Data and Participants
The data used in this study comes from the first wave (2016) of the IAB-BAMF-SOEP dataset, an annual, representative survey of 4465 adults (at least 18 years of age), predominantly refugees and asylum seekers who arrived in Germany between January 1, 2013 and January 31, 2016 (see [24, 25] for details). Respondents completed the survey in computer-assisted face-to-face interviews by trained interviewers using audio files in five different languages. Participation was voluntary.
We excluded 21 respondents from our analyses due to missing corresponding household interviews. A further 27 respondents were excluded on the basis that they were mandated to leave Germany within the coming month. In these cases, self-reported measures of mental health and well-being are unlikely to reflect the integration measures and living conditions we are interested in evaluating. We excluded 92 further respondents from our analysis on the basis that they were members of the sampled asylum seekers’ households who were not themselves refugees who had arrived in Germany between 2013 and 2016, resulting in an analysis sample size of 4,325 respondents.
To measure psychological distress, we used the Patient Health Questionnaire for Depression and Anxiety (PHQ-4), a very brief and well-validated measurement instrument [26,27,28]. This 4-item battery uses a 4-point Likert-type scale (scores 0–3, (0) meaning symptoms not at all experienced in past 2 weeks, (1) on several days, (2) on more than half the days, (3) nearly every day) to screen for the core symptoms of depression (depressed mood, anhedonia) and anxiety (uncontrollable worrying and feeling nervous) with two separate scores or to yield a single overall measure of the degree of psychological distress ranging from 0 (no distress) to 12 (severe distress) [26, 29]. We used the total score of the PHQ-4, measuring psychological distress characterized by symptoms of depression and anxiety, in order to capture the complete spectrum of variance . Despite its brevity, the PHQ-4 performs very similarly to the combined longer PHQ-8 and the GAD-7 , which, in turn, are well-established as excellent screening tools for depression and anxiety, respectively [30, 31]. Previous studies have shown that the two depression items in the PHQ-4 match outcomes of the DSM-IV Structured Clinical Interview with a sensitivity of 87% and a specificity of 78% for major depressive disorder . The two anxiety items perform very well at diagnosing generalized anxiety disorder (Area Under the Curve (AUC) = 0.91), panic disorder (AUC = 0.85), social anxiety disorder (AUC = 0.83), and PTSD (AUC = 0.8) . In another sample, the PHQ-4 diagnosed depression and anxiety disorders with AUCs of 0.84 and 0.79 . The PHQ-4 also shows good internal reliability with Cronbach’s alphas of 0.79 for a Tanzanian , 0.84 for a Colombian , and 0.78 for a German sample . In our sample, the internal consistency of the scale was equally acceptable (Cronbach's alpha = 0.77).
We assessed life satisfaction, understood as the cognitive-evaluative dimension of subjective well-being, using a standard single-item measure widely applied in large national surveys where the costs of administering more comprehensive multi-item scales are prohibitive [35,36,37]. This measure yields acceptable reliability (range of r scores: 0.68–0.74) when tested longitudinally , good criterion validity when compared to a well-established multi-item scale, and similar construct validity to the multi-item scale . Many studies have also demonstrated high correlations between judgments of global life satisfaction and more comprehensive measures of satisfaction in key life domains [40, 41].
Sociodemographic Control Variables
Levels of education were aggregated according to ISCED standards as follows: low (early childhood education, primary education, lower secondary education), medium (upper secondary, post-secondary non-tertiary education, short-cycle tertiary education), and high (bachelor’s or master’s degree or equivalent, doctoral or equivalent degree). Nationality was reduced to categories with at least 100 observations: Syrian, Afghan, Iraqi, Eritrean, Other. Time in Germany was measured in years passed between arrival in Germany and the time of the interview. Marital status was assessed with the categories ‘Married’, ‘Single’, and ‘Divorced or Widowed’, religious affiliation with the categories ‘Muslim’, ‘Christian’, ‘Other’, ‘None’.
