Research report

Commonalities in grief responding across bereavementand non-bereavement losses

Anthony Papa n, Nicole G. Lancaster, Julie KahlerDepartment of Psychology, University of Nevada, Reno, Reno, NV, United States

a r t i c l e i n f o

Article history:Received 25 February 2014Accepted 7 March 2014Available online 25 March 2014

Keywords:Prolonged grief disorderComplicated griefBereavementJob lossDivorce

a b s t r a c t

Background: Despite implications for theory and treatment, commonality in responding to non-bereavement and bereavement losses are not well explicated.Method: This study identified the factor structure of the three most common responses to bereavement,prolonged grief, posttraumatic stress, and major depression in a bereaved community sample (n¼151,59% female, 68% white) from the U.S. recruited from Amazon’s MTurk using a cross-sectional surveydesign, then cross-validated the structure in samples where people had lost other potentially self-defining roles; one’s employment (n¼157, 47% female, 69% white) and one’s marriage (n¼116, 62%female, 80% white).Results: Results indicated that symptoms of prolonged grief, posttraumatic stress, and major depressionwere distinct factors in the bereaved sample, the three-factor solution was a good fit for the job-loss anddivorce samples, and levels of grief in each sample appeared to be best predicted by time since loss andcentrality of the loss to one’s identity.Limitations: Limitations include potential sample bias due to convenience sampling, and the cross-sectional design did not allow examination of the stability of factors over time.Conclusions: These results suggest that grief is not a unique response to loss of loved one but instead maybe a common phenomenology across types of loss. This implies that facilitating meaningful engagementin self-defining activities that compensate for the disrupting loss might be efficacious in promoting griefresolution without the need for working through individuals’ emotional attachment to a specificindividual or processing one’s emotional responses to the loss.

& 2014 Elsevier B.V. All rights reserved.

Over the last decade in a growing body of the literature hasdemonstrated that reactions to bereavement include mood disrup-tions, posttraumatic stress responses, and/or a set responses uniqueto grief. In this paper, we examined if the pattern of responding tobereavement across these response clusters was similar in peoplewho had recently lost a job or got a divorce.

1. Grieving non-bereavement losses

Most grief theories fall into two broad overlapping areas. The firstset hypothesizes that grief is a result of loss of a biologically drivensocial bond with an attachment figure (Bowlby, 1980; Shear and Shair,2005). The second set of theories can be loosely described as cognitivestress theories relating grief severity to the degree that a loss violatesassumptive world views (Stroebe and Schut, 1999), with extensionsemphasizing the role of meaning making (Harvey and Miller, 1998;

Neimeyer et al., 2006) or identity continuity (Bonanno et al., 2001) indetermining grief severity. All these theories are consistent inhypothesizing that the loss of an important other has unique implica-tions for individuals’ ability to self-regulate and adaptation (see Papa etal., 2013a). As a result, none of these theories have looked at thepotential impact of other non-bereavement losses with the exceptionof Harvey and Miller (1998), who proposed that an individual mightexperience grief after any loss that alters an important self-aspect,such as losing one’s job or divorce (Carlson et al., 2000).

A large body of research supports the premise that non-bereavement losses can have significant impact on adjustment afterevents ranging from natural disasters (e.g., Hobfoll, 2002), chronicpain and illness (e.g., Palomino et al., 2007), disability (Roos andNeimeyer, 2007), to being diagnosed with a mental illness (e.g., Steinet al., 2005). The similarity in reactions to non-bereavement lossesand bereavement has long been described in the literature (e.g.,Parkes, 1972). However, the few studies that have examine reactionsto non-bereavement losses have tended to look at adjustment interms of either depression or trauma, and typically has not includedevaluation of prolonged grief (PG) symptoms integral to currentconceptualizations of reactions to bereavement, making it unclear if

Contents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/jad

Journal of Affective Disorders

http://dx.doi.org/10.1016/j.jad.2014.03.0180165-0327/& 2014 Elsevier B.V. All rights reserved.

n Corresponding author. Tel.: þ1 775 682 8666.E-mail address: apapa@unr.edu (A. Papa).

Journal of Affective Disorders 161 (2014) 136–143

these losses share the unique pattern of response observed afterlosses of loved ones. In fact, only two studies to date have linked non-bereavement loss to symptoms of modern conceptualizations ofpathological grieving.

In the first, Shear and colleagues (2011) used a brief screening toolto measure four key symptoms included in the diagnostic criteria forPG in survivors of Hurricane Katrina 5–16 months after the disaster.The screening tool assessed frequency of longing/yearning, feelings ofbitterness, perceptions of emptiness or meaninglessness, and difficultyaccepting the loss. Of those that participated in the study, 3.7%reported loss of a loved one and 50.8% reported other non-bereavement losses such as loss of property, close association withfamily and friends, well-being, sense of control, etc., as the mostsignificant loss associated with the hurricane. Of those who reportedloss of a loved one, 18.9% reported moderate to severe grief symptomlevels based on rationally derived cut-scores for the screening tool. Ofthose reporting non-bereavement losses, 6.7% reported moderate tosevere levels of grief, with the highest prevalence among those thatreported non-death related interpersonal losses due to the hurricane(10.6%).

A second study looked at the factor structure of PG, generalanxiety, and general depression symptoms after loss of a job (Papaand Maitoza, 2013). In this study, a principal components analysisfound that items measuring PG, depression, and anxiety eachloaded onto clearly distinguishable factors. The results indicatedthat the factor structure of grief, depression, and anxiety symp-toms were distinct in the job loss sample, allowing the inferencethat there is a common response pattern in bereavement and non-bereavement losses based on the findings of other studies showingsimilar factors structures in bereaved samples. However, a limita-tion of this approach is that, while Papa and Maitoza (2013) foundevidence of discriminant validity of PG in a job loss sample, theydid not directly assess commonality across different types of loss.

