Online Program Home
My Program

Abstract Details

Activity Number: 361 - SPEED: Biometrics - Methods and Application, Part 2
Type: Contributed
Date/Time: Tuesday, July 30, 2019 : 11:35 AM to 12:20 PM
Sponsor: Biometrics Section
Abstract #307761
Title: Spectral Parameterization, Diagnostics, and Remedies for Confounding of Fixed Effects by Random Effects
Author(s): Patrick Schnell* and Maitreyee Bose
Companies: Ohio State University and Amgen
Keywords: diagnostics; linear mixed models; spatial confounding; spectral decomposition
Abstract:

Linear mixed models (LMMs) are a commonly-used tool for fitting linear models with correlated errors across a wide variety of fields. However, adding random effects to a linear model can cause unexpected large changes in fixed effects estimates relative to the same model without random effects, and the process by which such changes occur is not well understood. This phenomenon has recently attracted attention in the spatial statistics community in the form of spatial confounding. We present the spectral parameterization of LMMs as a tool for understanding such confounding, and develop diagnostics for evaluating the effect on fixed effect estimates of collinearities between columns of the fixed effect and reparameterized random effect design matrices, as well as individual observations. We illustrate the diagnostic tools and possible remedies on a spatially-referenced dataset of crime rates in neighborhoods in Columbus, OH.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program