Pre- and Peri-migration Control Variables
Negative flight experiences were coded ‘yes’ if any of a list of seven possible negative experiences (financial scams or exploitation, sexual assault, physical assault, shipwreck, robbery, extortion, imprisonment) was reported. They were coded ‘no’ if none of these experiences were reported and ‘wished not to report’ if the respondent chose not to answer the section on flight experiences. To count the number of distressing flight reasons, we created a numeric variable summing up the number of the following flight reasons: ‘fear of violent conflict or war’, ‘fear of military draft or forced recruitment into armed groups’, ‘persecution’, ‘discrimination’, ‘bad personal living conditions’. We did not include the following flight reasons in this index because of their lack of an obvious stressor status: ‘my family sent me’, ‘because family members left this country’, ‘because friends/acquaintances left this country’, ‘general economic situation in the country of origin’, ‘other reasons’. Finally, we created a two-level categorical variable capturing whether respondents came to Germany by themselves, combining the categories ‘arrived with family members’, ‘with friends/acquaintances’, ‘with other persons’ into one level juxtaposed with the category ‘arrived alone’.
Integration Measures and Post-migration Living Conditions
The legal status variable was created by combining the report of a received refugee or asylum status into one category, and counting both reports of awaiting the outcome of the initial asylum procedure and reports of awaiting the outcome of an appeal against the initial asylum procedure decision as ‘awaiting outcome’. The family reunification variable was conceived as a binary variable assigning a ‘yes’-category to reports of having either a spouse or any number of children born after 1998 and planning to bring these family members to Germany. Currently in education includes any kind of education (school, university or doctoral studies, vocational training, professional development course). Our employment status variable comprises a ‘yes’ category for any form of employment reported (full or part time, marginally employed, internships or traineeships), a ‘no’ category for a report of no current employment but seeking employment and a ‘not seeking employment’ category. Course participation was measured as the total number of courses attended out of five integration courses and general language courses. Social contacts were measured as amount of time spent with members of different communities, ranging from ‘never’ to ‘daily’. German language ability was measured as the averaged self-reported speaking, reading, and writing ability. See the SI Appendix for details.
All statistical analyses were conducted using R version 3.5.0 . We imputed missing data in all of the variables used for analysis through multiple imputation using chained equations with the “mice” R package  (10 imputed datasets created, 10 iterations, seed = 41) (see SI Appendix Table A1 for missings per analysis variable). To improve the accuracy of the imputation, we used auxiliary variables selected for their theoretical relatedness to the to-be-imputed variables (see SI Appendix). Only auxiliary variables with a minimum correlation of r = 0.1 with to-be-imputed variables were used in the imputation .
We calculated descriptives, as shown in Appendix Tables A2 and A3, as means and standard deviations with 95%-confidence intervals or proportions with 95%-confidence intervals. The weighted values shown in the final two columns were produced using the survey weights supplied by the Socio-economic Panel of the DIW Berlin .
In our main analysis, we calculated and pooled 10 multiple, hierarchical linear regressions to estimate associations between psychological distress, life satisfaction, and variables reflecting integration measures as well as refugees’ post-migration living conditions. The baseline models (1a, 1b in Fig. 1) predict psychological distress and life satisfaction from the sociodemographic control variables federal state of residence (not included in Figure, see SI Appendix Tables A4 and A6), age, gender, education, nationality, marital status, religious affiliation, and time since arrival in Germany. Subsequent models (2a, 2b in Fig. 1) include variables representing pre- and peri-migration stressors as further controls: the number of flight reasons, whether the respondent fled alone, and negative experiences during flight. For the full models (3a, 3b in Fig. 1), we added all key predictors (a–h) mentioned above. We did not weight our regression, but included the main factors that went into Kroh and colleagues’  calculation of individual weights (gender, age, time, nationality, since arrival in Germany, legal status, and federal state of residence) as independent variables  (p. 57).
We assessed the statistical significance of the difference between Models 1 and 2 and Models 2 and 3, respectively, using Wald tests implemented using a function for the comparison of nested models fitted to imputed data [43, 46]. We used the same tests to confirm the joint significance of all categorical variables with significant differences between levels. Our SI Appendix includes the models using non-imputed data as robustness checks (Tables A5 and A7). A further robustness check shown in the SI Appendix (Table A8) replicates Model 3a as a proportional odds cumulative logit model using the PHQ-4 as a four-category ordered outcome (‘none’, ‘mild’, ‘moderate’, ‘severe’).
To investigate potential moderation effects between our control variables and the post-migration variables of interest, we computed interactions between key sociodemographics (gender, age, nationality, education) and each of our post-migration variables, and ran stratified regressions to examine significant interactions further. Following Chen and associates , we also examined possible interactions between the number of flight reasons (our best proxy for traumatic experiences in the country of origin) and post-migration living conditions in their relationship with psychological distress and life satisfaction. Because this part of the analysis is exploratory, we looked into all interactions significant at the α = 0.05 level, despite multiple comparisons (see SI Appendix for details).