also indicated that gender roles mediate sex
differences in psychopathology (Huselid & Cooper, 1994). Our findings are consistent with these hypotheses: Women reported significantly lower levels of mastery, which was predictive of worry and rumination within the context of a larger cognitive model. However, we did not directly assess gender-role socialization in the current study and thus are unable to determine whether alternative explanations (e.g., biological) account for the observed differences for men and women. Additionally, it is important to note that the mean differences in mastery, worry, and rumination represented small effects (d < .30). Given that participants were drawn from an elite university in which students are selected for high achievement, women in our sample should be more likely to perceive themselves as competent compared to the average woman. Thus, it is possible that our results underestimate sex differences in mastery and its relationship to psychopathology. Results from the current study also suggest that womens increased vulnerability for worry and rumination may be a direct consequence of the number of stressors they experience compared to men. Arguably, women may be more likely to experience a greater number of daily conflicts due to the nature of the female role (Kando, 1972). If womens lives are simply more stressful than mens, one might argue that stress alone would be a much more parsimonious explanation for womens increased risk for distress. However, results from the current study indicated that mastery and stress made independent contributions to the model. These findings are consistent with contemporary learning theories (Mineka & Zinbarg, 2006), which argue that perceptions of threat and control are both important determinants of anxiety. However, they also raise an important
Overallc Est. 0.78 0.56 0.90 1.04 0.19 0.17 0.27 0.21 0.15 0.14 0.12 0.00 0.00 0.13 29.26 57.57 13.90 15.74 15.02 Est. 10.12 448.13 12065.63 S.E. 0.14 0.06 0.16 0.18 0.02 0.02 0.02 0.04 0.04 0.10 0.10 0.00 0.00 3.58 3.06 181.57 40.36 18.78 4.49 S.E. 0.54 24.73 2045.02
0.23 0.24 0.03 0.03 0.03 0.06 0.05 0.14 0.14 0.01 0.00 4.50 4.36 252.12 34.76 28.12 6.60 S.E. 0.75 36.65 1710.01
Note. Relationships indicated by a one-way arrow represent directional paths. Relationships indicated by a two-way arrow represent correlations and correlated disturbances. Est. = Unstandardized parameter estimate, S.E. = Standard error, SMS = Self-Mastery Scale, CISST-C = Coping Inventory for Stressful Situations Task Orientation: centered around mean, MIBQ = Modified Interpretive Bias Questionnaire, ICSRLE-C = Inventory of College Students Recent Life Experiences: centered around mean, PSWQ = Penn State Worry Questionnaire, RRS = Ruminative Responses Scale, and SXC = Cross product of ICSLRE-C and CISST-C. Significance for each path is indicated in the S.E. column. a n = 377. b n = 304. c N = 681. z < .05 based on Est. / S.E.
question as to whether womens perceived lack of control is an exaggerated, subjective judgment or whether it reflects the reality of womens environments in which they experience less objective control over the multitude of stressors they encounter (e.g., workfamily conflict, sexual assault, and unintended pregnancy). Future studies that attempt to distinguish between these alternatives would help to clarify whether interventions that seek to enhance mastery beliefs would be effective in reducing risk for worry. A secondary goal of this study was to test the specificity of associations between cognitive factors and worry. Results showed that interpretive bias was not more strongly associated with worry than with rumination. It is possible that our modified self-report questionnaire may not have the necessary sensitivity to separate more subconscious threatening interpretations from general negative interpretations. Alternatively, it is possible that threatening interpretations are not specific to worry and play a role in ruminative thought
by identifying potential sources of loss and triggering hopelessness. Clearly, these issues of specificity are complex, and studies that seek to identify the specific mechanisms associated with worry and rumination are much needed. Our findings suggested that mastery may be a viable target for anxiety interventions. Seligman, Schulman, DeRubeis, and Hollon (1999) demonstrated that a relatively brief cognitive-behavioral group intervention for students with a pessimistic attributional style (i.e., low sense of control of the environment) helped to prevent anxiety and depression. They further demonstrated that depressive symptom prevention was more effective in women than men and was significantly mediated by explanatory style. These findings lend further support to our hypothesis that mastery may be an important etiological factor in the development of anxiety and womens increased vulnerability. Longitudinal research focused on identifying factors that contribute to sex differences in mastery, and clarifying the
