I need to perform several GLM's (i.e. ANCOVA’s, with a single continuos dependent variable and several predictors, one dichotomous and some other continuos). I was looking for both a significance on the R-square and the single predictor coefficients (OLS estimation)
Preliminary analyses identified extremely non-normal residuals in some models and relevant heteroscedasticy in some others. Transformation of data couldn't bring any improvement and non-linear fitting was not the determining issue here.
I was looking for a single approach that can handle both the assumption violations at the same time, in order to keep a single analytic strategy along all the analyses. I read bootstrapping is good in this situation, especially wild bootstrapping (i.e. bootstrap of the residuals instead of the original cases).
As I haven't widely used a bootstrap approach, I would like to have your opinion whether I’m on the right way.
Thank you all in advance for you help!