Published
Second-Order Beliefs and Gender (with Andrew Dustan & Greg Leo)
Journal of Economic Behavior & Organization, August 2022
Beliefs about beliefs—second-order beliefs—about the differences between populations are important to understanding differences in outcomes between those populations. To study their potential impact, we develop an incentive-compatible experimental framework for eliciting beliefs (first-order) and beliefs about beliefs (second-order) about the differences in any measurable characteristics between any two populations. We implement the procedure to study beliefs about the performance of men and women on math and abstract bargaining tasks. In the math task, 78% of participants believe that most men believe men outscore women. In contrast, 34% believe that most women believe men outscore women. Despite these differences in second-order beliefs, we observe no such difference in first-order beliefs. The pattern of results is similar in the bargaining task. These results have important labor market implications for the persistence of gender gaps.
Mechanism Performance under Strategy Advice and Sub-Optimal Play: A School Choice Experiment (with Andrew Dustan, Martin Van der Linden, & Myrna Wooders)
Journal of Behavioral and Experimental Economics, November 2021
We implement a laboratory experiment to study how strategy advice affects participant decisions in a school choice game. In the Deferred Acceptance (DA) mechanism, advice to choose the dominant strategy of truth-telling induces participants to do so. In the Immediate Acceptance (IA) mechanism, advice to implement one of two heuristic strategies induces participants to choose one of those strategies. We develop a new partially-ordered typology of DA strategies to study the sub-optimal strategies chosen by participants under advice versus no advice. Then, using the varying proportions of participants choosing sub-optimal strategies in our data, we perform exploratory analyses on mechanism performance. We find that DA outperforms IA in efficiency, stability, and proportion of participants assigned their most preferred school. These performance differences are larger under strategy advice.
Working Papers
The Heterogeneous Impact of Changes in Default Gift Amounts on Fundraising (with Susan Athey, Undral Byambadalai, Matias Cersosimo, & Dean Karlan)
When choosing whether and how much to donate, potential donors often observe a set of default donation amounts known as an “ask string.” In an experiment with more than 400,000 PayPal users, we replace a relatively unused donation amount ($75) on PayPal's Giving Fund Website ask string with either a lower ($10) or a higher ($200) reference point to evaluate the impact on charitable giving. Relative to the status quo, we find that a higher reference point increases the total amount of money raised, while the lower reference point increases the number of donors, two objectives important to non-profits. Both interventions drive more people to choose a default amount compared to the status quo, where the alternatives are not donating or writing in an amount. Examining treatment effect heterogeneity and changes in the distribution of donations, we provide suggestive evidence about the mechanisms. We use data-driven machine learning methods to learn personalized policies that identify who should be shown the lower versus higher reference point. Personalization can increase the probability of choosing a default amount, and it can also alleviate the trade-off to non-profits between the total amount of money raised and the number of donors.
Impact Matters for Giving at Checkout (with Susan Athey, Matias Cersosimo, & Dean Karlan)
We conducted two experiments on PayPal’s Give at Checkout feature to learn about the effect of 1) information about charity outcomes on donations, and 2) exposure to these point-of-sale microgiving requests on subsequent giving. In this “impulsive” giving context, quantifying the charity’s outcome generates positive treatment effects, larger than those for a narrative. Third-party validation can decrease giving when added to the quantified outcome treatment, and has at most small effects relative to no information. The second experiment finds neither crowd-in (e.g., via habit formation) nor crowd-out (e.g., via budgeting) from these microgiving requests on later donation behavior.
