Few-wave vs. continuous consumer panels: Some issues of attrition, bias and variance

In econometrics, the word "panel" refers to a research design in which respondents answer the same questions at least twice, with the reporting occasions separated by a suitable interval of time. Occasionally a panel fitting this minimal definition is used for ad hoc marketing study. But the greatest part of advertisers' panel research spending goes to ongoing (usually syndicated) panels of the diary or scanner type. In this paper we argue that continual-wave panels are qualitatively different from "econometric-type" panels. Consequently, there is danger in carrying over research results from two-wave studies into the world of continual-wave consumer panels. We argue that greater attention must be given to variance than to bias in analyzing consumer panel data of the latter type, and report some new formulae for the variance of market measures under different panel reporting options. Some guidelines for panel users ensue.