2. The current study

A number of factor analytic studies have shown that PG differsfrom depression, anxiety, and PTSD in bereaved adults from the U.S.and Canada (Barnes et al., 2012; Ogrodniczuk et al., 2003; Prigersonet al., 1996a), elders in the U.S. (Prigerson et al., 1995, 1996b), adults,adolescents, and children in the Netherlands (Boelen and van denBout, 2005; Boelen et al., 2003, 2010; Spuij et al., 2012), adolescentsin Belgium (Dillen et al., 2009), and adults in Croatia (Golden andDalgleish, 2010). However, despite indications that the most commonresponses to bereavement include not just PG symptoms butsymptoms of MDD and PTSD (Kristensen et al., 2012), almost allthe factor analytic studies completed to date have examined thedistinctiveness of general grief, anxiety and/or depression, or haveonly looked at differences between PG symptoms and depression orPTSD – giving an incomplete picture of how these symptoms clustersoverlap to describe the phenomenology of loss. Only two factoranalytic studies have examined the factor structure of PG, MDD, andPTSD symptoms, though in both cases goals of the studies did notlead the researchers to address cross correlations between items andthe potential that a model trimming approach might illuminate theunique pattern of MDD, PTSD, and PG symptoms over and above acommon distress response. In the first, Boelen and colleagues’ (2010)confirmatory factor analysis (CFA) found the best fit for a six factorsolution including factors for PG, MDD, and up to four factors forPTSD in two community samples from the Netherlands. Similarly,Golden and Dalgleish (2010) completed a principle componentsanalysis in a post-war Croatian sample which found that itemsmeasuring PG, MDD, and PTSD loaded onto separate factors.

The current study sought to extend the existing literature byidentifying the items that describe the latent factors underlying

PG, PTSD, and MDD symptom ratings that might be unique tobereavement, and then to examine whether this pattern ade-quately describes the reported experience of those who experi-enced non-bereavement loss. The first step of this process entailedexamining the distinctiveness of the symptom profiles of the threemost common responses to bereavement identified in the griefliterature in a community sample from the United States. Thisinvolved identifying the factor structure describing PG, MDD, andPTSD using EFA, in which correlated errors could not be specified,and then refining that model using CFA, in which the varianceattributable to correlated errors could be specified and used inmodel refinement. The second phase of this project involved cross-validating the final refined model from the bereaved sample intwo other samples where participants had lost other potentiallyself-defining roles; employee and spouse.

3. Method

3.1. Participants and procedures

An Internet-based survey was used to compare three differenttypes of loss (bereavement, job loss, and divorce) on different indicesincluding self-reported psychiatric symptoms. Four hundred twenty-four participants were recruited using Amazon’s Mechanical Turk(MTurk) service. To participate, MTurk workers logged into theiraccount, which automatically listed tasks for which they qualifybased on researchers’ specifications in order to earn money for theirAmazon account (see Mason and Suri, 2012 for detailed description).Inclusion criteria were (1) loss of a full time job of at least 6 monthsduration, death of a parent, child or spouse, or divorce all within thelast 12 months, (2) English proficiency, (3) 18þyears old, and (4) U.S.A. residence. The survey took about 30 min to complete(M¼28.53 min). The researchers paid Amazon $1.50 for each survey,which was then credited to the Amazon accounts of volunteersminus a 10% fee. This study was fully approved by the University ofNevada, Reno Institutional Review Board.

Studies comparing MTurk workers to U.S. population have foundthat MTurk workers to be slightly younger than the population of theU.S. (approximately 32–33 across studies vs. 37 in the U.S. in 2013 perhttps://www.cia.gov/library/publications/the-world-factbook/fields/2177. html). MTurk samples have also been found to be moreeducated, and more likely to under- or unemployed than thepopulation of the U.S. . The mean ages of participants in our studyranged from 33 to 35 across samples. About 68% of bereaved, 69%of job loss, and 80% of the divorced sample identified as “whiteonly,” compared to 79% from 2012 US Census data (http://quickfacts.census.gov/qfd/states/ 00000.html). Participants with bache-lor’s degree or above ranged from 41 to 45% compared to 31% inthe general population (http://www.census.gov/hhes/socdemo/education/data/cps/2012/ tables.html). MTurk workers reliablyreport symptoms of MDD and general anxiety comparable to thosefound by Kessler et al. (2005) in the NCS-R and as well as similarlevels of exposure to potentially traumatic events. However, sincethis study recruited individuals who had recently experienced anadverse life event, levels of probable PTSD (ranging from 30 to 38%across groups) and MDD (ranging from 22 to 25% across groups)were higher than found in the general population.

3.2. Measures

In addition to demographics, grief symptoms were assessed usingthe Prolonged Grief-13 scale (PG-13); a 13-item questionnaire thatassesses the proposed symptoms of Prolonged Grief Disorder pro-posed by Prigerson et al. (2009). For this study, questions werechanged to refer to the “job you lost” or to the person’s ex-spouse/

A. Papa et al. / Journal of Affective Disorders 161 (2014) 136–143 137

partner, or the relationship that was lost instead of “person you lost.”Cronbach’s alpha was high across all types of loss (αjob loss¼0.93,αdivorce¼0.92, αbereaved¼0.94). PTSD symptoms were measured usingthe PTSD Checklist-Specific (PCL-S; Blanchard et al., 1996), a 17-itemscale that evaluates the severity of DSM-IV PTSD symptoms using a5-point scale referenced to a criterion event (in this case the lossesfor which individuals were recruited; αjob loss¼0.97, αdivorce¼0.96,αbereaved¼0.96). Depression symptoms were assessed using thePatient Health Questionnaire-9 (PHQ-9; Kroenke et al., 2001; αjob

loss¼0.91, αdivorce¼0.92, αbereaved¼0.92). Finally, we measured thepersonal relevance of losses by adapting and combining items fromfour measures of role centrality (how much a particular social role oraspect of identity influences is central to ones sense of self). Thismeasure included a 5-item scale developed by Callero (1985), a 2-item scale developed by Norton et al. (2005), a 3-item scaledeveloped by Thoits (2012), and items from the MultidimensionalInventory of Black Identity (Sellers et al., 1997). Items were rated on a10-point scale (αjob loss¼0.88, αdivorce¼0.89, αbereavement¼0.92).

4. Results

4.1. Preliminary analyses

There were no significant differences in age, F(2, 427)¼0.91,p40.05, ethnicity, χ2(12)¼12.08, po0.05, or gender of respon-dents, χ2(1)¼0.05, p40.05, across samples (Table 1). Groups didnot differ in self-reported PTSD, F(2, 418)¼0.52, p40.05, ordepression symptoms, F(2, 419)¼0.18, p40.05, but did differ inreported PG symptoms, F(2, 421)¼9.22, po0.001. Post-hoc ana-lysis using Scheffé's test indicated the only significant differencewas between the job loss (Mjob loss¼24.89, SD¼8.64) and

bereavement groups (Mbereaved¼29.85, SD¼11.06; po0.001). Dif-ferences between the job loss and divorce groups (Mdivorce¼27.41,SD¼10.76; p¼0.13) and between the bereavement and divorcegroups (p¼0.15) were not statistically significant.

4.2. Exploratory factor analysis of symptom ratings in bereavedsample

The aims of the EFA were to (1) identify the factors underlyingPG-13, PHQ-9, and PCL-S symptom reports, (2) assess if the factorsidentified by the EFA replicate previous research which found thatgrief, depression, and posttraumatic stress symptoms constituteseparate clinical phenomenon in the context of bereavement in aU.S. sample, and (3) identify a simplified model for a subsequentCFA to assess if the identified factor structure found in a bereavedsample adequately describes the experience of those who recentlyexperienced divorce or job loss.

The EFA used a principal axis factoring approach with obliquerotation. In order to reduce multicollinearity, the item measuring“difficulty concentrating” was removed from the PHQ-9 due tohigh collinearity with a similar item on the PCL-S, as was the sleepdifficulties item from the PCL-S due to overlap with a similar itemon the PHQ-9. A parallel analysis was conducted prior to complet-ing the EFA in order to determine the number of factors to beretained for the EFA using the procedure described in O’Connor(2000) and the associated SPSS syntax. Results from the parallelanalysis indicated that three factors should be retained in thesubsequent analysis. The procedure for item trimming consisted ofretaining items if (1) the communality for an item was greaterthan 0.50, (2) the largest factor loading for an item was greaterthan 0.60, and (3) its second highest factor loading was smallerthan 0.30.

Table 1Sample characteristics.

Bereavement, N¼151 Job loss, N¼157 Divorce, N¼116

Age M¼34.84 M¼33.21 M¼33.59SD¼12.44 SD¼10.85 SD¼9.37Range¼18–71 Range¼18–65 Range¼19–61

% Female 58.9 46.5 62.1EthnicityWhite/Caucasian 68.10% 68.80% 80.17%Black/African American 7.98% 12.74% 8.62%Latino/Hispanic 4.29% 7.01% 5.17%Asian/Pacific Islander 4.91% 5.10% 5.89%Native American/Indian 2.45% 1.27% 0.86%Multiracial 4.91% 1.91% 2.59%Not identified 7.36% 3.18% —

Education (U.S. percentagesa)Some high school (8.58%) 0.62% 0.64% 1.72%High school diploma (30.01%) 13.20% 12.73% 12.07%Some college (19.46%) 28.30% 35.67% 25.00%Associate’s degree (18.44%) 16.98% 9.55% 16.40%4-year college degree 30.82% 33.12% 32.76%Master’s/professional degree (16.63%) 8.81% 5.73% 11.21%Doctoral degree (2.68%) 1.26% 2.55% 0.86%PG symptomsPG-13 total score (SD) 29.85 (11.06) 24.89 (8.64) 27.41 (10.76)Probable number that meet diagnostic criteria 26 (15.95%) 13 (8.28%) 2 (1.72%)GRIEF factor total score (SD) 12.65 (4.53) 9.31 (3.50) 9.33 (4.48)PTSD symptomsPCL-S total score (SD) 38.61 (16.98) 37.56 (17.35) 39.68 (16.88)Number over clinical cut-off of 50 42 41 32Probable number that meet diagnostic criteria 52 (31.9%) 47 (29.9%) 44 (37.9%)PTSD factor total score (SD) 11.32 (5.17) 11.01 (5.40) 11.83 (5.24)Depression symptomsPHQ total score (SD) 9.13 (7.08) 8.75 (6.72) 9.21 (7.09)Probable number that meet diagnostic criteria 39 (23.9%) 35 (22.3%) 29 (25%)MDD factor total score (SD) 4.52 (3.40) 4.28 (3.21) 4.64 (3.59)

a U.S. Census Bureau, Current Population Survey, 2012 Annual Social and Economic Supplement.

A. Papa et al. / Journal of Affective Disorders 161 (2014) 136–143138

In the final model, the average sampling adequacy scores fromthe anti-image correlation matrix were 0.91 (range 0.81–0.95),Kaiser–Meyer–Olkin measure of sampling adequacy¼0.92, Bartlett’stest of sphericity, χ2¼2030.39, po0.001, and an average commun-ality score of 0.70 (range 0.60–0.82; MacCallum et al., 1999) allindicated that the sample was appropriate for factor analysis. In thismodel, the first factor consisted of items from the PCL-S measuringPTSD symptoms and explained 54.20% of the variance (Table 2).Factor 2 consisted of items from the PG-13 explaining 9.56% of thevariance. Factor 3 consisted of MDD symptoms from the PHQ-9accounting for 5.87% of the variance. Only the three factors identifiedby the parallel analysis met the Kaiser criterion with eigenvaluesgreater than 1.00. Examination of the screen plot indicated a breakafter three factors; further confirming that three factors adequatelydescribed the data as indicated by the parallel analysis.

