Experimental Design
Participants in the first experiment serve as workers and answer trivia questions from a number of male-type trivia categories. Experiment 2 elicits beliefs on gender difference in performance of participants in the previous experiment. Of key interest here is finding a trivia category where evaluators hold biased beliefs on the gender gap in performance. The data collection for the first two experiment is already complete.
The first two experiments identify sports as the trivia category where there are no gender differences in average performance but where men are believed to outperform women. Workers' performance in the sports trivia quiz becomes the foundation for employer beliefs and hiring decisions which will nor be captured in the third and main experiment of this study.
Participants in the employer experiment will be presented with four randomly selected resumes of a gender-balanced set of workers from experiment 1. Employers’ beliefs will be elicited about expected performance of all presented worker resumes on the sport trivia quiz. They will then be asked to hire two out of this group of four workers. Finally, employers will get feedback about actual performance of both of their hired employees. Employers will proceed with these belief elicitations and hiring decisions for six rounds with different workers in each round, allowing us to see how employer beliefs and hiring decisions update based on feedback. The experiment has two between-subject treatments: a control treatment in which there will be no restrictions on who the employers can hire; and a temporary affirmative action treatment. In the temporary affirmative action treatment, the first three rounds will have a quota policy for women wherein at least one of the two hired employees must be a woman, and this policy is subsequently removed in the last three rounds.
At the end of the round 6 hiring decision stage, employers will enter a final decision stage. Here they will be offered a chance to get costless information on workers’ performance and to potentially revise hiring choices for round 6. This final stage is used to identify employers who do not opt for this costless information for any worker. Such employers are classified as being likely to disregard information about employee performance and this classification allows us to explore heterogeneous effects on hiring and beliefs. It also allows for potential reduction in noise in the data as participants who do not demand the costless information are also likely to be less attentive.
To identify participants more vs. less likely to be biased in their beliefs and discriminate against women in hiring, I will first use data from the control treatment to estimate a logit regression where dependent variable will be an indicator which takes value 1 if two men are hired in round 1, and explanatory variable will be the gender difference in beliefs about performance of workers in round 1. Using this estimation, I will then predict the probability of hiring two men for the control group of employers and determine a cutoff point based on the top 25\% of employers most likely to hire two men in round. This cutoff will then then be used to classify two subgroups within the control group of employers. To achieve a comparable classification for employers in the temporary affirmative action treatment, I will use the previously estimated coefficients to predict a probability of hiring two men in round 1. Finally, I will classify the two subgroups within the temporary affirmative action treatment group of employers based on the predicted probability of hiring two men in round 1 around this cutoff point. I will do sensitivity analysis to ensure that any results based on this classification of subgroups is not sensitive to a particular cutoff point.