Locations of labor market entry by German university graduates: is (regional) beauty in the eye of the beholder?

Regional wage differentials are in part compensated by variations in the cost of living. In contrast, different employment prospects should equal out in the absence of mobility restrictions.

Note, however, that individuals in their working ages may be more sensitive to economic factors than, e.g., retirees (Chen and Rosenthal 2008).

Experimental findings of endowment effects and status quo bias could be invoked to justify this conjecture (Kahneman et al. 1990, 1991).

Specifically we excluded 20 destinations that are relatively less often chosen as employment regions. We then conducted a Hausman test whether coefficients are systematically different from the model comprising all graduates. No systematic differences of coefficients were detected (results are available from the authors upon request). An alternative to the conditional logit framework is the nested logit model (McFadden 1981), which has the advantage of accounting for both individual- and region-specific determinants of location choice. The nested logit model conceptualizes migration as a sequential process. In the first stage, individual characteristics condition the decision of migrating away from the present location. In the second stage, regional characteristics condition the choice among potential destinations. Assuming this sequential choice seems less adequate in the context of university graduates. In addition, the nested logit is computationally demanding, and could not have been adopted for the large set of destinations we study.

Accordingly the ROR level is more fine-grained than the level of NUTS2 regions, of which there are 39 in Germany.

In Eq. (4) only the number of regional jobs per sector is time-varying. The matrix of graduation fields and employment sectors is averaged over all three waves.

Effects implied by the odds ratios are sizeable but not of unrealistic magnitudes given the large number of available alternatives. An odds ratio of 6.7 for the indicator of the university region (Model 1) indicates that, if regions were homogeneous in all other respects, the university region would be chosen by about 6.6 % of all graduates, while each other region would be chosen by about 0.98 %.

As a robustness check we replicated Model 5 for the individual waves of the KOAB survey. Reflecting smaller sample sizes, the precision of some of the estimated coefficients is reduced, but qualitatively the results are almost unchanged and consistent across the different samples. Full results are available from the authors.

In part, this may be explained by the fact that no interactions of university region similarity and settlement types 6 and 7 are included in the analyses, because there are no universities located in these regions in our sample.

In line with these considerations, employers in rural regions have been found to focus their efforts to recruit recent graduates on co-located universities in order to maximize the recruitment success as well as post-recruitment retention (Winterhager and Krücken 2015).