The Impact of Algorithmic Decision-Making in Labor Market

How to cite

Cai, Cielo; Khavas, Sarah Rezaei; Fumagalli, Elena. 2024. "The Impact of Algorithmic Decision-Making in Labor Market." AEA RCT Registry. AEARCTR-0013580. https://doi.org/10.1257/rct.13580-1.0

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

Cielo Cai* , Sarah Rezaei Khavas , Elena Fumagalli

* Primary investigator

Registry ID
AEARCTR-0013580
DOI
https://doi.org/10.1257/rct.13580-1.0
First registered
Last updated
Status
On going
Trial dates
2024-04-27 – 2024-08-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/13580

Abstract

The adoption of AI in decision-making has seen rapid advances, especially in the context of the hiring process (Chalfin et al., 2016; Raghavan et al., 2020). Algorithms can accelerate the hiring process as well as improve the efficiency. However, ethical concerns have emerged due to potential bias and discrimination in hiring, posing more risks to vulnerable groups (Obermeyer et al., 2019; Giermindl et al., 2022). This project aims to comprehend how individuals adapt to and interact with algorithms in the labor market, focusing on how workers and HR managers interact with hiring algorithms. To achieve this goal, we will conduct several online, and lab experiments and collect data from the students population in the Netherlands as well as workers in online labor markets (i.e. Mturk, Prolifics) from the US and UK. In particular, this study serves as a pilot to initiate the previously mentioned project.

Intervention

We want to explore the impact of willful ignorance on HR managers' decisions when they receive recommendations from hiring algorithms. In the experiment, participants need to select high-potential workers for a particular task, and they will be provided with a series of information, includes the workers' basic performance information and the recommendation from an algorithm. They are informed that the algorithm might be biased, and they can acquire more information about each worker's performance and overrule or accept the algorithmic decision without further examination. HR managers in the control group have free access to the detailed information about each worker's performance, while in the treatment group, accessing this information is costly.

Experimental design

This pilot includes two phases of experiments: the Workers' experiment and the HR Managers' experiment.

In the Workers' experiment, workers will perform a few real-effort tasks and we will generate workers' performance data based on their performance and their demographic data.

In the HR managers' experiment, HR managers need to select high-potential workers. They will receive the hiring table, includes the workers' demographic and task performance, the recommendation from the hiring algorithm, and information regarding the potential biases of the algorithm. The HR manager can acquire more information about each worker's performance and overrule or accept the algorithmic decision without further examination.

Design details

Randomization method
We will program on Qualtrics and randomization will therefore be done by Qualtrics
Randomization unit
We will randomize in the individual level, each participant in the HR managers' experiment will be randomly assigned to either the control group or the treatment group.
Clusters
We will have two groups, the control group and the treatment group.
Observations
We plan to recruit 250 participants for this pilot, 100 for workers' experiment, 75 each for the control group and the treatment group for the HR managers' experiment.
Arms
In Workers' experiment, we collect participants' performance to generate the performance data, therefore we will recruit 100 participants to act as workers. In HR managers' experiment, we will have two groups, one control group and one treatment group, and each group will have 75 participants.
Primary outcomes
1. The time each HR manager spent on the hiring task 2. The number of times each HR manager clicked to obtain additional information 3. The share of women of the workers selected by each HR manager 4. The comparison of the average real Raven performance for workers selected by humans and algorithms respectively 5. The role of social preference in the worker selection process

Cited by

No citation data available for this registration yet.

Keywords