Overconfidence and the Gender Application Gap: A Laboratory Experiment
How to cite
Jost, Gregor; Selz, Vincent. 2024. "Overconfidence and the Gender Application Gap: A Laboratory Experiment." AEA RCT Registry. AEARCTR-0014968. https://doi.org/10.1257/rct.14968-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
Women apply for jobs that yield relatively lower pay than men (Fluchtmann, Glenny, Harmon, Maibom, 2024). Despite its relevance for the gender wage gap, the mechanisms driving the application gap remain largely underexplored. This study investigates whether gender differences in overconfidence about one’s relative ability can partially explain this phenomenon.
From a directed search model, we derive two predictions about the role of overconfidence for the gender application gap. First, since overconfident job seekers overestimate their chances of securing high-wage jobs, they have a higher incentive to apply there – the direct effect. Second, job seekers who anticipate others' overconfidence also expect higher competition for high-wage jobs, discouraging them to apply there – the externality of overconfidence. Hence, if men are more overconfident than women, they apply more for high-wage jobs, while women’s anticipation of male overconfidence would further deter them.
To test these effects, we conduct a lab experiment. In part one, participants complete a logic task, followed by the measurement of their overconfidence and second-order beliefs about others' overconfidence. In part two, pairs of participants compete in a stylized labor market where job allocation probabilities depend on their relative performances in the logic task. Pairs are randomized into treatment and control groups: in treatment pairs, one participant is informed about its actual relative performance in the logic task, while the other remains uninformed but knows its competitor has this information. Control group participants receive no performance feedback. This design allows for identifying the direct effect of overconfidence by comparing informed treatment participants to controls, and the externalities of overconfidence by comparing uninformed treatment participants to controls.
Intervention
Groups of two compete for two different jobs, which differ in wages. The job allocation probabilities depend on the relative performances of the two group members in a prior logic task. In control group pairs, neither group member receives feedback on its relative performance. In treatment group pairs, one member is informed of its actual performance relative to the entire session. This allows us to identify the direct effect of overconfidence. The other group member is informed that its competitor received such feedback, enabling us to identify the externality of overconfidence.
Experimental design
Part 1: All participants have 15 minutes to solve up to 11 Raven matrices. Participants are not directly incentivized for this but are informed that their payoff in Part 2 depends positively on the number of correctly solved matrices and the time taken to solve them.
After this, participants are randomized into groups of two and assigned a rank High or Low based on their relative performance within their pair. Then, we elicit how likely they think they are rank High within their pair (Q1). Additionally, we ask participants about what they think was the average response of the other participants of the session to Q1 (Q2). We incentivize truthful responses by monetarily rewarding answers close to the actual values.
Part 2: Each pair competes for two jobs differing only in the wages they offer. Participants compete for these jobs by allocating application weights (x, 1-x) among them. Each job offers one position to one of the two participants in the group. The assignment rule is set up in a way that the likelihood of receiving an offer from a particular job increases with both the subject's weight assigned to that job and its rank. If a subject receives only one job offer, the wage from that job is realized. If a subject receives offers from both jobs, the higher wage is realized. If a subject receives no offers, its payoff for Part 2 is zero.
Afterwards, all participants are presented with two hypothetical scenarios. In these vignettes, we ask participants how they would have applied, if they had been given the information from each respective scenario. Within-subjects, we keep the participant's hypothetical performance and that of its competitor constant across both scenarios. The only variation within-subjects is the competitor's belief about its own performance. Across subjects, we vary the own performance and the order of the vignettes.
Part 3: We elicit risk preferences via a mutliple price list. Specifically, participants make decisions between a varying amount they receive with certainty and a lottery between 0€ and 20€. One decision is randomly drawn; conditional on the choice the lottery is played and the payoff is realized. After Part 3, a post-experimental survey is administered.
Treatments: In Part 2, pairs are randomized into a treatment and control group. Pairs in the control group do not obtain any feedback on performance in the logic task. In the treatment group, one of the two pair members receives information about its relative standing within the whole session. This removes any erroneous beliefs about its relative performance and thereby allows us to isolate the direct effect of overconfidence. The other person in the treatment group group does not get any information about its own performance but is informed that its competitor obtains this information, allowing us to evaluate the externality overconfidence imposes on others.
Measures: Our primary measures for overconfidence and anticipated overconfidence are
1. Overconfidence: we use the answers to Q1 and subtract the actual probability to be ranked High.
2. Anticipated overconfidence: we use the answers to Q2 and subtract the actual average probability to be ranked High of the other participants within the session.
In addition to our primary measures, we elicit how many matrices they think they solved correctly with which we can construct an additional measure of overconfidence. Also, we survey their general risk attitudes using the general risk question from Dohmen et al. (2011) in the post-experimental survey. These duplicate measures are elicited to account for measurement error (Gillen, Snowberg and Yariv, 2019).
Payoffs: The payoffs of either Part 2 or Part 3 are randomly drawn by the computer in addition to the show-up fee and the Part 1 payoffs.
Design details
Cited by
No citation data available for this registration yet.