Reducing outlier effects with robust regression