ABSTRACT

This study aimed to compare the estimation patterns, between-index differences and associations, and the agreement and disagreement in effect-level classifications among three nonoverlap-based effect size indices—PND, IRD, and Tau-U—in single-case studies of positive behavior support conducted in inclusive education settings. Twenty-three studies published in Korean academic journals between 2016 and May 2026 were selected, and 111 AB comparisons were derived to calculate the three indices. Descriptive statistics, between-index differences, Spearman rank correlations, and agreement rates in effect-level classifications were analyzed, and the data characteristics of discrepant cases were examined. The results were as follows. First, the mean values of PND, IRD, and Tau-U were .944, .955, and .963, respectively, and the median value was 1.000 for all three indices, indicating that positive behavior support interventions were generally highly effective. Second, the three indices showed similarly high effect sizes across general study characteristics, including school level, disability type, research design, level of support, and dependent variable. Third, the Spearman rank correlations among the three indices were very high, ranging from .997 to .999, and the agreement rates in effect-level classifications ranged from 88.3% to 93.7%. Fourth, some AB comparisons were classified as data-structure influence cases, in which effect size estimates differed because of baseline extreme values or overlap between phases, whereas others were classified as classification-criterion boundary cases, in which similar estimates were assigned different effect levels because of differences in the classification criteria of the three indices. These findings suggest that researchers should select effect size indices appropriate to the purpose of the study and the characteristics of the data and, when necessary, examine the results of multiple indices together.

Keyword :