In the last 15 years, tremendous progress has been made in improving our understanding of complicated or prolonged grief reactions (Prigerson et al. 1995b, 2009), ultimately resulting in a refined set of diagnostic criteria for Prolonged Grief Disorder (PGD) proposed for the 5th edition of the Diagnostic and Statistical Manual (DSM; Prigerson et al. 2008). PGD is characterized by a range of chronic and severe grief symptoms, including intense separation distress, intrusive thoughts about the lost relationship, a sense of meaninglessness, and functional impairment in day-to-day life (Prigerson et al. 2008). Notably, PGD symptoms have been shown uniquely to predict a range of negative outcomes (controlling for other psychiatric symptoms), including heart problems, high blood pressure, changes in eating and smoking habits, suicidal ideation, and global psychological adjustment (Bonanno et al. 2007; Latham and Prigerson 2004; Prigerson et al. 1997), highlighting the importance of accurate identification and assessment of these symptoms and their risk factors in clinical practice.

Despite the recent strides made in our understanding of PGD, bereavement researchers have pointed out that the distinction between traumatic aspects of loss (e.g., anger/bitterness, numbness/shock, shattered worldview) and separation-related aspects (e.g. yearning/pining, loneliness) as well as our understanding of the unique risk factors associated with these different manifestations of grief remain somewhat murky (Stroebe et al. 2001). Thus, the purpose of the present study is to examine cause of death (e.g., homicide, accident, natural causes) and relationship to the deceased (e.g., immediate family, friend) as factors that may differentially predict separation- and traumatic-related aspects of PGD.

A number of factor analytic studies have concluded that PGD symptoms are distinct from the symptoms of other psychiatric disorders (i.e., depression, generalized anxiety, posttraumatic stress; Boelen and van den Bout 2005; Boelen et al. 2003a; Chen et al. 1999; Prigerson et al. 1995a, 1996). In particular, one of the cardinal features of PGD is that it is characterized by intense separation distress—a cluster of symptoms that are not presently represented by any other psychiatric disorder in DSM-IV (Prigerson et al. 2008). Nevertheless, the symptoms of PGD do share some overlap with the symptoms of other disorders. Most significantly, researchers and clinicians have noted that trauma-like symptoms often follow particularly high-impact losses (Rubin et al. 2003; Stroebe et al. 2001), and some have found it useful to draw a distinction between separation and traumatic distress as two related components of PGD (e.g., Prigerson and Jacobs 2001; Neimeyer et al. 2006).

It has been suggested that different risk factors may vary in the degree to which they impact these two components of PGD, with relational and attachment-based aspects of the loss (e.g., kinship, closeness to the deceased) being more strongly related to separation distress and situational factors surrounding the death itself (e.g., violent vs. natural causes, whether or not it was an act of volition) being more strongly related to traumatic distress (Rynearson 1994; Stroebe et al. 2001). Support for such a hypothesis has been provided by Pynoos and Nader (1988) who studied the grief and posttraumatic stress reactions of school children following a sniper attack. Specifically, these researchers noticed a more pronounced posttraumatic stress reaction among children who had a greater degree of exposure to the violence, whereas greater separation anxiety and grief were observed for children who were closer to a victim of the shooting (Eth and Pynoos 1985). Considering the potentially strong presentation of traumatic distress among those who have experienced the violent death of a loved one, some have raised the possibility of treating “traumatic grief” as a distinct subtype of PGD (Stroebe et al. 2001).

It should be noted, however, that others have failed to find a clear link between loss by violent means and increased traumatic distress (Jacobs 1993; Prigerson et al. 2002). For example, Prigerson and her colleagues (2002) examined levels of separation and traumatic distress among a sample of 151 clients in a psychiatric outpatient clinic in Pakistan who had lost a first degree relative by different means of death. Somewhat counterintuitively, the cause of death (i.e., violence, drowning, accident, or health) was not found to predict traumatic distress symptoms. However, those bereaved by homicide showed higher levels of separation distress compared to those bereaved by accidents.

