Self-selection of Asylum Seekers: Evidence From Germany

I examine the pattern of selection on education of asylum seekers recently arrived in Germany from five key source countries: Afghanistan, Albania, Iraq, Serbia, and Syria. The analysis relies on original individual-level data collected in Germany combined with surveys conducted at origin. The results reveal a positive pattern of selection on education for asylum seekers who were able to flee Iraq and Syria, and the selection is neutral for individuals seeking asylum from Afghanistan and negative for asylum seekers from Albania and Serbia. I provide an interpretation of these patterns based on differences in the expected length of stay at destination, the migration costs faced by asylum seekers to reach Germany, and the size of migration networks at destination. Electronic supplementary material The online version of this article (10.1007/s13524-020-00873-9) contains supplementary material, which is available to authorized users.


A2.1 Level of education
The empirical analysis focuses on the selection of asylum seekers from the origin population with respect to education 1 . Two steps are implemented to combine available information on the educational attainment of individuals 2 . First, the answers about the level of education in each questionnaire are divided into six different categories, i.e. no formal education, primary, lower secondary, upper secondary education, vocational training and university. Tables A2.1.1 and A2.1.2 detail the procedure that has been followed to assign answers about education to each group for the asylum seekers and the origin population, respectively.
Nevertheless, the six categories are not included in every survey, e.g. studies carried out in Afghanistan and Syria do not contain information on vocational training. Therefore, the initial binary indicators are grouped, so that the final variable of interest is composed of three levels: (i) Primary education or less, which refers to cases without education and with primary education (ii) Secondary education, which contains individuals who attended lower, upper secondary education and vocational training and (iii) Tertiary education, which encloses those who went to university.  Notes: For Syria, an individual is assigned to one category if at least one grade has been completed at a given level of education, otherwise the level immediately below is attributed. For Serbia, the original questionnaire (2013) asked the acquired education level, with the following answers: (i) No school, (ii) 4 th grade of primary school, (iii)

A2.2 Insecurity in the home country
The subjective perceptions on insecurity (labelled Insecurity) in the origin site are reported in three samples (Afghanistan, Iraq and Serbia). The starting point to match information among the different data sources is the answers collected in the IAB-BAMF-SOEP Refugee Sample. Asylum seekers were asked the following semi-open query: "What were the reasons for leaving your country of origin?" and, among all propositions, four of them are retained to build a binary variable indicating whether people felt threatened (at least, by one of the selected propositions) before the migration to Germany. The possible answers are: (i) "Fear or violent conflicts or war", (ii) "Fear or forced recruitment by military or armed groups", (iii) "Persecution" and (iv) "Discrimination (ethnic, religious, etc.)". Then, the variable is combined with relevant individual characteristics encompassed in the three originspecific data sets. For each case, a binary indicator has been derived from the ordinal answers that were available for the respondents. Table A2.2 presents the question about insecurity and the procedure followed to assign the replies to to the variable. Notes: Variables with a star superscript denote characteristics of asylum seekers that refer to the pre-migration period. For instance, the occupational status corresponds to the position held before they left the origin country.

A5 Consistency of language proficiency
This section addresses a potential concern associated to retrospective (i.e., linked to the pre-migration period) questions that were asked in the Refugee Sample. For instance, the ability to speak German is defined from the following question: "How well could you speak the German language before you move to Germany?". An issue will arise if the answers are a function of the time spent in Germany, so that they will be contaminated by the current language aptitude of asylum seekers. To evaluate whether this problem might exists in the empirical investigation, the relevant variable is regressed on the years and months since arrival in Germany. The results obtained using the linear probability model are given in Table A5. Notes: All models are estimated using OLS. Robust standard errors in parentheses. ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively. All represents the five selected origin countries. Source: Author's elaboration based on SAP (2011-16), LSMS (2012) and IAB-BAMF-SOEP Refugee Sample (2017).
All coefficients are not significant, regardless of the time variable taken into account. This outcome supports the idea that the time spent in the host country is likely not to influence the estimates of the language variable included in the specifications associated to Afghanistan and Albania.   Notes: All models are estimated using logistic regressions. McFadden's R 2 = 1 − ln(L M )/ln(L 0 ), with L M , the likelihood of the estimated model and L 0 , the likelihood of the model without predictors. Robust standard errors in parentheses. ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively. Muslim is the benchmark category to analyse the religious affiliation of Albanians. Employee (both with and without supervision tasks) is the reference group to interpret the occupational status in Serbia. Information about insecurity in Serbia is only available in the 2013 wave, which explains the number of observations reported in column (6). Source: Author's elaboration based on LSMS (2012), EU-SILC (2013-15) and IAB-BAMF-SOEP Refugee Sample (2017).   Notes: All models are estimated using logistic regressions and the reported coefficients are the average marginal effects. McFadden's R 2 = 1 − ln(L M )/ln(L 0 ) with L M , the likelihood of the estimated model and L 0 , the likelihood of the model without predictors. Robust standard errors in parentheses. ***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively. Due to data limitations, column (6) of Table 8 can not be replicated.