Model Risk Ratio 95% CI CI Width Stratified Mantel-Haenszela 2.75 1.11, 6.84 5.73 Logistic Standardization by Marginal Method 2.79 1.32, 6.17 4.85 Poisson 2.73 0.95, 7.82 6.87 Robust Poisson 2.73 1.27, 5.87 4.60 Log-Binomial Did not converge Log-Binomial (COPY) 2.44 1.09, 5.46 4.37 Log-Binomialb (IWLS truncated algorithm) 2.67 1.29, 5.49 4.20   2.54 1.20, 5.41 4.21   2.50 1.16, 5.36 4.20 Log-Binomial (IPTW strata weights)       2.65 1.03, 6.82 5.79 Log-Binomial (IPTW PS weights) 2.85 1.17, 6.91 5.74 Bayesian Logistic 2.65   1.31, 6.63c 5.32 Bayesian Log-Binomial 2.67   1.11, 6.38c 5.27 Logistic 7.93d    1.55, 40.44 38.89

Notes: IWLS = iterative weighted least squares; IPTW= inverse-probability-oftreatment- weighted, PS = propensity score. aThe estimated RR and 95% CI were calculated using age in years. Results were 2.91 (1.22, 6.93) using the two age groups in Table 3. bThree sets of starting values were chosen when using IWLS: 1) using the fitted MLE values for the model without age and starting the coefficient for age at zero, 2) estimates from the raw proportions and starting the coefficient age at zero, and 3) starting all coefficients at zero. cCredible interval and dOdds ratio.

Table 4: Example II results for hypothetical clinical trial [3] comparing the risk of recovery in new and conventional therapy groups. The risk ratio and 95% confidence interval (CI) for recovery adjusted for age and extent of disease was estimated using different models.