Gravar-mail: Comparing treatments via the propensity score: stratification or modeling?