Double machine learning and automated confounder selection: A cautionary tale

Hünermund, Paul; Louw, Beyers; Caspi, Itamar

doi:10.1515/jci-2022-0078

Cited by 8 publications

(4 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In particular, this means we need to be careful not to include mediators that have an unobserved common cause with Y or that we introduce a common effect of T and Y. Both cases would open a new path and substantially bias the estimation [70].…”

Section: Dml For Hybrid Modeling-a Causal Perspectivementioning

confidence: 99%

Causal hybrid modeling with double machine learning—applications in carbon flux modeling

Cohrs,

Varando,

Carvalhais

et al. 2024

Mach. Learn.: Sci. Technol.

View full text Add to dashboard Cite

Hybrid modeling integrates machine learning with scientific knowledge to enhance interpretability, generalization, and adherence to natural laws. Nevertheless, equifinality and regularization biases pose challenges in hybrid modeling to achieve these purposes. This paper introduces a novel approach to estimating hybrid models via a causal inference framework, specifically employing Double Machine Learning (DML) to estimate causal effects. We showcase its use for the Earth sciences on two problems related to carbon dioxide fluxes. In the Q10 model, we demonstrate that DML-based hybrid modeling is superior in estimating causal parameters over end-to-end deep neural network (DNN) approaches, proving efficiency, robustness to bias from regularization methods, and circumventing equifinality. Our approach, applied to carbon flux partitioning, exhibits flexibility in accommodating heterogeneous causal effects. The study emphasizes the necessity of explicitly defining causal graphs and relationships, advocating for this as a general best practice. We encourage the continued exploration of causality in hybrid models for more interpretable and trustworthy results in knowledge-guided machine learning.

show abstract

Section: Dml For Hybrid Modeling-a Causal Perspectivementioning

confidence: 99%

Causal hybrid modeling with double machine learning—applications in carbon flux modeling

Cohrs,

Varando,

Carvalhais

et al. 2024

Mach. Learn.: Sci. Technol.

View full text Add to dashboard Cite

show abstract

“…The covariate set was selected based on economic reasoning on the causal effect paths between a farm's AES participation and the respective outcome (Uehleke et al, 2022) and previous studies on AES participation (e.g., Pufahl & Weiss, 2009). By choosing the covariate set in this manner, we reduced the possibility of matching on covariates that might have increased bias of the effect estimates (Hünermund et al, 2022;Wooldridge, 2016).…”

Section: Datamentioning

confidence: 99%

Counterfactual evaluation of two Austrian agri‐environmental schemes in 2014–2018

Uehleke,

Leonhardt,

Hüttel

2023

Agricultural Economics

View full text Add to dashboard Cite

This article investigates the causal effect of farm participation in two Austrian agri‐environmental schemes (AES), Immergrün (ground cover) and Zwischenfrucht (catch cropping), on fertilizer and plant protection expenditures in the 2014 programming period. Combining European Farm Accountancy Data Network data with information on scheme participation from administrative control data offers identifying farm participation in specific schemes targeted at reducing input intensity. Given the overall small sample, we maximized the utilizable sample size by combining difference‐in‐difference and kernel matching with automated bandwidth selection. To address the remaining post‐matching covariate imbalances, we used double machine learning (DML) techniques for a guided selection of potential confounding covariates. Our results suggest that, given the available sample, we cannot substantiate moderate effects of AES participation, and that guided covariate selection by DML offers no gain over non‐guided covariate selection for the small sample. Our results underline the need to increase the number of farms and the duration in available farm panels to substantiate future counterfactual‐based evaluations of policy.

show abstract

“…These estimators are robust from a statistical standpoint, but not necessarily a causal identification one. The researcher must know which variables are possible confounders and to include them in the appropriate models, while not including colliders or mediators (Hünermund et al, 2023). The simulations discussed in this paper assume conditional ignorability; rather than testing what happens when models are missing important covariates, it focuses on accurate specification of the functional form of the treatment and outcome models.…”

Section: Conceptual Overviewmentioning

confidence: 99%

Double Robust, Flexible Adjustment Methods for Causal Inference: An Overview and an Evaluation

Hoffmann

2023

Preprint

View full text Add to dashboard Cite

Double robust methods for flexible covariate adjustment in causal inference have proliferated in recent years. Despite their apparent advantages, these methods remain underutilized by social scientists. It is also unclear whether these methods actually outperform more traditional methods in finite samples. This paper has two aims: It is a guide to some of the latest methods in double robust, flexible covariate adjustment for causal inference, and it compares these methods to more traditional statistical methods. It does this by using both simulated data where the treatment effect estimate is known, and then using comparisons of experimental and observational data from the National Supported Work Demonstration. Methods covered include Augmented Inverse Propensity Weighting, Targeted Maximum Likelihood Estimation, and Double/Debiased Machine Learning. Results suggest that these methods do not necessarily outperform OLS regression or matching on propensity score estimated by logistic regression, even in cases where the data generating process is not linear.

show abstract

Double machine learning and automated confounder selection: A cautionary tale

Cited by 8 publications

References 35 publications

Causal hybrid modeling with double machine learning—applications in carbon flux modeling

Causal hybrid modeling with double machine learning—applications in carbon flux modeling

Counterfactual evaluation of two Austrian agri‐environmental schemes in 2014–2018

Double Robust, Flexible Adjustment Methods for Causal Inference: An Overview and an Evaluation

Contact Info

Product

Resources

About