It is a capital mistake to theorize before one has data