Which statement about Poisson regression is true in the context of cohort studies?

• A. It cannot model rate data
• B. It always provides superior fit to Cox models
• C. It assumes normal residuals
• D. It is used to model rate data and counts with a log link

Answer: (D)

Poisson regression is a modeling approach for count data and rates, using a log link to connect the linear predictor to the expected count or rate. In cohort studies, you often count events and have varying follow-up time (person-time). By including an offset equal to the log of person-time, the model directly estimates incidence rates, and the exponentiated coefficients give incidence rate ratios. This is precisely what researchers want when comparing groups over time.

The statement that it cannot model rate data is incorrect because the offset for person-time lets Poisson regression model rates. It does not always provide a superior fit to Cox models; Cox models handle time-to-event data and hazards, and Poisson models are not inherently better across all scenarios. It also does not assume normal residuals; it assumes the response follows a Poisson distribution and uses a log link, with residual behavior not being normally distributed. Therefore, the true description is that Poisson regression models rate data and counts with a log link.