LESS-THAN EXPECTED VARIANCE IN STUDIES OF SERIAL POSITION EFFECTS IS NOT A SUFFICIENT REASON FOR CAUTION

A number of reports of serial position effects have been criticized for displaying less than the variance that would be expected on the basis of the binomial theorem. The statistical evidence cited in support of this claim is reviewed and found to be far from conclusive. At least three problems with this statistical evidence are noted. First, typical patterns of variance in studies of serial position effects, which had not previously been established, are at odds with those predicted by the binomial theorem. Second, according to statistical theory, the variance observed in any particular study should not necessarily equal the variance predicted by the binomial theorem, and may do so only under a very limited number of conditions. Third, the assumptions underlying the binomial model have been violated in applications to data from experiments on serial position effects, causing severe and systematic error in the estimation of expected variance. Given that the burden of proof falls on those claiming that evidence from some experiments on nonhuman serial position performance is flawed, the doubts raised over their supporting evidence indicate that it would be prudent to suspend judgments regarding such claims pending further empirical data.