Employer Provided AI-Training: The Role of Worker Characteristics, Training Types and Managers.

How to cite

Huber, Katrin; Caliendo, Marco; Pfeifer, Harald. 2026. "Employer Provided AI-Training: The Role of Worker Characteristics, Training Types and Managers.." AEA RCT Registry. AEARCTR-0018671. https://doi.org/10.1257/rct.18671-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Katrin Huber* , Marco Caliendo , Harald Pfeifer

* Primary investigator

Registry ID
AEARCTR-0018671
DOI
https://doi.org/10.1257/rct.18671-1.0
First registered
Last updated
Status
On going
Trial dates
2026-04-01 – 2027-06-30
Countries
Private
JEL codes
J01, J16, J24
Registry record
http://www.socialscienceregistry.org/trials/18671

Abstract

This study analyzes how managers in German firms allocate AI-related training to employees and whether systematic differences arise across worker characteristics. The main outcomes are the probability of training selection and differences by attributes of the training candidate and the training, as well as the role of managerial and firm characteristics. The intervention is a randomized conjoint survey experiment embedded in the BIBB Training Panel 2026. Managers evaluate hypothetical employee profiles with randomly varied attributes and training characteristics, allowing for causal identification of determinants of training allocation. As a complementary secondary component in an additional experimental layer, managers are randomly assigned to one of three information treatments prior to decision-making. This additional intervention is designed to explore how informational contexts may influence training choices. Randomization occurs at the firm, manager, and vignette levels. The population consists of German establishments and managers responsible for training decisions. The sample includes approximately half of the panel’s gross sample (approximately 2,000 firms expected). Eligibility is restricted to managers responsible for training decisions in establishments participating in the BIBB Training Panel 2026. Treatment assignment is fully randomized, ensuring causal inference.

Intervention

A conjoint survey experiment (choice experiment) with a complementary information treatment. In a first step, managers of German establishments are randomly assigned a vignette describing two hypothetical training candidates and the characteristics of the training they would undergo. They then share their expectations of how beneficial the training of each candidate would be for the firm and choose one of the two candidates to undertake the training. In a second step, as a complementary feature, managers are randomly assigned an information treatment within the vignette.

Experimental design

This study implements a randomized conjoint survey experiment embedded in the BIBB Training Panel 2026. Decision-makers in German establishments evaluate a series of hypothetical employee profiles for participation in an AI-related training program. In each choice task, respondents select one candidate and assess expected productivity gains. As a complementary component, respondents are randomly assigned to different information conditions prior to the decision tasks. Randomization occurs at the respondent and profile levels, enabling causal identification of the effects of profile attributes and informational contexts on decision-making. The sample consists of approximately 2,000 managers, each of them making 6 choices.

Design details

Randomization method
Randomization is carried out by a computer-based randomization.
Randomization unit
Randomization occurs at multiple levels: (i) respondent level (information treatment), (ii) vignette level (profile attributes), and (iii) firm level (participation in vignette study).
Clusters
2,000 respondents ( = firm representatives)
Observations
approx. 2,000 resulting in 12,000 choices
Arms
Approximately equal allocation across three information treatment arms (660–670 respondents per arm).
Primary outcomes
Our first primary outcome is the binary choice regarding which of two hypothetical applicants a human resource manager would prefer for the AI training. Our second primary outcome is the manager’s expected benefit for the firm.
Secondary outcomes
Effect-heterogeneity concerning the information treatment, decision makers’ characteristics and firm characteristics.

Cited by

No citation data available for this registration yet.

Keywords