Robust Confidence Intervals for Effect Size in the Two-Group Case

Abstract

The probability coverage of intervals involving robust estimates of effect size based on seven procedures was compared for asymmetrically trimming data in an independent two-groups design, and a method that symmetrically trims the data. Four conditions were varied: (a) percentage of trimming, (b) type of nonnormal population distribution, (c) population effect size, and (d) sample size. Results indicated that coverage probabilities were generally well controlled under the conditions of nonnormality. The symmetric trimming method provided excellent probability coverage. Recommendations are provided.

A comparative study of robust tests for spread: Asymmetric trimming strategies

“…Ten percent has been considered by Hill and Dixon (1982), Huber (1977), Stigler (1977), and Staudte and Sheather (1990); results reported by Keselman, Wilcox, Othman, and Fradette (2002) also support 10% trimming. Keselman et al (2005) found good results with 5% symmetric trimming.…”

“…A second study also suggested that we adopt asymmetric trimming methods. Keselman, Algina, and Fradette (2005) have indicated that effect size statistics, and confidence intervals for them, can be constructed from robust estimators. In their investigation, they compared the probability coverage of intervals involving robust estimates of effect size, based on seven procedures for asymmetrically trimming data, in an independent two groups design, as well as a method that symmetrically trimmed the data.…”

The influence of nonnormality from primary studies on the standardized mean difference in meta-analysis

“…First, as we mentioned above, exponential distributions are the distributions conventionally examined in Monte Carlo simulations of nonnormal Fig. 5 Type I error rates of Q tests as a function of SD ratio, population effect size, average sample size, and study size under different distributions data (e.g., Harwell et al, 1992;Keselman et al, 2005). Second, exponential distributions have practical meaning and can be applied in order to model response time, physical and verbal violence in couples, and similar variables (see more examples in Bono et al, 2017).…”

“…This ceiling effects could be modeled by the reversed exponential distribution. Exponential distributions are also commonly examined in Monte Carlo simulations (e.g., Harwell, Rubinstein, Hayes, & Olds, 1992;Keselman, Algina, & Fradette, 2005). In this simulation, using both the exponential and reversed exponential distributions allowed us to examine cases with oppositely skewed distributions.…”