A. Urdangarin Iztueta, T. Goicoa Mangado, T. Kneib, M. Ugarte Martínez
Spatial confounding is generally understood as the difficulty to dissociate fixed and random effects in spatial models. The issue is relevant as fixed effects estimates may present large biases and high standard errors. Overcoming spatial confounding is crucial in areas such as spatial epidemiology where detecting risks factors is one of the main objectives. There are different proposals in the literature to deal with spatial confounding in univariate spatial models, but no solution is conclusive. In a multivariate framework, the problem is even more challenging as covariates can affect the responses differently. Here, we introduce a modified spatial+ procedure that partitions the covariate into two components to capture large- and small-scale spatial dependence. To illustrate the methodology, we jointly analyse two types of crimes against women in Uttar Pradesh, India.
Keywords: crimes against women, multivariate disease mapping, spatial confounding,
Scheduled
Spatio-Temporal Statistics I
June 11, 2025 3:30 PM
MR 3