Using Machine Learning Methods to Support Causal Inference in Econometrics

Achim Ahrens, Christopher Aitken, Mark Edwin Schaffer

Research output: Contribution to journalArticle


We provide an introduction to the use of machine learning methods in econometrics and how these methods can be employed to assist in causal inference. We begin with an extended presentation of the lasso (least absolute shrinkage and selection operator) of Tibshirani (1996). We then discuss the ‘Post-Double-Selection’ (PDS) estimator of (Belloni et al., 2012, 2014b) and show how it uses the lasso to address the omitted confounders problem. The PDS methodology is particularly powerful for the case where the researcher has a high-dimensional set of potential control variables, and needs to strike a balance between using enough controls to eliminate the omitted variable bias but not so many as to induce overfitting. The last part of the paper discusses recent developments in the field that go beyond the PDS approach.
Original languageEnglish
JournalStudies in Computational Intelligence
Publication statusAccepted/In press - 8 Jan 2020
Event13th International Conference of the Thailand Econometric Society 2020 - Chiang Mai University, Chiang Mai, Thailand
Duration: 8 Jan 202010 Jan 2020


  • Causal inference
  • Lasso
  • Machine Learning

Fingerprint Dive into the research topics of 'Using Machine Learning Methods to Support Causal Inference in Econometrics'. Together they form a unique fingerprint.

  • Cite this