Variable selection for propensity score models.

By M. Alan Brookhart, Sebastian Schneeweiss, Kenneth J. Rothman, Robert J. Glynn, J. Avorn, T. Stürmer. 2006

2,278 citations recorded by the source corpus.

Subfields: Economics and Econometrics, Statistics and Probability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Economics and Econometrics

In Statistics and Probability

Citations in the corpus

FLAME: A Fast Large-scale Almost Matching Exactly Approach to Causal Inference (2017, methodology): “(R2) represents the desire for estimating the most granular causal estimates possible while (R3) addresses the need for matching covariates to be associated with the potential outcomes (Rubin and Thomas, 1996; Brookhart et al., 2006).”
Propensity score estimation: machine learning and classification methods as alternatives to logistic regression (2010, background): “The relationship between model fit and bias is not as clear as this, in that inclusion of a strong predictor of treatment in a propensity score model might improve model fit without markedly affecting bias, if that predictor of treatment is a risk factor for the exposure but not the outcome [13].”
On model selection and model misspecification in causal inference (2012, positive): “A drawback is that such propensity-score based procedures can be inefficient and more prone to bias amplification when they include predictors that have (almost) no residual association with the outcome.(23,26) Second, in the Appendix, we study the asymptotic behaviour of exposure effect estimators ”
Review of inverse probability weighting for dealing with missing data (2013, result): “This is also in agreement with findings about using propensity scores to adjust for confounding.(37,38)”
The role of prediction modeling in propensity score estimation: an evaluation of logistic regression, bCART, and the covariate-balancing propensity score. (2014, methodology): “Ideally, one would estimate PSs that balance only risk factors for the outcome (7).”
Association Between Wait Time and 30-Day Mortality in Adults Undergoing Hip Fracture Surgery (2017, background): “Mediators of delay unrelated to mortality (such as surgery timing)(2) and common effects of the relationship betweendelay andmortality (such as discharge disposition) were not included as variables in the models.(30) Missing data, which was less than 1% for all variables considered, were excluded fr”
Why Propensity Scores Should Not Be Used for Matching (2019, methodology): “In fact, we show in Section 4.4 that PSM’s two-step procedure is an increasingly worse sum- mary of Xi as the number of elements k in the vector increase beyond one (see Brookhart et al., 2006).”
An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies (2011, background): “…in this journal was written as a companion to this article to illustrate the application of propensity score methods to estimate the reduction in mortality due to provision of in-hospital smoking cessation counseling to current smokers who had been hospitalized with a heart attack (Austin, 2011a).”
Internal Audit Outsourcing and the Risk of Misleading or Fraudulent Financial Reporting: Did Sarbanes-Oxley Get It Wrong? (2012, positive): “(See Appendix 1 for variable descriptions) In estimating the propensity score matching model, it is standard practice to employ all the covariates that determine the outcome model (Brookhart et al. 2006).”
An introduction to inverse probability of treatment weighting in observational research (2021, background): “In addition, covariates known to be associated only with the outcome should also be included [14, 15], whereas inclusion of covariates associated only with the exposure should be avoided to avert an unnecessary increase in variance [14, 16].”
Prognostic score–based balance measures for propensity score methods in comparative effectiveness research (2013, positive): “As a result, variables that are strongly related to treatment assignment but unrelated to outcome, such as an instrumental variable, may have undue influence in the propensity score, which can lead to decreased precision and increased bias [13,14].”