Effect Sizes and Intra-Cluster Correlation Coefficients Measured from the Green Dot High School Study for Guiding Sample Size Calculations When Designing Future Violence Prevention Cluster Randomized Trials in School Settings
Purpose: Cluster randomized controlled trials (cRCTs) are popular in school-based research designs where schools are randomized to different trial arms. To help guide future study planning, we provide information on anticipated effect sizes and intra-cluster correlation coefficients (ICCs), as well as school sizes, for dating violence (DV) and interpersonal violence outcomes based on data from a cRCT which evaluated the bystander-based violence intervention ‘Green Dot’.
Methods: We utilized data from 25 schools from the Green Dot High School study. Effect size and ICC values corresponding to dating and interpersonal violence outcomes are obtained from linear mixed effect models. We also calculated the required number of schools needed for future studies utilizing available methods that do and do not consider variation in school size.
Results: Observed effect sizes for DV outcomes range from 0.06 to 0.11. Observed ICC values for DV outcomes range from 0.0006 to 0.0032. The upper limit of 95% CIs for the true ICCs range from 0.0023 to 0.0070.
Conclusion: School-based evaluations with violence outcomes are expected to have small effect sizes. Observed ICCs are less than 0.005 and upper limit of of 95% CIs for the true ICCs are less than 0.01. Designing school-based cRCTs should account for the ICC, even if its value is assumed to be negligible. Furthermore, variation in school sizes should also be accounted for to avoid having too few schools to achieve the desired power.
Digital Object Identifier (DOI)
Azam, Md. Tofial; Bush, Heather M.; Coker, Ann L.; and Westgate, Philip M., "Effect Sizes and Intra-Cluster Correlation Coefficients Measured from the Green Dot High School Study for Guiding Sample Size Calculations When Designing Future Violence Prevention Cluster Randomized Trials in School Settings" (2021). Biostatistics Faculty Publications. 54.