REML = FALSE in lmm_f.R

[amael-ls]

In the R code, it is written (line 109 -- 125):

M0<-lm(Z_TP~Z_Length,data=data)
# Modèle complet avec différents intercepts
M1<-lmer(Z_TP~Z_Length + (1|Fish_Species) + (1|Lake), data=data, REML=FALSE)
# Modèle complet avec différents intercepts et pentes
M2<-lmer(Z_TP~Z_Length + (1+Z_Length|Fish_Species) + (1+Z_Length|Lake), data=data, REML=FALSE)
# Aucun effet Lac, intercept aléatoire seulement
M3<-lmer(Z_TP~Z_Length + (1|Fish_Species), data=data, REML=FALSE)
# Aucun effet Espèce, intercept aléatoire seulement
M4<-lmer(Z_TP~Z_Length + (1|Lake), data=data, REML=FALSE)
# Aucun effet Lac, intercept et pente aléatoires
M5<-lmer(Z_TP~Z_Length + (1+Z_Length|Fish_Species), data=data, REML=FALSE)
# Aucun effet Espèce, intercept et pente aléatoires
M6<-lmer(Z_TP~Z_Length + (1+Z_Length|Lake), data=data, REML=FALSE)
# Modèle complet avec intercepts et pentes qui variant par Lac
M7<-lmer(Z_TP~Z_Length + (1|Fish_Species) + (1+Z_Length|Lake), data=data, REML=FALSE)
# Modèle complet avec intercepts et pentes qui variant par Espèce
M8<-lmer(Z_TP~Z_Length + (1+Z_Length|Fish_Species) + (1|Lake), data=data, REML=FALSE)

Given only the random effect structure is changing, while the fixed structure remains the same, I think REML = TRUE should be used instead, especially if the aim is to select a random structure using $AIC_c$ (cf section 5.7, p120-122, Zuur 2009, Mixed effects models and extensions in ecology with R).

[KevCaz]

Hi @amael-ls, this code aims at comparing M1, M2, ..., M8 to M0. There is no comparison between models with different random effects (that is my understanding). If I am correct, then it looks like it is ok, isn't it? There is a comparison between 2 models with 2 different structure made latter on in the workshop and REML=TRUE is properly used.