Accounting for One-Group Clustering in Effect-Size Estimation

Y

8

Multivariate Analysis; Effect Size; Sampling; Sample Size; Computation; Statistical Significance; Data Analysis; Research Design; Error of Measurement; Error Correction; Accuracy; Evaluation Methods

In some instances, intentionally or not, study designs are such that there is clustering in one group but not in the other. This paper describes methods for computing effect size estimates and their variances when there is clustering in only one group and the analysis has not taken that clustering into account. The authors provide the effect size and sampling variance computations, along with the adjustments for the test of significance. In addition, they provide examples of how much the statistical results can be affected (or adjusted) once one-group clustering is taken into account depending on various data characteristics. In examining sample data, that authors found that not accounting for one-group clustering affects the variance and degrees of freedom more than the effect size. With those same degrees of freedom used in the t-test, it is easy to change the conclusion of the test from significant to non-significant once one adjusts for this clustering. Tables are appended.

ERIC

9

2016

ED563038