‘the Analysis of Groups of Experiments’ Lessons from Yates and Cochran

In 1938 Yates and Cochran addressed the problem of combining groups of experiments[1]. Although they were considering an agricultural context, the problem is closely analogous to that of conducting a meta-analysis of clinical trials, which in turn has close similarities to the business of analysing multi-centre trials[2]. In fact, it has been suggested that meta-analysis ought to be employed on multi-centre trials[3], although this suggestion has been disputed[4]. Two important differences should be noted, however. First, unlike the situation faced by meta-analysts in a clinical context today, many of the agricultural experiments whose results would require combination might be expected to be of similar precision. Second, degrees of freedom for estimating this precision would be very few. This latter situation applies to metaanalysis of cross-over trials today but not generally when parallel group trials are combined. The Yates and Cochran paper is described and its lesson for modern practice in analyzing multi centre trials and in applying meta-analysis is considered. References: [1] Yates F, Cochran WG. The analysis of groups of experiments. Journal of Agricultural Science 1938;28(4),556-580. [2] Senn SJ. The many modes of meta. Drug Information Journal 2000;34,535-549. [3] Salsburg D. Why analysis of variance is inappropriate for multiclinic trials. Controlled Clinical Trials 1999;20(5),453-468. [4] Senn SJ. Letter to the Editor: in defence of the linear model. Controlled Clinical Trials 2000;21,589 592.