Advances and Applications in Statistics
Volume 5, Issue 3, Pages 313 - 323
(December 2005)
|
|
CONSTRAINT MAXIMUM LIKELIHOOD ESTIMATION OF RELATIVE RISK
Shuangge Ma (U. S. A.)
|
Abstract: When studying prevalence outcomes in prospective clinical studies, it is often of interest to estimate relative risks instead of odds ratios. Log-binomial models are extensively used for this purpose. Unlike the logistic models, the log-binomial models place restrictions on the parameter spaces. Maximum likelihood estimators are likely to occur on the boundary for large datasets and/or high dimensional covariates, in which cases most existing software based on the maximum likelihood estimations cannot provide converging estimators. In this paper, we propose a constraint maximum likelihood approach that always converges, while having the same asymptotic efficiency as the maximum likelihood approach. Variance estimation based on asymptotic normality of the proposed estimators is also discussed. Numerical studies show the proposed approach performs better than the Poisson and the binomial regressions. |
Keywords and phrases: binomial regression, cohort studies, odds ratio, Poisson regression, relative risk. |
|
Number of Downloads: 399 | Number of Views: 1429 |
|