The present study seeks to clarify these divergent findings by replicating and expanding upon past research with a large community sample of bereaved young adults who lost a range of relationships by a variety of different causes of death. In particular, this study will examine the following hypotheses:

  1. 1.

    Symptoms of separation and traumatic distress can be distinguished from one another using factor analytic procedures.

  2. 2.

    Losses by violent means (i.e., homicide, suicide, accident) will result in prolonged grief reactions mostly characterized by traumatic distress. Furthermore, cause of death is predicted to be linked less strongly to levels of separation distress.

  3. 3.

    The loss of primary attachment figures (i.e., immediate family members) will result in prolonged grief reactions primarily characterized by separation distress (compared to those who lost a friend or extended family member). However, relationship to the deceased is not anticipated to strongly predict traumatic distress symptoms.

Method

Participants and Procedure

Participants in this study were drawn from a larger data set of bereaved college students who, following institutional review, were recruited in their introductory psychology courses across four waves of data collection at the University of Memphis, a large state university serving an ethnically and economically diverse student body (see Currier et al. 2008, for a complete description of this larger sample). Each participant was at least 18 years of age and also reported the death of a friend or loved one within the past 2 years. For each wave, eligible participants completed a single-session questionnaire that included measures of grief symptoms as well as questions concerning their demographic characteristics (e.g., age, ethnicity, gender) and the circumstances surrounding their loss (e.g., regarding the cause of death and their relationship to the deceased). If participants experienced multiple losses, they were asked respond to the questions with regard to the loss that “had the greatest impact” on them.

Some waves of the data collection used different measures and omitted others. Therefore, these analyses were restricted to only that subset of participants (n = 1022) who completed the Inventory of Complicated Grief-Revised (ICG-R; Prigerson and Jacobs 2001), which was used to create scores for separation distress and traumatic distress. Given the emphasis on relationship to the deceased and cause of death in this study, the analyses were further restricted to only those who clearly specified their relationship to the deceased and the cause of death of their loved one. Specifically, ambiguous responses (e.g., undetermined cause of death) or responses of “other” with no further elaboration were omitted. Thus, 947 bereaved individuals made up the sample for the present study.

The current sample ranged in age from 18 to 53 years with a mean of 21.0 years (Median = 19.0, SD = 4.9). Women made up 75.3% of the sample (n = 713), and 24.7% were men (n = 234). In addition, 56.5% of the participants were Caucasian (n = 535), 38.1% African American (n = 361), 1.6% Asian American (n = 15), and 3.8% were of another ethnicity (n = 36), reflecting the undergraduate distribution of ethnicities at the urban research institution. The most commonly reported types of loss were those due to natural, anticipated causes (e.g., cancer), which made up 47.8% of the losses in the sample (n = 453). Other losses in the sample were due to natural sudden causes (e.g., heart attack; 22.7%, n = 215), fatal accident (18.4%, n = 174), suicide (4.5%, n = 43), and homicide (6.5%, n = 62). Most participants had lost an extended family member (e.g., grandparent, uncle, cousin; 64.3%, n = 609). However, a sizable number of participants reported losing a friend (27.9%, n = 264) or an immediate family member (e.g., parent, sibling, child, spouse/partner; 7.8%, n = 78).

Measures of Separation and Traumatic Distress

Measures of separation and traumatic distress were created with items from the ICG-R. This measure is composed of 30 declarative statements to which responses are made on a five-point Likert-type scale describing the frequency of symptoms (e.g., from 1 = never to 5 = always). A Dutch version of the ICG-R has displayed high internal consistency (α = .94), concurrent validity (r = .71) with scores from the Texas Revised Inventory of Grief (Faschingbauer 1981), and good test-retest reliability (r = .92) over a period ranging from 9 to 28 days (Boelen et al. 2003b). The ICG-R has also been shown to predict a range of serious long-term health and mental health consequences of bereavement, justifying its interpretation as a measure of PGD symptomatology (Ott 2003; Prigerson et al. 1997; Prigerson and Jacobs 2001). In addition, a recent study suggests that PGD symptoms (as measured by the ICG-R) are distributed along a continuum, highlighting the potential utility of studying a range of PGD symptomatology (Holland et al. 2009b).

