J Soc Clin
Psychol. Author manuscript; available in PMC 2014 Nov 20. Published in final edited form as: PMCID: PMC4238306 NIHMSID: NIHMS602665 In a two-wave, longitudinal study, 221 nonreferred adolescents completed measures of guilt, shame, Behavioral self-blame (BSB), Characterological self-blame (CSB), depressive
symptoms measures, and attributional style. Goals were to examine similarities between Tangney’s (1996) conceptualization of guilt and shame and Janoff-Bulman’s (1979) conceptualization of BSB and CSB in adolescents. Specific aims were (1) to examine the
relation of depressive symptoms and depressive cognitions to shame, guilt, CSB, and BSB, (2) to estimate the longitudinal relations between depressive symptoms and measures of guilt, shame, BSB, and CSB, and (3) to assess the convergent and discriminant validity of shame/CSB measures and guilt/BSB measures. Results suggest that shame and CSB converge into a common construct, significantly related to depressive symptoms and cognitions. Convergence of guilt and BSB, however, was limited to
particular pairs of measures. One research tradition has focused on the relation of guilt and shame to depression, as well as anger, hostility, substance abuse, eating disorders, low self-efficacy, and poor problem solving (Covert, Tangney, Maddux, & Heleno, 2003;
Dearing, Stuewig, & Tangney, 2005; Ferguson, Stegge, Miller, & Olsen, 1999; Luyten, Fontaine, & Corveleyn, 2002;
Sanftner, Barlow, Marschall, & Tangney, 1995; Tangney, Burggraf, & Wagner, 1995; Tangney, Wagner, & Gramzow, 1992;
Woien, Ernst, Patock-Peckham, & Nagoshi, 2003). Another research tradition has studied the relation of behavioral and characterological self-blame to depression as well as loneliness, isolation, uncontrollability, helplessness, and stressful life events (Anderson, Horowitz, &
French, 1983; Cole, Peeke, & Ingold, 1996; Janoff-Bulman, 1979; Peterson, 1979;
Peterson, Schwartz, & Seligman, 1981; Stoltz & Galassi, 1989). Despite noteworthy conceptual similarities between these two pairs of constructs, almost no one has examined them conjointly (cf.
Lutwak, Panish, & Ferrari, 2003). Linking these two traditions could have implications for theory and research pertaining to a very wide range of psychological and mental health outcomes. The overarching goal of the current study is to examine the relation of the guilt/shame constructs and the behavioral/characterological self-blame constructs to each other, to depression, and to
depression-related cognitions, potentially paving the way for a clearer unification of these two research traditions. In this paper, we adhere to definitions of guilt and shame first articulated by Lewis (1971) and subsequently elaborated by Tangney (e.g.,
Tangney, Wagner, & Gramzow, 1992). Guilt is an emotion characterized by tension, regret, and remorse about a particular action or inaction. Guilt has a cognitive component, in that the negative emotion presupposes one’s responsibility for the problematic action or inaction. The attribution of responsibility to one’s own behavior (something that can be changed) is important,
insofar as it suggests and motivates different behaviors that can make amends for the previous transgression (Ferguson et al., 1999). Following a negative outcome, a guilt experience might begin with the thought, “Look at the horrible thing I have done.” Alternatively, shame is an emotion characterized by a sense of shrinking, smallness, worthlessness, and exposure
(Tangney, Wagner, & Gramzow, 1992). It too has a cognitive component in that the negative outcome derives from a helpless concern that the self will be revealed as defective (Tangney & Fischer, 1995). Feeling shame often involves the entire self, requires a high degree of
self-scrutiny, and can elicit a desireto escape while feeling frozen or unable to get away (Sabini & Silver, 1997). Following a negative outcome, a shame experience might begin with“Iam a horribleperson” (Lewis,1974;
Lindsay-Hartz, de-Rivera, & Mascolo, 1995; Tangney, 1995). By these definitions, we would expect guilt and shame to be somewhat correlated with one another but differentially related to emotional outcomes such as depression. If guilt motivates constructive, restorative behavior, it
should help the individual to avoid depression. Alternatively, if shame focuses attention on possible defects in oneself and involves a maladaptive sense of personal worthlessness, it may facilitate depression (Tangney, 1996). Several studies support these predictions. In general, measures of guilt correlate negatively or negligibly to outcomes such as depression, anger,
aggression, and withdrawal, whereas measures of shame correlate positively with these same problematic outcomes (Hoglund & Nicholas, 1995; Luyten et al., 2002; Tangney et al., 1995; Tangney, Wagner, & Gramzow, 1992; Woien et al., 2003). This relation of guilt and shame to depression may be understood in terms of attribution theory. According to Abramson, Seligman,
and Teasdale (1978), a helpless attributional style (AS) is one in which a person interprets negative events as having internal, global, and stable causes. The tripartite belief that the cause of a negative event is attributable to one’s self, is stable over time, and has global implications is a cognitive risk factor for depression (Abramson, Metalsky, & Alloy, 1989;
Hankin & Abramson, 2002). The cognitive components of both guilt and shame reflect internal attributions. Guilt and shame differ, however, in that shame involves beliefs that the internal causes are stable over time and have global implications, whereas guilt does not. Shame-related cognitions reflect a helpless AS. Several studies support these hypotheses
(Brown & Weiner, 1984; Covington & Omelich, 1981; Jagacinski & Nicholls, 1984; Pineless, Street, &
Koenen, 2006; Russell & McAuley, 1986; Tangney, 1992, 1998; Tangney & Dearing, 2002b;
Tangney, Wagner, & Gramzow, 1992; Tracy & Robins, 2006; Weiner, Graham, & Chandler, 1982). Janoff-Bulman (1979) first defined behavioral self-blame and characterological self-blame. Behavioral self-blame (BSB) reflects a cognitive style in which the individuals attribute the cause of negative events to the fact that they initiated (or failed to initiate) specific actions. As such, BSB is a control-related attribution that
focuses on the execution of behaviors that cause negative outcomes (or on the failure to execute behaviors that could have prevented negative outcomes). Although BSB focuses on prior behavior, it can also motivate behavioral change so as to avoid similar negative outcomes in the future. For example, Janoff-Bulman suggested that after the extreme negative outcome of rape, a BSB response may be, “I shouldn’t have walked down that street alone,” or “I should not have let that particular man into my
apartment.” In contrast, characterological self-blame (CSB) is a self-reflective cognitive style in which people blame their person and fault their character. Often characterized by self-criticism, CSB-responders often regard themselves as blameworthy or deserving of the negative outcome. Part of CSB is looking at the past and focusing on the negative implications about the self. Janoff-Bulman suggested that following a negative outcome, a CSB response may be “I am weak,” or “I am a careless
