Non normal residuals in multiple regression I used height, weight, gender and age to regress on BMR (basal metabolic rate) and obtained the following qq plot of residuals
I then regressed the above variables with log BMR and obtained the following qq plot for residuals
Shouldn't I be using log BMR here? If so, how do I interpret my results?
 A: Rates of various kinds are often quite skew, but it's not primarily the skewness that decides whether you should deal with log-scale or inverse-scale (or some other scale) over the original rate variable.
What's usually more important is the form of the model for the mean and the variance. Those are first- and second-order effects, skewness - and so in some sense, shape - is more like third-order.
That is, if you're fitting something that relates BMR to other variables, any transformation will change the shape of that relationship, and it will change the variation about the relationship, and it will change the degree of (and even presence of) interaction between the independent variables (predictors). Guessing the distribution about the relationship close to right when the form of the relationship itself is totally wrong would be little consolation!
You might like to consider instead using GLMs, which (at least partly) decouple the connection between the form of the model for the relationship between y and x's and the shape of the conditional distribution about that relationship (by allowing separation of the distribution choice - or at least the mean-variance relationship - from the link function).
So for example, you might consider the question of the form of the relationship between a (possibly transformed) mean-BMR and the $x$-variables (is it additive and linear on the original scale? Is it linear in the log of mean BMR? or inverse? or something else?). You can then look at 
whether (say) a gamma model (or normal model, or inverse Gaussian model, or perhaps even a Tweedie model, if you have the software for it) might be be able to describe the conditional distribution of your response. 