Although past research suggests that the ICG-R has a unidimensional structure (Boelen et al. 2003b), Prigerson and Jacobs (2001) have a drawn a distinction between symptoms of separation and traumatic distress and identified items on the ICG-R that tap into these two symptom clusters. Notably, past researchers have successfully used these groupings of items as separate measures to test hypotheses related to separation and traumatic distress (Hogan et al. 2004; Neimeyer et al. 2006; Prigerson et al. 2002).

Consistent with Prigerson and Jacobs’ (2001) suggested grouping of items, in the present study we summed items 2, 3, 5, 6, and 22 to create a measure of separation distress. Example separation distress items included, I feel myself longing and yearning for [the deceased] and I feel drawn to places and things associated with [the deceased]. Our measure of traumatic distress was created by summing items 4, 7, 8, 9, 11, 14, 17, 19, 21, 23, and 26; dividing by 11 (the number of items in the traumatic distress measure); and then multiplying by five (the number of items in the separation distress measure), in order to ensure that the two measures were on the same scale. The traumatic distress measure included items such as, I have lost my sense of security or safety since the death...and I feel like I have become numb since the death.... It should be noted that our measure of traumatic distress excluded item 13, I go out of my way to avoid reminders that [the deceased] is gone, as this study used a version of the ICG-R that did not include this item. In this study, our measures of separation distress and traumatic distress were internally consistent (α = .83 and α = .89, respectively) and were significantly correlated with one another (r = .75).

Plan of Analysis

Confirmatory Factor Analysis

As a preliminary step, we performed a confirmatory factor analysis to evaluate how well a 2-factor model of prolonged grief fit our data. In this analysis we tested a 1-factor solution and a 2-factor solution with the five separation distress ICG-R items loading on one factor and the 11 traumatic distress ICG-R items loading on the other. In evaluating these models, we relied upon a variety of fit indices, including the chi-square goodness-of-fit test, the comparative fit index (CFI; Bentler 1990), the standardized root mean square residual (SRMR), and the RMSEA (Browne and Cudeck 1993). The chi-square goodness-of-fit test assesses the discrepancy between the observed covariance matrix and the covariance matrix of the fitted model. With large samples, however, the null hypothesis of equivalence will be rejected for virtually any parsimonious model, and with a small sample model misfit may be undetected. Therefore, we relied primarily on the other fit indices. CFI values >.90 and SRMR values <.10 are generally regarded as favorable (Hu and Bentler 1999; Kline 2005). Likewise, RMSEA values ≤.05 are considered close approximate fit, values between .05 and .08 suggest reasonable fit, and values ≥.10 are indicative of poor model fit (Browne and Cudeck 1993). CFA was performed in Mplus (Version 4.1, Muthén and Muthén 2006). Because there was evidence of significant skew (Mardia’s multivariate skew = 67.16, p < .001) and kurtosis (Mardia’s multivariate kurtosis = 492.69, p < .001),Footnote 1 parameters were estimated with the maximum likelihood mean adjusted (MLM) procedure. MLM is robust even in the face of substantial non-normality.

Mean Comparisons

In order to test our hypotheses, we conducted two stages of analyses. First, we simply compared means on the separation and traumatic distress measures among participants who were mourning the loss of different types of relationships (i.e., extended family, friend, and immediate family) and losses due to different causes of death (i.e., natural, anticipated; natural, sudden; accident; suicide; and homicide). Specifically, using analysis of variance (ANOVA), we examined differences in separation and traumatic distress as a function of relationship to the deceased and cause of death, which provided information about which types of relationships and causes of death were associated with heightened separation and traumatic distress. Then, within each type of loss (e.g., those who lost an immediate family member or suicide survivors) we conducted a paired-samples t-test to see if there was a substantial discrepancy between individuals’ scores on the separation and traumatic distress measures, which provided information about whether or not a particular type of loss was primarily characterized by separation or traumatic distress.