person who is unable to stay out of trouble.” Like guilt and shame, BSB and CSB are linked to depression in opposite ways. On the one hand, BSB is adaptive insofar as it centers attention on potentially controllable causes of negative events, potentially circumventing depression. On the other hand, CSB focuses less on specific behaviors and more on underlying, generalized, personological attributes. As such CSB can be maladaptive insofar as it centers attention on causes that are
much more difficult to control. Empirical work generally supports these speculations (Anderson et al., 1983; Anderson, Jennings, & Arnoult, 1988; Cole et al., 1996; Graham & Juvonen, 1998; Janoff-Bulman, 1979; Peterson et al., 1981;
Stoltz & Galassi, 1989). The relation of BSB and CSB to depression can also be understood in terms of attribution theory. Janoff-Bulman (1979) suggested that CSB involves internal, global, and stable attributions (and therefore should relate to depression). Conversely, she
suggested that BSB involves internal, but not global or stable attributions (and therefore should not relate to depression). These predictions have been supported empirically insofar as Peterson et al. (1981) reported that CSB correlates with internal, stable, and global attributions, and that BSB correlates with internal, but not global or stable attributions. The
similarities of BSB and CSB to the cognitive components of guilt and shame suggest that these two literatures might be mutually informative. To date, however, only one study has simultaneously examined the constructs of guilt, shame, BSS, and CSB (Lutwak et al., 2003). Although Lutwak et al. relied on only one measure of guilt and shame (i.e., Test of Self-Conscious Affect;
Tangney, Wagner, & Gramzow, 1989) and only one measure of BSB and CSB (i.e., Characterological/Behavioral Self-Blame Scale; Janoff-Bulman, 1979), their results provide preliminary evidence of a connection between these constructs. After partialing out guilt, shame and CSB were
positively correlated. Likewise, after partialing out shame, guilt was positively correlated with BSB. Somewhat confusing the picture, however, Lutwak et al. also discovered small but significant correlations between shame and BSB and between guilt and CSB. In the current study, our goal is to provide a clearer picture by obtaining multiple measures of these constructs. Adolescence is a time of noteworthy cognitive and emotional development.
Kohlberg (1984) and Mitchell (1975) describe adolescent changes in moral development, in which teens begin to question earlier, more doctrinaire notions of right and wrong and develop a more complex and relativistic morality.
Lapsley (2006) describes the re-emergence of a kind of adolescent egocentrism, leading youth to feel as though they are the center of attention, potentially exacerbating feelings of self-consciousness and shame. Adolescence is also a time when the rate of depression increases rather dramatically. Preadolescent estimates of major depression range from 2 to 4 percent. During adolescence, point
prevalence rates rise to somewhere between 9 and 25 percent (Boyle et al., 1987; Hankin, 2006; Kessler, Avenevoli, & Merikangas, 2001). Given
the teenage rise in depression and given the concurrent cognitive developmental changes in factors related to depression, we elected to focus the current study on adolescence. The current study had three specific aims. First, we assessed the relations of guilt, shame, BSB, and CSB to depressive symptoms and depression-related cognitions. We expected all four constructs to correlate with the tendency to make internal attributions for negative events; however, we expected shame and
CSB to correlate more strongly than guilt and BSB with (a) the tendency to make stable and global attributions about negative events, (b) the expectation that negative events are likely in the future, (c) the inclination to perceive one’s self as flawed, and (d) the report of concurrent depressive symptoms. Second, we examined the prospective, longitudinal relations of guilt, shame, BSB, and CSB to symptoms of depression (and vice versa). Previous theory and research implicate cognitive styles
as risk factors for and as consequences of depression (Abramson et al., 1989; Abramson et al., 1978; Hankin & Abramson, 2002;
Rohde, Lewinsohn, & Seeley, 1990); consequently, we examined both possibilities. Finally, we examined the relation of various measures of guilt, shame, BSB, and CSB to one another. We expected some correspondence between measures of shame and CSB and between measures of guilt and BSB; however, we did not expect measures of shame and CSB to converge with measures of guilt and
BSB. A total of 221 adolescents participated in this study. Participants attended either a middle school or a high school just outside a midsize Southeastern city. To recruit students, we sent 641 informed consent forms and letters to student’s families in the two schools. Of the 641 students contacted, 226 (35%) obtained permission from parents to participate. In the five months between
the two time points of the study, a few students withdrew (11.8%) and a few were added (8.6%). Essentially, all attrition was the result of students moving out of the school district or being absent on the day of the assessment. All additions to the study were students who had been absent at the first assessment. We assessed the impact of “missingness” by comparing each of the 2 missing-data groups to the participants with no missing data on the nonmissing study variables. Out of 31 such
comparisons, 5 were statistically significant at p < .05. Student attrition was associated with slightly higher levels of TOSCA-A Guilt, TOSCA-A Shame, WIHQ-BSB, ACSQ-INT, and AGE. Most notably, there was a significant difference in age for those students who dropped out. Students who dropped were on average 1.7 years older than students who participated at both waves. Overall, the sample slightly overrepresented girls (60% females vs. 40% males) and was representative
of the racial make-up of the participating schools, including White (84.1%), African-American (5.3%), Hispanic (3.3%), and “Mixed Ethnicity/Other” (7.4%) teens. The participants ranged in age from 11 years to 18 years (M=12.8, SD=1.8). In the current study we used seven measures. Two were measures of guilt and shame. Two were measures of behavioral self-blame and characterological self blame. Two were measures of
depression. One wasa measureofattributionalstyleanddepression-relatedcognitions. One measure of guilt and shame was the Test of Self Conscious Affect— Adolescent version (TOSCA-A; Tangney, Wagner, Gavlas, & Gramzow, 1991). The TOSCA-A is a self-report measure comprised of 15 scenarios (10 negative and 5 positive) that
adolescents often encounter in everyday life. Each scenario is followed by four possible responses that assess guilt-proneness, shame-proneness, externalization, and detachment. Positive items also include responses that assess alpha-pride and beta-pride. An example scenario, “You trip in the cafeteria and spill your friend’s drink,” is followed by responses such as, “I would be thinking that everyone is watching me and laughing” (shame-proneness) and, “I would feel very sorry. I should have
