At the Competence-Centre on Microeconomic Evaluation (CC-ME) we advise and support EU policy making through ex-post causal evaluation and data-driven microeconomic analysis. The CC-ME hosts a Microeconometric Seminar Series to promote discussions with external researchers from academia and other institutions. Our Seminar Series is intended to disseminate advanced research methodologies and topics in the field of microeconomic evaluation. To further disseminate the benefits of our Series across the JRC, we post a summary of the presented papers together with the presenters' views and opinions on their research and the future of the field of Applied Economics. |
Last week we had the pleasure of having Professor Michal Kolesár from Princeton University present his paper on The Fragility of Sparsity.
In the context of statistics, econometrics and machine learning sparsity refers to the situation when an outcome variable expressed as the linear combination of other variables produces the number of estimated parameters to be comparable or larger to the sample size. Furthermore, the researcher believes that some of those parameters could be irrelevant and needs a method that helps separate the relevant ones from the rest.
Using three empirical applications, Kolesár argues that linear regression estimators which rely on the assumption of sparsity are fragile in two ways. He first starts by documenting that different choices of the regressor matrix that don't impact long regression---the one including all regressors---estimates, such as the choice of baseline category with categorical controls, move the post-double-selection estimates by one standard error or more. Second, he develops two tests of the sparsity assumption based on comparing sparsity-based estimators with long regression. He finds that both tests tend to reject the sparsity assumption in all three applications.
The paper concludes that unless the number of regressors is comparable to or exceeds the sample size, long regression yields more robust results at little efficiency cost.
To learn more about Professor Kolesár's work and opinions about the future of the field of Applied Economics we asked him to briefly answer a series of questions. You can find his answer to each of our questions below.
Q: What attracted you to research the topics in your paper? A: There has been an explosion of work in econometrics on developing casual methods that leverage sparsity assumptions (e.g. various variants of "double machine learning"). Our paper tries to understand the robustness of such methods to implementation details as well as the credibility of the sparsity assumption in concrete empirical applications we have seen essentially no work in this direction, and our paper tries to fill that gap. |
Q: Where is the research area where your paper fits moving? A: Our paper shows that the sparsity assumption is violated in the empirical applications we looked at and that seemingly innocuous implementation choices, such as the choice of baseline category with categorical controls can move the estimates around a lot. We would like to see more work on developing methods that are robust to such choices. |
Q: What, in your opinion, will the next breakthrough in Applied Economics be? A: Many of the innovations in recent years have come from access to detailed administrative data, and from being able to link several such datasets. I expect that innovations in what types of data we can access and make use of will continue to drive what types of questions we can make progress on. |
The CC-ME team would like to congratulate Professor Kolesár for his insightful research and thank him for presenting it in our Seminar Series.
For more information on the upcoming presentations and how to participate in our Seminar Series please visit our dedicated website.
| Originally Published | Last Updated | 13 Sep 2023 | 11 Mar 2024 |
| Knowledge service | Metadata | Microeconomic Evaluation |