Rate Ratio in Multiple Baseline Designs Using Hierarchical Poisson Regression Models

• In Event: Methods for Causal Inference and Estimation of Treatment Effects

Abstract

Although there are several effect sizes to quantify treatment effects in single case experimental designs, there are none that address the small data challenges in combination with count data scale of measurement, autocorrelations, and unequal treatment lengths in the time-series. This is particularly important when quantifying treatment effects in multiple baseline designs which are used more than any other design. We formulate and test the Bayesian estimate of a rate ratio effect size (BRR) for multiple baseline designs using a hierarchical Poisson regression model and Monte Carlo simulation. BRR functions well even for shorter time-series. Although it has less accuracy for higher population effect sizes, the relative bias still remained within 5% of the true value.

Authors

• Prathiba Natesan Batley, University of Louisville

• Larry V. Hedges, Northwestern University