2 Beyond the a priori Approach

As far as possible, the analyses we use are designed to be agnostic about the causes of variation in a particular variable. Unfortunately, the same absence of a priori bias is not always found among our scientific peers! A referee once wrote in a report on a manuscript describing a twin study:

It is probably alright to use the twin study to estimate the genetic contribution to variables which you know are genetic like stature and weight, and it's probably alright for things like blood pressure. But it certainly can't be used for behavioral traits which we know are environmental like social attitudes!

Such a crass remark nevertheless serves a useful purpose because it illustrates an important principle which we should strive to satisfy, namely to find methods that are trait-independent; that is, they do not depend for their validity on investigators a priori beliefs about what causes variation in a particular trait. Such considerations may give weight to choosing one study design rather than another, but they cannot be used to decide whether we should believe our results when we have them.