Résumé
Meta-analyses offer interesting tools for combining the results of studies carried out by several authors on thestrength of the observed effect of one variable on another, also called the effect size. The models proposed forthe meta-analysis are within the statistical framework, which makes it possible to make hypotheses on the parametricprobability laws related to the effect size variable. Then it is possible to use the classic statisticaltools as estimation, confidence intervals and tests. This choice is mainly explained by the fact that probabilitytheory is a well-known framework to model uncertainty for the meta-analysis users. However, probability theoryshows its limits for combining information from heterogeneous sources. In the absence of the assumption of homogeneity,the aggregation by a weighted average of the effect sizes, adopted in the meta-analyses to obtain the finaleffect size, does not seem to reflect the information provided by most sources. In this article, we propose ameta-analysis that takes advantage of possibility theory techniques to combine incomplete information from heterogeneoussources. The result is to distinguish between plausible and less plausible values for the effect size of a treatment, for example, given trials performed by different authors. To illustrate our proposal, we present an experiment on a classic case of meta-analysis which focuseson the effect of taking diuretics during pregnancy.