479 and alcohol use. Examination of a more diverse sample with a wider range of psychopathology is needed to establish the generalizability of the current model. Due to the large sample size required, large number of variables, and internal nature of the constructs, this study relied exclusively on the use of self-report forms. A concern of the current study was whether men and women were likely to report biased information based on socially acceptable gender roles. To reduce the potential influence of such reporting biases, students were overtly reminded of anonymity prior to survey administration and responded to questionnaires using computerized assessment. The interaction between social desirability and sex did not predict any of the measures for which we observed mean sex differences, suggesting that such biases did not influence our results. However, future studies would benefit from the use of alternative methods, such as experimental manipulation, direct observation, or outsider reports, to ensure that study findings are not restricted to self-report measurement. Due to the cross-sectional design of the current study, we cannot determine the direction of causality for observed associations. Although cognitive factors may contribute to the development of psychological symptoms, it is also possible that they are manifestations of preexisting disorders. Furthermore, there may be a bidirectional relationship between factors that contributes to a downward spiral of symptoms over time (i.e., lower mastery may increase anxiety, and greater anxiety may reduce mastery). We were also unable to assess all equivalent models that would challenge our theoretical interpretation of the data given that highly saturated models such as ours yield an extremely large number of alternatives (MacCallum, Wegener, Uchino, & Fabrigar, 1993). Future studies would benefit from longitudinal examination and/or experimental manipulation to test the causal relationship between variables in the model. Despite these limitations, the current study has a number of important strengths. Most notably, we tested a multivariate model to examine the relative contribution of multiple factors to sex differences in worry. Surprisingly few studies have examined the underlying processes that contribute to sex differences in worry (Robichaud, Dugas, & Conway, 2003), and studies that have examined why women are more prone to anxiety than men have often focused on a single domain of risk. It is evident that numerous influences contribute to vulnerability, and such factors do not operate in isolation. Broader perspectives that attempt to synthesize various risk factors are needed to establish a more nuanced and complete understanding of the causal mechanisms that lead to psychopathology. The current study represents one step toward this goal. Findings suggest that mastery beliefs are particularly important in the expression of psychological distress for women and may be one avenue of selected intervention for those at greatest risk.
Initial submission: December 15, 2007 Initial acceptance: April 12, 2008 Final acceptance: June 4, 2008
480
NOTES 1. Copies of the modified questionnaire may be obtained from the first author. 2. Significance was determined for individual paths by dividing the parameter estimate by the standard error. In large samples, this value is approximately normally distributed and is treated as a z score. The critical value for a two-tailed test at the .05 level is an absolute value greater than 1.96 (Muthen 1998). & Muthen,
REFERENCES
Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting interactions. Newbury Park, CA: Sage. Allison, P. D. (2003). Missing data techniques for structural equation modeling. Journal of Abnormal Psychology, 112, 545 557. American Psychiatric Association. (1994). Diagnostic and statistical manual of mental disorders (4th ed.). Washington, DC: Author. Billings, A. G., & Moos, R. H. (1981). The role of coping responses and social resources in attenuating the stress of life events. Journal of Behavioral Medicine, 4, 139157. Bollen, K. A. (1989). Structural equations with latent variables. New York: Wiley-Interscience. Bowen, R., South, M., Fischer, D., & Looman, T. (1994). Depression, mastery and number of group sessions attended predict outcome of patients with panic and agoraphobia in a behavioural/medication program. Canadian Journal of Psychiatry, 39, 283288. Brown, T. A., Antony, M. M., & Barlow, D. H. (1992). Psychometric properties of the Penn State Worry Questionnaire in a clinical anxiety disorders sample. Behaviour Research and Therapy, 30, 3337. Brown, T. A., Chorpita, B. F., Korotitsch, W., & Barlow, D. H. (1997). Psychometric properties of the Depression Anxiety Stress Scales (DASS) in clinical samples. Behaviour Research and Therapy, 35, 7989. Butler, G., & Mathews, A. (1983). Cognitive processes in anxiety. Advances in Behaviour Research and Therapy, 5, 5162. Chan, C. K., & Lovibond, P. F. (1996). Expectancy bias in trait anxiety. Journal of Abnormal Psychology, 105, 637647. Chorpita, B. F., & Barlow, D. H. (1998). The development of anxiety: The role of control in the early environment. Psychological Bulletin, 124, 321. Clara, I. P., Cox, B. J., & Enns, M. W. (2001). Confirmatory factor analysis of the Depression-Anxiety-Stress Scales in depressed and anxious patients. Journal of Psychopathology and Behavioral Assessment, 23, 6167. Cohan, S. L., Jang, K. L., & Stein, M. B. (2006). Confirmatory factor analysis of a short form of the Coping Inventory for Stressful Situations. Journal of Clinical Psychology, 62, 273283. Craske, M. G., & Zucker, B. G. (2002). Prevention of anxiety disorders: A model for intervention. Applied & Preventive Psychology, 10, 155175. DAngelo, B., & Wierzbicki, M. (2003). Relations of daily hassles with both anxious and depressed mood in students. Psychological Reports, 92, 416418.
