Backward model selection procedure for the fixed effect structure.

<p>All models had the same random effects structure (ID and year as random intercepts) and were fitted with ML estimation. The full model (Model 3) included the following variables as fixed effects: SIC, SIC², sex, Mark, TSM, BERG, Day, BERG : SIC, BERG : SIC². The fit of each successively less complex model was assessed using likelihood ratio tests. LR refers to log-likelihood ratio test statistics.