watched where I was going” (guilt-proneness). Each potential response is rated on a 5-point Likert scale (1 = Not at all likely to 5 = Very likely.) In adolescent samples, the measure has relatively high levels of internal consistency (α = .81 for the guilt subscale, α = .77 for the shame subscale), test-retest reliability, predictive, convergent validity (Tangney, 1996;
Tangney, Wagner, Hill-Barlow, Marschall, & Gramzow, 1996). Because the current study focused on guilt and shame, we dropped items assessing alpha-pride, beta-pride, externalization, and detachment. Our version of the measure consisted of 15 scenarios with two response items each (guilt-proneness and shame
proneness).1 Both factors had high levels of internal consistency at both time points (αs = .85 for the guilt subscale at both time points; α = .80 and .83 for the shame subscale at times 1 and 2, respectively). Our second measure of guilt and shame was the State Shame and Guilt Scale (SSGS;
Marschall, Sanftner, & Tangney, 1994). The SSGS is a self-report measure compromised of 15 items. Five items for each of three subscales measure state-feelings of shame, guilt, and pride. An example guilt item is, “I felt bad about something I did.” An example shame item is, “I want to sink into the floor and disappear.” Participants rate how they currently feel on 5-point Likert scales (1 = not
feeling this way at all to 5 = feeling this way very strongly). In college samples, the measure has high levels of internal consistency and convergent validity with αs ranging from .82 to .89 for each subscale (Ghatavi, Nicolson, MacDonalad, Osher, & Levitt, 2002; Tangney &
Dearing, 2002b). In the current study, we slightly altered the wording for each response item for clarity. For example, “1 = not feeling this way at all” was changed to, “I do not feel this way at all.” We dropped the pride items because of our focus on guilt and shame.2 Both subscales had high levels of internal consistency at both time points (αs = .79 and .83 for guilt;
αs = .80 and .83 for shame). Technically, the SSGS is a measure of state shame and guilt. Five-month stability coefficients, however, were .53 for the guilt scale and .57 for the shame scale. These were nearly as large as those for the TOSCA-A (.58 and .64, respectively) over a 5-month period. Apparently the states represented by this measure were relatively stable over time. One measure of BSB and CSB was the Why it
Happened Questionnaire (WIHQ; Cole et al., 1996). The WIHQ is a self-report questionnaire comprised of 15 scenarios (12 negative and 3 positive) that adolescents could imagine encountering in everyday life. Only the 12 negative scenarios are used to assess BSB and CSB. Half of these pertain to academic events and the other half pertain to social events. A sample academic item is, “You have turned
your homework in late many times this year. Your teacher takes points off every time homework is late.” Each scenario is followed by the sametwo questions, “Did this happen because of the kind of person you are?” (CSB) and “Did this happen because of something you did or didn’t do?” (BSB). Participants respond on 5-point scale (1 = No to 5 = Yes). In adolescent samples, the measure has acceptable levels of internal consistency (α = .70 to α = .85), test-retest reliability (r = .33 to
r = .57), and convergent validity (Cole et al., 1996).3 Both subscales had high levels of internal consistency at both time points (αs = .84, .87 BSB subscale, αs = .85, .86 for CSB). Our second measure of self-blame was the Attributional Blame Questionnaire (ABQ). The ABQ was created for the current study. Some scenarios were adopted from the Attributional Questionnaire (Graham & Juvonen, 1998) that assesses self-blaming tendencies of children in victimized situations. We add items in order to represent a wider variety of scenarios. An example scenario is, “Imagine that you are giving a report in front of the class. When you start to talk to the class, you say something that doesn’t make sense. The teacher and your classmates all look really confused. Some kids even laugh at you.” Each of 4 scenarios is followed by the same 12 responses inquiring how likely it would be for participants to have certain thoughts if this scenario happened to them. An example BSB response is, “This is my fault. I should have been better prepared.” An example CSB response is, “If I were a smarter kid, I wouldn’t have these problems in class.” Each item is rated on a 5-point Likert scale (1 = definitely would not think to 5 = definitely would think). For each of the 4 scenarios, six responses represent BSB and six represent CSB, for a total of 24 BSB items and 24 CSB items in all.4 The measure had high levels of internal consistency at both time points (αs = .85, .84 for BSB, αs = .85, .86 for CSB). A copy of this measure is included in an Appendix. Depressive SymptomsOne measure of depressive symptoms was the Children’s Depression Inventory (CDI; Kovacs, 1981). The CDI is a 27-item self-report measure that assesses cognitive, affective, and behavioral symptoms of depression in children. Each item consists of three statements graded in order of increasing severity from 0 to 2. Children select one sentence from each group that best describes themselves for the past two weeks. In nonclinic populations, the measure has relatively high levels of internal consistency, test-retest reliability, predictive, convergent, and construct validity (Carey, Faulstich, Gresham, Ruggiero, & Enyart, 1987; Kazdin, French, & Unis, 1983; Mattison, Handford, Kales, Goodman, & McLaughlin, 1990; Saylor, Finch, Spririto, & Bennett, 1984; Worchel, Hughes, Hall, & Stanton, 1990). In the current study, internal consistency was high at both time points (αs = .90 and .91). Our second measure of depression was the Center for Epidemiological Studies—Depression Scale for Children (CES-DC; Weissman, Orvaschel, & Padian, 1980). The CES-DC is a 20-item self-report scale designed to measure depressive symptomatology. The items reflect symptoms associated with depression (e.g., mood, irritability, sleep, and eating disturbances). On 4-point scales (0 = rarely or none of the time, 3 = most or all of the time), participants rate how often they have had the symptom during the past week (e.g., I did not feel like eating; I wasn’t very hungry). In various samples of clinic and nonclinic children and adolescents, the CES-DC has demonstrated high levels of convergent validity, good internal consistency (αs range from .78 to .93), moderate two-week test-retest reliability (r = .69), and significant construct validity (Faulstich, Carey, Ruggiero, Enyart, & Gresham, 1986; Fendrich, Weissman, & Warner, 1990; Hilsman & Garber, 1995; Weissman et al., 1980). In the current study, Cronbach’s α was .89 at both time points. In the current study, the CDI and the CES-DC correlated .75 and .78 at time 1 and time 2, respectively. For our analyses, we created a depression composite score by standardizing and summing the CDI and the CES-DC. The reliability of this depression composite (using Nunnally & Bernstein’s 1994 formula) was .94 at both time points. Attributional Style MeasureOur measure of attributional style was the Adolescent Cognitive Style Questionnaire (ACSQ; Hankin & Abramson, 2002). The ACSQ is a self-report measure consisting of 12 