Oksági kapcsolatok erejének mérése kontingenciatáblákban: az esélyhányados problémái és a hatásnagyság

Absztrakt

Recent methodological work places increasing emphasis on the causal interpretation of statistical evidence using the effect size measure, i.e. changes in expected values. However, data presented in the form of contingency tables are generally analyzed using the odds ratio since the actual value of effect size depends on the marginal distribution of the dependent variable. This paper argues that the odds ratio is not necessarily appropriate for the purpose of causal interpretation. First, the actual value of the odds ratio is difficult to interpret. Moreover, it always indicate strong association in tables with empty cells. Second, in tables where the dependent variable can take more than two values, local or spanning-cell odds ratios might indicate positive association even when the assumption of a positive association is incorrect. Odds ratios will lead to wrong qualitative conclusions if the independent variable affects the probability of the outcomes that are irrelevant in the calculation of odds ratios.