Correcting Regressor-Endogeneity Bias via Instrument-Free Joint Estimation Using Semiparametric Odds Ratio Models.

Area of Science:

• Econometrics
• Statistical Modeling
• Causal Inference

Background:

• Endogenous regressors can bias causal effect estimates when assuming regressor-error independence.
• Existing methods often rely on instrumental variables (IVs) with strict assumptions or struggle with certain regressor types.

Purpose of the Study:

• To propose a novel, flexible, and IV-free method for correcting endogeneity bias.
• To improve the accuracy of causal effect estimation, particularly for discrete endogenous regressors.

Main Methods:

• Utilized flexible semiparametric odds ratio conditional models to account for regressor-error dependence.
• Employed profile likelihood optimization for inference, avoiding parametric distributional assumptions and tuning parameters.
• The method does not require instrumental variables (IVs) or their associated exclusion restriction conditions.

Main Results:

• The proposed approach successfully corrects endogeneity bias without IVs.
• Demonstrated versatility in handling binary, count, and continuous endogenous regressors.
• Achieved improved accuracy in causal effect estimation compared to existing IV-free methods.

Conclusions:

• The novel semiparametric odds ratio model offers a flexible and powerful alternative for endogeneity correction.
• This IV-free approach expands the applicability of causal inference methods, especially for discrete endogenous variables.