Interpreting studies that compare high- and low-selected lines on new characters.

The attempt to characterize high- and low-selected lines on new variables poses serious interpretative problems when replicate lines are not available. Modest but significant line differences on new measures may be due to genetic drift totally irrelevant to the originally selected trait. Often these differences are exaggerated by inappropriate analysis using individual subject measurements rather than family means. Mean differences in high- and low-selected lines on new characters should not be ascribed to the originally selected trait unless (1) genetic drift can be estimated through the use of replicate lines, (2) the standardized mean difference exceeds 1/4 of the equivalent difference on the original selected trait, or (3) strong predictions involving multiple noncontingent measures are unconditionally supported. For most purposes of analysis, line means can be considered individual data points which can be used to compute correlations among measures. An alternative to selection with replicates--two-stage testing of commercially available inbred strains--should be considered when large genetic correlations between the characters are expected.