negative scenarios (6 academic and 6 social) that could be encountered in everyday life. An example academic item is, “You take a test and get a bad grade.” An example social item is, “You want a boyfriend/girlfriend but don’t have one.” After each scenario, participants are asked to write down the cause of the event. Participants then rate the degree to which this cause is (a) internal, (b) stable, (c) global, (d) likely to cause other negative events, and (e) a reflection of a flawed self. Each of these response items are rated on a 7-point Likert scale with higher numbers representing more of the specific characteristic. In adolescent samples, these scales have internal consistency ranging from .81 to .91, test-retest reliability ranging from .51 to .73, and reasonable evidence of predictive and convergent validity. In the current study, we dropped four items because they were inappropriate for younger adolescent populations (e.g., Your boss yells at you at work). Based on the remaining eight scenarios, α reliabilities were .79 for globality, .78 for stability, .59 for internality, .83 for consequences, and .84 for flawed self. PROCEDURESParticipating students completed a packet of questionnaires on two separate occasions, approximately five months apart. The first assessment occurred approximately two months into the school year, and the second assessment occurred two months before the end of the school year. The two packets were identical except that the ACSQ was only administered at time 1. Doctoral psychology students, advanced undergraduate students, and research assistants administered the questionnaires to the students. For middle school students, the research assistants read all questionnaires aloud, requiring all to proceed at the same pace, irrespective of their reading abilities. For the high school students, the questionnaires with more difficult formats (e.g., the ACSQ) were read aloud, but the students were allowed to work at their own pace on the rest. The sequence in which the questionnaires were administered was randomized by class-room in order to minimize the effects of order and fatigue on any instrument. Two to three additional research assistants circulated among the participants answering questions as needed. RESULTSPRELIMINARY ANALYSESTable 1 contains descriptive statistics for the study measures. Means and standard deviations for the subscales of each measure were comparable to those obtained in other studies with similar populations (e.g., Dallaire et al., 2006; Hankin & Abramson, 2002; Peeke, 1995; Tangney, 1992; Tangney & Dearing, 2002a). Table 2 contains correlations between guilt, shame, BSB, and CSB measures at both times. TABLE 1Means (and Standard Deviations) at Time 1 and Time 2
Note. SD = Standard Deviation; TOSCA–A = Test of Self–Conscious Affect–Adolescent measure; SSGS = State Shame and Guilt Scale; WIHQ = Why it Happened Questionnaire; ABQ = Attributional Blame Questionnaire; CDI = Children’s Depression Inventory; CES–DC = Center for Epidemiological Studies—Depression ScaleforChildren; ACSQ = AdolescentCognitive StyleQuestionnaire; BSB = behavioral self–blame; CSB = characterological self–blame. TABLE 2Correlations between guilt, shame, BSB, and CSB scales (Time 1 correlations below the diagonal, Time 2 correlations above diagonal)
GOAL 1Our first goal was to examine the relation of guilt, shame, BSB, and CSB to individual differences in depressive cognitions and concurrent depressive symptoms. We hypothesized that (1) guilt, shame, BSB, and CSB would all correlate moderately with the Internality subscale of the ACSQ, (2) shame and CSB would correlate more strongly than would guilt and BSB with the Globality, Stability, Consequences, and Flawed Self subscales of the ACSQ, and (3) shame and CSB would correlate more strongly than would guilt and BSB with concurrent measures of depressive symptoms. To test these hypotheses, we conducted a series of regressions in which a guilt, shame, BSB, or CSB subscale was regressed onto one of the ACSQ subscales or the depression composite. In each analysis, we also entered age as a control variable. (We also examined all Age × Predictor interactions, but as none of these were significant, we dropped the interaction terms from these regressions.) Examples of these regressions are TOSCAShame=β00+β01Age+β02DEP TOSCA Guilt = β10 + β11Age + β12 DEP. First we examined the size and significance of β02 and β12. Then we formally tested the null hypothesis that β02 = β12. We repeated such analyses using the subscales from the SSGS, ABQ, and WIHQ as the dependent variables and replacing DEP with other predictors such as the subscales of the ACSQ. These tests were accomplished using maximum likelihood structural equation modeling, comparing models with and without the specified equality constraint. The null hypothesis (that the two βs are equal) is rejected when the change in chi-square (Δχ2df = 1) is significant. The results in Table 3 generally supported our hypotheses. First, all measures of guilt, shame, BSB, and CSB were modestly related to the Internality subscale of the ACSQ (mdn β = .18). Furthermore, the strength of this relation did not vary significantly from subscale to subscale of the same instrument. That is, guilt and shame were equally internal, as were BSB and CSB. Second, our shame and CSB subscales were generally more strongly associated with stable attributions, global attributions, perceptions of negative consequences, and inferences of a flawed self than were our guilt and BSB measures (mdn βs were .40 and .20, respectively). Third, our shame and CSB subscales were generally more strongly associated with our composite measures of concurrent depressive symptoms than were our guilt and BSB subscales (mdn βs were .50 and .17, respectively). As the effects of Age were nonsignificant, they do not appear in Table 3. TABLE 3Partial Beta Weights (Standardized) from the Regressions of Various Guilt, Shame, BSB, and CSB Subscales onto Measures of Attributional Dimensions and Depressive Symptoms
We also tested for the incremental validity of shame versus CSB measures as well as guilt versus BSB measures in accounting for individual differences in concurrent depressive symptoms. To this end, we ran 4 sets of multiple regressions at each time point: 2 with shame and CSB measures and 2 with guilt and BSB measures. For each pair of regressions, we switched the order in which we entered the measures. Changes in R2 were larger for models in which guilt or shame measures were entered last (ΔR2s ranged from .21 to .30) than for models in which BSB or CSB measures were entered last (ΔR2s ranged from .02 to .04). GOAL 2Our second goal had two parts. One was to examine the degree to which guilt, shame, BSB, and CSB at time 1 predicted depressive symptoms at time 2, controlling for depressive symptoms at time 1. The other was to assess the degree to which depressive symptoms at time 1 predicted guilt, shame, BSB, and CSB at time 2, controlling for time 1 levels of the dependent variable. We approached these goals using a series of eight manifest variable structural equation models. In each model, we regressed our time 2 composite measure of depressive symptoms (DEP) and a time 2 measure of either guilt, shame, BSB, or CSB onto the same time 1 measures plus Age (as a control variable). An example of such a model is depicted in Figure 1. As the model is just-identified, goodness-of-fit is not an issue. Our goal was simply to estimate the magnitude, direction, and significance of the paths labeled a, b, c, and d. In total, we tested 8 such models, one for each of the two measures of guilt, shame, BSB, and CSB.  