4.3. Confirmatory factor analysis in divorce and job loss samples

In order to examine whether the EFA-derived factor structurefrom the bereaved sample adequately described reported experi-ences in the other types of loss, we completed a CFA in the job lossand divorce samples based on the factors identified in the EFA.Goodness of fit indices examined included the chi-square test andthe chi-square degrees of freedom ratio (χ2/df), which is lesssensitive to sample size than the chi-square test. Generally, aχ2/df of less than two is desirable (Tabachnick and Fidell, 2007). Inaddition, standardized root mean squared residual (SRMSR) wasused, which compares the residual correlation matrices derivedfrom the raw data versus the proposed model, and root meansquare error of approximation (RMSEA), which compensates formodel complexity. Values less than 0.05 are considered anindicator of good fit and values of 0.08 are considered adequatefor both SRMSR and RMSEA (Schumacker and Lomax, 2004). Wealso report the Comparative Fit Index (CFI) and the parsimony-corrected Tucker–Lewis Index (TLI). Values above 0.95 indicategood fit for these indices (Hu and Bentler, 1999). Finally, thecoefficient of determination (R2) is reported.

To begin, we examined cross-classifications, residuals, andcorrelated error in the EFA-derived model in an unrotated solutionwithin a CFA framework in the bereaved sample (for review ofprocedures see Brown, 2006, pp. 193–202) using maximum like-lihood estimation using the SEM module in Stata 12.0. No outlierswere identified in the sample. Examination of residuals and

modification indices in a unrotated SEM framework for the EFAmodel indicated significant correlated error among some indicators,as well as unacceptably high cross classifications with specificitems. To compensate, we addressed the unacceptable cross-classifications including the association of the PHQ item related tosleep with the PTSD factor, as well as between the PCL items relatedto acting or feeling like stressful event was happening again andavoiding activities or situations with the depression factor. See Fig. 1for the final refined model for the bereaved displaying the standar-dized solutions (χ2(58)¼68.439, p¼0.164, χ2/df¼1.180, SRMR¼0.028, RMSEA¼0.035, CI90 [0.000,0.064], CFI¼993, TLI¼0.991,R2¼0.998; Table 3). All factor loading estimates were significant(R2s¼0.54–0.83; pso0.001), as were all covariance estimates(pso0.035). Correlations between latent factors representing dif-ferent symptom clusters were all significant, but suggest adequatediscriminant validity between factors (Brown, 2006; rPGD.MDD¼0.45,rPGD.PTSD¼0.54, rMDD.PTSD¼0.58, pso0.001).

Moving to the second step of this process, examination of thefactorial validity of the refined model in the job loss sampleindicated that the model fit well (χ2(58)¼7.650, p¼0.048, χ2/df¼1.339, SRMR¼0.039, RMSEA¼0.047, CI90 [0.009,0.072],CFI¼0.987, TLI¼0.982, R2¼0.997; Table 3). Factor loading esti-mates showed that the items were strongly related to latentfactors (R2s¼0.50–84, pso0.001). All covariance estimates werealso significant (pso0.046), as were correlations between latentfactors (rPGD.MDD¼0.31, rPGD.PTSD¼0.50, rMDD.PTSD¼0.52, pso0.001).

Estimates of the refined model’s fit in the divorced sampleranged from good to adequate (χ2(58)¼91.197, p¼0.004, χ2/df¼1.572, SRMR¼0.048, RMSEA¼0.070, CI90 [0.041,0.097],CFI¼0.969, TLI¼0.958, R2¼0.998; Table 3). Factor loading esti-mates were significant (pso0.001), as were the correlationsbetween latent factors (rPGD.MDD¼0.45, rPGD.PTSD¼0.61, rMDD.

PTSD¼0.60, pso0.001). Correlated errors between the two griefitems measuring longing/yearning and emotional pain andbetween PCL-S items related to psychological distress at remindersand avoiding or thinking or talking about the event were notsignificant (ps40.30), but all other correlated error terms weresignificant (pso0.013). The factor loading estimates indicatedindicators were strongly related to latent factors (R2s¼0.56–0.83) with the exception of the PCL-S item measuring avoidanceof talking or thinking about the event (R2s¼0.34).

The results from these analyses support the premise that there isa similarity in response to both bereavement and non-bereavement

Table 2EFA results examining grief, posttraumatic stress, and depression symptoms in the bereaved sample after item selection.

Factor

1 2 3

Intense longing or yearning for deceased �0.02 0.91 0.06Intense feelings of emotional pain, sorrow, or pangs of grief �0.11 0.92 �0.07Feeling stunned, shocked, numb or dazed by the loss 0.12 0.72 �0.09Difficulty accepting the loss 0.20 0.63 �0.04Little interest or pleasure 0.14 0.01 �0.73Feeling down, depressed, or hopeless 0.12 0.05 �0.75Trouble falling or staying asleep, or sleeping too much �0.15 0.08 �0.87Feeling tired or having little energy 0.10 �0.02 �0.77Poor appetite or overeating 0.06 0.03 �0.71Acting or feeling as if a stressful experience from the past was happening again 0.83 0.09 0.08Emotional distress due to reminders of a stressful experience 0.78 0.05 �0.03Having physical reactions to reminders 0.77 �0.02 �0.10Avoiding thought and/or feeling associated with a stressful experience 0.76 0.01 �0.04Avoiding activities or situations associated with a stressful experience 0.93 0.05 0.12Irritability or angry outbursts 0.65 �0.02 �0.23Feeling jumpy or easily startled 0.75 �0.06 �0.18

A. Papa et al. / Journal of Affective Disorders 161 (2014) 136–143 139

losses. The final model derived from the bereaved group fit the jobloss sample closely. However, the fit with the divorce group rangedfrom adequate to good. MacCallum and colleagues (1999) note thatfactor analyses with communalities greater than 0.60 or with amean communality of at least 0.70 produce strong estimates littleinfluenced by sample size. In our EFA, communalities ranged from0.66 to 0.82, with an average of 0.70. In terms of factors-to-variablesratio, estimates of minimums for good estimation are generallybetween 1:3 and 1:4 with 1:6 being very good (Costello andOsborne, 2005; Fabrigar et al., 1999). In the EFA, our level ofoverdetermination was (3:16), interpreting the results based onthe Cattell criterion. Moreover, the degree that the variables load onfactors also has bearing on sample size in factor analytic studies. PerGuadagnoli and Velicer (1988) factors with four or more variablesloading over 0.6 are likely to the stable regardless of the samplesize, which is a criterion met in our EFA.

