Predisplacement and Postdisplacement Factors Associated With Mental Health of Refugees and Internally Displaced Persons:  A Meta-analysis FREE

The global refugee crisis has reached dire proportions. The world’s population of people forced to abandon their homes and communities because of war, political violence, or related threats includes not only those displaced outside their country of residence but also internally displaced persons, asylum seekers, stateless persons, and recently returned refugees. Although estimates of the total numbers of these persons vary, data from several international agencies indicate that these groups comprise a population that exceeded 42 million at the end of 2004.^1- 3 Severe and lasting psychological aftereffects have been extensively documented among them.^4- 6Additionally, the United Nations High Commissioner for Refugees alone returned approximately 15 million refugees to their countries of origin as recently as 1994 through 2003.⁷ Including all former refugees, the number of people who have been directly affected by forced displacement is estimated to exceed 1% of the world’s population.⁸ Analyses presented in this article specify relationships between subcategories of displaced persons and mental health outcomes. Because of important commonalities, the various subcategories have herein been subsumed under the general term refugee for simplicity.

Research on refugee mental health has often proceeded from a life events model^9- 10 in which psychopathology is understood as a posttraumatic reaction to the acute stressor of war.¹¹ Research in this tradition has established high rates of psychological disturbance among persons exposed to traumas through war and disaster.^12- 14 Theorists have recently argued that this model fails to capture important dimensions of the refugee experience,^15- 16 which is characterized by multiple events occurring in multiple contexts that persist over time. In addition, refugees are vulnerable to multiple dimensions of psychopathology beyond those that are narrowly posttraumatic.^16- 18

Unlike survivors of most discrete traumatic events, refugees experience diverse stressors that accumulate over the preflight, flight, exile, and resettlement/repatriation periods.¹⁹ Despite the historical focus on the acute stressors of war, the enduring contextual postmigration stress that refugees face—including marginalization, socioeconomic disadvantage, acculturation difficulties, loss of social support, and “cultural bereavement”²⁰—must be recognized. Such persisting forms of adversity reflect the broader sociopolitical context of the refugee experience.^15,21

Research on refugee mental health is fraught with practical obstacles. Populations are often physically, linguistically, and culturally inaccessible to researchers,²² and humanitarian aid usually has higher priority than scientific investigation. Studies are often exploratory and methodologically compromised, and the specificity of local circumstances makes it difficult to draw generalized conclusions.²³ As a result, research synthesis is needed to establish the magnitude of the mental health consequences of forced displacement and the variables that moderate these consequences. We have previously reported a meta-analytic investigation of refugee mental health,²⁴ but it was restricted to the former Yugoslavian Republic. We therefore conducted a meta-analysis that comprehensively examined studies of populations displaced in conflicts distributed throughout the world and across 5 decades. The meta-analysis tested several potential moderators of refugee mental health, including contextual variables subsequent to the acute stress of war.

Six postdisplacement condition moderators were investigated. Type of accommodation was examined in view of evidence that long-term institutional housing promotes dependency and demoralization.^25- 26 Postdisplacement economic opportunity was studied as a factor expected to be associated with better refugee outcomes.²⁷We predicted that greater cultural access (ie, the extent to which individuals are free to practice or engage in cultural, social, and religious customs and have access to the cultural institutions necessary for such practice), internal displacement (ie, being displaced within country of origin vs external displacement), and being repatriated would be associated with better outcomes because acculturative stress and cultural dislocation should have negative effects on mental health.^20,24,28 We also predicted that the status of the conflict would be another contextual condition affecting refugees’ mental health, with ongoing (vs resolved) conflicts being associated with worse outcomes.

In addition to these contextual variables, several refugee characteristics were investigated as potential moderators. The association between age and severity of posttraumatic response has been controversial,^13,29 with some evidence indicating worse outcomes for older people,^30- 33 some indicating greater vulnerability in children,^17,34- 37 and some mixed or null.^38- 40 Refugee sex was also examined,^17,41 with some evidence suggesting worse mental health outcomes for female refugees^38,42 and some indicating complex and culturally variable associations.²³ Predisplacement urban vs rural residence, region of origin (ie, Africa, Asia, Europe, Latin America, or Middle East), educational level, and predisplacement socioeconomic status were also investigated.

We also inquired about the moderating role of methodological characteristics of studies that might serve as confounders. We examined whether methodologically stronger studies⁴¹ and those whose control groups had greater exposure to adversity estimated less negative relative mental health outcomes for refugees. Finally, we assessed whether the time elapsed since displacement was associated with smaller effects, consistent with the decay of stress responses.^43- 46

A search of 2 databases was undertaken at the end of 2002 for all published English-language research reports that (1) investigated a refugee group (including all categories within UNHCR’s population of concern¹) and at least 1 nonrefugee comparison group and (2) reported 1 or more quantitative estimates of the relationship between forced displacement and indices of mental health (eg, symptom questionnaires, interview-based psychiatric diagnoses, maladaptive behavior checklists, neuropsychological tests, personality inventories, and self-esteem and well-being scales).

