How do we decide which [[Generalized Linear Models|GLM]] to use? For choosing the data model (i.e. the distribution the response follows), we have a few clues: - Check the support -- is it over all of $\mathbb{R}$ (Gaussian)? Is it integer-only (Poisson, negative binomial)? Or positive reals (Gamma)? Or 0-1 (Logistic/Binomial)? - Check the mean-variance relationship -- if $\sigma^{2}$ is constant in $\mu$, then OLS is good. Is $\mu \approx \sigma^{2}$ (Poisson)? Is $\mu \propto \sigma^{4}$ (Gamma)?