Copyright 2016 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.
In Reply We agree with Dr Garrido that, of the methods listed in our article, matching is likely to be most effective at balancing baseline characteristics between treated and untreated participants, thus eliminating to a greater extent systematic differences between groups.1 Although there appears to be a hierarchy (ie, matching over stratification over covariate adjustment) in terms of the effectiveness of balancing,2 we would like to emphasize the importance of accurate specification of the score prior to its ultimate use. Without proper specification (ie, inclusion of appropriate variables in the propensity score model), even propensity score matching is unlikely to effectively balance study groups. Although the intent of our article was to list common uses of the propensity score (all of which are better than no adjustment for confounding), we recognize certain uses of the propensity score are likely better than others; we did not mean to imply that all approaches were equally useful or effective.
Haukoos JS, Lewis RJ. Covariate Adjustment and Propensity Score—Reply. JAMA. 2016;315(14):1522. doi:10.1001/jama.2015.19093