Online Program

Friday, February 20
CS14 Interpretation Fri, Feb 20, 3:45 PM - 5:15 PM
Borgne

Using the Delta Method to Generate Means and Confidence Intervals from a Linear Mixed Model on the Original Scale, When the Analysis Is Done on the Log Scale (302899)

Edith C. Kieffer, University of Michigan School of Social Work 
*Brandy R. Sinco, University of Michigan School of Social Work 
Michael S. Spencer, University of Michigan School of Social Work 

Keywords: transform, delta method, mixed model, inverse transform

Common statistical software does not provide an option to report log-transformed analyses on the original scale. We implement the delta method to report the means and confidence intervals of difference scores from log-transformed longitudinal data so results can be reported on the original scale of the outcome. We demonstrate how this methodology can be implemented on output from LMM (linear mixed models) from common statistical software such as SAS Proc Mixed, and we compare results from the delta method to reporting percent change—another common approach to difference scores from log-transformed data.