Tau-U Revisited: Can Tau-U Combine Nonoverlap With Trend?

• In Event: Marriott Roundtable Session 52

Abstract

Tau-U was proposed by Parker, Vannest, Davis, and Sauber (2011b) to quantify the intervention effect in single-case experimental studies. According to its proponents, Tau-U can combine nonoverlap with trend assessments in both baseline and intervention phases. We systemically reviewed Tau-Us’ definitions, interpretations, and statistical properties from theoretical and empirical perspectives. Our investigation revealed that the assessment of nonoverlap cannot be combined with the assessment of trends in Tau-Us. Statistical tests and confidence intervals based on sample Tau-Us are flawed because the proper sampling distribution has not been developed in the literature. We caution researchers about interpreting these Tau-Us and computing tools that test them. Tau-U without trend is actually identical to nonoverlap of all pairs (Parker & Vannest, 2009).

Authors

• Chao-Ying J. Peng, Indiana University

• Li-Ting Chen, University of Nevada, Reno