I have a repeated measures design: a continuous dependent variable (measure), two fixed variables (treatments) and random effects (subjects). The dependent variable is not normal and the distribution differs between treatments.
I carried out GLMM (using spss), however I am unsure which distribution to choose. I compared the results from two GLMMs: GLMM with normal distribution with identity link and GLMM with normal distribution with log link. The significance of one fixed variable and the interaction of both is significant using log link and not using identity. I printed the Pearson residuals and predicted values when conducting the GLMMs and the log link residuals are (just)normal and identity link residuals are not normal.
Does the normality of the residuals in the GLMM normal with log link mean that this model is appropriate? If not, what can I do to decide for the best model and distribution?