Please rotate your device to landscape mode to view the charts.

Automatic Debiased Machine Learning of Causal and Structural Effects

Journal: Econometrica

Date: 20220501

Author: Chernozhukov, Victor; Newey, Whitney K.; Singh, Rahul

Abstract:
Many causal and structural effects depend on regressions. Examples include policy effects, average derivatives, regression decompositions, average treatment effects, causal mediation, and parameters of economic structural models. The regressions may be high-dimensional, making machine learning useful. Plugging machine learners into identifying equations can lead to poor inference due to bias from regularization and/or model selection. This paper gives automatic debiasing for linear and nonlinear functions of regressions. The debiasing is automatic in using Lasso and the function of interest without the full form of the bias correction. The debiasing can be applied to any regression learner, including neural nets, random forests, Lasso, boosting, and other high-dimensional methods. In addition to providing the bias correction, we give standard errors that are robust to misspecification, convergence rates for the bias correction, and primitive conditions for asymptotic inference for estimators of a variety of estimators of structural and causal effects. The automatic debiased machine learning is used to estimate the average treatment effect on the treated for the NSW job training data and to estimate demand elasticities from Nielsen scanner data while allowing preferences to be correlated with prices and income.

Link: Google Scholar

Key Findings

Automatic Debiasing Method

Developed a new automatic debiased machine learning (Auto-DML) method that corrects for bias in high-dimensional estimation while allowing any machine learning approach including neural networks, random forests, and Lasso.

Robust Performance

Method provides consistent results across different machine learning techniques for estimating average treatment effects, with similar estimates from neural nets, random forests, and Lasso approaches.

Economic Applications

Successfully applied to estimate job training effects and consumer demand elasticities, demonstrating practical value for economic analysis with high-dimensional data.

Treatment Effects Across Machine Learning Methods

  • Consistent ATET estimates across different ML methods using NSW control group
  • Neural networks show slightly lower estimates compared to Lasso and Random Forests
  • Standard errors are comparable across methods

Price Elasticity Estimates

  • Milk and soda show similar own-price elasticity patterns
  • Auto-DML estimates are lower than cross-sectional estimates
  • Results suggest importance of controlling for consumer heterogeneity

Contribution and Implications

  • Provides a practical method for debiasing machine learning estimates that can be applied to various economic analyses
  • Demonstrates robust performance across different machine learning methods, making it versatile for different applications
  • Offers particular value for analyzing high-dimensional economic data where traditional methods may be insufficient

Data Sources

  • Treatment effects visualization based on Table II: "ATET Using NSW Treatment and NSW Control, by Auto-DML"
  • Price elasticity visualization based on Table V: "Average Own-Price Elasticity, by Auto-DML"
  • Results focus on specification 3 (high-dimensional) from the treatment effects analysis and final elasticity estimates