The Dissociative Subtype of PTSD Scale: Initial Evaluation in a National Sample of Trauma-Exposed Veterans

Procedure

Data were collected by the survey research firm GfK Knowledge Networks, Inc., using an online research panel (GfK KnowledgePanel), which is designed to be demographically representative of the U.S. population. The panel includes over 80,000 households and was developed using address-based sampling methods; this approach improves on random-digit dialing sampling methods because it reaches households in which there is no home phone. The sample included households without Internet access, which were provided with Internet access and hardware. The sample is representative of approximately 98% of the U.S. population. After enrollment in the panel, potential participants are recruited to specific web-based studies via e-mail and are e-mailed a reminder if they have not responded to the survey in 3 days. GfK Knowledge Networks provides incentives for study completion, including points that can be redeemed for prizes, cash rewards, and raffles. This study resampled a subset (n = 2,715) of the 3,157 veterans who participated in a previous survey (i.e., the National Health and Resilience in Veterans Study; Pietrzak & Cook, 2013) and who endorsed trauma exposure on the Trauma History Screen (Carlson et al., 2011) as part of that original survey. A number of recent studies have characterized the veterans who took part in the National Health and Resilience in Veterans Study (Klingensmith, Tsai, Mota, Southwick, & Pietrzak, 2014; Tsai et al., 2015; Wisco et al., 2014). The sample is representative of the U.S. military veteran population. Of this group of trauma-exposed veterans, 1,126 were randomly selected and invited to take part in this study; 860 completed the survey (76.4% response rate). Poststratification weights were created to match veterans in the broader survey panel, which was matched to the October, 2010 U.S. population Census data; weights were applied in the data analyses where indicated below. The census data aligns well with veteran population estimates (Wisco et al., 2015). The survey assessed an array of psychological symptoms, only a subset of which are the focus of this study. The entire survey took, on average, 36.5 minutes (SD = 17.7) to complete and participants were awarded 50,000 points (approximately $50) for the time and effort they devoted to completing the survey. The survey was available online for approximately 2 weeks. Measure order was randomized with the exception that the National Stressful Events Survey (NSES; to assess trauma exposure and PTSD, see below) immediately preceded the DSPS given that the latter referred to the participant’s self-identified “worst” traumatic experience as determined by the NSES. Where indicated (see below), item order within a measure was also randomized. The local institutional review board reviewed and approved this study.

Participants

Of the 860 veterans who completed this study, the vast majority was male (91.5%) and reported their race and ethnicity as White, non-Hispanic (83.8%). The race and ethnicity of the remainder of the sample was as follows: 5.3% reported Hispanic ethnicity, 5.2% Black, 3.5% mixed race, and 2.1% identified as “other” race. Veterans ranged in age from 22 to 96 years (M = 63.0; SD = 12.3) and the majority was married/cohabitating (80.0%). Participants were drawn from across the United States: 39.7% of respondents were located in the South, 26.2% in the West, 23.4% in the Midwest, and 10.8% in the Northeast. Of those participants who reported one or more deployments (34.3%; n = 295), 65.6% reported involvement in the Vietnam War, 10.5% in the Korean War, 9.5% in the Persian Gulf War, 4.6% in the Iraq/Afghanistan, 2.6% in World War II, and 5.6% in another war/conflict.

Participants reported on their exposure to 16 different types of traumatic experiences on the NSES (Kilpatrick, Resnick, Baber, Guille, & Gros, 2011) and then self-identified their “worst” traumatic experience from this list. They were instructed to respond to the NSES PTSD assessment and to two psychogenic amnesia items on the DSPS in reference to this event. Index traumatic events included: sudden and unexpected death of a close friend or family member due to heart attack, stroke, or cancer (17.1%); combat exposure (13.8%); seeing dead bodies or body parts (7.6%); sudden death of a friend or loved one due to accident or suicide (7.0%); and sexual or physical assault (5.6%). Other index traumas were reported by fewer than 5% of the sample and are not detailed here. Participants reported exposure to a range of traumatic experiences, which occurred prior to their military service (26.9%; range: 0-13, M = 2.44, SD = 3.47), during their military service (44.7%; range: 0-4, M = 0.59, SD = 0.77), and/or after their military service (33.5%; range: 0-13, M = 1.12, SD = 2.06). On average, participants reported exposure to 2.44 different traumatic events across these time frames (range: 0-30; SD = 3.47).

