Random Variables and Correlational Overkill

Abstract

Research findings may be more publishable if significant results are reported. This type of publication bias would increase the likelihood of "chance" relationships being disseminated. The implications of these assumptions are empirically investigated in a correlational analogue study. A large number of significant relationships were found in several groups of subjects between their actual scores on 45 SVIB scales and scores on 10 "experimental" scales which were determined by a set of random numbers. Furthermore, "logical" factors were shown to underly relationships which existed among scores on a given random scale with its significant correlations to SVIB scales. Considerations in such overkill in simple correlational studies are the subject-to-variable ratio, variable independence, and more stringent probability levels.