Our analyses of factorial validity in the job loss and divorcesamples demonstrated strong intercorrelations among the vari-ables and the associated factors reflecting the high commonalitiesand strong loading seen in the EFA. Fit estimates suggest that theaccuracy of the estimates appears to be high and not negativelyaffected by sample sizes in the bereavement and job-loss samples.For the bereavement model, the confidence interval around theRMSEA estimate is tight and includes zero (RMSEA¼0.035, CI90[0.000,0.064]), and a PCLOSE value is 0.781. For the job losssample, the estimates are similarly accurate with the RMSEAconfidence interval lower than 0.08 at the upper end andapproaching zero on the lower end (RMSEA¼0.047, CI90[0.009,0.072]), and a PCLOSE value of 0.56. However, in the divorce

sample (our smallest sample size, N¼116), the picture is less rosy.The RMSEA estimate and the upper bound of the RMSEA CI is atthe level indicating mediocre fit (RMSEA¼0.070, CI90 [0.041,0.097]; MacCallum et al., 1996), and seen also in the lower PCLOSEvalue of 0.117 which approaches the cut off criteria for a closefitting model.

4.4. Post-hoc analysis of contributors to grief in the divorce sample

One potential reason why the CFA model fit less well in thedivorced samples could be related to a greater range of potentialreactions to divorce—but does this mean that divorce reactions aredifferent than job-loss or bereavement losses? Being married maybe more central to identity for some. People who initiated thedivorce, whose spouse committed adultery, or who faced financialtrouble as a result of the divorce might have experienced divorcedifferently than those who did not initiate the divorce, whosespouses did not commit adultery, and/or did not have financialdifficulties after the divorce. To examine this, we used a multipleregression model to examine the extent that the marriage was acentral part of the person’s identity, being the person to initiatethe divorce, partner’s infidelity, and having divorce-related finan-cial difficulty on the experience of grief indexed by the four itemsconstituting the grief factor. In this model we also included otherpotential contributors to the experience of loss after divorceincluding gender, age, whether participants were living with theirpartner before deciding to divorce, months separated, yearsmarried, and number of children.

Examination of residuals and multicollinearity indices indi-cated that the model met these assumptions for multiple regres-sion modeling. Six outliers were identified as having undueinfluence on results after examining leverage and discrepancyindicators and were excluded from the analysis. The omnibus testof the final model indicated that the predictors explained asignificant proportion of the variance in the grief index,R2¼0.440, F(8, 88)¼6.140, po0.001. Results indicated that thedegree that being married was central to individual’s identity wasthe strongest predictor of grief symptoms in our sample, β¼0.446,po0.001. In addition, longer marriages were associated withincreases in grief, β¼0.284, po0.05, while longer separations,β¼�0.199, po0.05, and increasing age, β¼�0.285, po0.05,

Fig. 1. Final model for CFA based on the bereaved group with standardized estimates.

Table 3Tests of measure invariance and population heterogeneity of PG, PTSD, anddepression symptoms in divorce and job-loss samples.

χ2 (df) df χ2/df SRMR RMSEA CFI TLI R2

Bereaved (150) 68.439 58 1.180 0.028 0.035 0.993 0.991 0.998Divorce (N¼116) 91.197nn 58 1.572 0.048 0.070 0.987 0.982 0.997Job loss (N¼156) 77.650n 58 1.339 0.039 0.047 0.969 0.958 0.998

n po0.05.nn po0.01.

A. Papa et al. / Journal of Affective Disorders 161 (2014) 136–143140

were associated with decreases in grief. Finally, the experience offinancial difficulties as a result of the divorce (coded 0¼no, 1¼yes)was associated with marginal increases in grief, β¼0.175, po0.10.

Repeating this analysis with the job loss and bereavementgroups, a similar pattern emerges with time since loss related todecreases in symptoms, and degree of role centrality associatedwith increases in symptoms. In the bereavement sample, gender,age, months since death, whether the death was expected,whether the deceased was ill for a month or more before dying,if the participant was a caretaker for the deceased, amount ofcontact prior to death, and role centrality was regressed on griefsymptoms, R2¼0.475, F(8, 119)¼13.433, po0.001. Results indi-cated that age, β¼�0.196, po0.01, months since the death,β¼�0.228, po0.01, and role centrality, β¼0.566, po0.001, weresignificant predictors of grief symptoms in this model. In the jobloss sample, a model including gender, age, primary wage earner,length employed at lost job, months unemployed, current house-hold income, and role centrality was assessed, R2¼0.340, F(7,128)¼8.901, po0.001. Significant predictors in his model weremonths unemployed, β¼�0.258, po0.001, current householdincome, β¼�0.174, po0.05, and role centrality, β¼0.485,po0.001. These results highlight the commonalities in factorscontributing to the experience of grief in these different types oflosses.

5. Discussion

The results of this study confirmed that the refined three-factorsolution found in the bereaved samples was a good fit when cross-validated in job-loss and divorce samples, despite the heterogeneityof response in the divorce group. This suggests that phenomenologyof grief is a common experience in job loss and divorce and notunique to the death of a loved one. The final model fit best in thejob-loss sample rather than the divorce sample, which surprisingbecause the divorce sample is more similar to bereavement inrepresenting the loss of a potentially important attachment rela-tionship. However, a number of factors appeared to qualify thislatter finding, with age, length of marriage, time separated beforethe divorce, and role centrality all predicting severity of griefresponding.