Of the 860 participants, we omitted 163 from these analyses leaving a total sample of 697: 142 were omitted because they completed the survey either so quickly (<15 minutes; n = 6) or so slowly (>2 hours, n = 136), that it raised doubt about the validity of their assessments, and 23 were eliminated because they achieved a T score of 90 or greater (Goodwin, Sellbom, & Arbisi, 2013) on an index of symptom overreporting (the revised Infrequency Psychopathology [Fp-r] validity scale on the Minnesota Multiphasic Personality Inventory–2 Restructured Form [MMPI-2-RF]; Tellegen & Ben-Porath, 2008), which was administered to address concerns about the lack of experimenter control inherent in Internet-based methods. These items were administered in random order along with a subset of other MMPI-2 items (for a total of 51 MMPI-2 items) that are not the focus of this investigation. Two participants were omitted from analyses because they achieved an Fp-r T score of 90 or greater and took more than 2 hours to complete the survey, leaving a final sample size of 697 for these analyses. Individuals excluded from analyses had significantly higher mean scores on lifetime DSM-5 total NSES PTSD severity (M = 26.44, SD = 12.62) and on total lifetime dissociation, as measured by the DSPS (M = 1.28, SD = 1.42) compared with those who were included in analyses, PTSD: M = 23.45, SD = 7.59; t(855) = 3.9, p < .001, d = .32; dissociation: M = 0.96, SD = 0.90; t(858) = 3.6, p < .001, d = .27; no other significant group differences emerged across demographic variables including sex, race, or age.

Measures

Dissociative Subtype of PTSD Scale

As described above, the DSPS is a new 15-item self-report measure of dissociative symptoms and the focus of this investigation. Item content was generated by the first author in consultation with the last author with the aim of covering multiple types of dissociative symptomatology (Spiegel & Cardeña, 1991). The measure was developed to be administered after the identification of the participant’s “worst” traumatic experience, as some items (14-15) reference the participant’s self-identified worst trauma. This was accomplished in this study using a routing structure which inserted the participant’s self-reported worst traumatic experience as identified on the NSES into the prompt for Items 14 and 15 (see Table 1). The structure of the DSPS is similar to the NSES in that the measure assesses both lifetime and current (past-month) symptoms. It also assesses the frequency and intensity of symptoms that occurred in the past month, as symptoms might fluctuate differentially on these dimensions such that, for example, an individual might experience mild derealization on a daily basis (i.e., without losing connection to the present) or, in contrast, might experience derealization on an infrequent basis, but lose all connection to the present moment for an extended period of time.

OPEN IN VIEWER

Table 1.

Dissociative Subtype of PTSD Scale Item Content and Lifetime Item Endorsement Prevalence.

Item	Full sample		By possible lifetime PTSD diagnosis
	Raw %, (n = 697)	Weighted %, (n = 674)	Possible PTSD, % (n = 91)	No possible PTSD %, (n = 602)	Effect size (Φ) for χ2
1. Have there ever been times when you felt disconnected from your body, as if your body were not your own?	9.9	9.5	34.1	6.3	.31
2. Have you ever felt “checked out,” that is, as if you were not really present and aware of what was going on around you?	14.3	13.6	48.4	9.3	.38
3. Have there ever been times when you felt like you were outside of your own body, as if you could look at yourself from the outside?	10.2	8.9	26.4	7.8	.21
4. Have you ever “lost time”—that is, been unable to account for large portions of your day or had trouble accounting for what you did for portions of your day?	12.2	11.6	36.3	8.6	.28
5. Have there ever been times when you looked in the mirror and did not recognize yourself physically?	4.2	3.29	15.4	2.5	.22
6. Have there ever been times when you were in a familiar place, yet it seemed strange and unfamiliar to you?	17.2	14.3	46.2	13.0	.30
7. Have there ever been times when your body did not feel real?	5.5	4.1	22.0	3.0	.28
8. Have there ever been times when the world around you (other people, objects, places) did not seem real?	8.9	8.4	36.3	4.8	.37
9. Have there ever been times when your body felt very strange and unfamiliar to you, as if it were not your own body?	4.4	5.3	17.6	2.5	.25
10. Have there ever been times when you felt lost, disoriented, or confused in a location that you know well?	15.9	13.8	50.5	10.9	.37
11. Have there ever been times (other than when you were very tired, sleepy, or on medications or drugs that made you drowsy) when you felt as if you were in a daze or a fog?	10.2	9.5	26.4	7.8	.21
12. Have there ever been times when you felt like you were watching the world around you as an outsider, as if it were a movie, but the world did not seem real?	9.0	7.7	25.3	6.7	.22
13. Have you ever had trouble remembering how you got somewhere (i.e., finding yourself at work, at home, at a store, or elsewhere without remembering how you traveled there)?	17.4	14.8	35.6	14.8	.18
14. Have you ever had trouble remembering important details about ________ (fill in with participant’s “worst” traumatic event from prior assessment)?	14.3	12.7	40.7	10.6	.29
15. Have you ever thought that you should be able to remember more about _______ (fill in with participant’s “worst” traumatic event) than you do?	16.8	12.5	40.7	13.5	.25

