David Palma attended the Transportation Research Symposium Conference in Rotterdam
Showcasing his poster titled: Semi-parametric approximation of spatial correlation matrix among observations in multinomial discrete choice models.
Abstract
Spatial correlation is a well-known fact in geography, yet its treatment in econometric models is far from simple, involving convoluted and computationally intensive implementations, and difficult-to-interpret results. This work proposes a simple and efficient method to account for spatial correlation in logit models. The approach avoids correlating all observations to each other by creating a small (compared to the number of observations) number of centroids. Then these centroids are correlated through the addition of random error terms in the utility to all observations. Furthermore, the error terms can be assumed to be non-random parameters to be estimated, dramatically reducing the computational complexity of the problem, while improving the fit and interpretability of the model
