Assessment of PTSD’s E2 Criterion: Development, Pilot Testing, and Validation of the Posttrauma Risky Behaviors Questionnaire

Introduction

Posttraumatic stress disorder (PTSD) is highly prevalent, with recent evidence suggesting that 8.30%, or approximately 27 million Americans according to the recent U.S. Census data (2018), will develop PTSD in their lifetime (Kilpatrick et al., 2013). Unsurprisingly, PTSD is both widely researched and a key consideration in clinical practice. One important clinical and empirically-established correlate of PTSD is engagement in reckless and self-destructive behaviors (RSDBs; Lusk, Sadeh, Wolf, & Miller, 2017; Tull, Weiss, & McDermott, 2015; Weiss, Tull, Sullivan, Dixon-Gordon, & Gratz, 2015).

Among theoretical explanations for the PTSD-RSDB link, the disinhibition viewpoint indicates that individuals with PTSD may have difficulties inhibiting rewarding RSDBs (Casada & Roache, 2005); the emotion dysregulation perspective indicates that individuals engage in RSDBs to reduce the negative affect or increase the blunted positive affect characteristic of PTSD (Baker, Piper, McCarthy, Majeskie, & Fiore, 2004); and the cognitive explanation suggests the trauma’s effects on decreasing attention span and information processing capacity may increase the likelihood of impulsive RSDBs (Ben-Zur & Zeidner, 2009). Of clinical salience, engagement in RSDBs among trauma-exposed samples has a detrimental impact on physical and mental health outcomes. For instance, in a study of veterans receiving treatment for PTSD, a significant number of deaths were related to RSDBs such as substance misuse and suicide (Drescher, Rosen, Burling, & Foy, 2003). Additionally, another study found that engagement in RSDBs was associated with greater psychological, emotional, and behavioral problems (Contractor, Weiss, Dranger, Ruggero, & Armour, 2017).

With this empirical, theoretical, and clinically-significant foundation, the E2 symptom assessing posttrauma RSDBs was added to the DSM-5 PTSD diagnostic criteria (American Psychiatric Association, 2013). However, the lack of a comprehensive (yet brief) validated screening measure to assess E2 poses a barrier to its measurement and consideration in treatment. In fact, existing assessments of PTSD’s E2 criterion lack clinical utility and empirical support in a number of ways. First, the assessment of PTSD’s E2 criterion in adults has relied either on a single-item (E2 symptom) included in PTSD assessments or requires a time-intensive battery of multiple distinct measures of a range of specific RSDBs (e.g., substance use, aggressive behaviors). With evidence supporting the co-occurrence of RSDBs (Cooper, 2002) representing an underlying unified latent factor (Shaw, Wagner, Arnett, & Aber, 1992; Weiss, Tull, Dixon-Gordon, & Gratz, 2016), using multiple RSDB measures may have less utility than a measure assessing a unified RSDB construct. Second, one could use existing comprehensive risky behavior measures such as the Risky Impulsive and Self-Destructive Questionnaire (Sadeh & Baskin-Sommers, 2017), the Risky Behaviors Questionnaire (Weiss et al., 2016), and the Risk-Taking Behavior Scale (Pat-Horenczyk et al., 2007). However, these are lengthy (e.g., 38, 29, and 87 items), assess more than just frequency of engaging in RSDBs (e.g., functionality of RSDBs), and/or are restricted to a specific developmental period (e.g., adolescence). Further, these measures do not include items to specifically examine posttrauma manifestations of RSDBs, a necessary criterion for evaluating PTSD’s E2. Indeed, while the types of RSDBs may be similar among trauma-exposed and non-trauma-exposed samples, there is a demonstrated uniqueness in the presentation, function, and course of RSDBs among trauma-exposed populations. For instance, trauma-exposed individuals may functionally engage in RSDBs as an emotion regulation strategy to cope with PTSD symptoms/distress (Marshall-Berenz, Vujanovic, & MacPherson, 2011; Weiss et al., 2015; Weiss, Tull, Viana, Anestis, & Gratz, 2012), implying their onset after trauma/PTSD symptoms.

Given the aforementioned limitations of existing measures, we need a comprehensive (yet brief enough to ensure clinical utility) and validated screening measure to examine the unidimensional E2 criterion. We developed the Posttrauma Risky Behaviors Questionnaire (PRBQ) to assess extent of engagement in posttrauma RSDBs and examined its factor structure, reliability, and validity (content, convergent, construct, incremental) in a trauma-exposed community sample recruited via Amazon’s MTurk platform. We further replicated this factor structure and examined its validity (construct and convergent) with a different trauma-exposed sample of college students (Hinkin, 1998; Holmbeck & Devine, 2009). We hypothesized good psychometric properties and ability to represent distinct RSDBs as a unidimensional construct.

Method

Procedure and Participants

Community Sample.

We recruited adult participants (≥ 18 years) from Amazon’s Mechanical Turk (MTurk), a platform which yields reliable data in terms of good-excellent internal consistency and high test-retest reliability of measures (Buhrmester, Kwang, & Gosling, 2011; Shapiro, Chandler, & Mueller, 2013). The study was described as a 45-60-minute survey to develop a measure of RSDBs following stressful life experiences. Inclusion criteria were (1) living in North America, (2) working English fluency, and (3) traumatic experience(s) screened with the Primary Care PTSD Screen for DSM-5 (Prins et al., 2015). Participants who met eligibility criteria, provided informed consent, and completed the survey on Qualtrics validly were compensated $1.25. These procedures were approved by the [removed for blind review] Institutional Review Board (IRB).

Student Sample.

We recruited undergraduate college students (≥ 18 years) from a large university in southwestern U.S. through an online participant pool from February 2018 to December 2018. The study was described as a 60-90-minute survey of ‘Stress, Substance Use, and Health,’ which included several measures of RSDBs. Inclusion criteria were reporting the experience of a traumatic event on the Life Events Checklist for DSM-5 (LEC-5; Weathers et al., 2013). Participants who met eligibility criteria, provided informed consent, and completed the survey on Qualtrics were compensated with 1.5 course credit. The procedures were approved by the [removed for blind review] IRB prior to participant contact.

Exclusions and Missing Data

Community Sample.

Of the obtained 891 responses, 47 responses of 18 participants who attempted to answer the questionnaire multiple times were excluded (remainder n = 844). We further excluded 150 participants who did not meet all inclusionary criteria, 122 participants who did not pass all four attention and comprehension validity checks (Meade & Craig, 2012; Thomas & Clifford, 2017), 97 participants who missed data on all measures, 11 participants who did not endorse a trauma/most distressing trauma on the LEC-5 (Weathers et al., 2013), and 47 participants who missed >30% item-level data on the PRBQ. The final sample included 417 participants (see Table 1 for demographics).

Table 1.

Demographic data for the community (N = 417) and student (N = 530) samples

	Community		Student
	M	SD	M	SD
Age	35.92	11.14	20.66	3.62
Years of education	15.27	2.38	14.42	1.12
	N	%*	N	%*
Gender
Women	236	56.60%	396	74.70%
Men	174	41.70%	134	25.30%
Other	7	1.70%	-	-
Ethnicity
Hispanic	46	13.00%	-	-
Non-Hispanic	308	85.30%	-	-
Unknown	6	1.70%	-	-
Race
White/Caucasian	305	71.20%	246	46.40%
African-American	34	8.50%	75	14.20%
Asian/Asian-American	41	10.50%	28	5.30%
Latino/Latina	-	-	101	19.10%
Other, unknown, or multiracial	37	3.40%	80	15.10%
Employment
Full-time	297	71.20%	-	-
Part-time	65	15.60%	-	-
Unemployed, retired, or student	55	13.20%	-	-
Student	-	-	530	100%
Income
< $50,000	215	51.60%	173	33.00%
≥ $50,000	202	48.40%	353	67.00%
Relationship status**
Single	-	-	144	57.80%
In a relationship	-	-	99	39.80%
Not dating	66	15.80%	-	-
Dating	135	32.40%	-	-
Married	183	43.90%	6	2.40%
Divorced, separated, widowed	33	7.90%	-	-
Probable PTSD (≥ 33)1	150	36.00%	126	23.80%

Note. M = mean; SD = standard deviation. PTSD = posttraumatic stress disorder.

¹Scores ≥ 33 are indicative of a probable PTSD diagnosis (Bovin et al., 2016);

^*valid percentages accounting for missing data are reported;

^**Relationship status descriptors were asked for part (47%) of the student sample

Student Sample.

Of the obtained 595 responses, we excluded 40 participants not endorsing a trauma on the LEC-5 (Weathers et al., 2013) and 25 participants with missing data on the PRBQ. The final sample included 530 participants (see Table 1 for demographics).

Measures

Life Event Checklist for DSM-5 (LEC-5; Weathers et al., 2013).

Administered to both samples, the LEC-5 is a 17-item self-report measure assessing exposure to traumatic events. Participants respond using the following response options: “Happened to me,” “Witnessed it,” “Learned about it,” “Part of my job,” “Not sure,” or “Doesn’t apply.” Participants who selected one of the first four response options were considered to have endorsed a traumatic event consistent with PTSD’s DSM-5 Criterion A (American Psychiatric Association, 2013).

Posttrauma Risky Behaviors Questionnaire (PRBQ; see Appendix A).

Administered to both samples, this 16-item self-report measure was developed to assess the extent of RSDB following the most traumatic event (PTSD’s E2 criterion). Items 1-14 assess the extent of past-month engagement in specific and clinically-relevant RSDBs (ranging from 0 [never] to 4 [very frequently]). The first 14 items are summed, with higher scores reflecting greater extent of RSDB engagement. The last two items examine functional impairment (ranging from 0 [not at all] to 3 [extremely]) and the relation of RSDBs (initiation/worsening) to the onset of the most traumatic event (yes/no responses). Participants take about 5-10 minutes to complete this survey.

PRBQ’s development followed recommended guidelines (Furr, 2011; Germain, 2006; Hinkin, Tracey, & Enz, 1997). At every step, the authors discussed and critically evaluated information/feedback and made decisions based on consensus. In Step 1 (domain generation), we used a deductive approach wherein theory, clinical relevance, and empirical evidence informed domain selection (Germain, 2006). Research assistants conducted a detailed literature review to identify RSDBs with empirical and theoretical relations to PTSD (e.g., Ben-Zur & Zeidner, 2009; Contractor, Frankfurt, Weiss, & Elhai, 2017; Lusk et al., 2017; Weiss et al., 2016). The initial pool comprised 12 domains (i.e., substance use, tobacco use, gambling, risky sexual behavior, disordered eating, criminal behavior, reckless spending, aggressive behaviors, reckless driving, non-suicidal self-injury (NSSI), suicidal behaviors, reckless/thrill-seeking behaviors) and 20 sub-domains (e.g., alcohol, drugs, unprotected sex, trading sex, binging/overeating, purging, shoplifting/stealing, destructive behaviors, physical intimidation, road rage, athletic activities such as rock climbing). For Step 2 (expert panel review), we selected five experts in the field of trauma/RSDBs based on a consensus among authors to critically review the initial PRBQ draft (Germain, 2006). The five experts were licensed clinical psychologists, faculty members in psychology/psychiatry departments, and have published extensively in the field of trauma/RSDBs (minimum of 40 published articles). These same experts were contacted for Step 4 of the scale development process (they are not authors of this scale). Based on their feedback, we eliminated (e.g., “reckless/thrill-seeking behaviors”) and modified (e.g., “bingeing/overeating,” “restricting,” and “purging” were combined into a problematic eating behaviors domain) domains, finalizing 15 RSDB domains.

In Step 3 (item generation/development/selection), we developed an initial item pool of 15 RSDB items (Germain, 2006; Hinkin et al., 1997) by reviewing measures assessing different RSDBs (e.g., Strom et al., 2012; Weiss et al., 2016). We ensured that each item addressed a single issue and was worded in simple and clear language (Hinkin et al., 1997). A Likert scale of five response options was created (Hinkin et al., 1997), adapted from the Clinician Administered PTSD Scale for DSM-5 (CAPS-5; Weathers et al., 2018). Additionally, two supplementary items were created; one assessed functional impairment modified from the Patient Health Questionnaire-9 (PHQ-9; Kroenke & Spitzer, 2002), and the other assessed the relation of the RSDBs to the onset of the most traumatic event modified from the CAPS-5 (Weathers et al., 2018). In Step 4 (content adequacy assessment), this 17-item measure was subjected to an expert panel review. We obtained feedback regarding several dimensions (i.e., clarity, conciseness, acceptability of grammar, face validity, redundancy with other items) for each of the 15 specific RSDB items (Germain, 2006; Hinkin et al., 1997), the instruction set, Likert scale, and the two supplementary items. In Step 5, the authors collaboratively modified the items, instruction set, Likert scale, and supplemental items to create a final 16-item measure. At this stage, the domain of “problematic health behaviors” was removed based on expert review. Finally, in Step 6 (pilot testing), the 16-item scale was administered with other measures to assess psychometric properties (Furr, 2011; Germain, 2006; Hinkin et al., 1997). The sample sizes of 417 (community sample) and 530 (student sample) were consistent with existing recommendations of scale development (DeVellis, 2003; Hinkin et al., 1997).

Risky Behaviors Questionnaire (RBQ; Weiss et al., 2016).

Administered to the community sample, the RBQ is an 89-item, self-report measure assessing the total frequency and function of 29 clinically-relevant RSDBs in the past year. Scores are based on the total frequency of RSDBs. The RBQ has adequate psychometric properties (Weiss et al., 2016). Cronbach’s α coefficient in the community sample of the current study was .72.

Aggression Questionnaire (AQ; Buss & Perry, 1992).

Administered to the community sample, the AQ is a 29-item self-report measure assessing aggression domains. Participants rate the extent to which items apply to them on a 7-point Likert-type scale (1 = extremely uncharacteristic of me, 7 = extremely characteristic of me). The AQ has demonstrated adequate psychometric properties (Buss & Perry, 1992). Cronbach’s α coefficients in the community sample of the current study were .81, .82, .80, and .91 for the physical aggression, verbal aggression, anger, and hostility subscales, respectively.

Drug Abuse Screening Test (DAST; Skinner, 1982).

Administered to the community sample, this 10-item self-report measure assesses drug misuse using 1 (yes) and 0 (no) response options. We used the summed score in the current study to represent drug misuse. It has adequate psychometric properties (Yudko, Lozhkina, & Fouts, 2007); Cronbach’s α coefficient in the community sample of the current study was .83.

Inventory of Statements About Self-Injury (ISAS; Klonsky & Glenn, 2009).

Administered to the community sample, the ISAS is a self-report measure assessing the frequency of 13 lifetime NSSI behaviors. Items are summed to represent the total frequency of NSSI. The ISAS has adequate psychometric properties (Klonsky & Glenn, 2009). As expected, a relatively low Cronbach’s α coefficient was found (α = .54) in our community sample given the checklist nature of this scale (Streiner, 2003).

Eating Attitude Test-26 (EAT-26; Garner, Olmsted, Bohr, & Garfinkel, 1982).

Administered to the community sample, the EAT-26 is a 26-item self-report measure assessing cognitions, emotions, and behaviors related to eating disorders with a 6-point Likert scale (0 = never, 5 = always). Items were recoded to a scale ranging from 0 to 3, as detailed in Garner et al. (1982), with higher scores being associated with more disordered eating behavior. Cronbach’s α coefficients in the community sample of the current study for the Dieting, Bulimia, and Oral Control subscales were .89, .61, and .72, respectively.

Media and Technology Usage and Attitudes Scale (MTUAS; Rosen, Whaling, Carrier, Cheever, & Rokkum, 2013).

Administered to the community sample, this 44-item self-report measure assesses technology-based media use with a 10-point Likert scale (1 = never, 10 = all the time). It has adequate psychometric properties (Rosen et al., 2013). For the community sample of this study, Cronbach’s α coefficients for the Email, Texting, Calling, Smartphone, TV, Media, Internet, Gaming, Social Media, and Facebook subscales were .81, .78, .82, .94, .65, .87, .88, .81, .93, and .84, respectively.

Problem Gambling Severity Index (PGSI; Ferris & Wynne, 2001).

Administered to the community sample, the PGSI is a 9-item self-report measure assessing problematic gambling behaviors over the past 12 months with a 4-point Likert scale (0 = never, 3 = almost always). We used a summed score to represent problematic gambling. It has adequate psychometric properties (Holtgraves, 2009); Cronbach’s α coefficient for the community sample of this study was .96.

Addiction Severity Index (ASI; McLellan et al., 1992).

Selected ASI items were used in the community sample to assess criminal behavior: (a) lifetime criminal arrests, (b) lifetime criminal convictions, and (c) number of days they engaged in criminal behavior during the past month. The ASI has adequate psychometric properties (McLellan, Cacciola, Alterman, Rikoon, & Carise, 2006).

Compulsive Buying Scale (CBS; Faber & O'guinn, 1992).

Administered to the community sample, the CBS is an 8-item measure assessing behaviors, motivations, and feelings related to compulsive buying using a 5-point Likert scale (1 = never/strongly disagree, 5 = very often/strongly agree). It has adequate psychometric properties (Faber & O'guinn, 1992); Cronbach’s α coefficient in the community sample of the current study was .87.

Sexual Risk Survey (SRS; Turchik & Garske, 2009).

Administered to the community sample, the SRS is a 23-item measure used to assess sexual risk taking behavior over the past six months. Respondents report the frequency with which each item applies to them. The SRS has adequate content, concurrent, and convergent validity (Turchik & Garske, 2009); Cronbach’s α coefficient in the community sample of the current study was .73.

Reckless Driving Habits Scale (RDHS; Taubman-Ben-Ari, Mikulincer, &Iram, 2004).

Administered to the community sample, the RDHS is a 14-item measure assessing the frequency of reckless driving behaviors with a 6-point Likert scale (1 = never, 6 = always). Cronbach’s α coefficient in the community sample of the current study was .94.

Alcohol Use and Disorders Identification Test Alcohol Consumption Questions (AUDIT-C; Bush et al., 1998).

Administered to both samples, the AUDIT-C is a 3-item self-report measure assessing heavy drinking. It has good psychometrics (Bush et al., 1998); Cronbach’s α coefficients were .74 and .76 for the community and student samples, respectively.

Suicide Behaviors Questionnaire-Revised (SBQ-R; Osman et al., 2001).

Administered to both samples, the SBQ-R is a 4-item self-report measure that assesses lifetime suicidality. The SBQ-R has adequate psychometric properties (Osman et al., 2001); Cronbach’s α coefficients for the community and student samples were .84 and .86, respectively.

PTSD Checklist for DSM-5 (PCL-5; Weathers et al., 2013).

Administered to both samples, this 20-item self-report measure assesses PTSD severity in the past month using a 5-point Likert scale (0 = not at all, 4 = extremely). The PCL-5 has excellent psychometric properties (Bovin et al., 2016). For the community sample, the Cronbach’s α coefficients were .96, .91, .87, .92, and .90 for the total, intrusion, avoidance, negative alterations in cognitions and mood (NACM), and alterations in arousal and reactivity (AAR) scales, respectively. For the student sample, the Cronbach’s α coefficients were .96, .90, .91, .92, and .89 for the total, intrusion, avoidance, NACM, and AAR scales, respectively. Participants completed the PCL-5 referencing the most distressing trauma endorsed on the LEC-5 (Weathers et al., 2013).

Patient Health Questionnaire-9 (PHQ-9; Kroenke & Spitzer, 2002).

Administered to both samples, this 9-item self-report measure assesses depression symptoms over the past two weeks using a 4-point Likert scale (0 = not at all, 3 = nearly every day). The PHQ-9 has good psychometrics (Kroenke, Spitzer, & Williams, 2001); Cronbach’s α coefficients of the total scale for the community and student samples were .91 and .90, respectively.

Statistical Plan

Community Sample.

We first examined normality statistics for all variables. Following recommended guidelines, we assessed PRBQ’s psychometric properties by examining its factor structure using exploratory (EFA; n = 222) and confirmatory (CFA; n = 195) factor analyses on two randomly selected subsamples from our data, internal consistency, and validity (content, construct, convergent, and incremental; Furr, 2011; Germain, 2006; Hinkin et al., 1997).

We used EFA (oblique geomin rotation) to examine 1-4 factor models with Mplus version 8. The PRBQ Likert-scale represents a ranked order of frequency-based responses with no assumption of equal intervals across response options (Jamieson, 2004). Hence, we treated the 14 PRBQ items as categorical (Bernstein & Teng, 1989; Finney & DiStefano, 2013) using the weighted least squares means and variance adjusted (WLSMV) estimation, which has been shown to be less biased and more accurate than robust maximum likelihood in estimating factor loadings for ordinal data (Li, 2016). Factor selection was guided on the basis of fit indices [non-significant χ², comparative fit index (CFI) and Tucker Lewis Index (TLI) values ≥ 0.95 (0.90-0.94), root mean square error of approximation (RMSEA) value ≤ 0.06 (0.07-0.08), and standardized root mean square residual (SRMR) value < 0.08 (0.09-0.10; Hu & Bentler, 1999)], Horn’s parallel analyses’ results, pattern and quality of factor loadings (≥ .40) and cross-factor loadings, the eigenvalue > 1 rule, scree plot, inter-correlations among factors, and the putative meaning of the factors (Costello & Osborne, 2005; Fabrigar, Wegener, MacCallum, & Strahan, 1999; Horn, 1965). On a second randomly selected subsample, we conducted a CFA treating the PRBQ items as categorical, using WLSMV as the estimator, scaling factor variances to 1, and estimating all factor loadings. We evaluated model fit based on the aforementioned fit indices.

Next, Cronbach’s alpha was calculated for internal consistency. After establishing content validity (expert panel review), convergent validity was examined using correlations (Pearson, Spearman’s, point-biserial) of the PRBQ score with measures of mental health symptoms and RSDBs. We used recommended benchmarks (Hinkle, Wiersma, & Jurs, 2003) to interpret effect sizes of a correlation coefficient (0.90-1.00 = very high; 0.70-0.90 = high; 0.50-0.70 = moderate; 0.30-0.50 = low; 0.00-0.30 = negligible). To evaluate construct validity, we computed correlations (Spearman’s, point-biserial) between the PRBQ and PTSD’s E2 Criterion scores; and analyses of variance (ANOVA) and chi-square tests to explore differences in PRBQ scores across participants with and without clinically significant levels of PTSD’s E2 Criterion (item-level score of ≥ 2 indicates clinical endorsement; Weathers et al., 2013). Finally, a binary logistic regression analysis was used to explore the incremental validity of the PRBQ score in predicting PTSD’s E2 Criterion above the RBQ (established and validated measure of RSDBs).

Student Sample.

We first examined normality statistics for all variables. Next, following recommended guidelines, we assessed PRBQ’s psychometric properties by examining the best-fitting factor structure using CFA, internal consistency coefficient, convergent validity (Pearson, Spearman’s, point-biserial correlations with measures of mental health symptoms and of RSDBs), and construct validity (Spearman’s, point-biserial, ANOVA, chi-square tests) in a replication sample of trauma-exposed students (Furr, 2011; Germain, 2006; Hinkin et al., 1997). We followed the same statistical procedures and guidelines as used for the community sample.

Results

Community Sample

Based on benchmarks of skewness > 2 and kurtosis > 7 (Curran, West, & Finch, 1996), four measures exhibited non-normal distributions (ISAS, RBQ, SRS, ASI). Applying transformations did not correct non-normality, thus, nonparametric tests (i.e., Spearman correlations) were used to assess the relation of these variables to PRBQ scores.

Descriptive Statistics and Demographic Correlates.

Supplemental Table 1 summarizes means and standard deviations for mental health and RSDB measures and Supplemental Table 2 indicates frequencies of endorsing each response option for the PRBQ items. We conducted ANOVAs and Pearson correlations to examine if the 14-item PRBQ score varied as a function of demographic variables. No significant relationship was found between years of education and the 14-item PRBQ score (r = .09, p = .078). The 14-item PRBQ score was significantly associated with age (r = −.28, p < .001); gender (F [2, 416] = 4.08, p = .02, η² = .02), with men (M = 8.53, SD = 11.26) reporting higher PRBQ scores than women (M = 5.86, SD = 8.31); employment (F [2, 416] = 3.98, p = .02, η² = .02), with individuals endorsing full-time employment (M = 7.69, SD = 10.52) reporting higher PRBQ scores than those who were unemployed/students/retired (M = 3.67, SD = 4.33); relationship status (F [3, 416] = 3.13, p = .03, η² = .02), with individuals in dating relationships (M = 9.04, SD = 11.27) reporting higher PRBQ scores than married individuals (M = 5.68, SD = 8.51); ethnicity (F [1, 416] = 6.87, p = .001, η² = .03), with Hispanic individuals (M = 11.30, SD = 12.35) reporting higher PRBQ scores than non-Hispanic individuals (M = 6.45, SD = 9.24); race (F [3, 416] = 2.64, p = .05, η² = .02); and income (F [1, 416] = 7.63, p = .01, η² = .02), with individuals earning < $50,000/year (M = 8.31, SD = 10.58) reporting higher PRBQ scores than individuals earning ≥ $50,000/year (M = 5.68, SD = 8.69).

PRBO’s Psychometric Properties.

Referencing EFA, the scree plot (see Supplemental Figure 1), eigenvalue > 1 rule, and Horn’s parallel analyses¹ indicated a one-factor solution; the two-factor solution had cross-loadings for two PRBQ items and the obtained factors lacked conceptual meaning. Thus, EFA results based on the recommended guidelines suggested a one-factor PRBQ model; χ²(77) = 113.72, p = .004, CFI =1.00, TLI = .99, RMSEA = .05, SRMR = .05. Review of factor loadings and communality estimates (see Table 2) indicated that items loaded well on the obtained factor and did not need removal for the subsequent analyses. CFA results indicated an excellent fit for the one-factor PRBQ model; χ²(77) = 144.31, p < .001, CFI = .99, TLI = .99, RMSEA = .07, SRMR = .04. See Table 2 for factor loadings. Further, results revealed excellent internal consistency of .94. The 14-item PRBQ score associated significantly with PRBQ supplemental items 15 (r_s = .72, p < .001) and 16 (r_pb = .51, p < .001). The supplemental items were significantly associated (r_pb = .50, p < .001).

Table 2.

Factor loadings and communalities for the 14 Posttrauma Risky Behaviors Questionnaire items

	EFA factor loadings (Community)	EFA communalities (Community)	CFA factor loadings (Community)	CFA factor loadings (Student)
Item 1. Problematic alcohol use	.68	.47	.76	.57
Item 2. Problematic drug use	.81	.66	.81	.62
Item 3. Problematic gambling	.88	.78	.90	.90
Item 4. Problematic technology use	.68	.47	.77	.52
Item 5. Impulsive or risky sexual behaviors	.88	.77	.89	.70
Item 6. Problematic eating behavior(s)	.75	.57	.69	.60
Item 7. Illegal behaviors	.88	.77	.91	.81
Item 8. Reckless spending	.73	.53	.81	.63
Item 9. Physically aggressive behavior(s)	.92	.84	.91	.88
Item 10. Verbally aggressive behavior(s)	.81	.66	.85	.75
Item 11. Property destruction	.93	.87	.93	.93
Item 12. Reckless driving	.91	.83	.90	.64
Item 13. Deliberately injuring yourself without intending to kill yourself	.94	.88	.88	.78
Item 14. Suicidal behaviors	.85	.73	.92	.84

Note. EFA is exploratory factor analyses (all factor loadings are significant atp < .05; CFA is confirmatory factor analyses (all factor loadings are significant at p < .001).

Referencing convergent validity (see Table 3), the 14-item PRBQ total score was positively associated with PTSD (total [moderate effect] and all of the subscale scores [moderate to low effects]) and depression severity (moderate to low effects). Further, the 14-item PRBQ total score was positively associated with measures assessing overall RSDBs (RBQ; low effect) and the specific RSDBs of aggression (each domain; moderate to low effects), alcohol misuse (low effect), drug misuse (low effect), compulsive buying (moderate effect), disordered eating (each domain; low effects), problematic technology use domains (texting, smartphone use, TV viewing, media sharing, gaming, social media, Facebook friendships; negligible effects), problematic gambling (moderate effect), suicidality (negligible effect), reckless driving (moderate effect), illegal behaviors (each domain; low to negligible effects), NSSI (low effect), and risky sexual behavior domains (each domain excluding risky sexual acts; negligible effects). PRBQ supplemental items 15 and 16 were positively related to PTSD (total [low effect] and all subscale scores [moderate to low effects]) and depression severity (moderate to low effects). PBRQ items 15 and 16 scores were positively correlated with overall RSDBs (RBQ; low to negligible effects) and the specific RSDBs of aggression (all domains; low to negligible effects), alcohol and drug misuse (low to negligible effects), compulsive buying (low effects), disordered eating (all domains; negligible effects), problematic gambling (low to negligible effects), suicidality (negligible effects), reckless driving (low to negligible effects), illegal behaviors (all domains; negligible effects), and NSSI (low to negligible effects). Additionally, PBRQ item 15 was positively correlated with problematic technology use domains (texting, smartphone use, media sharing, gaming, social media use, Facebook friendships; negligible effects), and risky sexual behavior domains of impulsive sexual behaviors and intent to engage in risky sexual behaviors (negligible effects).

Table 3.

