Categories → Causal Inference , Statistics , Method
Propensity score estimation methods are used in causal inference to reduce bias in observational studies, and include propensity score matching, stratification, and weighting.
In Causal Wizard, propensity score methods are offered during the Estimation process.
Propensity score methods are a popular approach for reducing bias in observational studies when estimating causal effects. A good (academic, but easy to read) introduction to the topic can be found in this article, which also covers their use in observational studies.
The basic idea is to estimate the probability (propensity score) of being in the treatment group for each individual in the study based on observed covariates. This probability can then be used to balance the treatment and control groups on measured covariates, reducing confounding and improving causal inference. There are several methods for using propensity scores, including propensity score matching, propensity score stratification, and propensity score weighting.
Propensity score matching involves pairing each individual in the treatment group with one or more individuals in the control group who have similar propensity scores. This creates a matched sample where the distribution of covariates is similar between the two groups. The treatment effect can then be estimated as the difference in outcomes between the treatment and matched control groups. Propensity score matching can be implemented using various algorithms, such as nearest neighbor matching, kernel matching, and exact matching.
Propensity score stratification involves dividing the study population into strata based on the estimated propensity score, such that individuals within each stratum have a similar propensity for treatment. The treatment effect can then be estimated within each stratum and combined using weighted averaging or regression adjustment. This approach can provide more precise estimates of treatment effects compared to matching when the sample size is small or the propensity score distribution is highly skewed.
Propensity score weighting involves assigning weights to each individual in the study based on their propensity score, such that the weighted distribution of covariates is similar between the treatment and control groups. The treatment effect can then be estimated using weighted regression models or inverse probability weighting. This approach is less restrictive than matching or stratification and can accommodate complex study designs, such as clustering or survey sampling.
The most common form of propensity score weighting you'll encounter is Inverse Probability of Treatment Weighting (IPTW).
In summary, propensity score methods are a flexible and widely used approach for addressing confounding in observational studies. Propensity score matching, stratification, and weighting are three commonly used methods that differ in their implementation and assumptions. The choice of method depends on the research question, the availability and quality of covariate data, and the underlying distribution of the propensity score.
The choice of propensity score method should be based on the research question, sample size, distribution of the propensity score, and availability and quality of covariate data, with propensity score matching preferred for small samples and stratification or weighting preferred for larger samples with skewed propensity score distributions or multiple treatment levels.