Prospects of Return: The Case of Syrian Refugees in Germany

Abstract

This study examines the factors that influence refugees’ decision to return to their country of origin. The research employs exploratory factor analysis and linear regression to investigate the individual-level factors that influence these decisions. Data for the analyses come from a structured survey on 889 Syrian refugees in five different German cities in late 2015. Findings indicate that refugees with higher levels of education are more likely to consider return, given that their country of origin adopts democracy. The decisions of those who experience direct threats to their safety, however, are influenced primarily by the restoration of stability.

Keywords:

• Displacement
• ISIS
• refugee movements
• Syria
• Germany
• violent conflict

Acknowledgments

An earlier version of this paper was presented at the annual meeting of the International Studies Association (ISA) in 2017. The authors would like to thank Ekrem Karakoc, James Milner, and the anonymous reviewers for their critical comments and helpful suggestions on earlier drafts.

Notes

1 One exception to this rule is Arias, Ibáñez, and Querubín (2014), which examines internal displacement in Colombia and suggests that violence, economic opportunities, and sociodemographic characteristics shape the decision to return.

2 We have reason to expect education to influence return decisions for other reasons as well. Research indicates that predisplacement characteristics affect refugee mental health during displacement and that refugees with higher education and socioeconomic levels score lower on mental health indices (Porter & Haslam, 2005). Most notably, a low level of education hinders reintegration into the country of origin, due in part to decreasing employment prospects (Black & Gent, 2006; Koser, 2015). These findings suggest that education and employment prospects should also influence decisions on sustainable return.

3 The decision to never return is not included in the analysis, as it cannot be an observed indicator of a return factor by definition.

4 Cattell’s (1966) scree test confirmed this result. Horn’s (1965) parallel analyses determine the number of factors to be extracted as four, but this is too high for seven indicators.

5 As it is likely for the factors to be correlated, an oblique rotation (promax) was preferred.

6 The uniqueness figure is very high for the Kurdish Autonomy variable 0.941, indicating that more than 94% of the variance in this variable is not associated with the two extracted factors. Put differently, the Kurdish Autonomy variable has the lowest communality value of 0.059. No dictatorship variable has the lowest uniqueness value at 0.005 or the highest communality value of 0.995. The remaining four indicators have fairly high uniqueness figures, ranging from 0.454 to 0.846.

7 The sum of squared loadings indicates that Factor 1 accounts for greater variance than Factor 2. That is, the accounted-for variance in the four observed indicators associated with Factor 1 is greater in comparison. The proportion of variance figures indicate that Factor 1 accounts for 27.2% of the variance in the four indicators associated with it, out of the seven potential factors that can be extracted from the seven indicators. The proportion of variance is 16.0% for Factor 2. (Each figure equals the sum of squared loadings in the first row, divided by seven, the number of indicators.) These figures add up to a cumulative variance figure of 43.2%. The factor correlation between Factors 1 and 2 is −.452.

8 The Kurdish Autonomy indicator loads with different signs on the two factors, indicating that the views regarding Kurdish autonomy do not have the same influence on the construct as the others. This may be due to a variety of reasons that have to do with the Syrian political context, but regardless, the loadings are not high in either factor. In fact, the absolute values of each loading is the lowest in the corresponding set. (This particular indicator still needs to remain in Factor 1, as indicators with negative signs must also be a part of factor score calculations.)

9 The same analysis returns similar results when performed with the principle axis factoring (PAF) method, using the fa function in the psych package in R. As expected, the uniqueness figures, factor loadings, the sum of squared loadings, proportion-of-variance figures, and factor correlations are not the same as those calculated with the ML method, but they are very similar. The fa function also reports the proportion of the variation each factor explains. In this case, Factor 1 explains 35% of the variation and Factor 2 explains 65%. More importantly, the fa function reports several goodness-of-fit statistics such as standardized root mean square residual (SRMR), root mean square error of approximation (RMSEA), and Tucker-Lewis Index (TLI), which correspond to 0.03, 0.077, and 0.908, respectively. In other words, the RMSR is below 0.05, the RMSEA is below .08, and the TLI is above .90, all indicating a good fit to the data.

Additional information

Funding

This project was funded by the Australian Research Council (Discovery Project Grant, DP150102453).