Effect size conventions for correlations in educational sciences using the example of early childhood education (masterthesis)

In the 1980s, Jacob Cohen developed effect size convetnions for correlations in behavioral psychology based on his own observations and studies. These measures were adopted in many areas of empirical social science and are still used today, including in empirical educational research. In psychology, these effect sizes have been the subject of critical debate for some time, which has now spread to other disciplines. In empirical educational research, however, such a discussion has been largely absent to date. Against this background, the present exploratory meta-analysis examines whether the effect size conventions proposed by Cohen for correlations in educational sciences, using early childhood as an example, can continue to serve as a suitable reference. Of 198 identified studies, 48 were included in the analysis according to defined criteria. A total of 1,297 correlations were extracted and evaluated using RStudio. The results suggest that Cohen's effect size convetnions tend to be set too high. For the analyzed data, threshold values of .10 for small, .20 for medium, and .35 for large correlations were obtained. A random effects model showed high heterogeneity of correlations between studies, which cannot be explained by sampling error alone. This heterogeneity suggests that contextual characteristics such as the age structure of the sample or the study design should be taken into account to a greater extent. However, due to the limited number of correlations and studies in several categories, no conclusive statements can be made about these characteristics. Overall, the proposed thresholds should therefore be understood as an initial guide. They can help researchers in empirical educational research to classify effect sizes appropriately and plan future studies on a more sound basis.