Convergent validity of the Posttrauma Risky Behaviors Questionnaire

Community Sample
Measure	14-Item PRBQ	PRBQ Item 15	PRBQ Item 16
PTSD Checklist for DSM 5	r = .57p < .001	rs = .49 p < .001	rpb = .36p< .001
Intrusions	r = .48p < .001	rs = .43p < .001	rpb = .35p < .001
Avoidance	r = .37p < .001	rs = .56p < .001	rpb = .37p < .001
Negative alterations in cognitions and mood	r = .55p < .001	rs = .57p < .001	rpb = .43p < .001
Alterations in arousal and reactivity	r = .59p < .001	rs = .57p < .001	rpb = .42p < .001
Criterion E2	rs =.57p < .001	rs = .54p < .001	rpb = .29 p = .001
Patient Health Questionnaire-9	r = .62p < .001	rs = .60p< .001	rpb = .40p < .001
Risky Behavior Questionnaire	rs = .38p < .001	rs = .29p < .001	rs = .37p < .001
Buss-Perry Aggression Questionnaire
Physically aggressive behaviors	r = .59p < .001	rs = .39p < .001	rpb = .16p =.001
Verbally aggressive behaviors	r = .43p < .001	rs = .32p < .001	rpb = .20p < .001
Anger	r = .56p < .001	rs = .40p< .001	rpb = .25p < .001
Hostility	r = 48p < .001	rs = .47p < .001	rpb = .33p < .001
Alcohol Use Disorders Indicator Test-Alcohol	r = .33p < .001	rs = .25p = .001	rpb = .23p < .001
Consumption Questions
Drug Abuse Screening Test	r = .44p < .001	rs = .34p < .001	rpb = .37p < .001
Compulsive Buying Scale	r = .57p < .001	rs = .48p < .001	rpb = .33 p < .001
Eating Attitudes Test
Dieting	r = .31p < .001	rs = .21p < .001	rpb = .28p < .001
Bulimia and food preoccupation	r = .37p < .001	rs = .29p < .001	rpb = .29p < .001
Oral control	r = .39p < .001	rs = .22p < .001	rpb = .26p < .001
Media and Technology Usage and Attitudes Scale
E-mailing	r = .03p = .51	rs = −.002p = .96	rpb = .03 p = .52
Text messaging	r = .13p = .01	rs = .10 p = .04	rpb = .04p=.43
Phone calling	r = 0.9p = .08	rs = .08p = .11	rpb = .02p=.68
Smartphone usage	r = .18p = .004	rs = .13p = .01	rpb = .07p=.19
TV viewing	r = 19p = .003	rs = .08p = .09	rpb = .01p =.84
Media sharing	r = .27p < .001	rs = .10p = .05	rpb = .01p =.77
Internet searching	r = 0.9p = 0.7	rs = .09p =.08	rpb = 0.4p = .40
Video gaming	r = .24p < .001	rs = .11p =.03	rpb = .04p =.42
Social media usage	r = .21p< .001	rs = .17p =.001	rpb = 0.7p =.18
Facebook friendships	r = .23p < .001	rs = .14p =.01	rpb = .03p =.62
Problematic Gambling Severity Index	r =.68p < .01	rs = .34p < .001	rpb = .14P =.01
Suicidal Behaviors Scale	r =.12p = .02	rs = .22p < .001	rpb = .25p < .001
Reckless Driving Habits Scale	r =.68p < .001	rs = .30p < .001	rpb = .16p =.001
Addiction Severity Index - Lifetime arrests	rs = .31p < .001	rs = .25p < .001	rs = .20p < .001
Addiction Severity Index Lifetime convictions	rs =.24p < .001	rs = .20P < .001	rs = .18p =.001
Addiction Severity Index Past 30-day illegal behaviors	rs = 31p < .001	rs = .17p =.001	rs = .15p =.003
Inventory of Statement About Self-Injury	rs =.32p < .001	rs = .29p < .001	rs = .30p < .001
Sexual Risk Survey	rs =.04p = .41	rs = −.01p =.89	rs = .08p =.11
Uncommitted partners	rs =.14p = .01	rs = .09p =.09	rs = .09p =.09
Risky sex acts	rs =−.01p = .93	rs = −.08p =.32	rs = .09p =.10
Impulsive sexual behaviors	rs = .29p < .001	rs = .23p < .001	rs = −.07p =.18
Intent to engage in risky sexual behaviors	rs = .23p < .001	rs = .12p =.02	rs = −.04p =.45
Risky anal sex acts	rs = .14p = .01	rs = .06p =.24	rs = .03p =.56
Student Sample
PTSD Checklist for DSM 5	r = .41p < .001	rs = .54 p < .001	rpb = .47p < .001
Intrusions	r =.32p < .001	rs = .45p < .001	rpb = .41p < .001
Avoidance	r =.30p < .001	rs = .42p < .001	rpb = .39p < .001
Negative alterations in cognitions and mood	r = .40p < .001	rs = .54p < .001	rpb = .45p < .001
Alterations in arousal and reactivity	r =.44p < .001	rs = .53p < .001	rpb = .43p < .001
Criterion E2	r =.50p < .001	rs = .45p < .001	rpb = .29p < .001
Patient Health Questionnaire-9	r = .41p < .001	rs = .53p < .001	rpb = .37p < .001
Alcohol Use Disorders Indicator Test-Alcohol	r = .33p < .001	rs = .04p = .39	rpb = .04p = .38
Consumption Questions
Suicidal Behaviors Scale	r =.33p < .001	rs = .44p < .001	rpb = .39p< .001

Note. r = Pearson Product Moment correlation coefficients; r_s = Spearman Rank-Order correlation coefficients; r_pb = point-biserial correlation coefficients; PBRQ = Posttrauma Risky Behaviors Questionnaire; PTSD = posttraumatic stress disorder.

Referencing construct validity in relation to PTSD’s E2 Criterion (see Table 3), the PRBQ scores were positively related to PTSD’s E2 Criterion on the PCL-5 (moderate to negligible effects). Clinical endorsement of PTSD’s E2 Criterion was associated with higher scores on the: (a) 14-item PRBQ scale, F(1, 415) = 228.93, p < .001, η² = .36, such that individuals endorsing clinically significant E2 (M = 17.48, SD = 12.89) reported higher trauma-related RSDBs than participants not endorsing clinically significant E2 (M = 3.79, SD = 5.44); (b) PRBQ item 15, F(1, 415) = 118.50, p < .001, η² = .22, such that individuals endorsing clinically significant E2 (M = 1.32, SD = 0.89) reported higher functional impairment stemming from trauma-related RSDBs than participants not endorsing clinically significant E2 (M = 0.40, SD = 0.68); and (c) PRBQ item 16, χ² (1) = 23.52, p < .001, such that more individuals endorsing clinically significant E2 (59.60%) reported that trauma-related RSDBs developed/worsened after their most traumatic event than participants not endorsing clinically significant E2 (32.40%).

Referencing incremental validity in predicting PTSD’s E2 Criterion, results indicated that the RBQ was not significantly related to PTSD’s E2 Criterion (B = .001, SE = .001, Wald χ² [1] = 2.66, p = .10, Exp[B] = 1.00) in the first step of the model, χ² (1) = 2.50, p = .11, R² = .01. In the second step of the model, the PRBQ was significantly and uniquely associated with PTSD’s E2 Criterion (B = .17, SE = .02, Wald χ² [1] = 66.75, p < .001, Exp[B] = 1.19) above and beyond the RBQ (B = .000, SE = .001, Wald χ² [1] = 0.02, p = .90, Exp[B] = 1.00); χ² (2) = 122.03, p < .001, R² = .44. In this final step, 84.9% of cases were correctly classified.

Student Sample

Based on benchmarks of skewness > 2 and kurtosis > 7 (Curran et al., 1996), no variables exhibited non-normal distributions, thus, parametric tests were used.

Descriptive Statistics and Demographic Correlates.

