log-binomial-log-poisson-regression

Use log-binomial regression instead of logistic regression to estimate risk ratios or prevalence ratios for binary outcomes, avoiding the divergence between odds ratios and risk ratios when outcomes are common.

If your log-binomial regression model fails to converge, you can use log-Poisson regression as an alternative to estimate risk ratios, provided you apply robust variance estimators to correct for distributional misspecification.

Estimate prevalence ratios for cross-sectional data by running log-binomial or log-poisson regression models on your dataset to calculate the ratio of probabilities of currently having the condition between groups.

Log-Poisson regression requires robust variance estimators because the Poisson distribution is misspecified for binary outcomes; applying robust variance estimators corrects this misspecification to provide valid standard errors.

Yes, you can use proc_genmod to estimate risk ratios for cohort studies by configuring log-binomial or log-poisson regression models to compare cumulative incidence probabilities between exposure groups.

Risk ratio estimation compares cumulative incidence probabilities in cohort studies, whereas prevalence ratio estimation compares the current probability of having a condition in cross-sectional data using log-binomial or log-poisson models.