Difference-in-differences analysis with repeated cross-sectional survey data.

Area of Science:

• Econometrics
• Public Health Policy
• Survey Methodology

Background:

• Traditional difference-in-differences (DiD) methods struggle with policy evaluation using repeated cross-sectional (RCS) data.
• Challenges include heterogeneous sample compositions over time and reliance on sampled, not population-level, data.
• Accurate estimation of the average treatment effect on the treated (ATT) is often compromised.

Purpose of the Study:

• To develop a robust method for estimating policy effects using RCS data.
• To address limitations of traditional DiD in the presence of time-varying sample compositions.
• To identify a policy-relevant target estimand and its identification conditions.

Main Methods:

• Proposed a novel weighting approach combining propensity score estimation and survey weights.
• Established theoretical properties of the new weighting method.
• Conducted simulations to evaluate finite-sample performance.

Main Results:

• The proposed weighting method successfully addresses challenges in DiD analysis with RCS data.
• Demonstrated accurate estimation of the average treatment effect on the treated (ATT) under specified conditions.
• Simulation results confirmed the method's finite-sample performance.

Conclusions:

• The developed method provides a reliable approach for policy effect evaluation using RCS survey data.
• Applicable in scenarios with evolving unit compositions and incomplete population data.
• Successfully applied to estimate the impact of a beverage tax on adolescent soda consumption in Philadelphia.