Supplemental Table 1 summarizes means and standard deviations for mental health and RSDB measures and Supplemental Table 2 indicates frequencies of endorsing each response option for the PRBQ items. ANOVAs and Pearson correlation results indicated no significant associations/differences for age (r < .01, p = .94); education (r = .01, p = .82); gender (F [1, 528] = .90, p = .35, η² < .01); race (F [5, 524] = 1.43, p = .21, η² = .01); or socioeconomic status (F [1, 524] = 0.19, p = .66, η² < .01).

PRBQ’s Psychometric Properties.

CFA results indicated an excellent fit for the one-factor PRBQ model; χ²(77) = 268.24, p < .001, CFI = .96, TLI = .95, RMSEA = .07, SRMR = .07. See Table 2 for factor loadings. Results revealed very good internal consistency of .82. The 14-item PRBQ score associated significantly with PRBQ supplemental items 15 (r_s = 56, p < .001) and 16 (r_pb = 31, p < .001). The supplemental items were significantly associated (r_pb = .44, p < .001). Referencing convergent validity (see Table 3), the 14-item PRBQ total score was positively associated with PTSD (total [low effect] and all subscale scores [low effects]) as well as depression severity (low effect), alcohol misuse (low effect), and suicidality (low effect). PRBQ supplemental items 15 and 16 were positively related to PTSD (total [moderate to low effects] and all subscale scores [moderate to low effects]) as well as depression severity (moderate to low effects), and suicidality (low effects); however, neither item was associated with alcohol misuse (negligible effects).

Referencing construct validity in relation to PTSD’s E2 Criterion (see Table 3), the PRBQ scores were positively related to PTSD’s E2 Criterion on the PCL-5 (moderate to negligible effects). Clinical endorsement of PTSD’s E2 Criterion was associated with higher scores on the: (a) 14-item PRBQ scale, F (1, 526) = 138.75, p < .001, η² = .21, such that individuals endorsing clinically significant E2 (M = 11.41, SD = 9.09) reported higher trauma-related RSDBs than participants not endorsing clinically significant E2 (M = 4.05, SD = 4.18); (b) PRBQ item 15, F (1, 526) = 130.69, p < .001, η² = .20, such that individuals endorsing clinically significant E2 (M = .90, SD = .78) reported higher functional impairment stemming from trauma-related RSDBs than participants not endorsing clinically significant E2 (M = .25, SD = .49); and (c) PRBQ item 16, χ² (1) = 28.93, p < .001, such that more individuals endorsing clinically significant E2 (50.60%) reported that trauma-related RSDBs developed/worsened after their most traumatic event than participants not endorsing clinically significant E2 (22.30%).

Discussion

The current study provides preliminary support for the utility and validity of the 16-item brief (yet comprehensive) self-report PRBQ in the assessment of PTSD’s E2 Criterion. First, EFA and CFA provided evidence for a unitary construct of RSDBs as assessed by the PRBQ in two distinct trauma-exposed samples, indicating a general overarching factor capturing common variance across diverse RSDBs (most items had high factor loadings). Thus, although RSDBs may be diverse in manifestation and in underlying etiologies (Baker et al., 2004; Ben-Zur & Zeidner, 2009), they constitute a unitary construct in the current study samples as also supported by prior research (Shaw et al., 1992; Weiss et al., 2016). These results indicate the utility of comprehensive assessments such as the PRBQ that summarize the extent of engagement in a wide range of clinically-relevant and trauma-specific RSDBs (versus administering multiple measures of specific RSDBs). Further, these results may also support treatments targeting common mechanisms underlying multiple and different (rather than specific) RSDBs.

Given the diversity of assessed RSDBs, it is not surprising that the factor loadings had a wide range in both samples; these findings concur with prior research on risky behavior measure development and validation (Sadeh & Baskin-Sommers, 2017; Weiss et al., 2016). The differing factor loadings may reflect differences in the amount of item variance explained by the retained factor as evident from communality estimates; RSDBs with the highest factor loadings had better communality estimates compared to RSDBs with the lowest factor loadings. Notably, factor loadings in the student sample were lower than the factor loadings in the community sample possibly attributed to differences in frequencies of endorsed items and sample characteristics (e.g., the student sample was younger and had more female participants). However, we do see a similar pattern of the strength of factor loadings across these samples. In the community and student samples, items with the highest factor loadings (≥ .90 and ≥ .80 for CFA/EFA respectively) were those assessing physically aggressive behaviors, property destruction, problematic gambling, illegal behaviors, and suicidal behaviors; additional items with high factor loadings in the community sample included those assessing reckless driving and NSSI. In the community sample, items with the lowest factor loadings (≤ .70 for CFA/EFA) were those assessing alcohol misuse, technology use, and problematic eating behaviors; we see a similar pattern of results for the student sample (≤ .60 for CFA). Future research needs to cross-validate PRBQ’s factor structure and pattern of factor loadings in diverse and clinical samples.

Additionally, our study provides preliminary evidence of PRBQ’s reliability and validity. The PRBQ demonstrated very good to excellent internal consistency in both samples, suggesting that it assesses the RSDB domain in a reliable manner. Further, the PRBQ correlated with assessments of mental health and RSDBs as expected in both samples (particularly with PTSD symptoms in the community and college samples, and with depression symptoms, aggression, compulsive buying, problematic gambling, and reckless driving in the community sample), demonstrating good convergent validity. Most correlations had low to moderate effect sizes. Referencing the observed strong relation of RSDBs with PTSD and depression severity, particularly in the community sample, and as evidenced in past work (Chang, Jiang, Mkandarwire, & Shen, 2019; Del Gaizo, Elhai, & Weaver, 2011; Holmes, Foa, & Sammel, 2005; Pat-Horenczyk et al., 2007; Swendsen & Merikangas, 2000), individuals with greater PTSD/depression severity may engage in RSDBs to modulate distress (Tull & Gratz, 2013; Weiss et al., 2015; Weiss et al., 2012), or the short-term pleasure associated with RSDBs may serve to reduce severity of these symptoms (Simpson, Stappenbeck, Luterek, Lehavot, & Kaysen, 2014). Further, shared distress (i.e., negative affect) across PTSD and depression constructs (Contractor et al., 2014; Watson, 2009) may have accounted for their strong relations with PRBQ scores; in other words, negative affect may have similarly influenced responses to PTSD, depression, and PRBQ measures. Notably, not all technology and risky sexual behavior domains correlated significantly with the PRBQ. Perhaps different technology use patterns (e.g., smartphone use vs. emailing) and different risky sexual behaviors (e.g., uncommitted partners vs. risky sex acts) have divergent relations with trauma exposure/PTSD (e.g., Strom et al., 2012; Vanderwerker & Prigerson, 2004). It is also possible that these relations are moderated by variables such as emotional regulation (Tull, Weiss, Adams, & Gratz, 2012) and different trauma types/patterns (Contractor, Caldas, Fletcher, Shea, & Armour, 2018), or that trauma exposure is not the most salient factor accounting for engagement in these types of RSDBs. Such hypotheses can be investigated with different measures of technology addiction and risky sexual behavior facets that capture more nuanced behaviors.

