Variants of Cohen’s d, in this instance dt and dadj, has the largest influence on U1 measures used with smaller sample sizes, specifically when n1 and n2 = 10. This study indicated that bias for variants of d, which influence U1 measures, tends to subside and become more manageable, in terms of precision of estimation, around 1% to 2% when n1 and n2 = 20. Thus, depending on the direction of the influence, both dt and dadj are likely to manage bias in the U1 measure quite well for smaller to moderate sample sizes.
Walker, David A.
"Bias Affiliated With Two Variants Of Cohen’s d When Determining U1 As A Measure Of The Percent Of Non-Overlap,"
Journal of Modern Applied Statistical Methods:
1, Article 11.
Available at: http://digitalcommons.wayne.edu/jmasm/vol4/iss1/11