|Saturday, February 22|
|PS3 Poster Session III & Continental Breakfast||
Sat, Feb 22, 7:30 AM - 9:00 AM
REML Sensitivity: A Common Situation When EMS Is Preferable (302793)*Christopher C. Breen, Eli Lilly and Company
Keywords: REML, EMS, Sensitivity, iid, Change over Time, Mean Shift
REML (restricted, residual, or reduced maximum likelihood) is a commonly recommended estimation approach for mixed models (models containing both fixed and random effects). Traditional EMS (expected mean squares) estimation is not recommended for this purpose by some popular statistical software. This poster highlights a common situation in which REML estimation may result in a biased estimate of the fixed effect slope (b) from a simple mixed model (Y = ci + bX + kZ + error) that is applicable in many industries (estimating change over time and initial process variability concurrently). The bias occurs under small excursions of the iid assumption (small mean shifted clusters). EMS is not biased in the presence of the same excursions, and the bias is shown to be a function of the magnitude of the random effect (k). All results are supported through simulation with a real-life example providing motivation for the problem.