Participants and Procedures
This study is a secondary analysis of a subset of data collected for a larger community-based longitudinal study (Pittsburgh Girls Study [PGS]). The PGS involves 2,450 girls (now women) who were initially recruited in 1999 and 2000 when they were ages 5 to 8 years old (see Keenan et al., 2010 for further details on PGS recruitment and study design). Participants for the sub-study, which focused on aggressive and self-harming behavior in young women, were identified from the larger PGS based on self-reports of recent aggressive behavior, suicidality, or self-injury (see [34] for additional details). A total of 166 young women were recruited and consented to participate in the sub-study. During initial assessments (baseline) of the sub-study, participants completed a battery of clinical interviews and self-report measures (see Measures section for details). Follow-up assessments (data not presented here) occurred at 6- and 12-months, respectively, after the initial assessments.
Given that our current focus is on trauma and ED, only those participants who completed the baseline assessment measures of trauma and ED (N=144) were included for the primary analyses and the results presented here. These participants were between the ages of 18 and 24 (M = 21.51, SD = 1.57), and were primarily African American or non-Hispanic White. The demographics of this sub-study sample were similar to those of the larger longitudinal study (redacted citation) from which participants were selected (see Table 1 for additional information).
Measures
BPD in the Linear Regression Models.
The Structured Interview for DSM-IV-TR Personality (SIDP-IV;[35]) was used to generate dimensional BPD scores for our linear models. The SIDP-IV is a semi-structured diagnostic interview for DSM-IV-TR personality disorders. Interviews were administered by research staff with a bachelor’s degree or higher who were trained to reliability by a doctoral-level clinical psychologist. SIDP-IV items are rated on a 0 to 3 scale (0 = not present, 1 = subthreshold, 2 = present, 3 = strongly present). Dimensional scores (a sum of all BPD item scores) were used as an index of BPD symptomatology severity. The BPD items demonstrated adequate internal consistency for dimensional BPD scores in this subsample (Cronbach’s α=.87).
BPD in the SEM model.
The Personality Assessment Inventory-Borderline Features Scale (PAI-BOR, [36] )is a 24-item self-report measure that assesses four dimensions underlying BPD: affective instability, identity problems, negative emotions, and self-harm. The PAI-BOR is used to measure subconstructs of BPD in our structure equation model (SEM) as it categorizes symptoms into four BPD domains which can be used to specify the measurement model in SEM in addition to assess the relationships among three latent factors (ED, trauma and BPD). Intraclass coefficients among this sample for subscales are as follows: Affect instability (α=.72), identity problems (α=.68), self-harm (α=.73), and negative relationships (α=.64).
Trauma.
The Childhood Trauma Questionnaire Short Version (CTQ-SF; [37]) items ask about experiences from early childhood to adolescence, which are rated on a 5-point scale with response options ranging from Never True to Very Often True. The CTQ-SF produces a total score and five trauma-related subconstructs—physical, sexual, and emotional abuse, and physical and emotional neglect. The CTQ-SF showed good reliability among this sample. Intraclass correlation coefficients for subscales are: Physical neglect (α=.71), emotional abuse (α=.82), emotional neglect (α=.84), physical abuse (α=.76), and sexual abuse (α=.93).
Emotion Dysregulation.
The Difficulties in Emotion Regulation Scale (DERS; [38] ) is a 36-item self-report measure that was developed to assess emotion dysregulation comprehensively, including items that reflect difficulties in six emotional dimensions: Non-acceptance, Goals, Impulse, Strategies and Clarity [38]. More specifically, Non-acceptance means non-accepting reactions to negative emotions or stress; the Goals dimension contains items reflecting difficulties in engaging in goal-directed behaviors (such as concentrating or accomplishing tasks); the Impulse dimension consists of items that describe difficulties with controlling behaviors under negative emotions; the Awareness (reverse-coded) scale assesses the ability to attend to and recognize emotions; the Strategies dimension includes items that evaluate limited access to regulation strategies; and Clarity measures lack of clarity about one’s own emotions(e.g. unable to identify one’s emotions). Each item of the DERS is rated on a 5-point scale ranging from 1 “almost never” to 5 “almost always”. DERS demonstrated good internal consistency among our sample as indicated by intraclass correlation coefficients for subscales of non-acceptance (α=.85), goals (α=.72), impulse (α=.81), strategies (α=.85) and clarity (α=.85).
Shame/Guilt.
The Guilt and Shame Proneness scale (GASP) is a 16-item self-report scale that assesses individuals’ tendencies to experience shame and guilt following embarrassing or offensive events across different settings [39]. The GASP consists of two shame subscales (negative behavior-evaluations and repair action tendencies) and two guilt subscales (negative self-evaluations and withdrawal action tendencies). For the two guilt subscales, negative behavior-evaluations items address bad feelings about one’s actions, whereas repair items describe behavioral intentions such as correcting one’s mistakes (e.g., “you would try to act more considerately toward your friends”). As far as the shame subscales, negative self-evaluations consist of items about feeling bad about oneself, whereas withdrawal items address tendencies to hide from the public (e.g., “you would avoid the guests until they leave”). Each item of the GASP is rated on a 7-point scale, with “1” indicating “very unlikely” and “7” indicating “very likely”. Finally, internal consistency for GASP was unsatisfactory among our sample. The intraclass correlation coefficients are: Negative behavior-evaluations (α =.69), repair action tendencies (α =.54), negative self-evaluations (α =.72) and withdrawal action tendencies (α =.55).
Anger.
The original State-Trait Anger Expression Inventory-2 (STAXI-2) is a 57-item self-report measure comprised of six subscales: State Anger, Trait Anger, Anger Expression‐In, Anger Expression‐Out, Anger Control‐In, and Anger Control‐Out [40]. We utilized an abbreviated anger scale that included only Trait Anger, Anger Expression-In, Anger Expression-Out, and Anger Control (we used mean scores of both Control-in and out scores, which were also reverse coded). In terms of each subscale, Trait Anger measures the disposition to experience anger with or without provocation; Anger Expression-In assesses the frequency of controlling one’s angry feelings; Anger Expression-Out measures how often one takes actions upon his/her anger; and Anger Control measures one’s ability to control one’s anger by utilizing positive outlets (Control-out) or calming oneself down (Control-in). The internal consistency of each subscale in this sample was adequate. The intraclass correlation coefficients are: Trait Anger (α =.88), Anger Expression-In (α =.71), Anger Expression-Out (α =.83), and Anger Control (α =.81).
Analyses
Linear Regression Models.
To examine the differential relationships between BPD features, ED, and specific traumas, we conducted stepwise multivariate linear models. The initial model was comprised of five single trauma types as main predictors. Step two included DERS constructs as additional independent variables. Step three added four anger variables: Trait Anger, Anger Expression-out, Anger Expression-in and Anger Control (this variable was reversed coded). The final step further included four subconstructs of shame/guilt. Finally, performance of different models (e.g. model R2) were evaluated and compared.
SEM.
In order to test the indirect effect from trauma to higher BPD features through ED, we conducted path analysis with latent factors using the structural equation modeling (SEM) method in R. The structural model was comprised of the latent predictor Trauma, the latent outcome variable BPD, and the mediator ED. The measurement model is specified as follows: Trauma is measured by five trauma subtypes (physical abuse, sexual abuse, emotional abuse, physical neglect and emotional neglect), BPD by four symptomatic categories (affective instability, identity problems, negative emotions and self-harm), and ED by six emotional subconstructs (non-acceptance, goals, impulse, awareness, strategies, and clarity) (Figure 1).