Manifest variable path diagram of longitudinal relations between guilt and depressive symptoms (DEP), controlling for children’s age. Results of these analyses appear in Table 4. In every model, the autoregressive paths (a and d) were highly significant. Their standardized beta weights ranged in size from .42 to .71, indicating moderate to strong temporal stability of the key constructs over this 5-month interval. In the models that included a measure of guilt or BSB, evidence of predictive relations between DEP and the guilt or BSB (paths b and c) was weak and inconsistent across the various measures. In only one out of four analyses (the one using the SSGS) did DEP predict either guilt or BSB. Likewise, in only one out of four analyses (the one using the WIHQ) did a measure of guilt or BSB predict subsequent DEP scores. Much stronger evidence of prediction emerged for the models of shame and CSB. In all four models, the effect of DEP on shame and CSB was positive and statistically significant, indicating that higher levels of depressive symptoms at one point in time predicted higher levels of shame and CSB five months later, even after controlling for prior levels of the dependent variable and age. The magnitude of these standardize betas ranged from .10 to .33, with stronger effects emerging for the SSGS measure of shame and the ABQ measure of CSB.5 TABLE 4Parameter Estimates from Longitudinal Models of Reciprocal Predictive Relations (as in Figure 1)
GOAL 3Our third goal was to examine the convergent and discriminant validity of multiple measures of guilt, shame, BSB, and CSB in youth. Our specific hypotheses were (1) that shame and CSB would show some degree of convergence insofar as CSB involves cognitions that are similar to those associated with shame and (2) guilt and BSB would also show some degree of convergence because BSB involves cognitions similar to those associated with guilt. To test these hypotheses, we conducted a series of maximum likelihood confirmatory factor analyses. For the first hypothesis, we began with a 2-factor model in which the two time 1 shame measures loaded onto a Shame factor and the two time 1 CSB measures loaded onto a CSB factor, as depicted in Figure 2. The model fit the data extremely well (see Table 5). The factor loadings were all relatively large and significant, indicating good convergence of the shame measures onto the Shame factor and of the CSB measures onto the CSB factor (see Table 6). As expected, the correlation between the two factors was extremely high (its 95% confidence interval contained 1.0), suggesting that the two factors were highly redundant of one another. Consequently, we also tested a 1-factor model in which all four measures loaded onto a single Shame/CSB factor. This model also fit the data extremely well, and was not significantly different from the 2-factor model (see Table 5). We replicated these findings with the time 2 measures. These results clearly support our first hypothesis. Latent variable path diagrams for two-factor and one-factor models for guilt, shame, BSB, and CSB. TABLE 5CFA Fit Statistics for 1– and 2–factor Models of Shame/CSB and Guilt/BSB
TABLE 6Factor Loadings from Confirmatory and Exploratory Factor Analyses
Note. TOSCA–A = Test of Self–Conscious Affect–Adolescent measure; SSGS = State Shame and Guilt Scale; ABQ = Attributional Blame Questionnaire; WHIQ = Why it Happened Questionnaire; BSB = behavioral self–blame; CSB = characterological self–blame. Loadings > .30 are in bold face. We used the same approach to test the second hypothesis about the guilt and BSB measures, with very different results. Both the 1-factor and the 2-factor model fit the data quite poorly. As shown in Table 5, the χ2 estimates were significant at both waves. Some of the alternative fit indices were quite low (e.g., the RFI was .60 at time 1 and .32 at time 2). Furthermore, the RMSEA was much larger than the recommended .05 cutoff. Not only did the guilt and BSB measures fail to converge onto a common latent variable, but the two guilt measures failed to converge onto a Guilt factor and the two BSB measures failed to converge onto a BSB factor. In a post hoc attempt to understand the structure of these four measures, we conducted an exploratory factor analysis, using a principal axis extraction and oblique rotation. In both the time 1 and time 2 data, the Kaiser criterion suggested the existence of two factors. Both sets of results appear in Table 6. The TOSCA-A measure of guilt and the ABQ measure of BSB loaded onto the first factor, and the SSGS measure of guilt and the WIHQ measure of BSB loaded onto the second factor. The factors were only moderately correlated with each other (.34 at time 1 and .26 at time 2). In other words, guilt and BSB measures converged to form two relatively distinct hybrid guilt/BSB factors. Examination of the content of the TOSCA-A guilt scale and the ABQ-BSB scale revealed that many of the items had a constructive or restorative theme, leading us to call factor 1 a “restorative guilt” factor. Similar examination of the SSGS-guilt scale and the WIHQ-BSB scalerevealed no such content, leading us to call factor 2 a “nonrestorative guilt” factor. In a second post hoc analysis, we examined the predictive utility of these restorative and nonrestorative guilt factors, entering them into the longitudinal analyses with depression. Although the models fit the data well, neither restorative nor nonrestorative guilt predicted later depression after controlling for prior depression.6 One significant effect did emerge, however, in that time 1 depression predicted nonrestorative guilt 5 months later (β = .38, p < .01), even after controlling for prior levels of nonrestorative guilt and depressive symptoms. DISCUSSIONThree main sets of findings emerged from this study. First, our hypotheses about the concurrent correlates of guilt, shame, BSB, and CSB were largely supported: (a) all measures were significantly (albeit weakly) associated with the tendency to make internal attributions about negative events, (b) shame and CSB measures were more strongly related to depressive cognitions than were the guilt and BSB measures, and (c) shame and CSB were more strongly related to self-reported depressive symptoms than were guilt and BSB. Second, longitudinal analyses revealed that depressive symptoms predicted subsequent levels of shame and CSB; all other prospective results were much less consistent. Third, our hypothesis about the convergence of shame and CSB measures onto a single underlying construct was strongly supported; however, support for our hypothesis regarding the convergence of guilt and BSB was limited to specific pairs of measures. We elaborate on each of these findings below. First, the pattern of concurrent correlations of guilt, shame, BSB, and CSB to various measures of depressive cognitions and symptoms was largely consistent with Lewis’ and Tangney’s description of the cognitions associated with guilt and shame (Lewis, 1974; Tangney, 1996) and with Janoff-Bulman’s conceptualization of BSB and CSB (Janoff-Bulman, 1979). Both guilt and shame focus attention inward—either in one’s actions/inactions or on one’s self. Likewise, both BSB and CSB involve internal attributions for negative events. Consistent with these conceptualizations, our data revealed that guilt, shame, BSB, and CSB were all positively associated with the tendency to attribute negative events to internal causes. Somewhat surprising, however, was that these correlations were relatively small. We speculate that this may have been due to relatively low reliability of the Internality subscale of the ACSQ (relative to the other ACSQ subscales). Most of our analyses revealed that shame and CSB were more strongly associated with other depression-related cognitions than were guilt and BSB. These negative cognitions included the tendency to attribute negative events to stable and global causes, the inclination to infer that negative events will have serious negative consequences, and the proclivity to perceive one’s self as fundamentally flawed. This last point is central to both the Lewis/Tangney and the Janoff-Bulman theories. Tangney explains that shame involves a concern that ones’ flaws and defects will be revealed (Tangney, 1995). Likewise, Janoff-Bulman defines CSB as a self-reflective attribution in which people blame their person and fault their character for negative outcomes (Janoff-Bulman, 1979). Not surprisingly, shame and CSB were much more strongly associated with adolescents’ depressive symptoms than were guilt and BSB. This finding is consistent with both theories: shame and CSB are maladaptive. Taken together, these results help to consolidate a number of previous findings. Our positive correlations between shame and measures of depressive symptoms and depressive attributional style replicate a number of previous studies (e.g., Harder, 1995; Luyten et al., 2002; Pineless et al., 2006; Tangney, 1992, 1996, 1998; Tangney & Dearing, 2002b; Tangney, Wagner, & Gramzow, 1992). Likewise, our positive correlations between CSB and measures of depressive symptoms and attributional style replicate a separate body of research (e.g., Janoff-Bulman, 1979; Peterson, 1979; Peterson et al., 1981). The measures of shame displayed greater incremental validity in relation to concurrent depressive symptoms than did the measures of CSB. Despite the substantial overlap between these constructs, this evidence suggests that shame has greater association with depression than does CSB. We suggest that the affective component of shame that is not present in CSB may be responsible for this difference. Second, our pattern of longitudinal results is not completely commensurate with either guilt-shame theory or self-blame theory. Tangney (1995) suggested that guilt has an adaptive quality insofar as it focuses attention on atonement, reparations, and restoration—implying that feelings of guilt would mediate recovery. Conversely, shame is maladaptive as it makes one more vulnerable to feelings of depression, anger, withdrawal, and various forms of psychopathology (Tangney et al., 1995; Tangney, Wagner, Fletcher, & Gramzow, 1992). Janoff-Bulman suggested that BSB focuses attention on one’s own behavior, on things that one could do (or not do) so as to avoid similar negative events in the future. In contrast, CSB is a problematic attribution that fosters nonproductive rumination and forestalls recovery. Despite the fact that both theories forecast quicker recovery for individuals who engage in guilt or BSB, and prolonged maladaptation (if not depression) for individuals who experience shame or CSB, our results provided no consistent longitudinal support for either the healthy effects of BSB and guilt or the detrimental effects of CSB and shame. These null results replicate some previous studies but not others. On the one hand, Peterson et al. (1981) found no relation of either BSB or CSB to future depression after controlling for previous depression. Andrews, Qian, Valentine (2002) also found no relation between TOSCA-A (shame or guilt) and change in depression over time. On the other hand, Andrews et al. found that a different measure of shame, the Experience of Shame Scale (ESS), did predict later depressive symptoms after controlling for the TOSCA-A Shame and Guilt scales and previous levels of depressive symptoms. We note, however, that many of the ESS items begin with phrases like “Have you worried …,” and may therefore be affected by the respondent’s level of anxiety. Considerable evidence suggests that anxiety often precedes depression (Cole, Peeke, Martin, Truglio, & Seroczynski, 1998; Orvaschel, Lewinsohn, & Seeley, 1995). Especially after controlling for TOSCA-A guilt and shame, Andrews et al.’s results may be reflective of the longitudinal relation between anxiety and depression. A third study by Stuewig and McCloskey (2005), using the Adolescent Shame Measure (Reimer, 1995) that is modeled after the TOSCA-A, found that shame predicted depression two years later. Unfortunately, their control for prior depressive symptoms was obtained six years prior to the other predictors. Incomplete control of prior depression may be partially responsible for the residual significant effect of shame on depressive symptoms. In sum, our null results are consistent with previous longitudinal studies that used similar measures and implemented good statistical control for prior levels of depression. Instead, we found that shame and CSB appeared to be more of a consequence to than a cause of depressive symptoms. That is, depressive symptoms predicted shame and CSB five months later, even after controlling for prior levels of shame and CSB. To our knowledge no other studies have addressed this question. The finding is reminiscent of Lewinsohn’s scar hypothesis, suggesting that depression begets certain cognitive frailties that remain evident even after the remission of depression (Rohde et al., 1990; Zeiss & Lewinsohn, 1988). Further research into the effect of depression on shame and CSB may be warranted. Third, we found substantial convergence of shame and CSB, but only weak, measurement-specific convergence of guilt and BSB. Our support for the convergence of shame and CSB consists of (a) the fact that 2 measures of shame and 2 measures of CSB all loaded onto a single latent variable, and (b) all four measures correlated in the same direction and to similar degrees with attributional style, the inclination to infer that current negative events have dire consequences for the future, the inference that the self is defective or flawed, and the experience of depressive symptoms. This pattern of results suggests that the cognitions associated with shame are commensurate with those that characterize CSB. This finding opens a door connecting two nearly independent literatures and enables us to extrapolate from one to the other. On the one hand, research shows that CSB relates to a variety of factors associated with depression, such as feelings of loneliness and isolation, uncontrollability and helplessness (Janoff-Bulman, 1979; Peterson, 1979), and stressful life events (Peterson et al., 1981). On the other