Applying a propensity score-based weighting model to interrupted time series data: improving causal inference in...

Area of Science:

• Econometrics
• Policy Analysis
• Statistical Modeling

Background:

• Researchers often use aggregated time series data for program evaluations and policy change studies.
• Basic analytic models with single groups can have remaining sources of bias.
• Control groups strengthen causal inference but require comparability on pre-intervention factors.

Purpose of the Study:

• To introduce a propensity score-based weighted regression model for analyzing aggregated time series data.
• To overcome limitations of basic models and improve causal inference in policy evaluations.
• To offer a flexible and accessible alternative to existing methods like Synthetic Control.

Main Methods:

• Propensity score-based weighted regression model.
• Weighting control groups to represent the counterfactual outcome of the treatment group.
• Application to cigarette sales data in California pre- and post-Proposition 99.

Main Results:

• The propensity score-weighted regression model was applied to analyze cigarette sales data.
• Results were comparable to the Synthetic Control method.
• The proposed weighting approach demonstrated technical simplicity and flexibility.

Conclusions:

• The propensity score-based weighted regression model offers a robust and accessible method for causal inference with aggregated time series data.
• This technique is less complex than the Synthetic Control method and easily implemented in standard statistical software.
• The model accommodates multiple treatment units and diverse treatment effect estimators, enhancing its utility in policy evaluation research.