Multilevel Analysis

In the second stage of the analysis, we tested our hypotheses in a single statistical model, which allowed us to account for potential confounds between relationship to the deceased and cause of death. For example, in this sample losses by violent causes of death were more likely to be friends; whereas losses by natural causes of death were most likely to be extended family members (e.g., grandparents). In this model, we also included demographic variables, such as age, gender, and ethnicity.

Because each participant contributed two data points (i.e., a score for separation distress and a score for traumatic distress) in this analysis, multilevel modeling was used to account for clustering in the data. Whereas standard linear regression and similar procedures only consider random variation at one level (e.g., most typically the “leftover” variability among participants), multilevel modeling allows for multiple sources of error. In this case, we considered two sources of variability: between participants (i.e., differences from one individual to another) and within participants (i.e., discrepancies between individuals’ separation and traumatic distress scores). Given this study’s focus on identifying factors that differentially predict separation and traumatic distress symptoms, we were primarily interested in arriving at a model that could account for a substantial portion of the variability within participants (i.e., that could account for the discrepancy between separation and traumatic distress scores).

We performed our analysis using SPSS mixed and followed the recommended guidelines by Singer (1998) and Peugh and Enders (2005) for analyzing nested cross-sectional data. Parameters were estimated using restricted maximum likelihood (REML). Prolonged grief scores (which consisted of one separation distress and one traumatic distress score for each individual) served as the dependent variable. Consistent with recommended guidelines for this type of analysis, we first analyzed an unconditional means model, which only included the intercept in the model. This “bare bones” model served as a point of comparison for a larger model that included the substantive predictors of interest (e.g., cause of death, relationship to the deceased). Specifically, as described by Bryk and Raudenbush (1992, p. 65), we were able to estimate the percentage of variance that was accounted for by substantive predictors (similar to R squared in standard linear regression) by examining the reduction of variance components (i.e., between and within individuals) in our larger model, compared to the unconditional growth model.

The larger model included six variables as main effects: 1) age (centered by subtracting the sample mean), 2) gender (−1 = male, 1 = female), 3) ethnicity (−1 = Caucasian, 1 = ethnic minority), 4) cause of death (dummy coded with natural, anticipated losses as the reference group), 5) relationship to the deceased (dummy coded with loss of extended family members as the reference group), and 6) PGD symptom cluster (−1 = separation distress; 1 = traumatic distress). As we were primarily interested in predicting discrepancies between separation and traumatic distress scores, we also examined interactions between the PGD symptom cluster variable and the other variables. These interaction terms allowed us to directly test whether or not a discrepancy between separation and traumatic distress scores in one group (e.g., among homicide survivors) was significantly larger or smaller than the discrepancy observed for another group (e.g., among those bereaved by natural, anticipated causes). Only statistically significant interactions were included in the final model.

Results

Confirmatory Factor Analysis

The CFA revealed that a 1-factor model did not provide a strong fit to the data (χ 2(104) = 1072.01, p < .001; CFI = .81; SRMR = .07; RMSEA = .10). Fit was significantly improved for the 2-factor model;Footnote 2 however, most fit indices were still not within the acceptable range (χ 2(103) = 966.42, p < .001; CFI = .83; SRMR = .07; RMSEA = .09). Inspection of the modification indices suggested that three correlated errors should be added between ICG-R items 5 and 6 (both of which tap into feelings/behaviors consistent with yearning), items 8 and 9 (both of which assess feelings of disbelief/shock), and items 21 and 23 (both of which inquire about lack of purpose/fulfillment regarding to the future). These correlated errors are likely due to significant overlap in wording and/or content beyond that measured by the two latent factors. When these errors were allowed to correlate the 2-factor model demonstrated reasonable fit (χ 2(100) = 606.75, p < .001; CFI = .90; SRMR = .06; RMSEA = .07). Thus, it appears that a 2-factor model is viable, although there may be some redundancy in the content/wording of the individual items.

