Meta-Analysis in Marketing when Studies Contain Multiple Measurements

Most meta-analyses in marketing contain studies which themselves contain multiple measurements of the focal effect. This paper compares alternative procedures to deal with multiple measurements through the analysis of synthetic data sets in a Monte Carlo study and a re-analysis of a published marketing data set. We show that the choice of procedure to deal with multiple measurements is by no means trivial and that it has implications for the results and for the validity of the generalizations derived from meta-analyses. Procedures that use the complete set of measurements outperform procedures that represent each study by a single value. The commonly used method of treating all measurements as independent performs reasonably well but is not preferable. We show that the optimal procedure to account for multiple measurements in meta-analysis explicitly deals with the nested error structure, i.e., at the measurement level and at the study level, which has not been practiced before in marketing meta-analyses.