Note. PTSD = posttraumatic stress disorder. Prevalence of each item by possible PTSD diagnostic status based on raw (unweighted) n of 693 with sufficient data to calculate possible PTSD diagnosis.

Verbatim item content is listed in Table 1 (subsequent tables and the text abbreviate the item content for brevity). Each item began with a “stem” prompt, which asked if the respondent had ever experienced the symptom (yes/no). If the initial response was positive, a follow-up item ascertained if the individual had experienced the symptom in the past month (yes/no). If the response to this item was positive, two additional subitems were then presented: the first assessed the frequency of the symptom in the past month on a 4-point scale (response options: once or twice, once or twice a week, 3 or 4 times a week, and daily or almost every day in the past month and the second assessed the intensity of the symptom in the past month on a 5-point scale (response options: not very strong, somewhat strong, moderately strong, very strong, and extremely strong). Lifetime items did not include these frequency and intensity scales in order to limit the time required to complete the measure and reduce overall participant burden. For one item assessing reductions in awareness (i.e., feeling as if in a daze or fog), the item prompt included instructions that the respondent should not rate the item if the symptom only occurred in the context of substance or medication use that induced drowsiness or in other instances of extreme fatigue. Given that we focused on lifetime PTSD symptoms (see above), we examined only the dichotomous lifetime DSPS items. Alpha coefficients for this measure are provided below. The DSPS is available from the first author.

Data Analyses

First, we calculated the raw and weighted prevalence of possible current and lifetime PTSD per the NSES self-report data. This scoring algorithm followed the DSM-5 definition of the disorder and required the presence of at least one PTSD Criterion B symptom, one Criterion C symptom, two Criterion D symptoms, and two Criterion E symptoms (at a symptom severity rating of “moderate” or higher for current symptoms); consistent with prior work with the NSES (Miller et al., 2013), the algorithm also required that the participant endorse significant distress or impairment as a result of the PTSD symptoms (i.e., DSM-5 PTSD Criterion F) and that nontrauma specific symptoms (e.g., PTSD Criteria D and E) began or became worse following trauma exposure. We then estimated the raw and weighted frequency of endorsement of each of the dichotomous lifetime symptoms on the DSPS (using survey procedures in SAS 9.3). Second, we conducted an EFA of raw scale items to determine the common factors underlying the measure. We chose the best class solution based on the scree plot, the pattern of item loadings, and the χ² difference test, which tests if a nested, more parsimonious, model with fewer free parameters is associated with degraded fit compared with a parent model with more free parameters (e.g., a two- vs. three-factor EFA model, respectively). Third, we tested how scores on the scales identified in the EFA correlated with our measure of absorption. Fourth, we conducted a latent profile analysis (LPA) using mean scores on the items contributing to each PTSD symptom cluster as defined in the DSM-5 (i.e., intrusions, avoidance, negative cognitions and mood, and arousal and reactivity) and on the items contributing to the factors identified in the EFA. We did this to determine if the DSPS could identify a discrete subset of individuals with marked dissociative symptoms, consistent with prior work on this subtype of PTSD. We used analysis of variance (in SPSS) to compare the pattern of mean scores on the indicators submitted to the LPA as a function of latent class (with a Bonferroni correction of .007 to guard against family-wise error) and compared the latent classes on a variety of indices including PTSD diagnosis, trauma history, and demographics using χ² and tests of mean difference, as appropriate.