Mean Comparisons

Cause of Death

Means for participants bereaved by different causes of death on the separation and traumatic distress measures can be seen in Table 1. An ANOVA test revealed that there were significant differences overall between individuals bereaved by different causes of death on the separation distress measure, F(4, 942) = 3.81, p = .004. However, Tukey’s posthoc test indicated that the only significant difference was between those bereaved by natural, anticipated death and those bereaved by accident, with deaths due to accident being associated with greater separation distress.

Table 1 Paired-samples t-tests comparing separation and traumatic distress scores
Full size table

With regard to traumatic distress, significant differences were found overall between participants bereaved by different causes of death, F(4, 942) = 27.04, p < .001. Tukey’s posthoc test revealed that those bereaved by violent causes of death (i.e., accident, suicide, and homicide) had significantly higher levels of traumatic distress compared to those bereaved by natural causes of death (anticipated and sudden). No significant differences were detected between those bereaved by different violent causes of death (e.g., homicide vs. suicide) or between those bereaved by natural, anticipated causes and those bereaved by natural, sudden causes.

Paired-samples t-tests are presented in Table 1 that compare relative levels of separation and traumatic distress within groups of individuals bereaved by different causes of death and within groups of individuals who lost different types of relationships. Overall, these analyses revealed that participants bereaved by natural, anticipated; natural, sudden; and accidental causes had higher levels of separation distress, compared to their levels of traumatic distress. However, those bereaved by suicide and homicide showed comparable levels of separation and traumatic distress. Values for Cohen’s (1977) d (also presented in Table 1) offer further support these conclusions, showing small to medium effect sizes (representing differences between separation and traumatic distress scores) for individuals bereaved by natural, anticipated; natural, sudden; and accidental causes and negligibly small effect sizes for those bereaved by suicide and homicide.

Relationship to the Deceased

Means for separation and traumatic distress are presented in Table 1 for losses of different types of relationships. An ANOVA test revealed significant differences in levels of separation distress between individuals grieving the loss of different relationships, F(2, 944) = 44.49, p < .001. Tukey’s post hoc test detected significant differences between all three groups, with individuals who lost an immediate family member having the highest levels of separation distress, followed by loss of friends and loss of extended family members, respectively.

Significant differences were also found on the traumatic distress measure for participants who lost different types of relationships, F(2, 944) = 44.32, p < .001. Again, Tukey’s posthoc test identified significant differences between all three groups, with individuals who lost an immediate family member having the highest levels of traumatic distress, followed by loss of friends and loss of extended family members, respectively.

Paired-samples t-tests (presented in Table 1) indicated that levels of separation distress were significantly higher than levels of traumatic distress for participants who lost an immediate family member, friend, or extended family member. This conclusion was also supported by Cohen’s (1977) d values (also presented in Table 1), which showed small to medium effects across all three of these relationship categories.

Multilevel Analysis

We first tested an unconditional means model that only included the intercept in the model. This analysis revealed that the intercept (or grand mean in this case) for prolonged grief scores overall was significantly greater than zero (Estimate = 9.12, p < .001). In addition, there was statistically significant variability within participants (Variance = 4.90, p < .001) as well as between participants (Variance = 9.95, p < .001), indicating that there is likely variability in grief scores that may be explained by substantive predictors. The intraclass correlation was .67, indicating that roughly 67% of the total variability occurred between participants.

We next tested a larger model with substantive predictors. Results for this model are presented in Table 2. This analysis revealed greater prolonged grief symptoms (considering scores for both the separation and traumatic distress measures) for: (1) women, Estimate = 0.27, p = .03; (2) those bereaved by natural, sudden causes, Estimate = 0.59, p = .03; accident, Estimate = 1.51, p < .001; suicide, Estimate = 1.56, p = .005; and homicide, Estimate = 1.97, p < .001 (compared to those bereaved by natural, anticipated causes); and (3) those who lost an immediate family member, Estimate = 3.73, p < .001 (compared to individuals who lost an extended family member).

