The Impacts of Soft Affirmative Action: Experimental Evidence

Baseline: No affirmative action for the minority group during the hiring process.
Treatment (Intervention): There is a quota for minority candidate during the hiring process

2021-06-30

2021-10-31

1) the probability of being selected by a majority/minority employer, controlled by the share of a minority among the pool of the candidate
2) the mean differences in estimated results between the majority candidates and the minority
candidates.

1) The interaction term: minority x relative adjusted scores (majority-minority) may have a negative relationship with 1) the probability of a minority to be selected, and positive relationship with 2) the differences in estimated scores between the majority and minority candidate.

2) a) AA treatment (NIC): We detect statistical discrimination if the mean differences in estimation (majority candidate – minority candidate) is positive and significant

3) we detect taste-based discrimination if the differences in estimation (majority candidate – minority candidate) are greater for the majority employers

The experiment consists of 3 parts. The first one is the preliminary phase to capture the demographic information of all the participants. All the participants will finish a questionnaire to continue the experiment, we elicit the information to classify in-group and out-group participants and capture their personal characteristics. In part 2, participants will engage in a 4-letter word anagram task and a risk attitude task. The second part is a hiring game that includes 3 stages with 2 tasks. One player from the majority group and one player from the minority group will play the role of the employer while the rest will play the role of potential employees. The first stage is the interview task where the intervention of affirmative action policy will be introduced and determine 4 potential employees to enter the next stage where the intervention treatment will be introduced. The second stage is the hiring game where employers will choose 2 candidates to form a group and finish an output task. The thrid stage is the estimation task where employers will estimate the performance of their selected candidates based on the group scores in the output task.

Part 1:
a) Practice task: 4-letter word anagram
b) Task 1: Piece-rate task: capture the differences in individual ability under the piece-rate settings ($0.25 per correct answer)
c) Task 2: risk attitudes

Part 2:
a) Preliminary stage: Each session has 8-12 subjects. 1 majority and 1 minority will be randomly selected as “employers”. The other 8-10 are potentials. A noisy signal will be introduced: 50% of participants will be "lucky" with a 1.25 multiplier and 50% of participants will be "unlucky" with a 0.75 multiplier. The noisy signal will be used in stage 1 of the hiring game

b) stage 1(interview task): Only the potentials will finish another task (task 3). And they will have an adjusted score (assigned multiplier times actual scores in task 3). Potentials get no payment from task 3, but the results of task 3 will determine if they can continue the experiment and earn more.

- Baseline treatment: Only the top 4 best performers enter stage 2. The other potentials will finish the experiment
- AA policy treatment: The best-performing (2) minority will enter stage 2. The rest potentials will finish the experiment.

c) stage 2 (output task): Employers are required to select 2 potentials as “candidates” to form a group. They will be given the identity, their adjusted scores in task 1 and the field of education). The candidates will be required to finish another 4-letter word anagram task (task 4). Those who are not selected will finish the experiment. Employers get paid based on the group scores of his/her candidates. Candidates get paid based on their actual scores in task 4.

d) stage 3 (estimation task): Employers will be asked to estimate the contribution of each candidate. Employers will earn a $5 - deviation for each estimation. There is a range for the estimation payment: the maximum will be a $10 bonus and the minimum will be a $10 penalty ($5 per each estimation).

Randomisation was done by a computer through O-tree

Individual participant

No

300-500 individual participants

300-500 observations

300-500 individual participants