In addition, though not a primary goal, this study was one ofthe first factor analytic studies to confirm the discriminant validityof PG symptoms from PTSD and MDD symptoms in a U.S. sample.Because we used an EFA framework with a bereaved sample todetermine the model to be tested in the CFA portion of the studyrather than testing an a priori three-factor solution, it allowed usto develop and test a simplified version of the three-factor modelin which our final model contains only items that capture theunique variance of each construct in our sample. In many ways,items not included in each factor were as interesting as the itemsincluded. The resultant PG factor consisted of four items in ouranalysis: longing/yearning, feeling stunned/shocked/dazed by theevent, and trouble accepting what happened. Avoidance of remin-ders of the loss loaded on the PTSD factor was not included in thePG factor. Damage to self-concept (role confusion, meaningless-ness, foreshortened future, guilt, and worthlessness), social pro-blems (distrust of others and detachment/estrangement fromothers), and cognitive disruptions (problems with recall andconcentration) were not uniquely described by the PG, PTSD, orMDD factors, possibly reflecting common experiences acrossdiagnostic categories. The final solution for the MDD factorconsisted of depressed mood, loss of interest, loss of energy,disrupted appetite/weight gain or loss, and hypersomnia/insom-nia. Psychomotor disruption appeared to be captured better byitems from the PTSD hyper-arousal criterion than MDD. Also of

note is that suicidal ideation/suicidality did not load on anyspecific factor, which is unsurprising given known increases insuicidality in both traumatized (Krysinska and Lester, 2010) andbereaved samples (Latham and Prigerson, 2004). The final PTSDfactor consisted of acting or feeling as if the event was reoccurring,reactivity to reminders avoidance of reminders, and chronicarousal consisting of irritability/outbursts of anger and exagger-ated startle response. Re-experiencing items not included in thisfactor were distressing recollections and dreams, which are oftendescribed by people who have been bereaved as triggers forlonging/yearning and for feelings of emotional pain/sorrow. Thislast finding in particular highlights that these results may exhibit apattern unique to post-loss reactions. However, given the relativescarcity of studies looking at PG symptoms in contrast to PTSD andMDD symptoms, the ability to look across studies to see if this is apattern that is replicated across bereaved samples is a goal for thefuture.

5.1. Limitations

A number of caveats qualify these results. It is important tonote that our sample had higher levels of education than reportedby the U.S. Census Bureau (see Table 1). The biasing effect on ourresults as a result of this discrepancy on reported levels ofpsychopathology is difficult to ascertain. Meta-analysis has indi-cated that higher education levels have a strong association withdecreased MDD (Lorant et al., 2003) and somewhat weakerassociation with decreased PTSD prevalence (Brewin et al.,2000). This might be offset by the self-selected nature of thesample and a corresponding bias to over-report disruptions afterloss. This highlights the limitations of using of a conveniencesample in general and one recruited from Amazon’s MTurkparticipant pool in particular. Research indicates that samplesfrom MTurk tend to roughly conform to population norms interms of age, ethnicity, and severity of mental health symptoms aswell as mean responses on outcomes compared to samplesrecruited using traditionally social science research methods, butthese studies have also found higher education and unemploy-ment rates. Our sample was similar to the findings for other MTurksamples in many ways, with the exception of higher symptomsreporting as a result of recruiting participants who had recentlyexperienced adversity. While symptom levels in this sample werehigh compared to population norms and MTurk samples ingeneral, the self-selecting nature of convenience sampling makesit uncertain if those who did not volunteer might have been moreor less symptomatic.

In addition, this study examined the distinctiveness of thesymptoms of Prolonged Grief Disorder proposed by Prigerson et al.(2009) rather than other proposed nosologies (Complicated Grief,Shear et al., 2011; Persistent Complex Bereavement, APA, 2013).We did not examine DSM-5 diagnostic criteria for PersistentComplex Bereavement because the criteria had not been madepublic at the time of this data collection. In the case of Compli-cated Grief, there were fewer empirical studies assessing thedistinctiveness of the proposed symptom profiles than there werefor PG at the time of this data collection. The inclusion ofadditional criteria from either of these nosologies would poten-tially have changed the relationships found in this study.

Finally, the cross-sectional design of this study did not allowexamination of the stability of the factors in each of the groupsover time. The study recruited participants who experienced a losswithin the last 12 months. Looking at the job loss and bereavedsamples, there are indications that the stability of PG criteriadiffered by type of loss. In the bereaved sample, 37.4% (N¼55) hadexperienced their loss between 6 and 12 months. Of these, 25.5%met PG criteria. Of those whose loss was less than 6 months, 13%

A. Papa et al. / Journal of Affective Disorders 161 (2014) 136–143 141

met criteria for PG when disregarding the time criteria, χ2¼3.64,p¼0.06. In the job loss sample, 47.4% (N¼72) had experiencedtheir loss between 6 and 12 months. Of these, 9.5% met PG criteria.Of those whose loss was less than 6 months, 7.3% met criteria forPG when disregarding the time criteria, χ2¼0.23, p¼0.63.