Searches were conducted using PsycINFO, a broad psychological database, and PILOTS, a comprehensive trauma-specific database maintained by the US Department of Veterans Affairs National Center for PTSD.⁴⁷ PILOTS was queried using a Boolean operator (refugee* OR displaced) and PsycINFO with narrower criteria: key word = {[(refugee*) OR (displaced person*) OR (displaced people)] AND [(control*) OR (compar*) OR (“nonrefugees”) OR (“nonrefugee”) OR (nonrefugee*) OR (“non-displaced”) OR (nondisplaced)]}. No restrictions were placed on the time period covered by the search, so all indexed literature from 1871-2002 for PILOTS and from 1840-2002 for PsycInfo was included.

The search of PILOTS produced 1010 citations, and PsycINFO yielded 273 additional nonoverlapping records. All abstracts were read and those that unambiguously failed to meet inclusion criteria were eliminated. Full texts were obtained for the 141 remaining articles. Manual searches through reference lists of identified articles were conducted and prominent authors were contacted informally but no additional published references were obtained. After eliminating all redundant research reports, 56 reports^48- 103 (published from 1959 through 2002; median year of publication, 1996) were identified as meeting inclusion criteria (Table 1 and Table 2). Two of these reports provided complementary information about the same comparison and 3 reported multiple independent comparisons, bringing the total meta-analytic sample to 59 independent comparisons (studies), detailed in Table 1 and Table 2. Of these, 20 used multiple comparison groups, yielding 85 stochastically dependent comparisons from which subsets of statistically independent comparisons were selected for analysis. The studies contained 67 294 participants (22 221 refugees and 45 073 nonrefugees), with a range of 34 to 29 468 (median, 164) participants per study. Together, the studies yielded 190 effect sizes, using primarily self- and parent-report questionnaires (83.2%), including standard^104- 106 and ad hoc instruments and structured (3.7%) and unstructured (12.6%) interviews.

Coding of the studies (M.P.) was conducted in accordance with established procedures,^107- 108 encompassing report and study characteristics (eg, methods), participant characteristics (eg, attributes and circumstances of each independent sample), and statistical outcome information. Information pertaining to all study groups and mental health measures was coded. A coding manual was developed a priori and revised iteratively.¹⁰⁸ Coded variables of interest are presented in Table 3. After extensive spot checking, a randomly selected sample of 6 reports was blindly recoded (M.P.), yielding perfect coding agreement, indicating strong reliability.

Two layers of dependencies existed in the coded database. When multiple measures were used on a single sample, their effect sizes were averaged.^107- 108 When multiple comparison groups were used, only 1 effect size from each study was allowed in any given analysis, as described below.

All statistical analyses followed established procedures,^107- 108 with the parameter of interest, δ, quantifying the degree of overlap between refugee and nonrefugee distributions on mental health measures.¹⁰⁹ The δ was estimated by the Cohen d, the effect size equal to the difference between 2 group means divided by the pooled standard deviation. Direct effect size calculation from published means and standard deviations was possible only in roughly half of the studies. For the 24 studies (41%) reporting categorical data (eg, group differences in prevalence of a diagnosis), effect sizes were calculated as the difference between the probits, or cut points on the normal distribution below which the 2 proportions fell. For studies failing to report pertinent statistical data, d was calculated using the other standard procedures for effect size estimation.¹⁰⁸ Each independent effect size estimate was weighted with the inverse of its variance (roughly in proportion to its study’s sample size).¹⁰⁷ Effect sizes derived from studies with less than 20 participants were corrected for small-sample bias according to a standard algorithm.¹¹⁰ Effect sizes were windsorized (artificially adjusted) to 2 SD units from the mean and inverse variance weights to the nearest cluster so that extreme outliers would not eclipse observed variance in the rest of the distribution.^108,111

Homogeneity analysis using the Q statistic¹¹² was performed to determine if effect size variability exceeded that due to sampling error, implying the systematic effect of moderator variables. Analysis indicated significant variability (Q_total = 1329.73; P<.001): 96% of the variance in effect size point estimates was due to heterogeneity rather than sampling error (I² = 0.96).¹¹³ Proposed moderator variables were examined by testing the probability of their systematic association with observed variance in the effect size distribution. Variables differing only between studies were investigated by averaging effect sizes across multiple comparison groups within individual studies. Some variables also varied between groups within multiple comparison group studies. For analyses of these variables, data sets consisted of effect sizes from all codable single comparison group studies plus 1 effect size from each codable multiple comparison group study, selected using a standard algorithm to yield maximally equal representation of all levels of the grouping variable, a requirement to optimize statistical power.¹⁰⁸ When only 1 group from a multiple comparison group study contained the required level of the variable, its effect size was included in the analysis and the remaining groups were excluded. When more than 1 group from such a study contained the required level of the variable, effect sizes from those groups were combined in a weighted average.

