Compositional or Contextual Effects? Neighbourhoods and Teenage Parenthood in Stockholm, Sweden

Abstract

Selection bias constitutes a major problem in neighbourhood effect research but perhaps especially so for studies of effects on teenage birth rates. Results from both the U.S. and Europe are highly inconsistent: where some find substantial effects and others no effects. This inconsistency in combination with the clear correlations between neighbourhood environment and teenage birth rates where teenage mothers and mothers-to-be are overrepresented in low SES neighbourhoods makes it unclear if and to what extent neighbourhoods exerts a causal influence on teenage birth rates or whether correlations are mere reflections of differences in choices on the housing markets. This study adds to the (mainly American and British) body of literature on neighbourhood effects on teenage birth but focuses on a context—Stockholm, Sweden—where teenage birth rates are substantially lower. It offers a discussion on the potential impacts of selection bias on estimates of neighbourhood effects on teenage birth rates and empirically tests the existence of such neighbourhood effects. To account for selection bias I use a random effects model with a Mundlak correction (a hybrid model), a model that corrects for selection similarly to a fixed-effects model but also allows the inclusion of fixed parameters. The hybrid model produces coefficients that are substantially smaller compared to an OLS model that does not control for selection; selection thus biases results upwards. The neighbourhood effects estimated by the hybrid model are so small that they can be ignored.