New Technologies and Productivity: A Study with Vocational Students

How to cite

Brosch, Hanna; Langer, Christina; Lergetporer, Philipp; Pfeifer, Harald; Tschöpe, Nele. 2025. "New Technologies and Productivity: A Study with Vocational Students." AEA RCT Registry. AEARCTR-0016936. https://doi.org/10.1257/rct.16936-1.0

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

Hanna Brosch* , Christina Langer , Philipp Lergetporer , Harald Pfeifer , Nele Tschöpe

* Primary investigator

Registry ID
AEARCTR-0016936
DOI
https://doi.org/10.1257/rct.16936-1.0
First registered
Last updated
Status
In development
Trial dates
2025-10-06 – 2026-12-31
Countries
Private
JEL codes
I21, J24, O33
Registry record
http://www.socialscienceregistry.org/trials/16936

Abstract

Artificial intelligence (AI) has the potential to transform the nature of work, yet little is known about how generative AI (GenAI) affects middle-skilled workers. Apprentices, who epitomize middle-skilled workers, provide an ideal setting to study these effects during the formative stage of skill development. In this study, we examine the impact of GenAI on apprentices’ productivity in Germany across three occupations: IT specialists for system integration, IT specialists for application development, and clerks in digitalization management. We design a randomized controlled trial (RCT) in which final-year apprentices are randomly assigned to either a treatment group with access to GenAI tools or a control group with no internet resources. All participants first solve a common baseline task, which allows us to compare performance across occupations and provides us with our own measure of individual ability. Subsequently, each occupation receives a middleand two high-skilled task specific to their field. The experiment evaluates whether GenAI enhances apprentices’ ability to complete middle- and especially higher-skilled tasks and whether it affects other task-related outcomes, like their understanding of the tasks at hand. In this pre-analysis plan, we present the study design, research questions, and empirical strategy for this project.

Intervention

We conduct a RCT (lab-in-the-field experiment) in vocational training schools across Germany. The whole experiment takes place during regular school hours and lasts approximately 1.5 hours (equivalent to two school hours).

Apprentices are randomly assigned to either the treatment group or the control group with equal probability (p = 1/2). Randomization is implemented by drawing a lot, after which apprentices complete the experiment on a computer provided by the research team and work on tasks with different difficulty levels.

Experimental design

Our experiment proceeds in the following steps:
1. Apprentices draw a lot to determine their treatment status.
2. Based on the draw, apprentices are assigned to a computer labeled with a number
indicating whether it belongs to the treatment or control group. Whenever capacity
allows, we separate treatment and control apprentices and let them participate in
the experiment in two different rooms to minimize spillovers.
3. Each apprentice then participates in the experiment on the computer, consisting of
the following components:
• Pre-survey, consisting of a demographic questionnaire (5 minutes)
• Baseline-task (10 minutes)
• Treatment information: apprentices in the treatment group are informed that
they will have access to ChatGPT for the subsequent tasks
• One medium- and two high-skilled tasks (20 minutes each), presented in randomized
order After each of the four tasks (baseline-task, medium- and two high-skilled),
apprentices answer questions about their task experience and respond to comprehension
questions
• Post-study questionnaire (10 minutes)
The treatment group receives access to ChatGPT (with access to GPT-4o and GPT-5). Treatment computers are pre-logged into ChatGPT Plus accounts. The control group does not have access to GenAI, and their computers are restricted from accessing the internet and any generative AI tools.

Design details

Randomization method
Randomization done by drawing a lot.
Randomization unit
Individual level (student)
Clusters
800 apprentices (400 per treatment arm)
Observations
800 apprentices (400 per treatment arm)
Arms
800 apprentices (400 per treatment arm)
Primary outcomes
1. Productivity (a) Task completion time (b) Correctness of answers 2. Task comprehension 3. Positive task experience Further, we plan to perform two main heterogeneities analysis by: - AI Literacy - Educational Attainment and Ability
Secondary outcomes
We may conduct additional exploratory heterogeneity analyses by apprentices’ effort they put into each task and their confidence about the correctness of their result.

Cited by

No citation data available for this registration yet.

Keywords