Abstract:
|
Consider a special case of a nested mixed model where there are N physicians, with n_i patients each, i=1,..., N. A question is answered, on a continuous scale, by both physicians and their patients. We are interested in the difference between physicians' and patients' answers. We assume that there is a correlation between the answer of a patient and the one of his or her physician and a correlation among the answers of a given physician. Using simulated data, we evaluate type I and type II errors for two model specifications in various scenarios: (A) a Mixed Model, in which the response is the difference between the physician and patient's answers, using a random effect for physician clusters; (B) a Mixed Model with Repeated Measures on both answers with a random effect for the physician's answers. This last model restricts the covariance to zero between the patients' answers of a given physician. We found that model B is always more powerful than model A, but the difference is rather small in most cases.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.