Effects of heteroscedasticity assumptions on CAR models.

T. Goicoa Mangado, M. Á. Martínez-Beneito, A. Adin Urtasun, L. Ugarte

The use of Conditional autoregressive (CAR) models have been common practice to analyze areal count data, particularly incidence and/or mortality data. Despite their well-grounded theoretical basis, they still present some not well understood effects. Their attractive and easy to interpret conditional definition does not bring clear marginal properties. For example, areas with the same number of neighbours have the same conditional but may have different marginal variance. In this work we focus on some aspects of CAR models related to edge effects, the geometry of the study region and heteroskedasticity assumptions. These issues may be relevant but often go unnoticed unless they are carefully examined. We propose a new CAR distribution to deal with some of these problems and introduce extensions to improve the proposal. We use Spanish mortality data to illustrate the techniques.

Palabras clave: eccentricity, border effects, homoscedastic CAR

GT Estadística Espacio - Temporal
3 de septiembre de 2026  09:00
Aula 29

I. Fuentes Santos, J. Cubas, J. Rey Medina, J. R. Aboal Viñas, R. M. Crujeiras Casais

A. Medialdea Villanueva, J. M. Angulo Ibáñez, J. Mateu Mahiques, C. Comas Rodríguez

A. Martínez Puig, A. Blanch Plana, C. Comas Rodríguez

J. Marcen-Gutierrez, J. Castillo-Mateo, A. E. Gelfand, J. Asín, A. C. Cebrián