Samantha (Sami) Horn

Human capital investment decisions often rely on beliefs about how long skills are retained. We test the accuracy of these beliefs by asking participants to learn a new skill and predict their performance after a period of non-use. Participants substantially underestimate skill decay, with average errors ranging from 28% to 59% across tasks. Beliefs adjust after experiencing decay, suggesting that misprediction is specific to delayed performance. Miscalibrated beliefs predict lower demand for refresher training and appear partly motivated: participants are more accurate when forecasting others’ skill decay. Exploratory variable-importance analyses further suggest that participants underweight age, a strong predictor of performance decline. These findings point to systematic errors in forecasting skill retention.

New technologies increase opportunities and pressures to learn new skills. Research on "desirable difficulties" has shown, however, that learners are often poor judges of the efficacy of alternative strategies for learning—for example, they fail to appreciate that spacing experiences with a task out over time will enhance learning. Focusing specifically on the choice of task difficulty, we examine, using two tasks (mirror-tracing in Studies 1-3, and learning Morse code in Study 4), whether learners choose tasks of appropriate difficulty to maximize their own learning, and if not (as we find), why they fail to do so. Study 1 (N=1,809) documents the choice of suboptimally easy tasks. Study 2 (N=1,916) varies the incentives for accuracy to examine whether effort-aversion drives practice task preferences, finding limited evidence that this is the sole driver of choice. Studies 3 (N=1,164) and 4 (N=1,449), measure beliefs about the optimal level of task difficulty for maximizing learning, finds that participants are not aware they would do better with harder practice tasks, and that these beliefs are predictive of choice of practice task difficulty. In Study 5 (N=760), we explore the impact of providing flexibility in adjusting task difficulty, and find that while participants initially prefer easier tasks, many gradually shift to harder ones. However, this adaptation does not lead to greater performance improvements than practicing with hard tasks from the start.

We compare two misinformation interventions designed to target distinct components of Bayesian updating: a factual training that refines priors about vaccine claims and a cue-based training that teaches diagnostic features of misinformation. In a field experiment in Sierra Leone, both interventions improve discernment between true and false vaccine headlines by 0.18 standard deviations, with effects persisting after one month. Factual training improves evaluation of taught content only, while cue-based training generalizes to untaught topics. Effects on information sharing are modest, and we find no effects on belief updating in response to new vaccine information or health behaviors.

We study how social anxiety shapes behavior, beliefs, and performance in evaluative interviews. In a controlled online experiment (N=933), applicants decide whether to participate in a live, five-minute interview with a trained interviewer that determines a monetary hiring bonus. Despite identical objective task ability and indistinguishable interviewer assessments, socially anxious applicants are substantially less willing to interview and hold markedly more pessimistic expectations about task performance, likability, and hiring. These beliefs strongly predict avoidance. Moreover, a non-discriminatory interview experience does not eliminate pessimism. Specifically, socially anxious applicants update more negatively than socially secure individuals, in a way that we show to be inconsistent with Bayesian updating under similar average signals. Applying machine learning techniques to transcripts and video data, we show that belief gaps arise from differential interpretation (updating) rather than differential treatment by interviewers (signal). Consistent with inequities emerging through self-selection rather than discrimination, we find that reducing frictions to showing up is more effective than altering interview environments in reducing mental health inequities.

We experimentally evaluate group-based financial education, savings account access, or both for members of Ugandan youth groups. We measure both short- and long-run impacts with one- and five-year endline household surveys. Education, but not account access, increases measured financial knowledge and trust at one year. At five years, knowledge effects essentially disappear, and trust effects weaken. However, savings and income increase for each treatment at both endlines, which is noteworthy given the interventions’ low cost and the long time horizon of our second endline. Exploring potential mechanisms, we find evidence consistent with multiple pathways to behavior change and outcome improvement.

Many economic decisions, such as whether to invest in developing new skills, change professions, or purchase a technology, benefit from accurate estimation of skill acquisition. We examine the accuracy of such predictions by having study participants predict the speed at which they will master unfamiliar tasks. Across three studies involving two types of tasks and two levels of difficulty, we find systematic underestimation of learning, even after receiving feedback. In a fourth study, participants predicting others' performance showed significantly less underestimation, suggesting that projection bias—overreliance on immediate perceptions of effort and difficulty—may drive prediction errors.

Phishing emails cost companies millions. In the absence of technology to perfectly block phishing emails, the responsibility falls on employees to identify and appropriately respond to phishing attempts and on employers to train them to do so. We report results from an experiment with around 11,000 employees of a large U.S. corporation, testing the efficacy of just-in-time feedback delivered at a teachable moment—immediately after succumbing to a phishing email—to reduce susceptibility to phishing emails. Employees in the study were sent an initial pseudo-phishing email, and those who either ignored or fell victim to the phishing email were randomized to receive or not receive feedback about their response. Just-in-time feedback for employees who fell victim to or ignored the initial pseudo-phishing email reduced susceptibility to a second pseudo-phishing email sent by the research team. Additionally, for employees who ignored the initial email, feedback also increased reporting rates.

The novelty-seeking model (NSM) does not offer a compelling unifying framework for understanding creativity and curiosity. It fails to explain important manifestations and features of curiosity. Moreover, the arguments offered to support a curiosity–creativity link—a shared association with a common core process and various superficial associations between them—are neither convincing nor do they yield useful predictions.

Information that is beneficial for health decision-making is often ignored or actively avoided. Countering information avoidance can increase knowledge of disease risk factors and symptoms, aiding early diagnoses and reducing disease transmission. We examine whether curiosity can be a useful tool in increasing demand for, and engagement with, potentially aversive but useful health information. Four pre-registered randomized online studies were conducted with 5795 participants recruited from online survey platforms. Curiosity for aversive health information was manipulated by providing a ‘curiosity incentive’—identity-related information alongside aversive information—(Study 1), obscuring information (Studies 2 and 3), and eliciting guesses about the information (Studies 2 and 4). Willingness to view four types of aversive health information was elicited: alcohol consumption screening scores (Study 1), colon cancer risk scores (Study 2), cancer risk factors (Study 3), and the sugar content of drinks (Study 4). In Study 1, the curiosity manipulation increases the likelihood that participants view information about the riskiness of their drinking. Studies 2 and 3 show that curiosity prompts counter people's reluctance to learn about and assess their cancer risk. And Study 4 shows that using curiosity prompts to encourage engagement with aversive information—sugar content of drinks—also improves health-related choices (opting for a sugar-free drink alternative).

We use data on the universe of COVID-19 vaccines in Sierra Leone to examine the relationship between COVID-19 vaccination take-up and support for the party in power and whether interruptions to vaccine supply reduced take-up of second doses. We find that COVID-19 vaccine take-up is higher in areas that support the ruling party, but this mirrors long-term vaccination patterns and not the politicization of COVID-19 vaccines. People whose second dose was due just before and after a vaccine stockout had similar second take-up rates (around 50 percent)—that is, delayed access to second doses did not deter eventual take-up.

We often talk about interacting with information as we would with a physical good (e.g., “consuming content”) and describe our attachment to personal beliefs in the same way as our attachment to personal belongings (e.g., “holding on to” or “letting go of” our beliefs). But do we in fact value information the way we do objects? The valuation of money and material goods has been extensively researched, but surprisingly few insights from this literature have been applied to the study of information valuation. This paper demonstrates that two fundamental features of how we value money and material goods embodied in Prospect Theory—loss aversion and different risk preferences for gains versus losses—also hold true for information, even when it has no material value. Study 1 establishes loss aversion for noninstrumental information by showing that people are less likely to choose a gamble when the same outcome is framed as a loss (rather than gain) of information. Study 2 shows that people exhibit the endowment effect for noninstrumental information, and so value information more, simply by virtue of “owning” it. Study 3 provides a conceptual replication of the classic “Asian Disease” gain–loss pattern of risk preferences, but with facts instead of human lives, thereby also documenting a gain–loss framing effect for noninstrumental information. These findings represent a critical step in building a theoretical analogy between information and objects, and provide a useful perspective on why we often resist changing (or losing) our beliefs.