481
MacLeod, C., Mathews, A., & Tata, P. (1986). Attentional bias in emotional disorders. Journal of Abnormal Psychology, 95, 1520. Marshall, G. N., & Lang, E. L. (1990). Optimism, self-mastery, and symptoms of depression in women professionals. Journal of Personality and Social Psychology, 59, 132139. Matheson, K., & Anisman, H. (2003). Systems of coping associated with dysphoria, anxiety and depressive illness: A multivariate profile perspective. Stress, 6, 223234. Mathews, A., & Mackintosh, B. (2000). Induced emotional interpretation bias and anxiety. Journal of Abnormal Psychology, 109, 602615. Matlin, M. W. (1987). The psychology of women. New York: Holt, Rinehart, & Winston. McCann, S. J., Stewin, L. L., & Short, R. H. (1991). Sex differences, social desirability, masculinity, and the tendency to worry. Journal of Genetic Psychology, 152, 295301. McLaughlin, K. A., Borkovec, T. D., & Sibrava, N. J. (2007). The effects of worry and rumination on affect states and cognitive activity. Behavior Therapy, 38, 2338. Meyer, T. J., Miller, M. L., Metzger, R. L., & Borkovec, T. D. (1990). Development and validation of the Penn State Worry Questionnaire. Behaviour Research and Therapy, 28, 487495. Miller, J. B. (1976). Toward a new psychology of women. Boston: Beacon Press. Mineka, S., & Zinbarg, R. (2006). A contemporary learning theory perspective on the etiology of anxiety disorders. American Psychologist, 61, 1026. Miranda, J. (1995). Feminism in psychology: Accelerating paradigmatic shifts (Vol. 1). Bombay, India: Research Centre for Womens Studies, S.N.D.T. Womens University. Mizell, C. A. (1999). African American mens personal sense of mastery: The consequences of the adolescent environment, self-concept, and adult achievement. Journal of Black Psychology, 25, 210230. Mogg, K., Bradley, B. P., Miller, T., Potts, H., Glenwright, J., & Kentish, J. (1994). Interpretation of homophones related to threat: Anxiety or response bias effects? Cognitive Therapy and Research, 18, 461477. Morrison, R., & OConnor, R. C. (2005). Predicting psychological distress in college students: The role of rumination and stress. Journal of Clinical Psychology, 61, 447460. Muris, P., Roelofs, J., Meesters, C., & Boomsma, P. (2004). Rumination and worry in nonclinical adolescents. Cognitive Therapy and Research, 28, 539554. B., Kaplan, D., & Hollis, M. (1987). On structural equaMuthen, tion modeling with data that are not missing completely at random. Psychometrika, 52, 431462. B., & Muthen, L. (1998). Mplus users guide. Los AngeMuthen, & Muthen. les: Muthen National Institute of Mental Health. (2006). Anxiety disorders. Bethesda, MD: U.S. Department of Health and Human Services. Nolen-Hoeksema, S. (2000). The role of rumination in depressive disorders and mixed anxiety/depressive symptoms. Journal of Abnormal Psychology, 109, 504511. Nolen-Hoeksema, S., Larson, J., & Grayson, C. (1999). Explaining the gender difference in depressive symptoms. Journal of Personality and Social Psychology, 77, 10611072.
482
Nolen-Hoeksema, S., & Morrow, J. (1991). A prospective study of depression and posttraumatic stress symptoms after a natural disaster: The 1989 Loma Prieta Earthquake. Journal of Personality and Social Psychology, 61, 115121. Osman, A., Barrios, F. X., Longnecker, J., & Osman, J. R. (1994). Validation of the Inventory of College Students Recent Life Experiences in an American college sample. Journal of Clinical Psychology, 50, 856863. Parker, J. D. A., & Endler, N. S. (1992). Coping with coping assessment: A critical review. European Journal of Personality, 6, 321344. Pearlin, L. I., Menaghan, E. G., Lieberman, M. A., & Mullen, J. T. (1981). The stress process. Journal of Health and Social Behavior, 22, 337356. Pearlin, L. I., & Schooler, C. (1978). The structure of coping. Journal of Health and Social Behavior, 19, 221. Robichaud, M., Dugas, M. J., & Conway, M. (2003). Gender differences in worry and associated cognitive-behavioral variables. Journal of Anxiety Disorders, 17, 501516. Ruscio, A. M., Borkovec, T. D., & Ruscio, J. (2001). A taxometric investigation of the latent structure of worry. Journal of Abnormal Psychology, 110, 413422. SAS Institute Inc. (1992). The CALIS procedure extended users guide. Cary, NC: Author.