How do you use propensity score matching to reduce selection bias in observational studies?

1 What is propensity score matching?

Propensity score matching is a method of matching units (such as individuals, schools, or regions) based on their propensity scores, which are the estimated probabilities of receiving the treatment given their observed characteristics. The idea is to create a balanced sample of treated and untreated units that have similar distributions of covariates, or the variables that may affect the outcome or the treatment assignment. By doing so, you can reduce the confounding effects of the covariates and isolate the causal effect of the treatment.

2 How to estimate propensity scores?

The first step in propensity score matching is to estimate the propensity scores for each unit in your sample. You can use any statistical model that predicts the probability of receiving the treatment as a function of the covariates, such as logistic regression, probit regression, or machine learning algorithms. The choice of the model depends on your research question, the availability and quality of the data, and the assumptions you are willing to make. You should check the accuracy and validity of your model by assessing its fit, predictive performance, and balance.

3 How to match units based on propensity scores?

The second step in propensity score matching is to match the units based on their propensity scores. There are different methods of matching, such as nearest neighbor, caliper, stratification, or weighting. The choice of the matching method depends on your research design, the size and composition of your sample, and the trade-off between bias and variance. You should check the quality of your matching by assessing the balance of the covariates between the matched groups, the common support of the propensity scores, and the sensitivity to unobserved confounders.

4 How to estimate causal effects after matching?

The third step in propensity score matching is to estimate the causal effects of the treatment on the outcome after matching. You can use any statistical model that accounts for the matched structure of the data, such as difference-in-differences, regression adjustment, or inverse probability weighting. The choice of the model depends on your research question, the type and level of the outcome, and the assumptions you are willing to make. You should check the robustness and validity of your model by assessing its fit, assumptions, and sensitivity.

5 What are the advantages and limitations of propensity score matching?

Propensity score matching is a technique used to reduce selection bias in observational studies, and it has several advantages. For example, it can reduce the dimensionality of the covariates and improve the balance of the covariates, allowing for flexible and transparent matching methods. However, it also has some limitations. It requires a large and representative sample of treated and untreated units, and it relies on the assumption of unconfoundedness. Additionally, imperfect matching or model misspecification can lead to bias or inefficiency.

6 How to use propensity score matching in practice?

Propensity score matching can help improve observational studies, but it requires thoughtful planning, implementation, and evaluation. To use it in practice, start by defining your research question, treatment, outcome, and covariates. Next, collect and clean your data to check for missing values, outliers, and errors. Then choose and estimate a propensity score model to assess its accuracy and validity. After that, select a matching method and evaluate its quality and balance. Finally, choose and estimate a causal effect model to verify its robustness and validity. Once all of these steps are complete, you can report the results and discuss their limitations and implications.

7 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?