|
Abstract:
|
Biomedical research studies often involve slightly skewed, positive, and bonded quantitative data. Such data are generally analyzed on the actual or log-transformed scale with ordinary regression analysis. Depending on skewness, a departure from Gaussian distribution of residuals, and sample size, the linear regression model may produce biased findings. Moreover, the analysis of log-transformed data does not produce appropriate interpretations. Since gamma regression is a preferred method of analysis for skewed data and converges to normal distribution under specific shape parameters, we propose to analyze quantitative outcome data in biomedical research using gamma regression with the identity link function. We provide descriptive and inferential comparisons of gamma and normal regressions for determining associations of age and sex on mini-mental state exam, clinical dementia rating sum, global depression scale scores utilizing our study data from TARCC. We also provide simulation-based evidence for gamma and normal regressions under different distributional and sample size settings. In many instances, gamma regression can be an alternative choice to normal regression.
|
