Primacy,recency, and the variability of data in studies of animals’ working memory

The performance measures in many experiments on animal memory are expected to have an underlying binomial distribution, with additional variance contributed, for example, by between- subject differences. This paper examines whether the data from published studies of serial position effects (primacy and recency) in animals’ working memory conform to that expectation. In most cases, the variance, when it can be estimated, is consistent with those statistical assumptions, but in certain studies, it is significantly smaller than expected. This is usually a sign of faulty procedure or analysis, and possible causes are discussed. The conclusion is that much of the evidence for primacy in animals is unsatisfactory, on statistical or other methodological grounds. The analytic approach outlined here might usefully be applied to detect potential problems with other experiments of a similar type, especially when manually operated apparatus is employed, and to improve their statistical power.