Table 2 Multilevel model predicting prolonged grief symptoms
Full size table

This analysis also indicated that scores on the separation distress measure generally tended to be higher than scores on the traumatic distress measure, Estimate = −0.89, p < .001. However, the interaction terms revealed that this discrepancy between separation and traumatic distress scores was significantly less pronounced for those who were bereaved by accident, Estimate = 0.58, p < .001; suicide, Estimate = 0.76, p = .001; or homicide, Estimate = 0.91, p < .001 (compared to individuals bereaved by natural, anticipated causes). In contrast, the loss of an immediate family member was associated with a more pronounced difference between separation and traumatic distress scores, Estimate = −0.73, p < .001 (compared to those who lost an extended family member). Unexpectedly, we also found a significant interaction between race/ethnicity and the PGD Cluster variable. In particular, we found that the overall trend for separation distress scores to be higher than traumatic distress scores was significantly less pronounced for ethnic minority individuals.

These variables as a whole accounted for 29.64% of the variance within participants and 8.16% of the variance between participants, above and beyond the unconditional means model (i.e., the “bare bones” model with no substantive predictors presented earlier).

Discussion

The results of this study generally provided support for our hypotheses. In particular, the uniqueness of separation and traumatic distress symptoms was generally supported in the confirmatory factor analyses. More substantively, those who lost a primary attachment figure (i.e., an immediate family member) exhibited prolonged grief symptomatology that was primarily characterized by separation distress (rather than traumatic distress), more so than for individuals who lost a friend or an extended family member. In addition, those who lost a loved one to violent causes (i.e., accident, suicide, homicide) tended to have more similar levels of separation and traumatic distress, compared to individuals who lost someone to natural, anticipated or natural, sudden causes. Taken together, these findings provide some support for the notion that symptoms of separation distress primarily stem from the disruption of a strong attachment; whereas, symptoms of traumatic distress may be influenced more by the traumatic circumstances surrounding the death itself (Rynearson 1994; Stroebe et al. 2001).

However, a couple of qualifications to this overall conclusion should be noted. First, as a whole, participants were more likely to have higher levels of separation distress compared to their levels of traumatic distress. Even among those who lost a loved one to violent means, levels of separation and traumatic distress were comparable. Thus, traumatic distress certainly did not dominate the grief experience of any of the groups of bereaved individuals examined in this study. This pattern of findings supports the idea that separation distress is the core symptom of PGD and highlights the uniqueness of this new diagnostic label, given that symptoms of separation distress are not presently represented by any diagnostic category in DSM-IV (Prigerson et al. 2008).

Secondly, it should be noted that, even though participants who experienced a loss by violent means had more similar levels of traumatic and separation distress (compared to those who experienced a loss by natural causes), relationship to the deceased was found to significantly predict traumatic distress symptoms. In fact, the group with the highest absolute levels of traumatic distress was individuals who lost an immediate family member (as shown in Table 1)—a group bereaved by a variety of different causes. This finding suggests that traumatic distress symptoms are influenced by a variety of factors, including aspects of the relationship with the deceased. Such an explanation would support claims that the loss of a close attachment can disrupt one’s sense of being protected and produce a shattered sense of security (Green 2000; Rubin et al. 2003), similar to the shattered assumptions and cherished beliefs observed among individuals who have experienced a traumatic event (Janoff-Bulman 1992).

A somewhat surprising finding of this study was that ethnic minority participants (who were mostly African American) tended to show a more balanced presentation of separation and traumatic distress symptoms after a loss compared to Caucasians, who tended to have particularly high separation distress scores and particularly low traumatic distress scores. This finding may provide an important clarification to past research that has found higher PGD rates and ICG-R scores for African Americans compared to Caucasians (Goldsmith et al. 2008; Laurie and Neimeyer 2008). In particular, the present study would suggest that African Americans (and perhaps other racial/ethnic minority individuals) may, on average, exhibit higher levels of traumatic distress but perhaps show somewhat lower levels of separation distress. This finding fits with past studies that have shown higher rates of posttraumatic stress disorder among racial/ethnic minority individuals following a traumatic life event (Frueh et al. 1998; Kulka et al. 1990; Perilla et al. 2002). Although the rationale behind these racial/ethnic differences remains unclear, greater severity of exposure to past/present stress and increased vulnerability (e.g., socioeconomic strain, racism, acculturative stress, previous losses or traumas) have been suggested as possible explanations (Perilla et al. 2002).

