Abstract This study analyzes the determinants of commuting between cities in the Londrina Metropolitan Region (LMR) using a Structural Gravity model with Fixed Effects, estimated by Poisson Pseudo-Maximum Likelihood (PPML), and an XGBoost model. The two approaches were used for predictive performance comparison, and spatial aspects were incorporated into both models by introducing an interaction between distance and labor market development as a regressor. The results suggest that distance acts as a repelling factor for commuting flows and while labor market dynamics in the origin city serve as a repelling factor, it acts as an attractive factor in the destination city. Although spatial effects were significant, their inclusion did not greatly enhance the predictive power of the models. Moreover, the Structural Gravity model and the XGBoost showed identical performances, suggesting that the Gravity model remains relevant for commuting research.
Nepomuceno et al. (Tue,) studied this question.