Once the appropriate data sets were constructed, effect sizes were grouped according to the levels of each variable. Variability within and between levels was analyzed with homogeneity tests. Q statistics were calculated for each level and were summed for Q_within. The difference between Q_within and Q_total tests the significance of a potential moderator variable’s effect on the overall variance of effect sizes.

Publication bias in the meta-analytic sample was assessed as a potential threat to validity¹¹⁴ using funnel plots.^115- 116 Publication bias implies that studies with less statistical power are less likely to be published than larger, more powerful studies and, hence, may be underrepresented. Smaller samples should yield more variable estimates of the population parameter than larger studies, and if publication bias exists, their estimates should also tend to be larger. Funnel plots indicated that a relatively large number of small studies were published and revealed no relationship between effect size and sample size. The Begg test^117- 118 revealed no association between effect size estimates and their variances (ρ = 0.19; P>.05), indicating that there was no significant publication bias.

The weighted mean effect size d across the 59 studies was 0.41 (range, −1.36 to 2.91; SE, 0.01) (Figure 1). Across studies, despite large variations in effect sizes, refugees scored 0.41 SDs lower on indices of mental health than nonrefugees.

Systematic associations between the proposed moderator variables and effect size variance were investigated and are presented separately for postdisplacement conditions, refugee characteristics, and study methodological characteristics.

Effect size variation as a function of postdisplacement conditions is summarized in Table 4, which demonstrates several significant moderators of the mental health of refugees relative to nonrefugees. As expected, postdisplacement accommodation was associated with mental health (Q = 86.82; P<.001; R² = 0.06) (Figure 2), with refugees resettled in permanent, private accommodations having significantly better mental health than those resettled in institutional and temporary private accommodations. Economic opportunity (right to work, access to employment, maintenance of socioeconomic status) had a linear relationship with better mental health (Q = 217.96; P<.001; R² = 0.13) (Figure 2). Access to cultural practices was not associated with mental health (Q = 1.45; P = .23; R² = 0.00).

The position of refugees vis-à-vis their initiating conflicts was also significantly associated with mental health. Internally displaced persons scored lower on mental health indices than externally displaced refugees (Q = 65.47; P<.001; R² = 0.05) (Figure 2). Repatriated refugees scored lower than those in exile, many of whom had been permanently resettled (Q = 52.79; P<.001; R² = 0.04). Refugees from conflicts that remained ongoing scored lower than those from conflicts that had been resolved (Q = 92.00; P<.001; R² = 0.07).

All of the coded refugee characteristics were significantly associated with the mental health of refugees relative to nonrefugees (Table 5). Age had a strong association (Q = 227.28; P<.001; R² = 0.13). Child and adolescent refugees had better relative outcomes than adults. Those aged 65 years or older scored lower on mental health than adults younger than 65 years, although only 2 studies assessed this population. Refugee sex also had a weak but significant moderator effect (β = 0.15; R² = 0.02; P<.001). Studies containing a higher proportion of female refugees indicated poorer mental health outcomes. Refugees displaced from rural areas had poorer outcomes than those from urban areas (Q = 17.64; P<.001; R² = 0.03). Region of origin was also significantly associated with mental health (Q = 505.96; P<.001) and was probably confounded by comparison group region of origin. Refugees from Europe (most often compared with nonrefugee residents of peaceful European countries) had the poorest relative outcomes, and those from Asia and the Middle East had mental health outcomes that were the least different from their comparison groups, which were most often nonrefugee residents of the respective regions, often developing countries with histories of civil unrest. More-educated refugees scored lower on mental health indices than less-educated refugees (Q = 319.68; P<.001; R² = 0.28), as did those with higher predisplacement socioeconomic status (Q = 177.71; P<.001; R² = 0.16).

Several study characteristics were associated with effect size (Table 6). Greater time between date of study and time of displacement was associated with better mental health for refugees relative to nonrefugees (β = −0.28; P<.001; R² = 0.08), although durations were often short (median, 2 years; range, 1 month to 41 years after displacement) and more research is needed on long-term effects. Refugees had less negative outcomes compared with nonrefugees who had undergone displacement (ie, immigrants) vs nonrefugees who had not been displaced (Q = 23.14; P<.001; R² = 0.01). This expected effect also held true for nonrefugee exposure to violence (Q = 106.82; P<.001; R² = 0.03) and nonrefugee war experience (Q = 14.24; P<.001; R² = 0.02). For these variables, the highest levels of contamination in the comparison groups were clearly associated with smaller effect sizes. Refugees showed mental health comparable with nonrefugees who were tortured or living in active war zones.

