I have sampled fish populations inside and outside of a protected area. I sampled 4 sites inside and 4 sites outside. Each site was sampled 6 times over a year. I am running generalised linear mixed models with Protection as a fixed factor, and Site and Time as random factors. I have tested the effect of the random factors using LRTs. However, I am reluctant to remove random effects that are not near significance (p > 0.25) from the model because I am concerned this may be committing 'sacrificial pseudoreplication'.

Can anyone provide guidance on this?