hand, the shame literature reveals positive correlations with anger/hostility (Tangney, Wagner, & Gramzow, 1992), substance abuse (Dearing et al., 2005), eating disorders, low self-efficacy (Sanftner et al., 1995), and poor problem solving (Covert et al., 2003). The combination of these literatures also raises a new question: might the cognitive, self-blaming component of shame be more stable and more problematic than its affective component? The question is worthy of future empirical consideration. Unlike shame and CSB, the convergence of guilt and BSB was limited to specific pairs of instruments. Guilt and BSB as measured by the TOSCA-A and ABQ converged onto one factor, whereas guilt and BSB as measured by the SSGS and the WIHQ loaded onto a separate factor. Although the measurement specificity of these findings prevents our making global conclusions, it does give rise to some interesting observations and speculations. Noting that these results derived from post hoc, exploratory analyses, we cautiously speculate that this set of measures represents two different kinds of guilt. In keeping with Tangney, Wagner, & Gramzow’s (1992) conceptualization of guilt as a mediator of constructive, restorative, or apologetic behavior, the TOSCA-A guilt scale includes such items as, “I would apologize and make sure my friend feels better,” “I would think: ‘This is making me anxious. I need to either fix it or replace it,’” and “I would feel unhappy and eager to correct the situation.” Commensurate with Janoff-Bulman’s (1979) conceptualization of BSB as “a functional response to a traumatic event” that fosters resilience and coping, the BSB scale of the ABQ includes such items as “I should have studied harder,” “I should have reacted differently when I got the assignment,” and, “I should have asked the teacher to let me do the report another time.” In contrast, the TOSCA-A and ABQ measures of these constructs were either unrelated to or negatively correlated with depression. Interestingly, previous research shows that BSB is positively related to controllability and willingness to accept responsibility for negative events (Peterson, 1979; Peterson et al., 1981), thus possibly buffering feelings of loneliness, isolation, and hopelessness (Anderson et al., 1983; Anderson et al., 1988), protecting individuals from certain types of psychopathology, and fostering healthy relationships (Covert et al., 2003; Leith & Baumeister, 1998; Sanftner et al., 1995; Tangney, 1991). Likewise, constructive guilt is related to problem-solving, empathy, and perspective-taking (Leith & Baumeister, 1998; Tangney, 1991), which can protect people from a variety of disorders (Dearing et al., 2005; Sanftner et al., 1995; Tangney, Wagner, Fletcher et al., 1992). In contrast, the SSGS measure of guilt and the WIHQ measure of BSB appear to converge upon a construct that is not particularly constructive or restorative. Items from the guilt subscale of the SSGS focus on the state-internal feeling: “I feel tension about something I have done” and “I feel bad about something I have done.” Similarly, items from the WIHQ focus on internal attributions, but without the anticipation of what might be done differently next time: “Did this happen because of something you did or didn’t do?” Previous research supports the existence of nonadaptive or nonrestorative guilt both in adults (Harder, Cutler, & Rockart, 1992; Kugler & Jones, 1992b; O’Connor, Berry, & Weiss, 1999) and in youth (Ferguson et al., 1999). Furthermore, results of the current study revealed that the guilt and BSB subscales of the SSGS and WIHQ were both significantly related to depressive symptoms. Taken together, these results lead us to speculate that two types of guilt/BSB may exist. One, as measured by the TOSCA-A and the ABQ, reflects Tangney’s definition of guilt and Janoff-Bulman’s definition of BSB (Janoff-Bulman, 1979; Tangney, 1995). It focuses on problem-solving, constructive behavior, and restoration following a negative event. The other, measured by the SSGS and WHIQ, is more reflective of Harder’s definition (Harder, 1995; Harder et al., 1992; Harder & Greenwald, 1999; Harder & Zalma, 1990; Kugler & Jones, 1992b). It focuses on internal attributions and negative affective that follow a negative event but do not necessarily foster constructive behavior or restoration. We offer these characterizations cautiously, however, as they are not consistent with our original hypotheses about these measures and derive from exploratory, post hoc examination of our data. Continued research into measures of guilt and related constructs is clearly needed. As such research progresses, investigators should consider an even wider set of theoretical frameworks. Although many theorists agree with Tangney that guilt pertains to concrete events, focuses on specific behaviors, and can lead to constructive behavior (Lewis, 1974; Zahn-Waxler, Kochanska, Krupnick, & McKnew, 1990), other theorists do not. Some suggest that guilt is trait-like and maladaptive (Ferguson, Stegge, Eyre, Vollmer, & Ashbaker, 2000; Harder & Greenwald, 1999; Jones & Kugler, 1993). Others suggest that separating guilt from shame takes the emotionality out of guilt. For example, Sabini and Silver (1997) believe that Tangney’s concept of shame-free guilt is “anemic.” Still others insist that guilt entails a moral transgression or the violation of societal norms (Abramson, Mosher, Abramson, & Woychowski, 1977; Harder & Greenwald, 1999; Lewis, 1974; Zahn-Waxler et al., 1990). Finally, O’Connor, Berry, and Weiss (1999) emphasize that behavior can be motivated by the desire to avoid future feelings of guilt, not just by the desire to expiate guilt feelings that already exist. As the currently available measures of guilt reflect this wide range of theoretical backgrounds, perhaps it is not surprising that they do not manifest a high degree of convergence. Several shortcomings of the current study suggest avenues for future research. First, our study used measures that jointly assessed either guilt and shame or BSB and CSB. We did not use any of the guilt-only or shame-only measures that are available, such as the Experiencing Shame Scale or the Guilt Inventory (Kugler & Jones, 1992a), the Adolescent Shame Measure, or the Adapted Shame and Guilt Scale (Hoblitzelle, 1982). Future studies using a broader range of guilt, shame, BSB, and CSB measures could further clarify points of convergence between different theoretical perspectives. Second, we administered a nonstandard version of the TOSCA-A absent of the externalization, detachment, alpha-pride, and beta-pride subscales. Although factor analyses revealed that the solid guilt and shame factors did remain, it is possible that our modification of this measure affected it in more subtle ways. Third, the poor convergence of SSGS guilt and TOSCA-A guilt subscales may be due to our nonstandard use of the SSGS. Although we found high stability with the SSGS, it is still a state measure of guilt and shame, whereas the TOSCA-A is a trait measure of these constructs. Our results should be interpreted in light of this observation. Fourth, we found some support for the existence of two types of guilt; however, we only had 2 measures of each and our analyses were post hoc. Future studies should use