We appreciate the comments of Field and colleagues regarding the use of random-effects models to account for physician effects in our study of racial disparities in diabetes care. As an alternative, the authors suggest treating physician effects as fixed rather than random. The primary advantage of treating physician effects as fixed rather than random is to avoid the assumption that the physician effects are uncorrelated with the measured predictors in the model. We acknowledge that models with fixed rather than random effects have the potential to avoid certain types of confounding by physician-level variables and are an appealing analytic approach when the scientific focus is exclusively on patient-level factors. However, a notable feature of models that treat physician effects as fixed rather than random is that they cannot incorporate physician-level predictors in the model; these models base estimation exclusively on the patient-level variation, ignoring the physician-level variation in the data. Because the goal of our study was to examine the impact of both patient characteristics (eg, patient age and insurance status) and physician characteristics (eg, volume of black patients treated by a physician) on racial disparities, a random-effects modeling approach was preferred. A fixed-effects modeling approach would not have allowed us to address all of the main questions of interest in our study.1
Sequist TD, Fitzmaurice GM, Marshall R, Shaykevich S, Safran DG, Ayanian JZ. Editor's Correspondence: COMMENTS AND OPINIONS. Arch Intern Med. 2009;169(2):199–205. doi:10.1001/archinternmed.2008.569
Customize your JAMA Network experience by selecting one or more topics from the list below.
Create a personal account or sign in to: