Nutrition and Machine Vision ($650,000)

Nutrition and Machine Vision ($60,000)

Motivated Reasoning, Paradoxical Thinking, and COVID-19 Vaccination ($180,000)

Employer beliefs, employee training, and labor market outcomes: A field experiment in Uganda ($452,000)

Employer beliefs, employee training, and labor market outcomes: A field experiment in Uganda ($16,660)

Examining the impact of waiting periods on improving the use of food subsidies for healthier consumption while maintaining choice ($198,940)

Improving Community College Outcomes through Performance Incentives ($312,000)

Improving Community College Outcomes through Performance Incentives ($25,000)

On the Elicitation of Willingness to Pay for Stigmatized Goods ($4,700)

Healthy food choices are a canonical example used to illustrate the importance of time preferences in behavioral economics. However, the literature lacks a direct demonstration that they are well-predicted by incentivized time preference measures. We offer direct evidence by combining a novel, two-question, incentivized time preference measurement with data from a field experiment that includes grocery purchases and consumption. Our present-focus measure is highly predictive of food choice, capturing a number of behaviors consistent with self-control problems, which provides direct evidence for the common assumption that important aspects of nutrition are driven by time preferences. [NBER Link]
Press: [Becker Friedman Institute]

Importance: By early 2022, COVID-19 vaccine uptake plateaued, resulting in unnecessary hospitalisations and deaths. With the widespread adoption mRNA vaccines, new messaging strategies are needed for those who remain highly hesitant to accept these new-age immunizations.
Objective: Our experiment tested whether those who remained unvaccinated late into the pandemic were willfully avoiding vaccine information, or if they would voluntarily seek vaccine education that could allow them to make more informed choices.
Setting & Participants: For this single-blind, between-subjects, randomised controlled trial, we recruited 7090 US residents who remained unvaccinated in 2022 to participate in our online survey-experiment.
Interventions & Design: We developed informational interventions designed build people’s understanding of COVID-19 vaccines by means of detailed yet easy-to-grasp videos that explain mRNA vaccine technology, research and development, testing and safety protocols, and potential side effects. Participants were randomised into three groups: (1) a control group who watched an unrelated placebo video, (2) a “video optional” group for whom every informational video was optional, and (3) a “video required” group for whom the vaccine technology video was mandatory, but all other videos were optional.
Main Outcomes: Primary endpoints were future vaccine uptake intentions, vaccine efficacy beliefs, vaccine safety beliefs, and trust in healthcare institutions (measured on a 100-point scale). Secondary endpoints measured information seeking behaviour.
Results: 60% (95% CI: 58.6–62.2) of the participants in the video required group chose to watch optional videos, and 55% (95% CI: 51.9–57.3) of participants in the video optional group watched at least one video. The video required group exhibited a 6% (95% CI: 2%–10%), improvement in stated vaccine intentions, an 11% (95% CI: 8%–15%) improvement in vaccine efficacy beliefs, an 8% (95% CI: 3%–14%) decrease in side effects concerns, and a 7% (95% CI: 4%–10%) percent increase in the trust of health care institutions. The video optional group yielded improvements of 7%, 6%, 16%, and 5% in the respective categories (95% CIs: 2%–13%, 2%–11%, 8%–24%, 0%–9%, respectively).
Conclusion & Relevance: Participants displayed high willingness to voluntarily seek information on mRNA vaccine technology. Further, our informational videos softened antivaccination sentiments and improved vaccine intentions. These results demonstrate that even the most vaccine-resistant individuals are willing to learn about mRNA vaccines and that a disinterested, educational approach may be best suited to reach them. [SSRN Link]
Press: [The New Scientist]

We conduct a field experiment with low-income shoppers to study how behavioral interventions can improve the effectiveness of healthy food subsidies. Our unique design enables us to deliver subsidies both before and during grocery shopping. We examine the effects of two non-restrictive changes to the choice environment: giving shoppers agency over the subsidy they receive and introducing a waiting period before a subsidized shopping trip to prompt deliberation about upcoming purchases. These interventions increase healthy food spending by 61% more than a healthy food subsidy alone, resulting in 199% greater healthy spending than in our unsubsidized control group. [SSRN Link]
Press: [Chicago Booth Review] ‍

Poll respondents often attempt to present a positive image by overstating virtuous behaviors. We examine whether people account for this "socially desirable responding" (SDR) when drawing inferences from poll data. In an experiment, we incentivize ``predictors'' to guess others' choice behaviors across eight actions with varying social desirability. To aid guessing, predictors observe random subsamples of (i) incentivized choices or (ii) hypothetical claims from polls. Predictors show reasonable skepticism towards hypothetical claims, which exhibit predictable SDR. However, their skepticism is not tailored to the direction or magnitude of SDR. This under-correction occurs even though subjects' explicit responses can predict SDR. [SSRN Link]

We experimentally examine whether a policy targeting college summer school enrollment can accelerate degree progress and completion. We randomly assign summer scholarships to community college students and find a large impact on degree acceleration, increasing graduation within one year of the intervention by 32% and transfers to four-year colleges by 58%. We elicit preferences for the scholarships and find that treatment effects are concentrated among students with a preference against summer school. Our results suggest that educational impacts do not drive enrollment preferences. And, that many more students could benefit from summer school than the  minority who currently enroll. [SSRN Link]

In a field experiment, we examine the impact of performance-based incentives for community college instructors. Instructor incentives improve student exam scores, course grades, and credit accumulation while reducing course dropout. Effects are largest among part-time adjunct instructors. During the program, instructor incentives have large positive spillovers, increasing completion rates and grades in students’ courses outside our study. One year after the program, instructor incentives increase transfer rates to 4-year colleges with no impact on 2-year college degrees. We find no evidence of complementarities between instructor incentives and student incentives. Finally, while instructors initially prefer gain-framed contracts over our loss-framed ones, preferences for loss-framed contracts significantly increase after experience with them. [PDF]
Press: [Econimate YouTube Video] [Tea for Teaching Podcast] [Behavioral Science Uncovered Podcast]

Disentangling effort and luck is critical when judging performance. In a principal-agent experiment, we demonstrate that principals' judgments of agent effort are biased by luck, despite perfectly observing the agent's effort. We find that two potential solutions to this "outcome bias"—the opportunity to avoid irrelevant information about luck, and outsourcing judgment to independent third parties—are ineffective.  When we give control over information about luck to principals and agents in separate treatments, we find asymmetric sophistication: agents strategically manipulate principals' outcome bias, but principals fail to recognize their own bias. Independent third parties are just as biased as principals. These findings indicate that the scope of outcome bias may be larger than previously understood and that outcome bias cannot be driven solely by emotional responses nor distributional preferences. Instead, we hypothesize that luck directly affects beliefs, and we test this hypothesis by eliciting the beliefs of third parties and principals. Lucky agents are believed to exert more effort than identical, unlucky agents. We propose a model of biased belief updating explaining these results.

Social scientists have observed that socially desirable responding (SDR) often biases unincentivized surveys. Nonetheless, media, campaigns, and markets all employ unincentivized polls to make predictions about electoral outcomes. During the 2016 presidential campaign, we conducted three list experiments to test the effect SDR has on polls of agreement with presidential candidates. We elicit a subject's agreement with either Hillary Clinton or Donald Trump using explicit questioning or an implicit elicitation that allows subjects to conceal their individual responses. We find evidence that explicit polling overstates agreement with Clinton relative to Trump. Subgroup analysis by party identification shows that SDR significantly diminishes explicit statements of agreement with the opposing party's candidate, driven largely by Democrats who are significantly less likely to explicitly state agreement with Trump. We measure economic policy preferences and find no evidence that ideological agreement drives SDR. We find suggestive evidence that local voting patterns predict SDR. [PDF]
Press: [Press Release] [KUAF (NPR)] [KNWA 24] [KARK 4] [The British Psychological Society] [Bloomberg]

Grading on the curve is a form of relative evaluation similar to an all-pay auction or rank-order tournament. The distribution of students drawn into the class from the population is predictably linked to the size of the class. Increasing the class size draws students' percentile ranks closer to their population percentiles. Since grades are awarded based on percentile ranks in the class, this reallocates incentives for effort between students with different abilities. The predicted aggregate effort and the predicted effort from high-ability students increases while the predicted effort from low-ability students decreases. Andreoni and Brownback (2017) find that the size of a contest has a causal impact on the aggregate effort from participants and the distribution of effort among heterogeneous agents. In this paper, I randomly assign "class sizes" to quizzes in an economics course to test these predictions in a real-stakes environment. My within-subjects design controls for student, classroom, and time confounds and finds that the lower variance of larger classes elicits greater effort from all but the lowest-ability students, significantly increasing aggregate effort. [PDF]

We model contests with a fixed proportion of prizes, such as a grading curve, as all-pay auctions where higher effort weakly increases the likelihood of a prize. We find theoretical predictions for the heterogeneous effect auction size has on effort from high- and low-types. We test our predictions in a laboratory experiment that compares behavior in two-bidder, one-prize auctions with behavior in 20-bidder, 10-prize auctions. We find a statistically significant 11.8% increase in aggregate bidding when moving from the small to large auction. The impact is heterogeneous: as the auction size increases, low-types decrease effort but high-types increase effort. Additionally, the larger auction provides a stronger rank-correlation between effort and ability, awarding more prizes to the higher-skilled and improving the efficiency of prize allocation. [NBER link]

