We use cookies to provide you with the best possible website experience. This includes cookies that are necessary for the operation of the site, as well as cookies used for anonymous statistics, comfort settings, or displaying personalized content. You can decide which categories you want to allow. Please note that depending on your settings, some features of the website may not be available.

Cookie settings

These necessary cookies are required to enable the core functionality of the website. Opting out of these cookies is not possible.

cb-enable
This cookie stores the user's cookie consent status for the current domain. Expiry: 1 year.
laravel_session
Stores the session ID to recognize the user when the page reloads and to restore their login session. Expiry: 2 hours.
XSRF-TOKEN
Provides CSRF protection for forms. Expiry: 2 hours.
IZA Discussion Paper No. 19106
November 2026
Variable Importance Recovery in Tabular Foundation Models: A Simulation Study

In machine learning, identifying which variables contribute predictive information can be as important as predictive accuracy. Tabular foundation models have matched or exceeded the predictive performance of tuned gradient boosting trees on many small- and medium-sized tabular datasets. Yet their ability to recover population-level variable importance has received little attention. We design and conduct a Monte Carlo simulation study to compare variable importance recovery across tabular foundation models and established machine learning baselines. Our framework distinguishes multiple population estimands, pairs each one with an aligned model-agnostic importance estimator, and evaluates recovery in terms of ranking, support, and relative magnitude. Our results show that tabular foundation models are generally competitive with established machine-learning baselines in recovering population variable importance, and often more accurate. We thus extend the evidence for their capabilities to a previously underexamined dimension, further strengthening their appeal.

Kommunikation
Mark Fallak
mark.fallak@liser.lu
+352 585-855-526
World of Labour
Olga Nottmeyer
olga.nottmeyer-ext@liser.lu
+352 585-855-501
Netzwerkkoordination
Christina Gathmann
christina.gathmann@liser.lu

Das IZA@LISER-Netzwerk ist eine weltweite Gemeinschaft für exzellente Forschung in der Arbeitsmarktökonomie und angrenzenden Fachgebieten. Nach dem Wechsel von Bonn wird das Netzwerk nun am Luxembourg Institute of Socio-Economic Research (LISER) koordiniert.

Über das IZA@LISER Network
Contact
IZA@LISER NETWORK (Current Site Operator):

Luxembourg Institute of Socio-Economic Research (LISER)
11, Porte des Sciences
Maison des Sciences Humaines
L-4366 Esch-sur-Alzette / Belval, Luxembourg

IZA Institute (In Liquidation):

Forschungsinstitut zur Zukunft der Arbeit GmbH i. L.
Schaumburg-Lippe-Str. 5-9, 53113 Bonn. Germany
Phone: +49 228 3894-0 | Fax: +49 228 3894-510
E-Mail: info@iza.org | Web: www.iza.org
Represented by: Martin T. Clemens (Liquidator)