Combating unmeasured confounding in cross-sectional studies: evaluating instrumental-variable and Heckman selection...

Area of Science:

• Econometrics
• Biostatistics
• Social Sciences

Background:

• Unmeasured confounding poses a significant challenge to estimating treatment effects in nonexperimental research.
• It arises from unmeasured individual characteristics influencing both treatment assignment and outcome.
• Existing methods often struggle to adequately address this bias.

Purpose of the Study:

• To introduce and compare econometric techniques for controlling unmeasured confounding.
• To evaluate the performance of the Heckman selection model (HSM) and instrumental-variable regression (IVR) against ordinary least squares (OLS).
• To demonstrate the application of these methods in analyzing real-world data.

Main Methods:

• Monte Carlo simulation to compare OLS, IVR, and HSM under various confounding conditions.
• Application of HSM and IVR with OLS to analyze cross-sectional data.
• Empirical analysis using General Social Survey data (2006-2010) to examine the marriage-well-being association.

Main Results:

• HSM generally outperformed IVR in terms of mean-square-error and power to detect treatment effects or confounding.
• Both HSM and IVR require large sample sizes to be effective.
• The study demonstrated the practical utility of HSM and IVR in addressing unobserved confounding bias.

Conclusions:

• HSM is a valuable tool for mitigating unmeasured confounding bias in nonexperimental research.
• IVR is also effective but generally less performant than HSM.
• Both methods necessitate substantial sample sizes for optimal results and can be used in conjunction with OLS.