Factor analyses were evaluated using the standard fit indices (e.g., root mean square error of approximation, confirmatory fit index, Tucker–Lewis index, standardized root mean square residual) using the criteria outlined by Hu and Bentler (1999). Latent class models were compared with the Bayesian information criterion (smaller values are preferred; Schwartz, 1978), the bootstrap likelihood ratio test, and the Lo–Mendell–Rubin–adjusted likelihood ratio test (Lo, Mendell, & Rubin, 2001). With these latter two statistics, a significant p value indicates that the class model under evaluation is preferred over a model with one fewer classes. Classes were also compared on entropy values (an index of class discrimination) and on their size and interpretability. Classes containing fewer than 5% of the sample were rejected based on concerns about the representativeness of such classes and the likelihood that this was an indication that the model had extracted too many classes (McLachlan & Peel, 2000; Nylund, Asparouhov, & Muthén, 2007). All latent variable analyses were conducted with the Mplus 7.11 statistical modeling software (Muthén & Muthén, 1998-2012). The mean and variance adjusted weighted least squares estimator was used in the factor analysis, given the use of dichotomous indicators, and the robust maximum likelihood (maximum likelihood ratio) estimator was used for the LPA, as this is standard for this type of analysis. All multivariate analyses incorporated missing data directly under the standard procedures for mean and variance adjusted weighted least squares and direct maximum likelihood estimation. The maximum amount of item-level missing data was 1.9%. Univariate analyses by latent class assignment were conducted in SPSS based on the sample with complete data (n = 688).

Prevalence Estimates

Using the DSM-5 scoring algorithm, the prevalence of possible lifetime PTSD per self-report assessment in the sample was 13.1% (weighted estimate: 14.4%) and the prevalence of possible current (past-month) PTSD was 2.0% (weighted estimate: 1.8%). The lifetime dissociation items were endorsed by 4.2% to 17.4% of participants (see Table 1). Dissociation items were more frequently endorsed by those with possible lifetime PTSD as compared with those without; all χ² statistics were significant at the p < .001 level and exceeded a Bonferroni correction for family-wise error of p < .003 (see Table 1, for estimates of effect size, Φ).

Factor Analyses and Associations of the DSPS Factor Analytically Derived Subscales

We investigated the fit of exploratory factor models with solutions ranging from one to four factors. The scree plot suggested three or four factors, and the χ² difference tests suggested that each successive model provided better fit than a model with one less factor (see Table 2). Evaluation of the pattern of factor loadings, however, clearly suggested that the three-factor model was more interpretable than the four-factor one: the four-factor solution included a factor that had only one item (Item 3: out of body) with its strongest loading on that factor and this item loaded significantly with other derealization and depersonalization items in the three-factor solution. The factors that emerged in the three-factor solution also closely approximated the hypothesized structure of the DSPS. Specifically, Factor 1 indexed derealization and depersonalization, Factor 2 indexed a more general loss of awareness of surroundings, and Factor 3 indexed psychogenic amnesia (see Table 3). In this solution, there was one instance in which an item intended to measure derealization (Item 12: the world seems like a movie) evidenced its strongest association with the loss of awareness factor (Factor 2; β = .51, p < .05); however, this item also showed a statistically significant loading of a similar magnitude on Factor 1 (β = .40, p < .05); therefore, this item was used to index the derealization/depersonalization subscale for all subsequent analyses. In addition, Item 8 (the world does not seem real) showed equivalent and significant loadings on Factor 1 (β = .49, p < .05) and Factor 2 (β = .47, p < .05). Given that it was intended to be an indicator of derealization, this item was retained on Factor 1 in subsequent analyses. Internal reliability (Cronbach’s alpha coefficient) for each factor analytically derived lifetime subscale on the DSPS was as follows: α = .79 for the seven-item derealization/depersonalization subscale, α = .79 for the six-item loss of awareness subscale, and α = .74 for the two-item psychogenic amnesia subscale. Alpha for the 15 lifetime items together was .85. Mean summary scores on each lifetime subscale were as follows: 0.52 (SD = 1.21), for the derealization/depersonalization subscale; 0.87 (SD = 1.46), for the loss of awareness subscale; and 0.31 (SD = .65), for the psychogenic amnesia subscale. The mean summary score across all lifetime items was 1.70 (SD = 2.67, range = 0-15).

OPEN IN VIEWER

Table 2.

Fit of Exploratory Factor-Analytic Models.

