Expertise, Personal Experience and Algorithm Aversion: Study 2

Last registered on August 23, 2021

View Trial History

Pre-Trial

Trial Information

General Information

Title

Expertise, Personal Experience and Algorithm Aversion: Study 2

RCT ID

AEARCTR-0008072

Initial registration date

August 18, 2021

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published

August 23, 2021, 9:15 AM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Locations

Country

China

Region

Primary Investigator

Name

Yu Gao

Affiliation

Contact Primary Investigator

Other Primary Investigator(s)

PI Name

Chong (Alex) Wang

PI Affiliation

Peking University

Contact Investigator

PI Name

Cong Wang

PI Affiliation

Peking University

Contact Investigator

Additional Trial Information

Status

On going

Start date

2021-07-27

End date

2021-08-28

Keywords

Behavior

Additional Keywords

Algorithm aversion

JEL code(s)

Secondary IDs

Prior work

This trial does not extend or rely on any prior RCTs.

Abstract

Artificial Intelligence (AI) outperforms humans in many tasks. However, accumulating evidence suggests AI aversion – a tendency to distrust and thus ignore AI advice, especially among experts. In a two-stage experiment, we test if trained experts are more acceptive of AI advice in forming predictions and are more sensitive to the AI’s incidental performance. Our results will help design algorithms that are better adopted by human decision-makers, and mitigate the biases that experts hold on algorithms.

External Link(s)

Registration Citation

Citation

Gao, Yu, Chong (Alex) Wang and Cong Wang. 2021. "Expertise, Personal Experience and Algorithm Aversion: Study 2." AEA RCT Registry. August 23. https://doi.org/10.1257/rct.8072-1.0

Sponsors & Partners

Experimental Details

Interventions

Intervention(s)

Intervention Start Date

2021-07-27

Intervention End Date

2021-08-27

Primary Outcomes

Primary Outcomes (end points)

weight on advice (WOA) and change in WOA

Primary Outcomes (explanation)

Weight on advice: the difference between the initial and revised judgment divided by the difference between the initial judgment and advice. WOA of 0% occurs when a participant ignores advice and WOA of 100% occurs when a participant abandons his or her prior
judgment to match the advice.

Secondary Outcomes

Secondary Outcomes (end points)

confidence in self, confidence in AI and self-predicted performance in the task

Secondary Outcomes (explanation)

Experimental Design

In Stage 1, each subject will be randomly assigned to the expert or the laypeople group, where they will receive immediate feedback or delayed feedback regarding their performance in a prediction task, respectively.
In Stage 2, each subject will be exposed to three randomly drawn levels of AI’s performance using the strategy method approach. Each subject will be firstly randomly assigned into the positive (+1 to +18) or the negative feedback (-1 to -18) group, and then we draw three numbers in the corresponding range.

Experimental Design Details

Randomization Method

Randomization will be done by the survey platform.

Randomization Unit

In stage 1, randomization is done at the individual level.
In stage 2, each individual will be assigned into a positive or negative performance condition, and then we draw three feedback levels in each corresponding condition.

Was the treatment clustered?

Experiment Characteristics

Sample size: planned number of clusters

1200 individuals

Sample size: planned number of observations

In stage 1, each individual will provide 10 observations (10 tasks). In stage 2, each individual will provide 3 (aggregate into one Baseline WOA) + 3*3 observations (aggregate into 3 Endline WOA).

Sample size (or number of clusters) by treatment arms

In Stage 1, we will assign 600 individuals to the expert and 600 to the laypeople condition.
In Stage 2, we will assign 600 individuals to the positve and 600 to the negative condition.

Minimum detectable effect size for main outcomes (accounting for sample design and clustering)

Supporting Documents and Materials

IRB