5.2. Implications

Overall, the results suggesting that there is a commonality inthe responses across types of loss and in the predictors of severityin grief reactions have distinct implications for theories of grievingand development of treatments to promote adaptive grief resolu-tion. It suggests that in understanding grief, researchers mightlook beyond bereavement and consider grief as a response to lossin general as proposed by Harvey and Miller (1998) . Thissupposition challenges the assumption that social bonding/attach-ment plays a unique role in the genesis of grief reactions. It alsochallenges the derivative assumption that grief resolution requiresemotional processing (in which individuals review and workthrough their emotions related to the death and the deceased inorder to reduce their attachment to the deceased by redirectingone’s emotional investment from the person who died), assumedin many modern grief treatments (Papa et al., 2013b), despiteevidence suggesting emotional processing does not necessarilypromote adaptation to bereavement (Bonanno et al., 2005).Instead, this study suggests that the experience of grief may berelated to loss of self-defining roles like relationships with parents,being married, or being a person employed as a professor. Grief,from this perspective, would be the experience of the loss of acentral self-defining role and the process of accommodating tothat loss. This suggests that grief resolution may be a process ofidentity reconsolidation facilitated by meaningful engagement inself-defining activities that compensate for disrupting losses with-out the explicit need for emotional processing.

Role of funding sourceThis project was funded by start-up funding given to Anthony Papa by the

University of Nevada, Reno.

Conflict of interestNone of the authors has a conflict of interest to disclose.

AcknowledgementsWe thank Danielle Cox, who assisted with the proof-reading of the paper.

References

American Psychiatric Association, 2013. Diagnostic and Statistical Manual of MentalDisorders, 5th ed. American Psychiatric Association, Arlington, VA.

Barnes, J., Dickstein, B.D., Maguen, S., Neria, Y., Litz, B.T., 2012. The distinctiveness ofprolonged grief and posttraumatic stress disorder in adults bereaved by theattacks of September 11th. J. Affect. Disord. 136 (3), 366–369.

Blanchard, E.B., Jones-Alexander, J., Buckley, T.C., Forneris, C.A., 1996. Psychometricproperties of the PTSD checklist (PCL). Behav. Res. Ther. 34 (8), 669–673.

Boelen, P.A., van den Bout, J., 2005. Complicated grief, depression, and anxiety asdistinct postloss syndromes: a confirmatory factor analysis study. Am. J.Psychiatry 162 (11), 2175–2177.

Boelen, P.A., van de Schoot, R., van den Hout, M.A., de Keijser, J., van den Bout, J.,2010. Prolonged grief disorder, depression, and posttraumatic stress disorderare distinguishable syndromes. J. Affect. Disord. 125 (1–3), 374–378.

Boelen, P.A., van den Bout, J., de Keijser, J., 2003. Traumatic grief as a disorderdistinct from bereavement-related depression and anxiety: a replication studywith bereaved mental health care patients. Am. J. Psychiatry 160 (7),1339–1341.

Bonanno, G.A., Papa, A., Lalande, K., Zhang, N., Noll, J.G., 2005. Grief processing anddeliberate grief avoidance: a prospective comparison of bereaved spouses andparents in the United States and the People’s Republic of China. J. Consult ClinPsychol. 73 (1), 86.

Bonanno, G.A., Papa, A., O’Neill, K., 2001. Loss and human resilience. Appl. Prevent.Psychol. 10, 193–206.

Bowlby, J., 1980. Loss: Sadness and Depression. 3. Basic Books, New York, NY.Brewin, C.R., Andrews, B., Valentine, J.D., 2000. Meta-analysis of risk factors for

posttraumatic stress disorder in trauma-exposed adults. J. Consult. Clin.Psychol. 68 (5), 748–766.

Brown, T.A., 2006. Confirmatory Factor Analysis for Applied Research. Guilford, NewYork, NY.

Callero, P.L., 1985. Role identity salience. Soc. Psychol. Q 48 (3), 203–215.Carlson, H., Johnston, A., Liiceanu, A., Vintila, C., Harvey, J., 2000. Lessons in the

psychology of loss: accounts of middle-aged Romanian women. J. Loss Trauma5, 183–200.

Costello, A.B., Osborne, J.W., 2005. Best practices in exploratory factor analysis: fourrecommendations for getting the most from your analysis. Practical Assess-ment, Research & Evaluation 10 (7) 1-9.(Available from: http://pareonline.net/pdf/v10n7.pdf).

Dillen, L., Fontaine, J.R., Verhofstadt-Denève, L., 2009. Confirming the distinctive-ness of complicated grief from depression and anxiety among adolescents.Death Stud. 33 (5), 437–461.

Fabrigar, L.R., Wegener, D.T., MacCallum, R.C., Strahan, E.J., 1999. Evaluating the useof exploratory factor analysis in psychological research. Psychol. Methods 4 (3),272–299.

Golden, A.J., Dalgleish, T., 2010. Is prolonged grief distinct from bereavement-related posttraumatic stress? Psychiatry Res. 178 (2), 336–341.

Guadagnoli, E., Velicer, W.F., 1988. Relation of sample size to the stability ofcomponent patterns. Psychol. Bull. 103, 265–275.

Harvey, J.H., Miller, E.D., 1998. Toward a psychology of loss. Psychol Sci 9 (6),429–434.

Hobfoll, S.E., 2002. Social and psychological resources and adaptation. Rev. Gen.Psychol. 6, 307–324.

Hu, L., Bentler, P.M., 1999. Cutoff criteria for fit indexes in covariance structureanalysis: conventional criteria versus new alternatives. Struct. Equat. Model. 6(1), 1–55.

Kessler, R.C., Chiu, W.T., Demler, O., Merikangas, K.R., Walters, E.E., 2005. Pre-valence, severity, and comorbidity of 12-month DSM-IV disorders in thenational comorbidity survey replication. Arch. Gen. Psychiatry 62 (6), 617–627.

Kristensen, P., Weisæth, L., Heir, T., 2012. Bereavement and mental health aftersudden and violent losses: a review. Psychiatry 75 (1), 76–97.

Kroenke, K., Spitzer, R.L., Williams, J.B.W., 2001. The PHQ-9: validity of a briefdepression severity measure. J. Gen. Intern. Med. 16 (9), 606–613.

Krysinska, K., Lester, D., 2010. Post-traumatic stress disorder and suicide risk: asystematic review. Arch. Suicide Res. 14 (1), 1–23.

