This project studies whether written self-descriptions and prosocial signals such as internships can reveal an applicant's true cooperativeness, even when applicants have strategic incentives to fake being a good team player.
Combining tools from experimental economics with machine learning and survey methods, the project shows that both AI-based text analysis and prosocial signals retain predictive power for cooperativeness where self-reported personality tests fail.
Organizations have long tried to assess job applicants' personality using self-reported psychometric tests such as the Big Five, but these tests are not robust to incentives to fake desirable traits. In a controlled online experiment, we test whether machine-learning classifiers trained on written self-descriptions (in the spirit of a cover letter) predict people's true, incentivized cooperativeness better than psychometric tests. We find that when people have incentives to fake, linguistic classifiers significantly outperform psychometric classifiers, and that a fine-tuned language model can detect the presence of incentives to fake in a person's self-description.
A companion study examines a related channel of self-presentation: prosocial signals such as internships or volunteering. In a survey experiment with professional recruiters, we show that applicants with internships at companies with a salient social mission are perceived as more prosocial, and that this raises their chances of being invited to a job interview. In a complementary lab experiment, we show that such prosocial signals remain predictive of a person's actual cooperativeness even when they are used strategically and cannot be verified.
Together, the two studies show that, unlike easy-to-fake psychometric self-reports, both written language and prosocial signals carry genuine information about applicants' cooperativeness, even in the presence of strategic incentives to misrepresent it.
Conference proceeding
Working paper
