Propensity score interval matching: using bootstrap confidence intervals for accommodating estimation errors of...

Area of Science:

• Medical research
• Causal inference
• Observational studies

Background:

• Propensity score methods are popular for reducing selection bias in medical research.
• Propensity score matching relies on point estimates, lacking a criterion for estimation errors.
• Evaluating matched unit closeness is challenging without knowing propensity score estimation errors.

Purpose of the Study:

• Introduce interval matching to accommodate propensity score estimation errors.
• Develop a more scientifically sound criterion for matching units.
• Reduce selection bias in causal inference from observational studies.

Main Methods:

• Interval matching utilizes bootstrap confidence intervals for propensity scores.
• Units are matched if their confidence intervals overlap between treatment and comparison groups.
• The method is demonstrated using a national dataset from the Centers for Medicare and Medicaid Services.

Main Results:

• Interval matching demonstrated superior reduction in selection bias compared to caliper matching.
• The empirical example showed promising evidence for interval matching's effectiveness.
• Interval matching provides a more meaningful and scientific criterion for unit matching.

Conclusions:

• Interval matching is a promising alternative for reducing selection bias in observational studies.
• This method is particularly useful for secondary data analysis on large national databases.
• The approach enhances causal inference by accounting for estimation errors in propensity scores.