Text this: Testing for normality in regression models: mistakes abound (but may not matter)