Emotion- versus Reasoning-based Drivers of Misinformation Sharing: A field experiment using text message courses in Kenya (with Susan Athey, Matias Cersosimo, & Zelin Li)
Two leading hypotheses for why individuals unintentionally share misleading information online are that 1) they are unable to recognize that a post contains misinformation, and 2) they make impulsive, emotional sharing decisions without thinking about whether a post contains misinformation. The strategies to counter each of these drivers of misinformation sharing differ by the techniques that they are designed to address. We categorize techniques according to whether they use misleading reasoning to make recognizing misinformation more difficult (reasoning-based) or manipulate emotions to encourage impulsive sharing decisions (emotions-based). To learn whether interventions designed to counter reasoning- or emotion-based techniques are more effective or whether the approaches are complementary, we evaluate three distinct versions of a low-cost and scalable five-day text message educational course. We assess the impact of the courses in a field experiment with approximately 9,000 participants in Kenya. We measure outcomes using a pre-post survey design that elicits intentions to share and find that all treatment courses work, decreasing misinformation sharing 28% on average relative to no text message course. The treatment designed to counter emotion-based techniques, the “Emotions” course, is more effective than teaching about reasoning-based techniques either alone in the “Reasoning” course or in combination with emotion-based techniques in the “Combo” course. Moreover, the Emotions course performs best on misinformation posts that use emotional manipulation, and does no worse than the Reasoning or Combo courses on misinformation posts that use reasoning-based techniques. In a follow-up experiment approximately two months later, 88% of the treatment effect of the three courses on misinformation sharing persists.
This paper investigates how workers' job application decisions are affected by their beliefs about hiring managers' beliefs regarding the relative productivity of women and men. To this end, I combine a natural field experiment with a lab experiment. In the field experiment, I partner with a firm to solicit approximately 5,000 job applications using ads that randomize over the gender of the hiring manager and the gender associations of the product sector. I then recruit the same job-seekers to a structured online lab experiment to elicit their beliefs about hiring managers' beliefs, based on the manager's gender and product sector. Truth-telling is incentivized with the Binarized Scoring Rule, using a procedure I adapt from Dustan, Koutout, & Leo (2022). I find that men are more likely to apply for a job with a manager whom they believe has beliefs that favor men more. A one standard deviation increase in beliefs about the manager's beliefs increases the probability a man applies by 30%. On the other hand, women are unresponsive to their beliefs about managers' beliefs. These results have important implications for the sorting by gender behavior driving a large part of the gender wage gap.
Reduction in Belief Elicitation (with Andrew Dustan & Greg Leo)
The state-of-the-art in eliciting probabilistic beliefs, the Binarized Quadratic Scoring Rule (BQSR), relies on an easily overlooked preference assumption: the reduction of compound lotteries. In a lab experiment, we find evidence that a large majority of people violate the reduction assumption for at least some compound lotteries involved in the BQSR. We show that people whose preferences are consistent with reduction are 33% more likely to report accurate beliefs compared to those whose preferences are not consistent with reduction. Lastly, we implement a novel Rank-Ordered Elicitation (ROE), which does not rely on the reduction of compound lotteries, to test whether eliminating the need for reduction increases the accuracy of reported beliefs. We find no evidence for this last hypothesis, suggesting that preferences inconsistent with the reduction of compound lotteries could be proxying for other participant characteristics that affect accuracy.
Gendered Beliefs among Peers (with Luis C. Carvajal-Osorio & Andrew Dustan)
Available on request.
We use an incentivized online lab-in-the-field experiment to measure what unemployed youth believe about others' abilities, preferences, and beliefs. Such beliefs may influence how social networks match people with employment and training opportunities through referrals and word-of-mouth information sharing. In particular, if women are systematically believed to be less capable or interested in particular fields, these beliefs have implications for gender-based inequality. The experiment targets applicants to an information technology (IT) career training program, using state-of-the-art elicitation methods to measure their beliefs about the differences between women's and men's general cognitive ability, IT-specific ability, and preferences for IT tasks. We find that both women and men believe that women have higher general cognitive and IT-specific ability, but believe that men have stronger preferences for IT tasks. This mixed pattern of gendered beliefs may explain why we find no differences in the rate at which women and men are referred to the program by applicants. In addition, we elicit beliefs about what others believe (second-order beliefs) are the differences in women's and men's general cognitive ability. Both women and men correctly believe women are women-favoring, but men incorrectly believe that men are men-favoring.
In Progress
Combating Misinformation on Social Media (with Ruth Appel, Susan Athey, Dean Karlan, Mike Luca, Nils Wernerwelt, & Rachel Zhou)
Experiment analysis in progress
Emotional Donation Appeals: An experiment to increase charitable giving on Facebook (with Patricia Andrews Fearon, Susan Athey, Dean Karlan, Mike Luca, Paige Tsai, & Nils Wernerwelt)
Experiment analysis in progress