The residual deviance is higher for the more complex model. Why is this the case? Can I compare the residual deviance of a GLM and a multi-level model in this way or should I use something else instead like the AIC? (which in this case gives a lower AIC for the more complex model, m2).
Poisson GLM model:
m1 <- glm(Count ~ Month + Year + Region , data = df, family = poisson(link = "log")) summary(m1)
residual deviance = 2465.0
Poisson multi-level model:
library(lme4) m2 <- glmer(Count ~ Month + Year + Region + (1|Year), data = df, family = poisson(link = "log")) summary(m2)
residual deviance = 3921.4