Latham, A.E., Prigerson, H.G., 2004. Suicidality and bereavement: complicated griefas psychiatric disorder presenting greatest risk for suicidality. Suicide LifeThreat. Behav. 34 (4), 50–62.

Lorant, V., Deliege, D., Eaton, W., Robert, A., Philippot, P., Ansseau, M., 2003.Socioeconomic inequalities in depression: a meta-analysis. Am. J. Epidemiol.157, 98–112.

MacCallum, R.C., Browne, M.W., Sugawara, H.M., 1996. Power analysis and deter-mination of sample size for covariance structure modeling. Psychol. Methods 1,130–149.

MacCallum, R.C., Widaman, K.F., Zhang, S., Hong, S., 1999. Sample size in factoranalysis. Psychol. Methods 4, 84–99.

Mason, W., Suri, S., 2012. Conducting behavioral research on Amazon’s mechanicalTurk. Behav. Res. Methods 44 (1), 1–23.

Neimeyer, R.A., Baldwin, S., Gillies, J., 2006. Continuing bonds and reconstructingmeaning: mitigating complications in bereavement. Death Stud. 30 (8),715–738.

Norton, T., Gupta, A., Stephens, M.A., Martire, L., Townsend, A., 2005. Stress,rewards, and change in the centrality of women’s family and work roles:Mastery as a mediator. Sex Roles 52, 325–335.

O’Connor, B.P., 2000. SPSS and SAS programs for determining the number ofcomponents using parallel analysis and Velicer’s MAP test. Behav. Res MethodsInstr. Comput. 32 (3), 396–402.

Ogrodniczuk, J.S., Piper, W.E., Joyce, A.S., Weideman, R., McCallum, M., Azim, H.F.,Rosie, J.S., 2003. Differentiating symptoms of complicated grief and depressionamong psychiatric outpatients. Can. J. Psychiatry 48 (2), 87–93.

Palomino, R.A., Nicassio, P.M., Greenberg, M.A., Medina, E.P., 2007. Helplessness andloss as mediators between pain and depressive symptoms in fibromyalgia. Pain129 (1 and 2), 185–194.

Papa, A., Maitoza, R., 2013. The role of loss in the experience of grief: the case of jobloss. J. Loss Trauma 18 (2), 152–169.

Papa, A., Rummel, C., Garrison-Diehn, C., Sewell, M.T., 2013a. Behavioral activationfor pathological grief. Death Stud. 37 (10), 913–936.

Papa, A., Sewell, M.T., Garrison-Diehn, C., Rummel, C., 2013b. A randomized opentrial assessing the feasibility of behavioral activation for pathological griefresponding. Behav. Ther.

Parkes, C.M., 1972. Components of the reaction to loss of a limb, spouse, or home.J. Psychosom. Res. 16 (5), 343–349.

Prigerson, H.G., Bierhals, A.J., Kasl, S.V., Reynolds, C.F., Shear, M.K., Newsom, J.T.,Jacobs, S., 1996a. Complicated grief as a disorder distinct from bereavement-related depression and anxiety: a replication study. Am. J. Psychiatry 153 (11),1484–1486.

Prigerson, H.G., Horowitz, M.J., Jacobs, S.C., Parkes, C.M., Aslan, M., Maciejewski, P.K.,2009. Prolonged grief disorder: psychometric validation of criteria proposed forDSM-V and ICD-11. PLoS Med. 6, e1000121.

A. Papa et al. / Journal of Affective Disorders 161 (2014) 136–143142

Prigerson, H.G., Shear, M.K., Newsom, J.T., Frank, E., Reynolds, C.F., Maciejewski, P.K.,Kupfer, D.J., 1996b. Anxiety among widowed elders: is it distinct fromdepression and grief? Anxiety 2, 1–12.

Roos, S., Neimeyer, R.A., 2007. Reauthoring the self: chronic sorrow and posttrau-matic stress following the onset of CID, Coping with Chronic Illness andDisability: Theoretical, Empirical, and Clinical Aspects. Springer, New York,pp. 89–106.

Schumacker, R.E., Lomax, R.G., 2004. A Beginner’s Guide to Structural EquationModeling, 2nd ed. Lawrence Erlbaum, Mahwah, NJ

Sellers, R.M., Rowley, S.A., Chavous, T.M., Shelton, J.N., Smith, M.A., 1997. Multi-dimensional inventory of black identity: a preliminary investigation of relia-bility and construct validity. J. Pers. Soc. Psychol. 73 (4), 805–815.

Shear, K., Shair, H., 2005. Attachment, loss, and complicated grief. Dev. Psychobiol.47 (3), 253–267.

Shear, M.K., Simon, N., Wall, M., Zisook, S., Neimeyer, R., Duan, N., Keshaviah, A.,2011. Complicated grief and related bereavement issues for DSM-5. Depress.Anxiety 28 (2), 103–117, http://dx.doi.org/10.1002/da.2078.

Spuij, M., Reitz, E., Prinzie, P., Stikkelbroek, Y., de Roos, C., Boelen, P.A., 2012.Distinctiveness of symptoms of prolonged grief, depression, and post-traumaticstress in bereaved children and adolescents. Eur. Child Adolesc. Psychiatry 21(12), 673–679.

Stein, C.H., Dworsky, D.O., Phillips, R.E., Hunt, M.G., 2005. Measuring personal lossamong adults coping with serious mental illness. Commun. Ment. Health J. 41(2), 129–139.

Stroebe, M., Schut, H., 1999. The dual process model of coping with bereavement:rationale and description. Death Stud. 23 (3), 197–224.

Tabachnick, B., Fidell, L., 2007. Using Multivariate Statistics, 5th ed. Allyn & Bacon,Boston, MA.

Thoits, P.A., 2012. Role-identity salience, purpose and meaning in life, and well-being among volunteers. Soc. Psychol. Q 75, 360–384.

A. Papa et al. / Journal of Affective Disorders 161 (2014) 136–143 143