Related Experiment Videos
A note on R2 measures for Poisson and logistic regression models when both models are applicable
1Department of Medical Computer Sciences, Section of Clinical Biometrics, University of Vienna, Spitalgasse 23, A-1090 Vienna, Austria.
Journal of Clinical Epidemiology
|February 13, 2001
Summary
Poisson and logistic regression models yield similar parameter estimates for epidemiological studies. However, Poisson regression
Area of Science:
- Epidemiology
- Biostatistics
Background:
- Epidemiological studies often analyze dichotomous outcomes using regression models.
- Poisson regression is a common alternative to logistic regression for such analyses.
Purpose of the Study:
- To compare the R-squared measures of Poisson and logistic regression models.
- To clarify the interpretation of R-squared in Poisson regression for epidemiological data.
Main Methods:
- Comparative analysis of Poisson and logistic regression models.
- Illustration using a specific epidemiological example.
- Theoretical explanation of R-squared differences.
Main Results:
- Both models provide similar parameter estimates and significance levels.
- A notable difference arises when calculating the R-squared (explained variation).
- Poisson R-squared reflects event rate predictability, not individual outcome predictability.
Conclusions:
- The R-squared measure in Poisson regression is suitable for event rates.
- It is inadequate for assessing the predictability of individual outcomes in epidemiological studies.
- Careful consideration of R-squared interpretation is crucial when choosing between Poisson and logistic regression.