Encouraging vaccination is a pressing policy problem. To assess whether text-based reminders can encourage pharmacy vaccination and what kinds of messages work best, we conducted a megastudy. We randomly assigned 689,693 Walmart pharmacy patients to receive one of 22 different text reminders using a variety of different behavioral science principles to nudge flu vaccination or to a business-as-usual control condition that received no messages. We found that the reminder texts that we tested increased pharmacy vaccination rates by an average of 2.0 percentage points, or 6.8%, over a 3-mo follow-up period. The most-effective messages reminded patients that a flu shot was waiting for them and delivered reminders on multiple days. The top-performing intervention included two texts delivered 3 d apart and communicated to patients that a vaccine was “waiting for you.” Neither experts nor lay people anticipated that this would be the best-performing treatment, underscoring the value of simultaneously testing many different nudges in a highly powered megastudy.

The causal effect of a doctor’s recommendation for Human Papillomavirus (HPV) vaccination on parents’ decisions in low-resource settings is not well understood. This study investigates how doctors’ endorsement of the HPV vaccine communicated through a public health poster affects parents’ decisions to vaccinate their daughters in Kenya. In January and February 2021, 600 parents of daughters eligible for the HPV vaccine but not yet vaccinated were recruited and completed a randomized survey. Participants saw a poster from a national campaign about HPV vaccination and either nothing further (Control) or an additional poster containing an HPV vaccine recommendation from a female (FDR) or male doctor (MDR). Primary outcomes are intentions to vaccinate and perceived safety of the HPV vaccine. Both recommendation arms increased the likelihood that participants reported the highest levels of vaccine intentions compared to control (FDR: 33.7% p = 0.01; MDR: 30.5%, p = 0.05, compared to Control (22.4%)) and safety perceptions (FDR: 24.2%. p = 0.09; MDR: 28.0%, p = 0.01, compared to Control (17.1%)) but there was no statistically significant increase in the likelihood to report above moderate vaccine intentions (FDR: 72.6%, p = 0.76; MDR: 72.5%, p = 0.77, compared to Control (71.4%)) or safety perceptions (FDR: 68.9%, p = 0.91; MDR: 75.0%, p = 0.17, compared to Control (68.6%)). We find no differential treatment effect by the recommending doctor’s gender. In conclusion, our results suggest that visual communication of a doctor’s support for the HPV vaccine can strengthen above-moderate intentions and safety perceptions but may not be enough to persuade the vaccine hesitant to vaccinate.

Policy-makers are increasingly turning to behavioural science for insights about how to improve citizens’ decisions and outcomes. Typically, different scientists test different intervention ideas in different samples using different outcomes over different time intervals. The lack of comparability of such individual investigations limits their potential to inform policy. Here, to address this limitation and accelerate the pace of discovery, we introduce the megastudy—a massive field experiment in which the effects of many different interventions are compared in the same population on the same objectively measured outcome for the same duration. In a megastudy targeting physical exercise among 61,293 members of an American fitness chain, 30 scientists from 15 different US universities worked in small independent teams to design a total of 54 different four-week digital programmes (or interventions) encouraging exercise. We show that 45% of these interventions significantly increased weekly gym visits by 9% to 27%; the top-performing intervention offered microrewards for returning to the gym after a missed workout. Only 8% of interventions induced behaviour change that was significant and measurable after the four-week intervention. Conditioning on the 45% of interventions that increased exercise during the intervention, we detected carry-over effects that were proportionally similar to those measured in previous research. Forecasts by impartial judges failed to predict which interventions would be most effective, underscoring the value of testing many ideas at once and, therefore, the potential for megastudies to improve the evidentiary value of behavioural science.

Introduction. We evaluate whether a combination of financial incentives and deposit contracts improves cessation rates among low- to moderate-income smokers.

Methods. We randomly assigned 311 smokers covered by Medicaid at 12 health clinics in Connecticut to usual care or one of three treatment arms. Each treatment arm received financial incentives for two months and either (i) nothing further ("incentives only"), (ii) the option to start a deposit contract with incentive earnings after the incentives ended ("commitment"), or (iii) the option to precommit any earned incentives into a deposit contract starting after the incentives ended ("precommitment"). Smoking cessation was confirmed biochemically at two, six, and twelve months.

