Meta-analysis of experimental data : application in animal nutrition

Abstract

Research in animal sciences and in nutrition in particular, requires more and more important treatment of databases. Indeed, for subjectsof interest, the number of publications and of results per publication is largely increasing. It is thus more and more necessary to be ableto extract quantitative data from the literature. As a consequence, the methods of statistical meta-analysis of experimental databases havebecome essential and it is important to adequately implement them. The management of the meta-analysis is done in several phases. Thefirst phase concerns the definition of working objectives and specifications that will be decisive for the choice of the applicant publications.These should be scrupulously evaluated before being integrated into the database. During their integration, it is important to carefullyencode (experiments, treatments, …) which will be important reference points for the rest of the analysis. The databases that havebeen built this way give rise to interpretation difficulties; they contain missing data and do not represent a classical experimental system.It is recommended to include a first step of careful graphical interpretation in order to have a global and a specific view of the data. Thisphase is followed by a study of the meta-system made up of the database to be interpreted. These different steps condition the definitionof the applied statistics model. This should allow to differentiate the variations of inter- and intra-experiments; it can integrate the qualitativeor quantitative factors. It also includes fixed or random effects. Finally, it may also consider systems of weighed data. After theadjustment of the model, it is important to develop a post-analytic study that consists in studying the residual variations and the roles ofthe different treatments and experiments in the results that were obtained. At this stage, it is often necessary to return to one of the earliersteps. This way, the meta-analysis becomes a heuristic step.