[Paper Review] A two-dimensional propensity score matching method for longitudinal quasi-experimental studies: A focus on travel behavior and the built environment

Haotian Zhong, Wei Li+1 more|arXiv (Cornell University)|Nov 2, 2019

Economic and Environmental Valuation19 references4 citations

TL;DR

This paper introduces a two-dimensional propensity score matching (2DPSM) method to estimate treatment effects in longitudinal quasi-experimental studies, particularly for travel behavior and built environment research. By matching treatment and control groups across both time (before/after) and cross-sectional units, 2DPSM improves synthetic control group quality and demonstrates superior performance in Monte Carlo simulations for causal inference in non-randomized settings.

ABSTRACT

The lack of longitudinal studies of the relationship between the built environment and travel behavior has been widely discussed in the literature. This paper discusses how standard propensity score matching estimators can be extended to enable such studies by pairing observations across two dimensions: longitudinal and cross-sectional. Researchers mimic randomized controlled trials (RCTs) and match observations in both dimensions, to find synthetic control groups that are similar to the treatment group and to match subjects synthetically across before-treatment and after-treatment time periods. We call this a two-dimensional propensity score matching (2DPSM). This method demonstrates superior performance for estimating treatment effects based on Monte Carlo evidence. A near-term opportunity for such matching is identifying the impact of transportation infrastructure on travel behavior.

Motivation & Objective

• To address the scarcity of longitudinal studies on built environment and travel behavior.
• To overcome limitations of standard propensity score matching in quasi-experimental longitudinal designs.
• To develop a method that enables synthetic control group creation across both time and cross-sectional dimensions.
• To improve causal effect estimation in non-randomized studies using matched longitudinal data.
• To provide a robust method for assessing infrastructure impact on travel behavior.

Proposed method

• The method extends standard propensity score matching to two dimensions: time (before/after treatment) and cross-sectional units (treatment vs. control groups).
• It matches treated units to control units based on observed covariates in both time and spatial dimensions to create balanced synthetic control groups.
• The approach uses propensity scores estimated from pre-treatment covariates to match units across time periods and across individuals or locations.
• Matching is performed simultaneously on time and cross-sectional units, ensuring similarity in both dimensions.
• The method leverages Monte Carlo simulations to validate its performance in estimating average treatment effects.
• It enables researchers to mimic randomized controlled trials by constructing synthetic control groups that are comparable in both time and space.

Experimental results

Research questions

• RQ1 How can propensity score matching be extended to handle longitudinal quasi-experimental data with both time and cross-sectional dimensions?
• RQ2 What is the performance of two-dimensional matching in estimating treatment effects compared to standard methods?
• RQ3 Can 2DPSM effectively estimate the impact of transportation infrastructure on travel behavior?
• RQ4 How well does 2DPSM balance covariates across time and space in synthetic control groups?
• RQ5 What are the conditions under which 2DPSM outperforms traditional matching approaches in longitudinal settings?

Key findings

• The 2DPSM method significantly improves the balance of covariates across both time and cross-sectional dimensions compared to standard methods.
• Monte Carlo evidence shows that 2DPSM produces more accurate and less biased estimates of average treatment effects.
• The method effectively reduces selection bias in longitudinal quasi-experimental studies by matching on two dimensions simultaneously.
• 2DPSM enables more reliable causal inference in studies of built environment effects on travel behavior where RCTs are infeasible.
• The approach is particularly effective for assessing infrastructure impacts on travel behavior using non-randomized data.
• The method demonstrates robustness in simulations, maintaining low bias and mean squared error in treatment effect estimation.