Just a quick update to recommend the book ‘The Cult of Statistical Significance’ by Ziliak and McCloskey. This is an important book. One key issue is about how we think of impact. Just because something is 95% significance with a weak effect on, say, spend it does not mean that we should ignore something with say 70% significance with a strong effect on the same spend, CSAT or whatever. 

An interesting and better way to approach this is not to cut our models down to 95% significance but take a multiple of ‘significance x effect’ (thanks to Mark from Cranfield for this advice). I would recommend this book to blow away some of the continuing problems of quantitative understanding in the market research profession.