# Is a mixed/ random effects model required if fixed effects model shows no pattern in the residuals?

I've got some data on discrete flood events (response variable - duration) on several rivers.

I've modelled the response variable by site and time using a Generalized Linear Model:

 mod = glm(duration ~ site + time)


I've then looked at the residual plots and cannot find any evidence for temporal correlation in the residuals. Does this mean that this model approach is valid (and that using a mixed/ random effects model is unnecessary) (or should one always use mixed/ random effects models for time-series data)?

• for what it's worth, this is really just a linear model (lm in R); since you're not specifying family, it defaults to Gaussian family. Technically you're right that it's a generalized linear model (since linear models are a subset of GLMs) but it may muddy the waters slightly. – Ben Bolker Oct 30 '13 at 13:12