Use propensity score matching when you want to compare outcomes between a treated group and a control group in observational data. This method helps you create a 'fair' comparison by pairing individuals who share similar characteristics, reducing the risk that pre-existing differences skew your results.
Understanding the Matching Process
Propensity score matching operates by condensing multiple confounding variables into a single score. By using a 1:1 nearest neighbor greedy algorithm, the method pairs each treated individual with their closest match from the control pool. This process is restricted by a caliper to ensure matches are of high quality, preventing the pairing of subjects that are fundamentally different.
The effectiveness of this method relies on the 'ignorability' assumption, which posits that all factors influencing both the treatment and the outcome have been measured. While the tool manages the technical matching, you should ensure that your chosen covariates capture the major drivers of the treatment assignment.
Validation Through Post-Checks
The platform includes built-in post-checks to ensure the integrity of your results. The tool evaluates common support—the degree to which the treated and control groups overlap—and flags the process if a significant portion of the treated group cannot be matched. This prevents the extrapolation of effects to groups where insufficient data exists.
Additionally, the tool produces a balance table detailing the standardized mean difference (SMD) for each covariate before and after matching. If any variable fails to reach the threshold of |SMD| < 0.1, the tool provides a warning that the conditions for causal inference may be compromised.
Interpreting Your Results
Once matching is complete, the tool calculates the Average Treatment Effect on the Treated (ATT) by comparing the matched outcomes of both groups. This estimate is accompanied by bootstrap-calculated confidence intervals and p-values to clarify the statistical significance of the findings.
When reading your output, always prioritize the balance and support reports provided by the engine. If the tool triggers a concern flag, the causal interpretation of the ATT should be approached with caution, as it implies that the matching process did not successfully eliminate the bias introduced by your confounding variables.
1 · Intent → method
An LLM picks causal_psm from a fixed catalog.
2 · Method → numbers
Deterministic Python engine runs the math. Same input → same output.
3 · Numbers → plain language
A second LLM translates the result into your domain’s vocabulary.
How does this method handle group differences?
Propensity score matching calculates a probability of receiving treatment for every individual based on your covariates. It then pairs individuals with similar probabilities, ensuring that the treated and control groups look as alike as possible before calculating the treatment effect.
What happens if the balance is not achieved?
Lattice automatically performs a post-check on covariate balance. If the standardized mean difference (SMD) remains above 0.1 after matching, the tool flags a balance concern, indicating that your conditions for independent assignment are not sufficiently met.
Tool input schema
Schema for causal_psm not exported yet (run pnpm export:registry).