Raymond Duch

Social media affect financial markets, but public posts by financial media personas are voluntary disclosures. What is not disclosed is therefore usually unobserved. We address this measurement problem by conducting repeated, real-time interviews of "digital twins" built from monitored finfluencers' X accounts under a fixed protocol. The interviews recover stocklevel public-persona belief proxies even when no public recommendation is made. Because the interviews are generated and archived before the relevant return windows, the design avoids the look-ahead bias that arises when LLMs are queried ex post. The evidence shows that information obtained from these digital-twin interviews predicts the cross section of large-cap stock returns in the expected direction. Repeated real-time interviews therefore show how selective disclosure can be turned into measurable panels of market views.

The study of public opinion occupies an uneasy space between theory and measurement, referring simultaneously to a latent set of beliefs held by members of a political community and to the responses elicited through imperfect, context-sensitive instruments. The integration of AI into public opinion research makes this tension unavoidable. AI offers tools that are faster, cheaper, and more flexible than anything available, but the effects also extend further in ways that can blur distinctions between measuring and manufacturing opinion. AI raises questions the field has long grappled with: What exactly is public opinion, and what counts as evidence of it? Whether AI enriches or distorts the study of public opinion depends on choices that are as much theoretical as technical. In the limit, substituting AI-generated responses for human voices risks a form of epistemic collapse in which public opinion research loses its anchor to the people it purports to represent.

We examined variation in attitude toward mandatory COVID-19 vaccination in early life by sociodemographic characteristics, personal COVID-19 experience, health risk attitude, political ideology, and other COVID-19 vaccine mandate attitudes. For this purpose, we used data from 19 928 participants from 16 countries surveyed in March–November 2022 for the second wave of the COVID-19 vaccine preference and opinion survey (CANDOUR). Analyses were adjusted for poststratification weighting. Participants who disagreed with early life mandatory COVID-19 vaccination were more likely to decline vaccination against COVID-19 (14.5%) compared with those who were neutral (1.8%) and those who agreed (0.5%). Disagreement with early life mandatory vaccination was associated with more unwillingness to take risks with their own health, being centre or left-wing on the left-right political spectrum, believing COVID-19 vaccination should be a personal choice, and being opposed to vaccine mandates for schoolchildren and the public. Neutrality or agreement with early life mandatory vaccination was associated with neutrality or agreement with a vaccine mandate for schoolchildren or a governmental COVID-19 vaccine mandate for everybody. Pandemic preparedness governance needs to focus on attitudes toward vaccine mandates. Further research and commitment by governments at various levels are needed to identify social, cultural, and system-level factors that could inform vaccination strategies to be implemented for the next pandemic.

The recent COVID-19 pandemic offers a rare opportunity to understand how citizens attribute responsibility for governments’ responses to unanticipated negative—and in this case, systemic—exogenous shocks. Classical accounts of responsibility are complicated when crises are pervasive, involve multiple valence dimensions, and where individuals can make relative assessments of performance. We fielded a conjoint experiment in 16 countries with 22,147 respondents. In this experiment, subjects made re-election decisions regarding 178,184 randomly assigned incumbent profiles. We find that incumbents’ performance along both health and economic dimensions drives these hypothetical reelection decisions. Using machine learning techniques, we find only muted heterogeneity in the magnitude and distribution of these treatment effects. This result suggests that these widely reported performance signals have consistent political effects across countries. In a complementary analysis, we also find that subjects’ intentions to vote for incumbent governments are positively correlated with subjective and relative evaluations of the government’s pandemic performance, along both health and economic dimensions. These results provide consistent evidence that evaluations of pandemic performance matter politically.

The initial predictions presented in this article confirm that presidential candidate vote-share estimates based on AI polling are broadly exchangeable with those of other polling organizations. We present our first two biweekly vote-share estimates for the 2024 US presidential election and benchmark them against those being generated by other polling organizations. Our post–Democratic National Convention top-line estimates for Trump (47%) and Harris (46%) closely track measurements generated by other polls during the month of August. The subsequent early September (post-debate) PoSSUM vote-share estimates for Trump (47%) and Harris (48%) again closely track with other national polling being conducted in the United States. An ultimate test for the PoSSUM polling method will be the final preelection vote-share results that we publish before Election Day on November 5, 2024.

We implemented a clustered randomized controlled trial with 6,963 residents in six rural Ghana districts to estimate the causal impact of financial incentives on coronavirus disease 2019 (COVID-19) vaccination uptake. Villages randomly received one of four video treatment arms: a placebo, a standard health message, a high cash incentive (60 Ghana cedis) and a low cash incentive (20 Ghana cedis). For the first co-primary outcome—COVID-19 vaccination intentions—non-vaccinated participants assigned to the cash incentive treatments had an average rate of 81% (1,733 of 2,168) compared to 71% (1,895 of 2,669) for those in the placebo treatment arm. For the other co-primary outcome of self-reported vaccinations 2 months after the initial intervention, the average rate for participants in the cash treatment was 3.5% higher than for participants in the placebo treatment (95% confidence interval (CI): 0.001, 6.9; P = 0.045): 40% (602 of 1,486) versus 36.3% (672 of 1,850). We also verified vaccination status of participants: in the cash treatment arm, 36.6% (355 of 1,058) of verified participants had at least one dose of the COVID-19 vaccine compared to 30.3% (439 of 1,544) for those in the placebo—a difference of 6.3% (95% CI: 2.4, 10.2; P = 0.001). For the intention and the vaccination outcomes, the low cash incentive (20 Ghana cedis) had a larger positive effect on COVID-19 vaccine uptake than the high cash incentive (60 Ghana cedis). Trial identifier: AEARCTR-0008775.

The rapid advancement of ‘deepfake’ video technology — which uses deep learning artificial intelligence algorithms to create fake videos that look real — has given urgency to the question of how policymakers and technology companies should moderate inauthentic content. We conduct an experiment to measure people’s alertness to and ability to detect a high-quality deepfake amongst a set of videos. First, we find that in a natural setting with no content warnings, individuals who are exposed to a deepfake video of neutral content are no more likely to detect anything out of the ordinary (32.9%) compared to a control group who viewed only authentic videos (34.1%). Second, we find that when individuals are given a warning that at least one video in a set of five videos is a deepfake, only 21.6% of respondents correctly identify the deepfake as the only inauthentic video, while the remainder erroneously select at least one genuine video as a deepfake.