The Jackknife and Multilevel Modeling: A New Application of an Old Trick

Abstract

In this article the authors demonstrate two instances where the jackknife can be used to enhance hierarchical linear model (HLM) analyses. The jackknife was used to improve the HLM estimates of composite measures by jackknifing over items. The first study examined fixed-effects and variance component estimation. The jackknife appeared to reduce the bias in the estimates both of slopes and of variances by implicitly adjusting for item-by-person and item-by-group interactions. The second study examined the utility of the jackknife as a multilevel item analysis tool. The results suggest that pseudovalues offer a unique opportunity for isolating item variability in multilevel data. The jackknife seems to offer enhancements and insights to conventional HLM analyses.