Resumen de Comparaison de six logiciels pour l'analyse de la variance d'un plan S(A2*B2> déséquilibré

Denis Corroyer, Emmanuel Devouche, Jean Marc Bernard, Philippe Bonnet, Yannick Savina

• We compare six statistical software packages (EyeLID-2, PAC, SPSS, Statislica, Statview, Var3) for the analysis of variance of a relatively simple data set (unbalanced design S). The comparison involves both descriptive and inductive aspects and focuses on the following points : 1 I which of the various options for comparisons (weighted vs unweighted comparisons, specific vs general comparisons) are available ; 2 I do the programs integrate new procedures and indices that are now recommended by the APA due to recent methodological advances (effect size measures, Bayesian inference) ; 3 I how are the graphical and statistical procedures accessed? It appears that the various desirable options or procedures are not always available. For instance, calibrated effects or Bayesian procedures are rarely found, and most software packages do not compute the more classical confidence intervals, which are necessary for the study of the effect sire in the absence of Bayesian procedures. To have access to all desirable procedures, the joint use of several software programs appears necessary.

  For some of the software programs studied, we noticed a lack of explanation of the proposed choices (e.g. the choice between unweighted and weighted comparisons), a shortage of information in the output, the use of expressions that express inadequately the relevant statistical concepts and methodological choices (e.g. the various « types » of analysis, simply labelled IIIIIIIII1V), and even some incoherencies between the various displayed results. In some cases, these problems might lead the researcher to state invalid conclusions.