AEA RCT Registry currently lists 12677 studies with locations in 171 countries.
We use a series of survey experiments embedded over multiple rounds of a panel survey to study the sources, causes, and remedies of measurement error in survey data. The topical focus of our surveys is on living standards, extreme climate events, and their interaction with welfare at the household and intra-household level. Within each survey round, we randomize survey mode (in-person vs. over the phone), collect a series of data quality proxies, and conduct a range of experiments to study mechanisms.
The northern region of Sweden faces a strong demand for employees across multiple industries and the Swedish government is actively trying to create incentives for unemployed individuals to relocate to these regions. The Swedish public employment service has been assigned to evaluate if vacancy referrals to job openings in the northern regions have a positive effect on the probability that unemployed individuals apply to these jobs and also to a higher degree relocate to these regions. A randomized controlled trial has been implemented where job-seekers in the treatment group are given vacancy referrals to job-openings in the northern regions. Job-seekers in the treatment group are randomly divided between vacancy referrals mandatory or only suggestive to apply for. The control-group re...
Artificial intelligence is increasingly integrated into high-stakes decisions—investment, hiring, healthcare, military targeting—making human oversight both practically important and increasingly mandated by regulation. Whether people supervise an autonomous artificial agent differently from another human is therefore a crucial question: excessive scrutiny undermines the efficiency gains that motivated AI integration, while overreliance defeats the purpose of the human-in-the-loop. This project investigates whether individuals exhibit AI aversion when overseeing decisions with real consequences for others, and disentangles two mechanisms: the black-box effect, arising from uncertainty about the AI's decision-making process, and intrinsic AI aversion, a reluctance to rely on algorithm...
This study evaluates a lightweight strategy to improve fall-prevention awareness and safety habits among older adults. A group of seniors (aged 60+) and their caregivers in China will participate. The control group will receive standard, general safety advice. The treatment group will receive personalized risk feedback (with one treatment group receives positive framed feedback and another group receives negative framed feeedback) and choose one simple, low-pressure safety habit to try (such as clearing a walkway or using a nightlight). Following a tracking period, we will compare both groups to see if this practical method effectively motivates seniors to overcome risk denial, seek help, and adopt safer daily habits.
This pre-analysis plan describes a randomized controlled trial evaluating an AI-powered mental health chatbot among undergraduate students in India’s public universities. Using baseline data from 4,489 students across seven colleges in Delhi, we document a high prevalence of depression, anxiety, and loneliness, and low rates of mental health care-seeking. Classrooms are randomized to a control group or to receive four weeks of access to the chatbot. Within treated classrooms, students are cross-randomized to alumni endorsements and habit-formation streaks. The plan pre-specifies intent-to-treat estimates of the effects of chatbot access and the additional interventions on adoption and engagement, care-seeking behavior, stigma (both own and perceived), willingness to pay for therapy, and...
This study uses a survey experiment to measure how students and workers in Brazil perceive and value public employment relative to private employment. Participants report baseline beliefs about wages, job stability, and probabilities of entering public and private employment under different search strategies. They are then randomly assigned to receive aggregate labor-market information about one of these topics or to a no-information control condition. After the information module, the survey elicits expected wages, reservation wages, cross-sector switching thresholds, career-path probabilities, willingness to pay for additional information through a Becker-DeGroot-Marschak (BDM) mechanism, and choices over hypothetical jobs in a discrete choice experiment (DCE). The main hypotheses ...
Do people account for payoff-irrelevant but signal-relevant variables (i.e., nuisance variables) when interpreting information, and does ego-relevance influence this inference process? We study nuisance neglect in a transparent two-component signal-generating structure in which a payoff-relevant rank state (X) and a payoff-irrelevant but signal-relevant nuisance variable (Y) jointly determine the observed signal (S). In the experiment, each subject completes two parallel tasks in randomized order: an IQ-style task, which is ego-relevant, and a temperature-guessing task, which is ego-irrelevant. In each task, we elicit subjects’ prior beliefs about their rank state, provide a signal generated by a known rule that combines the true state and the nuisance variable, and then elicit po...
This study collects survey data on how the anticipated sex of a first child affects prospective parents’ stated intentions regarding labour supply and childcare. Respondents aged 22 to 35, cohabiting and without children, are randomly assigned to imagine that their first child will be a daughter or a son. Randomisation is stratified by respondent sex. We then elicit expected and desired weekly working hours for the respondent and their partner, the anticipated division of childcare and household work, open-ended text on apprehensions and excitement, beliefs about raising sons and daughters, and gender role attitudes.
This study proposes an experiment that examines how the magnitude of underlying likelihoods influences likelihood judgments and posterior beliefs. The experiment constructs information environments that hold the likelihood ratio—and therefore the Bayesian posterior—fixed while varying the absolute magnitude of the underlying likelihoods. By eliciting both likelihood judgments and posterior beliefs, the study investigates whether likelihood-reporting distortions depend on likelihood magnitude, whether these distortions propagate to posterior beliefs, and whether pure and mixed samples with identical likelihood ratios generate different likelihood judgments or posterior beliefs.
AI chatbots increasingly offer memory features that personalize responses based on users' past conversations. We study the impact of memory on users' beliefs. The core idea is that when AI systems have memory of users' prior conversations, they may produce outputs that overly confirm users' prior beliefs. If users do not account for the correlation between the model output and their prior beliefs, they may act as if they exhibit confirmation bias. We run a large-scale experiment in which participants first discuss their views on political topics with an AI interviewer, then make probabilistic predictions about future events with AI assistance. We vary whether the AI prediction assistant has access to the participant's interview conversation. We analyze how the AI's predictions differ wh...