more measures of each type of guilt to test this hypothesis more rigorously. Fifth, our sample was primarily middle-class and Caucasian. Future studies of more diverse populations could test the generalizability of these findings. Sixth, our study examined the implications of guilt, shame, BSB, and CSB for dimensional depression in a nonclinical population. The relation of these constructs to depressive disorder in clinical populations could be different and warrant future investigation. Finally, our study focused on adolescence, a time when conceptions of morality are in flux, feelings of egocentricity abound, the motivation to self-evaluate runs high, and rates of depression are on the rise. If future studies included a wider age range, developmental changes in the relations among guilt, shame, BSB, CSB, and depression could be examined. Such research could contribute to our understanding of developmental differences in the very nature of depression. AcknowledgmentsThis research was supported in part by grants R01MH64650 from the National Institute on Mental Health, P30HD15052 from the National Institute of Child Health and Human Development, and the Warren Family Foundation. We are also grateful to Alanna E. Truss, Katie Von Canon, Melissa Arias, and Elijah Mendoza for their help in many aspects of this project APPENDIX. ATTRIBUTIONAL BLAME QUESTIONNAIRE (ABQ)Each of the four scenarios is followed by 12 response options. We only list these options once because they are quite similar (although not identical) across scenarios. Scenario 1: Imagine that you are giving a report in front of the class. When you start to talk to the class, you say something that doesn’t make sense. The teacher and your classmates all look really confused. Some kids even laugh at you. Scenario 2: Imagine that one day in math, your class breaks into groups to play a math game. You get graded on how many answers you get right. During the math game kids in your group keep talking to you. As a result, you miss a lot of the questions. Your performance earns you a bad grade. Scenario 3: Imagine that you are getting something out of your locker just as the bell rings. It is pretty quiet in the halls because most of the kids have already gone to class. Just then you see another group of kids breaking into a locker near where you are. They see you and one of them pins you against the locker and threatens you. Scenario 4: Imagine that you’ve just bought your lunch after waiting in line for a long time. As you are walking away, someone bumps into you on purpose. You’re not hurt, but most of your food spills on your clothes. The other kids in the line start laughing at you. Response options rated on 1-5 scales (1 = “definitely would not think” to 5 = “definitely would think”).
Scoring: Characterological self-blame: 1, 3, 4, 5, 9 Behavioral self-blame: 2, 6, 7, 8, 10 Footnotes1Because we modified the TOSCA-A by dropping certain items, we examined the factor structure of the abbreviated measure. Using principal axis extraction and oblique rotation, we found evidence of two strong factors. Guilt items loaded onto factor 1, and shame items loaded onto factor 2. Primary factor loadings ranged from .57 to .83, whereas cross loadings were all less than .20. The factors correlated with each other .26 and .10 at time 1 and time 2, respectively. The resultant factors corresponded very closely to Guilt and Shame in the original measure. 2Given our changes to this measure, we examined its structure. Factor analysis of the measure, using oblique rotation and principal axis factoring, revealed two strong factors. With only two exceptions, guilt items loaded onto factor 1, and shame items loaded onto factor 2. We dropped two items that did not load properly at both time points (I feel regret, remorse; I feel small). For the remaining items, primary factor loadings ranged from .50 to .86, whereas cross loadings were all less than .30. The factors correlated with each other .52 and .65 at time 1 and time 2, respectively. 3In the current study, principal axis factor analysis with oblique rotation revealed two strong factors. Only BSB items loaded onto factor 1, and only CSB items loaded onto factor 2. Primary factor loadings ranged from .48 to .89, whereas cross loadings were all less than .28. The factors correlated with each other .60 at time 1 and.66 at time 2. 4In the current study, principal axis factor analysis with oblique rotation revealed two strong factors. Only BSB items loaded onto factor 1, and only CSB items loaded onto factor 2. Primary factor loadings ranged from .48 to .89, whereas cross loadings were all less than .28. The factors correlated with each other .60 at time 1 and.66 at time 2. 5Previous research has sometimes focused on guilt-free shame and shame-free guilt, operationalized simply by controlling for guilt when examining shame, and vice versa (e.g., Lutwak et al., 2003; Tangney & Dearing, 2002b; Tangney, Wagner, & Gramzow, 1992; Tracy & Robins, 2006). Therefore, we repeated each analysis of each guilt (or shame) subscale while controlling for the time 1 shame (or guilt) subscale from the same measure. Results of these analyses were remarkably similar to those in Table 4. The magnitudes of the beta weights were almost unchanged whereas increases in standard error caused a few previously significant results to become nonsignificant (as a consequence of the multicolinearity induced by the addition of the control variable). We also opted not to use residual analysis (which has sometimes been used with the TOSCA instruments) because of the potential for systematically biased results as detailed in Darlington and Smulders (2001). 6The model with “restorative guilt” also displayed a Heywood case, in which an error variance was negative and a correspondent factor loading was greater than one. To correct this, we set the error variance for the TOSCA-A to zero. Contributor InformationCARLOS TILGHMAN-OSBORNE, Vanderbilt University. DAVID A. COLE, Vanderbilt University. JULIA W. FELTON, Vanderbilt University. JEFFREY A. CIESLA, Department of Psychology, Kent State University; Vanderbilt University. REFERENCES
Which perspective suggests that explaining our own failures in terms that are global stable and internal contributes to depression *?Which perspective suggests that explaining our own failures in terms that are global, stable, and internal contributes to depression? emotional extremes.
Which perspective has emphasized the impact of learned helplessness on depression *?4. The social-cognitive perspective has linked the experience of depression to: A) learned helplessness.
What is an example of hopelessness theory?This theory would predict, for example, that an individual who has an argument with an acquaintance is more likely to become depressed if they interpret this event as a pro- duct of their poor interpersonal ability (internal), which they believe will never change (stable) and will nega- tively influence all their other ...
What are the flat affect and catatonia associated with?Catatonia Diagnosis
For a catatonic schizophrenia diagnosis in particular, it is essential for physicians to differentiate between negative symptoms of schizophrenia (such as lack of emotional expression or flat affect), and true catatonic symptoms.
|
