Testing environmental and genetic effects in the presence of spatial autocorrelation

• DOI: 10.1111/ecog.00566
• Publication Date: 2014-02-18T12:21:28Z
• Authors: François Rousset; Jean‐Baptiste Ferdy
• Keywords: Laplace's method; Spatial Dependence; Quasi-likelihood

Spatial autocorrelation is a well‐recognized concern for observational data in general, and more specifically spatial ecology. Generalized linear mixed models (GLMMs) with spatially autocorrelated random effects are potential general framework handling these correlations. However, as the result of statistical practical issues, such GLMMs have been fitted through undocumented use procedures based on penalized quasi‐likelihood approximations (PQL), under restrictive correlation. Alternatively, they often neglected favor simpler but questionable approaches. In this work we aim to provide validated means inference GLMMs, that overcome limitations. For purpose, new software developed fit GLMMs. We it assess performance likelihood ratio tests fixed autocorrelation, Laplace or PQL likelihood. Expectedly, approximation performs generally slightly better, although variant was better binary case. show previous implementation methods R language, glmmPQL, not appropriate applications. Finally, illustrate efficiency bootstrap procedure correcting small sample bias tests, which applies also non‐spatial models.