About

Barbara Mellers is the I. George Heyman University Professor at the University of Pennsylvania, with cross-appointments in the School of Arts and Sciences and the Wharton School. Her training is in cognitive psychology, and her research focuses on how and why people form beliefs, judgments, and preferences. She employs an experimental approach, manipulating and controlling variables to build mathematical models that describe underlying decision-making processes. Her work explores variables that influence behavior but are often absent from traditional theories of rational decision making, such as context effects and response mode effects. Additionally, she studies the impact of social and emotional variables on judgments and decisions, including how emotions influence choice, perceptions of fairness, and cooperative behavior in economic games. Her goal is to understand deviations from rational behavior and to improve judgment and decision-making processes. Mellers has contributed to understanding decision errors, such as those made by juries, and has worked on enhancing the accuracy of human predictions of geopolitical and economic events through large-scale experiments like the Good Judgment Project. Her research demonstrates that forecasting skills can be cultivated and that certain training, environments, and cognitive profiles lead to more accurate predictions.

Research topics

• Psychology
• Mathematics
• Medicine
• Applied psychology
• Nursing
• Environmental health

Selected publications

• Prompt Engineering Large Language Models' Forecasting Capabilities

  ArXiv.org · 2025-06-02

  preprintOpen accessSenior author

  Large language model performance can be improved in a large number of ways. Many such techniques, like fine-tuning or advanced tool usage, are time-intensive and expensive. Although prompt engineering is significantly cheaper and often works for simpler tasks, it remains unclear whether prompt engineering suffices for more complex domains like forecasting. Here we show that small prompt modifications rarely boost forecasting accuracy beyond a minimal baseline. In our first study, we tested 38 prompts across Claude 3.5 Sonnet, Claude 3.5 Haiku, GPT-4o, and Llama 3.1 405B. In our second, we introduced compound prompts and prompts from external sources, also including the reasoning models o1 and o1-mini. Our results show that most prompts lead to negligible gains, although references to base rates yield slight benefits. Surprisingly, some strategies showed strong negative effects on accuracy: especially encouraging the model to engage in Bayesian reasoning. These results suggest that, in the context of complex tasks like forecasting, basic prompt refinements alone offer limited gains, implying that more robust or specialized techniques may be required for substantial performance improvements in AI forecasting.

  PublisherOA PDFDOI

• Reply to Rohrer and Wenz and Arslan: The association between income and emotional well-being

  Proceedings of the National Academy of Sciences · 2024-10-31

  articleOpen accessSenior author
  PublisherOA PDFDOI

• Crowd prediction systems: Markets, polls, and elite forecasters

  International Journal of Forecasting · 2024-01-22 · 3 citations

  article
  PublisherDOI

• Human and Algorithmic Predictions in Geopolitical Forecasting: Quantifying Uncertainty in Hard-to-Quantify Domains

  Perspectives on Psychological Science · 2023-08-29 · 9 citations

  articleOpen access1st authorCorresponding

  Research on clinical versus statistical prediction has demonstrated that algorithms make more accurate predictions than humans in many domains. Geopolitical forecasting is an algorithm-unfriendly domain, with hard-to-quantify data and elusive reference classes that make predictive model-building difficult. Furthermore, the stakes can be high, with missed forecasts leading to mass-casualty consequences. For these reasons, geopolitical forecasting is typically done by humans, even though algorithms play important roles. They are essential as aggregators of crowd wisdom, as frameworks to partition human forecasting variance, and as inputs to hybrid forecasting models. Algorithms are extremely important in this domain. We doubt that humans will relinquish control to algorithms anytime soon-nor do we think they should. However, the accuracy of forecasts will greatly improve if humans are aided by algorithms.