In addition, the PRBQ was found to be associated with severity and clinically significant levels of PTSD’s E2 Criterion, providing support for its construct validity in both samples; the PRBQ does assess the construct of posttrauma RSDBs. Finally, in the community sample, the PRBQ was found to demonstrate incremental validity in predicting PTSD’s E2 Criterion above and beyond an established measure of RSDBs, indicating its utility in specifically assessing post-trauma RSDBs. Such findings are consistent with the lower correlations between the PRBQ and the RBQ; the latter distinctly assesses non-trauma specific RSDBs using a frequency rating scale. Additional research is needed to evaluate psychometric properties of the PRBQ, including its construct validity (e.g., via clinician-administered interviews), clinical cut-off scores, test-retest reliability, and changes in scores corresponding to administration of interventions.

Further, although there were demographic differences in PRBQ scores in the community sample, there were no differences in the student sample; future studies need to cross-validate the presence/absence in demographic differences in PRBQ scores and examine the equivalence of the construct of PRBQ-assessed RSDBs across the assessed subgroups. For instance, younger participants endorsed a higher frequency of RSDBs compared to older participants, consistent with prior research (Brener, McMahon, Warren, & Douglas, 1999; Jonah, 1990; Rosenberg et al., 2001), and possibly attributed to greater trauma experiences in this age group (Hatch & Dohrenwend, 2007). Additionally, men reported a higher frequency of RSDBs than women consistent with existing research (Brady & Randall, 1999; Nolen-Hoeksema, 2004; Pat-Horenczyk et al., 2007). Further, individuals who are employed may have more opportunities to engage in some RSDBs due to greater financial means (e.g., reckless spending, gambling, and substance use); future research needs to test this hypothesis.

In research and clinical settings, the PRBQ may be reliably and validly used to assess PTSD’s E2 Criterion. Clinicians could use the PRBQ as a screening instrument to identify the extent of engaging in and nature of RSDBs to be included as intervention targets (thus individualizing treatments) as well as to monitor changes in RSDBs. Given that RSDBs frequently co-occur (Cooper, 2002) and engagement in RSDBs increases the risk for subsequent traumas (Sadeh, Miller, Wolf, & Harkness, 2015) and mental health concerns (e.g., PTSD; Lusk et al., 2017), a timely assessment with the PRBQ can aid in preventing future adverse outcomes.

Study results need to be considered in the context of limitations. The tendency to provide socially desirable responses and the possibility of misinterpreting questions are inherent in self-report measures (Rosenman, Tennekoon, & Hill, 2011); PRBQ’s validity should be further examined using a multi-method assessment integrating performance-based measures such as the Balloon Analogue Risk Task (Lejuez et al., 2002) to objectively assess RSDB propensity and clinician-administered measures such as the CAPS with the ability to minimize misinterpretations (Foa & Tolin, 2000; Weathers et al., 2018). The cross-sectional nature of the data and lack of a clinician-administered RSDB measure prevented us from identifying a cut-off score sensitive to clinical change. Further, our findings need to be interpreted considering the demographic and psychopathology characteristics of both samples (e.g., RSDB base rates). Cross-validation of the PRBQ with diverse, high-risk, and clinical samples is needed. While the MTurk platform is a notable strength of our study, we need to consider cautionary caveats. Past work suggests that MTurk’s subject pool reflects the U.S. population in several demographic characteristics such as gender, age, marital status, full-time employment status, and ethnicity data (Buhrmester et al., 2011; Shapiro et al., 2013; U.S. Census Bureau, 2016b, 2016c); however, only some aspects of the current study sample demographics were comparable to the U.S. population: gender distribution (49.20% male, 50.80% female; U.S. Census Bureau, 2016a), mean age (37.70 years; U.S. Census Bureau, 2016a), number of individuals identifying as White (73.30%; U.S. Census Bureau, 2016a), and number of married individuals (46.40-49.80%; U.S. Census Bureau, 2016c). Relatedly, using validity checks and excluding individuals missing substantial amount of data improved MTurk data quality (Aust, Diedenhofen, Ullrich, & Musch, 2013; Buhrmester et al., 2011; Oppenheimer, Meyvis, & Davidenko, 2009) and the extent of our sample truncation (~52%) is comparable to other MTurk studies (57%; van Stolk-Cooke et al., 2018); however, such steps may have biased our sample, thereby limiting generalizability. Future research may benefit from using quality checks such as restricting participation to MTurk workers with high reputation (>95% approval ratings; Peer, Vosgerau, & Acquisti, 2014). Finally, several associations for convergent validity had low/negligible effect sizes; thus, replication with other RSDB measures is important while simultaneously examining the mediating/moderating influence of variables such as demographic characteristics.

Despite these limitations, our study provides preliminary and promising evidence of the validity of a novel, brief, and comprehensive measure of PTSD’s E2 Criterion for use by clinicians and researchers to assess the nature and frequency of posttrauma RSDBs. The PRBQ had a unitary factor structure; very good to excellent internal consistency; and good convergent, construct, and incremental validity. In addition, items tapping into functional impairment and relation to the onset of the trauma can yield valuable information indicating severity and course of RSDBs. Overall, the PRBQ can help refine clinical treatment targets and provide pointers for areas of additional assessment. Given that RSDBs occur outside the context of trauma, researchers may prefer flexibility in administering a RSDB measure to all eligible participants (including non-trauma exposed) to answer certain planned critical questions. Thus, future research may benefit from examining the psychometric properties/applicability of a modified PRBQ scale (non-trauma specific instructions and administering the only 14 items) in non-trauma exposed samples.