A propensity score represents the probability that a subject will receive a specific treatment given their observed characteristics. Researchers use propensity score matching to pair treated and untreated subjects with similar scores. This creates comparable groups and reduces bias from confounding variables. The propensity score model is especially useful in observational studies where selection bias can skew results.
One key propensity score technique is propensity score regression, where the propensity score is included as a covariate in a regression model. This approach helps to reduce the impact of confounding variables, resulting in a more accurate estimation of treatment effects. Another approach is propensity score weighting, which assigns weights to subjects based on their propensity scores to create a synthetic sample where the treatment assignment is independent of the observed covariates.
In situations where direct matching is not feasible, propensity adjustment can be used. This involves using the propensity score to adjust the treatment effect estimation. Usually, propensity scores are estimated using logistic regression, which ensures they accurately reflect the likelihood of treatment assignment based on the observed data.
Advantages of propensity scoring include reducing selection bias, improving causal inference, and enhancing the validity of observational study results. These methods allow researchers to draw more reliable conclusions about treatment effects, ultimately improving the quality of observational research.
List of recommended resources #
For a broad overview
Propensity Scores: Uses and Limitations
This article by Kristin L. Sainani reviews propensity score methods, including what they are, how they work, and their advantages and limitations. It also takes two examples to illustrate the use of propensity scores clearly.
Getting Started with Personalization through Propensity Scoring
This blog post by Databricks gives a brief on how propensity scores helps in personalization of data as well as provides a good on building a propensity scoring workflow.
For in depth understanding
This paper by Peter C. Austin provides an in-depth understanding of propensity scores, focusing on the four different propensity score methods: matching on the propensity score, stratification on the propensity score, inverse probability of treatment weighting using the propensity score, and covariate adjustment using the propensity score.
Understanding Propensity Score Analyses
This paper by Nafisha Lalani, Rachel B. Jimenez, and Beow Yeap gives a broad overview of the observational method of propensity score analysis. The paper takes the help of an example of conducting a study to assess the association between radiation therapy technique and partial breast irradiation and the risk of local recurrence for patients with ductal carcinoma in situ of the breast to explain the method.
Case study
Are Global Value Chains Women Friendly in Developing Countries?: Evidence from Firm-Level Data
This paper by Marize Kalliny and Chahir Zaki aims at examining the impact of global value chains on women’s trade participation as entrepreneurs and employees. Propensity score estimation method is used, among others, to analyze the data.
Joseph Shapiro and Jorge Moreno Trevino use propensity score matching to evaluate the effectiveness of CONAFE, a compensatory education program in Mexico, in improving student test scores and lowering repetition and failure rates.
References #
Propensity Score: Construction & Evaluation