Social science researchers often use categorical scales (e.g., extremely unlikely, …, extremely likely) to measure behavioral intentions and assume proportional cardinality between the ordinal categories. However, respondents may have different interpretations, providing a potential source of measurement error in predicting intentions. This study asks respondents to associate a probability (i.e., 0 to 100) and variability (i.e., low, medium, or high) with response categories for intentions to exercise more or eat more vegetables. Results show that perceived probabilities of ‘unlikely’ categories are higher than under the assumption of proportionality, and perceived probabilities of ‘likely’ categories are lower than under the assumption of proportionality. Likely categories also featured relatively higher variability. Thus, measures of intention using categorical-response scales may overestimate positive intentions and underestimate negative intentions.

Outcome bias occurs when evaluators overemphasize realized outcomes relative to the information they possess about agents' actions. We use a principal-agent experiment to identify outcome bias in punishment decisions. Despite perfectly-observable effort, unlucky agents receive 0.27 SD more punishment than their equally diligent but lucky counterparts. This bias persists into subsequent interactions. Both principals and independently-recruited third parties evaluate agents by reporting beliefs about their typical effort, selecting agents for a dictator role, and delegating a real-effort task to agents. Outcome bias has only a modest effect on principals' beliefs about effort and no effect on the beliefs of third parties. In contrast, evaluators consistently favor lucky agents in dictator-selection and delegation decisions. We propose two mechanisms driving this bias: confusion and moral luck. Our findings suggest that outcome bias stems primarily from moral luck--a phenomenon by which outcomes determine the blameworthiness of the actor--rather than confusion. This distinction matters for the design of performance evaluation mechanisms in any dynamic environment.

Polarization has increased the importance of persuasion in public health. We study how to influence beliefs among vaccine-hesitant adults in the U.S. through a novel "paradoxical reasoning" intervention that operationalizes reductio ad absurdum techniques to weaken arguments against vaccination. We experimentally compare this approach with two standard strategies: 1) medical messages from public authorities and 2) emotional appeals through personal narratives. All three produce similar positive average treatment effects on vaccination intentions but reach different groups: official messages primarily persuade Democrats, personal narratives resonate with non-white individuals, and paradoxical reasoning shifts beliefs among Republicans. Heterogeneity analysis further shows that paradoxical reasoning is especially effective among individuals with higher baseline mistrust, highlighting its capacity to change intentions among those least responsive to conventional messages. Quantile regressions further show that paradoxical reasoning is most effective in the lower tail of the intentions distribution, unlike the other strategies which mainly affect the upper tail. These findings indicate that standard messaging achieves average gains largely by "preaching to the choir," while paradoxical reasoning achieves similar effects through the more difficult task of persuading the most resistant.

Classic models of workforce training predict that the inability of employers to recoup training costs will produce a quantity distortion whereby training provision is inefficiently low. We explore an additional inefficiency in training provision: a selection distortion whereby training may be targeted to workers with relatively lower skill gains from training. Using experimental evidence from the Ugandan metalworking sector, we show how misaligned preferences for training between employers and workers may result in such a distortion. We develop a labor market model that produces two main predictions: 1) worker demand for training depends on anticipated skill gains and 2) employer demand for training depends on the ability to extract those skill gains. We confirm these predictions, finding that employer and worker preferences have opposing relationships with key variables. Employers under-select workers with high anticipated skill gains and over-select workers with strong attachments to the firm. Misaligned preferences do not stem from asymmetric information: we find employer beliefs about skill gains from training are predictive of realized gains.

Children who spend time in foster care have substantially worse short- and long-term outcomes. Placement stability can mitigate some of these harms, but it depends on an adequate supply of resource families. Yet many areas face severe shortages of such families, in part because of high turnover and short lengths of service. We study an intensive, faith-based wraparound support program, Hands of Hope (HoH), designed to expand the supply of resource families through recruitment and retention. Using administrative data on all Indiana resource families and foster children, we examine whether HoH support affects foster care supply at both the extensive and intensive margins. At the extensive margin, the stock of resource families in a ZIP code rises as the network of HoH-affiliated churches grows closer. At the intensive margin, relative to a propensity-score matched control group, families receiving HoH support remain active longer, support more placements, provide more child-placement days, and are more likely to still be fostering up to two years after their first placement. At the child level, we find no evidence of declines in placement stability or related measures of care quality, suggesting that the program does more than simply retain marginal families who would otherwise exit. Overall, our results suggest that non-monetary support for resource families can meaningfully expand foster care supply without detectable declines in care quality.