Limitations and Implications

Several limitations to the present study should be noted. First, although this sample was diverse in terms of race/ethnicity, types of relationships lost, and causes of death, it was primarily made up of young adults, which limits the generalizability of the findings. However, past research has not revealed a clear association between age and severity of grief reactions for comparable types of losses (Ball 1977; Ringdal et al. 2001; Sanders 1981), indicating that age in itself may not impact the grieving process as much as other factors (e.g., the objective circumstances of the loss, one’s subjective interpretation of it). In addition, young adults are most likely to lose a peer to motor vehicle accidents and are also at increased risk for other types of violent losses (Miniño et al. 2002). As a result, young adulthood may represent a developmental stage when certain types of losses are most common. Nevertheless, future research would do well to replicate this study with children/adolescents, older adults, or mixed age samples that have an equal representation of age groups.

In addition, even though the primary independent variables of interest in this study (i.e., cause of death and relationship to the deceased) would presumably remain static across time (strengthening the claim that these variables precede the dependent variable in time), the cross-sectional design employed in this study prevents us from making any definitive causal or temporal claims. For example, we cannot rule out the possibility that the heightened traumatic distress symptoms found among homicide survivors might be due to long-standing posttraumatic symptomatology resulting from years of living in a community prone to violent crime. Finally, although we were able to account for a substantial portion of the within-participant variability (i.e., the discrepancy between participants’ separation and traumatic grief scores), there was still a good deal of unexplained variability in our final model. This remaining variability suggests that there are likely other unmeasured factors that can help explain why some individuals primarily show elevations in separation distress after a loss and others show elevations in both separation and traumatic distress.

Notwithstanding these limitations, the results of the present study have relevance for future research and for the assessment and treatment of bereaved individuals. From a methodological standpoint, these findings highlight the potential utility of considering separation and traumatic distress symptoms separately, as the present study along with past investigations (e.g., Neimeyer et al. 2006) suggest that different factors may be responsible for triggering these two types of grief reactions. The present study also indicates that a 2-factor model of PGD with separation and traumatic distress loading on separate factors may provide a somewhat better fit than a unidimensional model. Future studies would do well to examine the factor structure of these ICG-R items further in different samples.

In addition, these findings suggest that those who have experienced a loss by violent means may be unique in that these individuals are most likely to present with a combination of separation and traumatic distress symptoms at more comparable levels than for losses by natural causes (and at roughly equivalent levels in the case of loss by suicide or homicide). If these results were consistently replicated in other samples, it might support the inclusion of a PDG subtype of “traumatic grief” (Stroebe et al. 2001). At the very least, it would suggest that at-risk groups, like those who have experienced a loss by suicide or homicide, should be carefully screened for traumatic symptomatology.

Importantly, the identification of trauma-like symptoms following the loss of a loved one could influence the course of treatment. For example, some clinicians have suggested that for clients with a strong comorbid presentation of traumatic and grief symptoms, therapy proceeds best when management of the trauma and trauma-related anxiety (e.g., through relaxation exercises and guided imagery) precedes more grief-focused interventions (e.g., imaginal conversations with the deceased, revisiting the painful details of the loss; Lindy et al. 1983; Pynoos and Nader 1988; Rynearson 1994).

Alternatively, others distinguish between the challenges to meaning-making posed by the effort to process the “event story” of the death on the one hand versus the “back story” of the relationship to the deceased on the other hand, arguing that different narrative strategies might be used in therapy to promote each of these aspects of loss integration (Neimeyer and Sands 2011). Accordingly, clinical interviews or formal grief assessments that discriminate between traumatic and relational disruption in bereavement could prove helpful to practitioners as well as researchers in more precisely targeting relevant symptomatology (Rubin et al. 2009). Although some preliminary empirical evidence suggests that distinguishing between different types grief-related symptoms may appropriately influence the choice of intervention (Holland et al. 2009a), future research that investigates the distinction between separation and traumatic distress and the clinical implications of such a distinction seems warranted.