In research on information integration by children, integrational capacity of a child is ascertained by the number of statistically significant main effects in analysis of variance for the child. Such an approach has an obvious bias, for the error variance may be bigger for younger but smaller for older children. Analyses of error variance from individual child analyses of four previous experiments on prediction of performance yielded mixed results. The mean error term sometimes decreased, sometimes increased, and sometimes remained the same over increasing age of children. The number of cues in the integration task and the nature of judgment seemed to be better predictors of error variance than the age of children. It seems that a study of integrational capacity must as a rule check on response variability along with the number of statistically significant main effects over ages.