  PublisherDOI

• Income and emotional well-being: A conflict resolved

  Proceedings of the National Academy of Sciences · 2023-03-01 · 168 citations

  articleOpen accessSenior author

  Do larger incomes make people happier? Two authors of the present paper have published contradictory answers. Using dichotomous questions about the preceding day, [Kahneman and Deaton, Proc. Natl. Acad. Sci. U.S.A. 107 , 16489–16493 (2010)] reported a flattening pattern: happiness increased steadily with log(income) up to a threshold and then plateaued. Using experience sampling with a continuous scale, [Killingsworth, Proc. Natl. Acad. Sci. U.S.A. 118 , e2016976118 (2021)] reported a linear-log pattern in which average happiness rose consistently with log(income). We engaged in an adversarial collaboration to search for a coherent interpretation of both studies. A reanalysis of Killingsworth’s experienced sampling data confirmed the flattening pattern only for the least happy people. Happiness increases steadily with log(income) among happier people, and even accelerates in the happiest group. Complementary nonlinearities contribute to the overall linear-log relationship. We then explain why Kahneman and Deaton overstated the flattening pattern and why Killingsworth failed to find it. We suggest that Kahneman and Deaton might have reached the correct conclusion if they had described their results in terms of unhappiness rather than happiness; their measures could not discriminate among degrees of happiness because of a ceiling effect. The authors of both studies failed to anticipate that increased income is associated with systematic changes in the shape of the happiness distribution. The mislabeling of the dependent variable and the incorrect assumption of homogeneity were consequences of practices that are standard in social science but should be questioned more often. We flag the benefits of adversarial collaboration.

  PublisherOA PDFDOI

• Reference-Point Theory: An Account of Individual Differences in Risk Preferences

  Perspectives on Psychological Science · 2023-09-14 · 5 citations

  articleOpen access1st authorCorresponding

  We propose an account of individual differences in risk preferences called "reference-point theory" for choices between sure things and gambles. Like most descriptive theories of risky choice, preferences depend on two drivers-hedonic sensitivities to change and beliefs about risk. But unlike most theories, these drivers are estimated from judged feelings about choice options and gamble outcomes. Furthermore, the reference point is assumed to be the less risky option (i.e., sure thing). Loss aversion (greater impact of negative change than positive change) and pessimism (belief the worst outcome is likelier) predict risk aversion. Gain seeking (greater impact of positive change than negative change and optimism (belief the best outcome is likelier) predict risk seeking. But other combinations of hedonic sensitivities and beliefs are possible, and they also predict risk preferences. Finally, feelings about the reference point predict hedonic sensitivities. When decision makers feel good about the reference point, they are frequently loss averse. When they feel bad about it, they are often gain seeking. Three studies show that feelings about reference points, feelings about options and feelings about outcomes predict risky choice and help explain why individuals differ in their risk preferences.

  PublisherDOI

• Are markets more accurate than polls? The surprising informational value of “just asking” – CORRIGENDUM

  Judgment and Decision Making · 2023-01-01

  erratumOpen accessSenior author

  An abstract is not available for this content. As you have access to this content, full HTML content is provided on this page. A PDF of this content is also available in through the ‘Save PDF’ action button.

  PublisherOA PDFDOI

• Fair Skill Brier Score: Evaluating Probabilistic Forecasts of One-Off Events with Different Numbers of Categorical Outcomes

  SSRN Electronic Journal · 2023-01-01 · 1 citations

  articleOpen access
  PublisherDOI

• Decomposing the effects of crowd-wisdom aggregators: The bias–information–noise (BIN) model

  International Journal of Forecasting · 2022-02-02 · 12 citations

  articleSenior author
  PublisherDOI

• Predicting the future with humans and AI

  Consumer Psychology Review · 2022-12-07 · 4 citations

  article1st authorCorresponding

  Abstract We review the classic clinical versus statistical prediction debate as well as related modern work on humans versus. algorithms. Despite the successes of statistical prediction over clinical prediction, there is still widespread resistance to algorithms. We discuss recent attempts to understand that resistance. Current research focuses on when people use algorithmic predictions, how people perceive algorithms, and how algorithms can be made more appealing. We also examine attempts to boost human forecasting accuracy, either by spotting talent, cultivating talent via training, or developing algorithms that aggregate individual forecasts. We hypothesize that hybrid models with both human and algorithmic predictions may encounter less resistance than algorithms alone, especially when the algorithm is “humanized” (with anthropomorphic features) and the human is “algorithmized” (by reducing noise, decreasing bias and increasing signal).

  PublisherDOI

Recent grants

• Forecasting and Political Discourse

  NSF · $490k · 2016–2020

Frequent coauthors

• Philip E. Tetlock

  49 shared

• Lyle Ungar

  California University of Pennsylvania

  24 shared

• Michael H. Birnbaum

  California State University System

  23 shared

• Alan Schwartz

  Research Network (United States)

  22 shared

• Pavel Atanasov

  IE University

  17 shared

• A. Peter McGraw

  University of Colorado Boulder

  17 shared

• Anne Cooke

  16 shared

• Philip E. Tetlock

  University of Pennsylvania

  16 shared

Labs

• Solomon LabsPI