Results. At two, six, and twelve months after baseline, our estimated treatment effects on cessation are positive but imprecise, with confidence intervals containing effect sizes estimated by prior studies of financial incentives alone and deposit contracts alone. At two months, the odds ratio for quitting was 1.4 in the incentive-only condition (95% CI: 0.5 to 3.5), 2.0 for incentives followed by commitment (95% CI: 0.6 to 6.1), and 1.9 for incentives and precommitment (95% CI: 0.7 to 5.3).

Conclusions. A combined incentive and deposit contract program for Medicaid enrollees, with incentives offering up to $300 for smoking cessation and use of support services, produced a positive but imprecisely estimated effect on biochemically verified cessation relative to usual care and with no detectable difference in cessation rates between the different treatment arms.

Introduction. Socioeconomic disparities have been established for conventional cigarette use, but not for electronic cigarettes. This study estimates socioeconomic gradients in exclusive use of conventional cigarettes, electronic cigarettes, and dual use (ie, use of both products) among adults in the United States.

Methods. Analyses consider nationally representative data on 25–54-year-old respondents to the 2014–2016 National Health Interview Surveys (N = 50,306). Demographically adjusted seemingly unrelated regression models estimate how two socioeconomic status measures—respondent education and household income—relate to current exclusive use of conventional cigarettes, electronic cigarettes, and dual use.

Results. Conventional cigarette use exhibits negative education and income gradients, consistent with existing research: −12.9 percentage points (confidence interval [CI]: −14.0, −11.8) if college educated, and −9.5 percentage points (CI: −10.9, −8.1) if household income exceeds 400% of the federal poverty level. These gradients are flatter for dual use (−1.4 [CI: −1.8, −0.9] and −1.9 [CI: −2.5, −1.2]), and statistically insignificant for electronic cigarette use (−0.03 [CI: −0.5, 0.4] and −0.3 [CI: −0.8, −0.2]). Limiting the sample to ever-smokers, higher education is associated with a 0.9 percentage point increase in likelihood of exclusive electronic cigarette use at interview (CI: 0.0, 1.9).

Conclusions. Education and income gradients in exclusive electronic cigarette use are small and statistically insignificant, contrasting with strong negative gradients in exclusive conventional cigarette use. Furthermore, more educated smokers are more likely to switch to exclusive e-cigarette use than less educated smokers. Such differential switching may exacerbate socioeconomic disparities in smoking-related morbidity and mortality, but lower the burden of tobacco-related disease.

Implications. Research has not yet established whether socioeconomic disparities in electronic cigarette (e-cigarette) use resemble those observed for conventional cigarettes. This article uses nationally representative data on US adults aged 25–54 to estimate income and education gradients in exclusive use of conventional cigarettes, e-cigarettes, and dual use. Both gradients are steep and negative for conventional cigarette use, but flat and statistically insignificant for e-cigarette use. Repeating the analysis among ever-smokers indicates that more educated smokers are more likely to transition toward exclusive e-cigarette use than less educated smokers. Such differential substitution may exacerbate disparities in smoking-related morbidity and mortality.

We partnered with a large online auction website to test differing messages’ effects on the decision to donate to charity at checkout. Our setting, where impulsive decisions are likely to be driving donations, allows us to evaluate intuitive responses to messages prompting a donation. We find that shorter messages, matching grants, and descriptions of a charity’s mission increase both the likelihood that a user donates, as well as the average amount donated. Conversely, displaying the impact of the donated amount, the popularity of the charity, and that a charity uses scientific evidence do not improve donation rates. These results contribute to our understanding of how framing requests drives the decision to donate and are practically relevant to the many retail sites which promote giving at point of sale.

We use a simulation study to compare three methods for adaptive experimentation: Thompson sampling, Tempered Thompson sampling, and Exploration sampling. We gauge the performance of each in terms of social welfare and estimation accuracy, and as a function of the number of experimental waves. We further construct a set of novel "hybrid" loss measures to identify which methods are optimal for researchers pursuing a combination of experimental aims. Our main results are: 1) the relative performance of Thompson sampling depends on the number of experimental waves, 2) Tempered Thompson sampling uniquely distributes losses across multiple experimental aims, and 3) in most cases, Exploration sampling performs similarly to random assignment.