Model	χ2 (df)	RMSEA	SRMR	CFI	TLI	Model comparison	Δχ2 (Δdf)
One-factor	463.38*** (90)	.08	.12	.92	.91
Two-factor	197.08*** (76)	.05	.08	.97	.97	1 vs. 2	207.44*** (14)
Three-factor	133.83*** (63)	.04	.06	.99	.98	2 vs. 3	57.73*** (13)
Four-factor	87.93** (51)	.03	.05	.99	.98	3 vs. 4	42.92*** (12)

Note. RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual; CFI = confirmatory fit index; TLI = Tucker–Lewis index; df = degrees of freedom. Analyses are based on n = 697.

**

p < .01. ***p < .001.

OPEN IN VIEWER

Table 3.

Standardized Factor Loadings From Three-Factor Exploratory Factor Analysis.

Item	Factor 1	Factor 2	Factor 3
1. Disconnected from body	.932*	.02	−.094
2. Checked out	.239	.641*	−.042
3. Outside body	.888*	.005	−.164
4. Lost time	.115	.693*	.075
5. Not recognize self	.431*	.308	.004
6. Strange or unfamiliar place	.071	.734*	.167*
7. Body not feel real	.859*	.031	.008
8. World not seem real	.485*	.468*	.021
9. Body is strange or unfamiliar	.980*	−.145	.116
10. Lost or disoriented in place you know well	−.047	.918*	.006
11. Daze or fog	−.119	.739*	.09
12. World is a movie	.395*	.510*	−.081
13. Not remember how got somewhere	−.013	.782*	−.038
14. Trouble remembering trauma details	.006	.031	.993*
15. Should remember more about the trauma	.001	.008	.824*

Note. Correlations between factors were as follows: Factor 1 with Factor 2, r = .73; Factor 1 with Factor 3, r = .41; Factor 2 with Factor 3, r = .44. Item content is abbreviated in this table and full item wording is shown in Table 1. The placement of items on subscales used in subsequent analyses is denoted in bold. Analyses are based on n = 697.

*

p < .05.

The pattern of correlations between the DSPS subscales and the MPQ-BF Absorption scale suggested discrimination between the subscales and this normal-range trait with the subscales evidencing small- to medium-sized associations with Absorption (per the Cohen, 1988 definitions wherein r ≥ .10, .30, and .50 are used as benchmarks for small, medium, and large effects, respectively). Specifically, Absorption correlated with the Derealization/Depersonalization subscale at r = .38 (p < .001), with the Loss of Awareness subscale at r = .35 (p < .001), and with the Psychogenic Amnesia subscale at r = .17 (p < .001). In other words, the maximum amount of shared variance between MPQ Absorption and any DSPS subscale was 14.4%. Associations between these three DSPS scales and total lifetime PTSD severity were r = .44, .46, and .42 (all p < .001), respectively. ¹

Latent Profile Analyses

We next calculated mean scores on the items contributing to the four DSM-5 PTSD symptom clusters and the three subscales of the DSPS, as determined through the EFA, and used these seven mean scores as indicators in a LPA. We compared the fit of solutions with 2 to 5 latent classes. The fit of each model is shown in Table 4. Models with increasing class sizes showed improved fit (lower Bayesian information criterion values and significant bootstrap likelihood ratio test p values with relatively equivalent and high entropy values indexing good classification), but the five-class model was rejected because the best log likelihood value was not replicated and because the model yielded two classes composed of fewer than 5% of the sample. The four-class model was also rejected as it included a small class defined by moderate PTSD symptom severity and relatively greater dissociative symptom severity, but it was not evident that this class differed meaningfully from the high symptom class and this solution was also less parsimonious than the three-class one. Given this, the three-class solution was selected as the best fitting model. Class 1 in this solution comprised 8.3% (weighted estimate: 7.4%) of the participants and showed high mean scores on all the PTSD symptom clusters and the three dissociation subscales; Class 2 comprised 18.5% (weighted estimate: 18.5%) of the participants and showed high scores on the PTSD symptom clusters, low scores on the derealization/depersonalization subscale, a moderate elevation on the loss of awareness subscale, and scores on psychogenic amnesia that were equivalent to those in Class 1; finally, Class 3 comprised 73.3% (weighted estimate: 74.0%) of the sample and evidenced low scores on all indicators submitted to the analysis. Based on this pattern of results, Class 1 was identified as the high PTSD/high dissociation class (i.e., the dissociative subtype), Class 2 as the high PTSD group (without dissociation), and Class 3 as the asymptomatic group. We could not conduct the LPA among only those with possible lifetime PTSD due to the limited sample size. However, based on the analysis in the larger sample of trauma-exposed veterans, 39.6% of those with possible lifetime PTSD were assigned to the dissociative subtype of PTSD.^2,3,4

OPEN IN VIEWER

Table 4.

