T. Goicoa, M. Ugarte, A. Urdangarin Iztueta
Spatial confounding can impact fixed effects inference in spatial models. Despite the lack of a general definition, it is widely recognized that spatial confounding occurs when fixed effect estimates change after including spatial random effects that are collinear with the covariate of interest. Different methods have been proposed to address this issue, but their suitability for producing fixed effect estimates remains unclear. This study evaluates several methods, including restricted regression, spatial+, and transformed Gaussian Markov random fields, and compares their performance in a simulation study. Our results demonstrate that the spatial+ approach provides accurate fixed effect estimates. To illustrate the practical implications of our findings, we revisit three datasets: Dowry deaths in Uttar Pradesh (2012), stomach cancer incidence in Slovenia (1995-2001), and lip cancer incidence in Scotland (1975-1980).
Keywords: bias, collinearity, fixed effects, random effects
Scheduled
GT08.ESPATIEMPO1 Invited Session
November 9, 2023 11:40 AM
CC4: Room 2