I am a bit confused when to use, and how to identify the need to use the following transformations: log, quadratic and inverse, in a linear model.
Usually the models I'm looking at have around 8-10 quantitative variables.
My main issue is that I'm not sure what to look out for. For e.g in log, is the shape of the distribution and skew/kurtosis values sufficient? So if the skew/kurtosis indicates a distribution closer to a normal distribution, then it makes sense to use log on the variable?