Fit of Latent Profile Analyses.

Model	LL	BIC	LMRA p	BLRT p	Entropy
Two-class	41.99	−677.95	<.001	<.001	.94
Three-class	668.04	−1139.67	.07	<.001	.95
Four-class	839.44	−1430.10	.16	<.001	.95
Five-class	1006.99	−1712.82	.15	<.001	.96

Note. LL = log likelihood value; BIC = Bayesian information criterion; LMRA p = p value associated with Lo–Mendell–Rubin–adjusted likelihood ratio test; BLRT p = p value associated with the bootstrap likelihood ratio test. Analyses are based on n = 697.

Figure 1 shows the means scores on the indicators in the analysis by latent class. Analyses of variance evaluating overall mean score differences on the items submitted to the LPA as a function of latent class were all statistically significant at the p < .001 level, which surpassed the Bonferroni p value threshold of .007 for the omnibus F test. We conducted Tukey’s post hoc pairwise comparisons and found that all pairwise comparisons yielded statistically significant mean differences (at the p < .05 level), with the exception of the difference in mean avoidance symptoms between the dissociative class and the PTSD class (p = .21) and the difference in mean psychogenic amnesia symptoms between these same two classes (p = .83). Table 5 shows the effect sizes for each of the pairwise comparisons; the dissociative class differed from the PTSD class (and the asymptomatic class) primarily on the basis of mean scores on the derealization/depersonalization subscale, although effect sizes were also large (per Cohen, 1988) for the difference in means across these classes on the loss of awareness subscale. The notion that the dissociative subtype was most heavily influenced by scores on the derealization/depersonalization subscale was also supported by evaluation of the pattern of correlations between the posterior probability scores from the latent class analysis (a dimensional index of the likelihood of membership in each class) and the item mean scores on the DSPS subscales. Specifically, the posterior probability score for Class 1 (the dissociative subtype class) evidenced its strongest association with mean scores on the derealization/depersonalization subscale (r = .84, p < .001), relative to the mean scores on the loss of awareness subscale (r = .59, p < .001) and the psychogenic amnesia subscale (r = .20, p < .001); its strongest association with any PTSD subscale was with the arousal symptoms (r = .44, p < .001). In contrast, Class 2 (the PTSD latent class) was not highly correlated with any of the DSPS subscales (strongest r = .36, p < .001, with the psychogenic amnesia subscale), and evidenced its strongest association with the PTSD avoidance symptom cluster (r = .62, p < .001). The asymptomatic class was negatively correlated with all the DSPS and PTSD subscales (full correlation matrix available from the first author).

OPEN IN VIEWER

Figure 1.

Mean severity on PTSD and dissociation by latent class.

OPEN IN VIEWER

Table 5.

Effect Size Estimates for Mean Differences in PTSD and Dissociation Subscales as a Function of Latent Class Assignment.

Subscale (item means)	Effect size (d) for pairwise comparisons
	Dissociative vs. PTSD	Dissociative vs. asymptomatic	PTSD vs. asymptomatic
Intrusions	0.41	2.85	2.44
Avoidance	0.29	2.17	2.44
Negative Alterations in Cognition/Mood	0.39	2.54	2.15
Arousal	0.37	2.75	2.38
Derealization/Depersonalization	4.94	5.22	0.28
Loss of Awareness	2.09	2.71	0.62
Psychogenic Amnesia	0.09	0.98	1.07

Note. PTSD = posttraumatic stress disorder. Multivariate data analyses to derive classes were based on n = 697 using full information likelihood estimation to handle missing data, but evaluation of means as a function of latent class were conducted in SPSS using only complete data where total n = 688 such that n for dissociative class = 57, n for high PTSD class = 125, and n for asymptomatic class = 506.

