Proportionally Reducing Sample-Size Requirements by Increasing Dependent-Variable Reliability

Abstract

Increasing dependent-variable reliability ρ(y,y) proportionally reduces both sample-size requirements and costs of independent-groups studies. This result is analytically considered in terms of the classical Z-test comparison of two means representing independent samples of size ( N). Analysis reveals that the dependent-variable reliability directly trades-off with sample-size in its impacts on sensitivity, so that constant statistical-power is maintained by reducing sample-size with a proportionally increased ρ(y,y). Two illustrations (reflecting sample-size and associated cost reductions of up to 50% or more) are presented of the value of systematic efforts to enhance reliability: (a) Integrating performance measures in evaluation of worker exposure-effects, and (b) Quality and preferences in new-product development. Test-Retest Reliability analyses are recommended as means to evaluate (1) Potential for enhancing statistical-power via increases in ρ(y,y) and (2) Impacts of attempts to increase statistical-power via increases in dependent-variable reliability (ρ(y,y)).