what are the advantages and disadvantages of IPTW (Inverse Probability of Treatment Weighting) comparing to PSM (propensity score matching) in dealing with confounding variables?


2 Answers 2


Despite some similarities, propensity score matching (PSM) and inverse probability of treatment weighting (IPTW) behave differently, mainly because matching selects some cases/controls and discards others, while IPTW includes all study units.

The scholarly literature suggests indeed that PSM and IPTW have similar accuracy in many cases, but in some specific scenarios PSM behaves better. However, in my experience when there are discrepancies between these methods, eventually the data collection approach and the study itself ends up being less credible and externally valid.

In any case, you can peruse the following works on the subject (it is only a quick selection):





The choice between propensity score matching and weighting seems to be a widely debated topic among statistical sholars. Some thoughts, after having read through many papers of infuriated statisticians:

Propensity score matching


  • Often used, thus familiar to non-statisticians
  • Simple and intuitive subject to subject comparison by reducing the multidimensional covariate space to one dimension


  • The imperfect balance of covariates is often ignored (two individuals with the same propensity score are considered equal while they may have a strongly different set of covariates)
  • A measure of proximity between propensity scores may be arbitrary
  • Some subjects have to be excluded because no match can be found and the analysis is therefore restricted to a sub-population that is not explicitly described.

For a more detailed analysis by the main critics of this method: [1], [2].

Propensity score weighting / Inverse probability weighting


  • Explicit global population (if no clipping is used)
  • Can be easily combined with more advanced methods (see below)


  • Extreme weights at the tails of the propensity score distribution increase the variance and decrease the balance between covariates

Finally, both methods are subject to significant biases when the propensity score model is misspecified.

Therefore, the use of doubly-robust estimators, a combination of propensity score adjustement and covariable outcome estimation, seems to be becoming standard practice [3]. This combination aims to reduce the risk of bias due to suboptimal specification of the models used to estimate the propensity score or outcome regression. The possible cost of this method to reduce bias is an increase in variance.


  1. King, G., & Nielsen, R. (2019). Why Propensity Scores Should Not Be Used for Matching. Political Analysis, 27(4), 435-454. doi:10.1017/pan.2019.11
  2. Pearl, J. (2000). Causality: Models, Reasoning, and Inference. New York: Cambridge University Press. ISBN 978-0-521-77362-1.
  3. Causal Inference: What If. Miguel A. Hernán, James M. Robins. 2020.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.