We also assessed the extent to which individual PTSD items were differentially endorsed among individuals assigned to the high PTSD versus high PTSD and dissociation classes so we conducted χ² analyses for each of the 20 PTSD NSES items, and given this number of statistical tests, we used a Bonferroni correction to adjust the threshold for statistical significance to p < .0025. A number of these comparisons achieved p values that were less than .05, the most significant of which was for DSM-5 Criterion B3 (flashbacks). This item was endorsed by 71.9% of those in the dissociative subtype compared with 49.6% of those in the PTSD group, χ²(1, n = 182) = 7.95, p = .005; however, neither this nor any of the other comparisons were significant after correction for multiple testing.

We next tested for demographic and trauma-history differences between the dissociative versus PTSD classes, and found that the dissociative class was less likely to self-report as Caucasian, 68.4% versus 81.6%, respectively, χ²(1, n = 182) = 3.90, p = .048, Φ = .15, and on average, was somewhat older than the PTSD class, M_age for dissociative class: 61.33 years, SD = 10.36 years versus M_age for PTSD class: 56.90 years, SD = 13.42 years, t for unequal variances (137.98) = 2.43, p = .016. There were no significant differences between these two groups as a function of sex, self-reported history of childhood trauma or sexual trauma (at any age), or the number of different types of trauma the participant was exposed to during premilitary, military, or postmilitary service time frames. The dissociative class was more likely to meet criteria for possible lifetime PTSD per self-report assessment (63.2%) relative to the high PTSD class (44.0%); χ²(1, n = 182) = 5.75, p = .017, Φ = .18.

Limitations

Results of this study should be interpreted in light of several limitations. First, we did not analyze the current (past-month) items because of the low prevalence of possible current DSM-5 PTSD and infrequent endorsement of current dissociative symptoms, which would consequently limit generalizability of results. Future work testing the measure in clinical samples, where the prevalence of such symptoms is likely to be greater, is required to comprehensively evaluate the current (frequency and intensity) and lifetime scales and to determine their clinical utility for the reference population.

Second, we did not administer a structured diagnostic interview of dissociation to use as a criterion variable for the DSPS, nor was our assessment of PTSD based on a diagnostic interview. This is important because the web-based self-report methodology employed here is not as methodologically rigorous as that of the gold-standard assessment approach (i.e., structured interviews). Given the inclusion of the subtype in the newly released DSM-5 and the need for measures to reliably assess it, we prioritized the speed and efficiency associated with a self-report web-based methodology (and the large sample size this can achieve) over clinical interviewing.

Third, our use of a web-based methodology afforded less control (i.e., internal validity) over the administration of the measures relative to an in-person approach. Nevertheless, we made efforts to mitigate this by conservatively eliminating participants who showed evidence of an invalid style of responding. A related point is that we did not administer the full MMPI-2-RF due to time constraints and instead included select scales with item order randomized; it is unclear how doing so might have affected the ability of Fp-r to accurately identify invalid response profiles.

Fourth, we did not examine the test–retest reliability of the measure, nor did we conduct a comprehensive examination of the construct validity of the DSPS. It is possible that symptoms or problems other than dissociation may have influenced responses to the measure. For example, cognitive errors associated with advancing age or traumatic brain injury (e.g., memory loss) or substance misuse could conceivably result in symptoms similar to those included on the DSPS; these areas are important for future research. Along these lines, we have already revised the measure to expand instructions to rule out the role of substances and extreme fatigue for all items on the measure.

Fifth, the measure does not assess whether dissociative symptoms began or got worse during or following trauma exposure; thus, it is possible that some symptoms predated trauma exposure. The DSM-5 subtype criteria, however, do not specify that dissociative symptoms must be temporally linked to trauma exposure; this represents an additional area for further study. Sixth, the results of latent variable analyses are specific to the variables submitted to the analysis, and it is possible that findings would differ if we had included additional indicators of psychiatric disturbance (e.g., comorbid symptoms) in the models. That said, our aim was to test existing models of the association between PTSD and dissociation, and the evaluation of the structure of dissociation with other psychiatric comorbidity is outside the scope of this study.

Finally, the sample was composed primarily of older male veterans, and while representative of the broader veteran population, this limits the generalizability of the results to this group and precludes us from examining possible demographic moderators (e.g., sex) of DSPS item endorsement and of the results more generally; cross-validation of the DSPS in both clinical and community samples, that include a broader range of ages and greater representation of women, is necessary to examine the replicability of these results and to further test the psychometric performance and clinical utility of the measure.