We propose a new statistical model of locally interdependent hazards for a hazard model with social interactions. This interaction gives rise in turn to local social multipliers. Such neighbourhood effects are particularly strong for determining the length of immigration spells. Our simulation study demonstrates the good behaviour of the new spatial maximum likelihood estimator. In order to address concerns about the endogeneity of location choices, we also implement a weighting scheme for the maximum likelihood procedure that combines insights from inverse propensity score weighting and Bartik-shift-share instrumentation.

We estimate this statistical model on Dutch administrative data of the entire population of recent Turkish labour immigrants and establish empirically the importance of social interactions at the neighbourhood level.

The estimated interaction parameter is positive and sizeable, thus leading to a multiplier effect that substantially accelerates returns for all durations. We quantify the local social multipliers in several factual and counterfactual experiments, and
demonstrate that these are substantial.