Population admixture and stratification are potential sources of confoundin
g in a variety of statistical analyses of genetic data, Many approaches to
adjusting for such confounding have been developed in the context of associ
ation analyses. These approaches may generally be viewed as relying on nonp
arametric null hypotheses that specify certain, sometimes incompletely obse
rved, conditional distributions; the conditional distributions are used to
normalize test statistics to have expectation zero under the null hypothesi
s. Here, with a very simple example, it is shown that there is potentially
information that is free of confounding, but that is not recovered by such
normalized statistics. An approach presented here in the context of the sim
ple example might be extended to methods for recovering information in more
complex settings. Am. J. Med. Genet. (Neuropsychiatr. Genet.) 105:57-59, 2
001. (C) 2001 Wiley-Liss, Inc.