Abstract The present study explores students’ decisions to either relocate or commute to their place of study depending on the availability of student tickets. By leveraging regional variations in the coverage of these subsidized public transport tickets, we explore whether their availability decreases students’ likelihood of moving. We investigate how the importance of subsidized tickets changes with commuting time and how their relevance varies based on students’ financial resources, social backgrounds, and risk attitudes. To do so, we use the MESARAS 2013 (Mobility, Expectations, Self-Assessment, and Risk Attitude of Students) survey, which queries university entrants in the field of economics at seven German universities. We link locational identifiers from the survey with the Google Distance Matrix API to assess commuting times and regional administrative data. Our logistic regression models suggest that the availability of student tickets may decrease the likelihood of moving. The association appears to be less pronounced as commuting time increases and seems to depend on parental academic status as well as students’ budget. Thus, our study offers three contributions: First, we provide policy-relevant evidence on the importance of affordable public transportation. Second, we address (social) inequalities in the moving versus commuting decision. Third, we demonstrate how accurate spatial modeling of commuting times and integration of survey data with external data sources benefits socio-spatial analysis.

Abstract Local labor markets in Germany are interconnected through mobile workers who commute or migrate across borders. Therefore, the aggregated number of new matches between employers and employees in one region is influenced by spillovers from neighboring markets, but might also be disproportionately higher due to easier access to suitable jobs in agglomerated areas. Vocational education and training (VET) students in company training, so called apprentices, are younger and therefore less mobile than the general workforce, suggesting that regional spillovers may be less pronounced when looking at VET labor markets. In addition, occupational heterogeneities in the spatial distribution of employers might further amplify or decrease these spillovers. This paper provides the first estimates of spatial matching functions for VET students in general and for multiple profession groups. Using a matching function model with spatially lagged stocks of applicants and vacancies, it analyses efficiency, elasticities, regional influences, and spatial spillovers. To do so, it develops a novel set of spatial weights and ways to measure agglomeration. The findings show that matching efficiency for VET is higher in more agglomerated regions and that measurable spillovers do exist, even for the group of less long-distance mobile adolescents. This underscores the importance of accurate spatial modeling: only indices based on realistic travel times or commuting data show these influences. Finally, this paper finds significantly lower elasticities in the stock of unemployed applicants for regions that are well connected to their surroundings, leading to a more balanced matching process.

Abstract At the end of compulsory education, the decision to either continue school or pursue training represents a critical juncture widely analyzed in the social sciences. Choices for or against a career direction are shaped by interdependent individual, institutional, and regional-level factors. To disentangle these relationships, researchers have typically relied on (multilevel) regression techniques, focusing on mediating and moderating effects among distinct predictors. In this study, we employ artificial neural networks (ANNs) to analyze survey data from the German National Educational Panel Study, enriched with regional indicators. Focusing on school-to-vocational training transitions as established research topics in social sciences we build on recent studies that demonstrate how machine learning methods can identify complex, non-linear, and higher-order patterns among explanatory variables. To enhance the interpretability of complex patterns based on ANNs, we employ explainable artificial intelligence (XAI) techniques, specifically Shapley Additive exPlanations (SHAP) and Rule Extraction (RE) algorithms. These methods allow us to identify multiple non-linear interactions within and across analytical levels. We argue that the adoption of ANNs in the social sciences provides the potential to inspire new substantive research questions, generates novel insights into established relationships, and makes complex patterns of influence more accessible.