As expected, higher methodological quality was associated with slightly smaller effect sizes.⁴¹ An index based on the sum of 5 dichotomously coded indicators of quality: study, blindness, sample randomization, researcher qualifications (psychiatrist, psychologist, or other mental health professional vs student or layperson), type of setting (academic [universities or university-affiliated hospitals] vs nonacademic), and quality of study measurement instruments (surveys and scales)¹⁶ revealed a weak negative association between methodological quality and effect size (β = −0.10; P<.001; R² = 0.01).

Publication decade was weakly associated with effect sizes (Q = 302.31; P<.001; R² = 0.01) but lacked a discernible pattern. Studies for which comparing mental health in refugees and nonrefugees was not a primary focus yielded larger effect sizes (Q = 12.98; P<.001; R² = 0.01), providing additional evidence against the influence of publication bias in the meta-analytic sample.¹⁰⁷ Proximity of researcher ethnicity to refugee ethnicity was not significantly associated with effect size (Q = 5.3718; P = .07; R² = 0.00), although the data demonstrated a marginal trend toward researchers finding larger effects when examining refugees of ethnicities dissimilar to their own.

Given the collinearity of some coded variables in the meta-analysis, the 10 substantive variables (ie, postdisplacement conditions and refugee characteristics) measured dichotomously or ordinally were factor analyzed to yield statistically independent predictors of effect size. Three factors emerged: material welfare (postdisplacement private accommodation and economic opportunity, high predisplacement socioeconomic status, conflict resolved), predisplacement cultural capital (greater education and age, predisplacement urban residence), and distance from original conflict (external displacement, not being repatriated, greater cultural freedom). All 3 factors independently predicted effect size in a least-squares regression analysis (β = −0.16, 0.22, and −0.14, respectively; all P<.05; R² = 0.33). Better relative mental health among refugees was associated with greater material welfare, lesser predisplacement cultural capital, and greater distance from the original conflict.

This meta-analysis of refugee mental health clarifies the magnitude and determinants of the psychological consequences of the refugee experience. The magnitude of these consequences, relative to those experienced by control populations, is within the range considered medium-sized.¹¹⁹ It remains substantial even when the comparison group has been displaced or directly exposed to war and violence, attesting to the depth of adversity that refugees experience.

As predicted, postdisplacement conditions were associated with mental health outcomes. Materially secure conditions, indexed by economic opportunities and permanent private accommodation, were associated with superior outcomes. By implication, psychopathology among refugees is not an inevitable posttraumatic consequence of acute wartime stress but reflects contextual factors that can be significantly remediated by generous material support on the part of governments and agencies. Similarly, the status of the initiating conflict appears to be a significant feature of the sociopolitical context, because refugees had much better outcomes when the conflict that displaced them had been resolved.

Other postdisplacement conditions had unexpected effects. Acculturative stress, cultural dislocation, and bereavement^20,24,28 were predicted to yield better outcomes for refugees who had unrestricted cultural access and were internally displaced and repatriated, but none of these predictions were supported. Externally displaced and unrepatriated refugees had relatively positive mental health outcomes, a finding that runs counter to the emphasis on repatriation in the humanitarian aid community. Repatriation, internal displacement, and cultural continuity are likely to have positive implications, but these may be overshadowed by the harmful effects of unstable political and economic circumstances within the nation of origin.

Refugee characteristics also moderated mental health outcomes. Consistent with some past work on posttraumatic responses,^38,42 female refugees had slightly worse mental health outcomes than male refugees. The apparently greater resilience of younger refugees supports previous findings,^30- 33 implying that children and adolescents are less affected by the enduring stresses of displacement.²⁹ However, higher levels of education and socioeconomic status before displacement, considered by some to have buffering functions,¹⁹ were associated with worse mental health outcomes in the analysis. Greater predisplacement intellectual and economic resources may imply a greater subsequent loss of status rather than a protective effect on refugees against their predicament. Refugee mental health outcomes were associated with region of origin, although no predictions were tested. European refugees had the poorest relative outcomes, perhaps in part because their comparison groups tended to be residents of peaceful and economically privileged countries. Refugee outcomes were also relatively poor in Africa, but because only 2% of the meta-analytic sample was drawn from Africa compared with 23.6% of the UNHCR’s population of concern,¹ the continent’s refugee crises demand more investigation. At present, the global distribution of refugees is not adequately represented in the mental health literature.

In sum, this study supports the role of enduring contextual factors before and after displacement as moderators of mental health among the world’s refugees. The psychological aftereffects of displacement by war cannot be understood simply as the product of an acute and discrete stressor, but depend crucially on the economic, social, and cultural conditions from which refugees are displaced and in which refugees are placed.^15,22