response to this problem, guidelines for replication in genetic association studies now call for exact replication. The same marker—or, if technical difficulties preclude this, a perfect or near-perfect proxy for the original marker—should be genotyped across all studies and analyzed using the same genetic model. In this section we discuss prerequisites for exact replication. We use the term “exact replication” cautiously, recognizing that this is an unattainable goal in epidemiology (e.g. studies conducted by different investigators at different times let alone places will sample from different populations) and that in some sense it is the “inexactness” of replication studies that increases credibility of the observed association (it is less likely to be an artifact due to a bias that is unique to the initial study). We use the term to emphasize the danger of “moving the goalposts” so far that claims of replication carry little weight.
