The Trade-Offs in Promoting Equity-Focused Initiatives in Crowdfunding

Journal of the Association for Information Systems · 2025-01-01

Organizations interested in supporting matters of diversity and equity need guidance on whether to be transparent about their equity-focused initiatives and appeal to users’ sense of social justice. This paper explores how information transparency affects backer responses to equity-related initiatives. The educational crowdfunding platform DonorsChoose launched an “equity-focused” initiative to highlight the platform’s commitment to equity in education and encourage donations towards that cause. Using an observational analysis and an experimental study, I found evidence that information transparency, i.e., explicitly promoting equity-focused projects, dampens overall donations because any change in donations associated with the equity-focused initiatives is insufficient to offset the decrease in other donations. However, I found evidence that distant donors are more responsive to equity-focused recommendations, consistent with the hypothesis that a holistic mindset moderates the effectiveness of promoting equity-focused initiatives. This study provides insight into the trade-offs associated with information transparency for equity-related initiatives and the complicated relationship between organizations and equity-related initiatives.

Engagement Diversification and the Effectiveness of Platform Content Moderation: An Instrumental Variable Approach

SSRN Electronic Journal · 2025-01-01

Fighting Misinformation on Social Media: An Empirical Investigation of the Impact of Prominence Reduction Policies

SSRN Electronic Journal · 2025-01-01

One world, one opinion? The superstar effect in LLM responses

2025-01-01 · 1 citations

As large language models (LLMs) are shaping the way information is shared and accessed online, their opinions have the potential to influence a wide audience.This study examines who is predicted by the studied LLMs as the most prominent figures across various fields, while using prompts in ten different languages to explore the influence of linguistic diversity.Our findings reveal low diversity in responses, with a small number of figures dominating recognition across languages (also known as the "superstar effect").These results highlight the risk of narrowing global knowledge representation when LLMs retrieve subjective information.

Fairness principles across contexts: evaluating gender disparities of facts and opinions in large language models

AI and Ethics · 2025-12-05

Abstract This paper examines how fairness principles differ when evaluating large language model (LLM) outputs in fact-based versus opinion-based contexts, focusing on gender disparities in responses related to notable individuals. Using prompts designed to elicit either factual information (identifying Nobel Prize winners) or subjective judgments (identifying the most accomplished figures in a field), we analyze responses from GPT-4, Claude, and Llama-3. For fact-based tasks, fairness is assessed through correctness and refusal rates, revealing minimal gender disparities when models achieve high accuracy, although refusal patterns can vary by model and gender. For opinion-based tasks, where no single correct answer exists, fairness is operationalized through representational metrics such as demographic parity and disparate impact. Results show substantial gender disparities in opinion-based outputs across all models, with representation shaped by prompt wording (e.g., “important” vs. “prestigious”), subject domain, and inclusion of secondary answers. However, the highly skewed context makes the final assessment about fairness challenging. Our findings highlight that fairness metrics and interpretations must be contextualized by output type. Performance parity is an appropriate goal for fact-based outputs, whereas representational inclusivity is central for opinion-based outputs. Representational inclusivity alone may not be sufficient when the context for the LLM’s task differs from the population. We discuss theoretical implications for fairness evaluation, noting that high performance can mitigate disparities in factual contexts but that opinion-based contexts require more nuanced, values-driven approaches.

Where Does the Hate Flow? The Impact of Multihoming on User Responses to Content Moderation

SSRN Electronic Journal · 2025-01-01

How Do LLMs Impact Human-Provided Mental Healthcare Services? A Study of a Mental Health Forum

Academy of Management Proceedings · 2025-07-01

Mental healthcare has become a concern globally, and the challenges of providing and accessing appropriate care are only magnified by a dearth of available providers. Large Language Models (LLMs) are anticipated to hold transformative potential for numerous industries including healthcare. LLMs demonstrate superior performance in medical expertise, communication skill, as well as emotional and social intelligence, and are expected to alleviate the problem of workforce undersupply in mental healthcare. However, the question of how human-provided services will evolve in response to the emergence of LLMs remains critical but unanswered. We focus on an online mental health forum wherein mental healthcare counselors offer support to seekers through question-and-answer (Q&A). In early 2023, an LLM-powered chatbot was integrated into the forum to automatically leave responses to seekers. We investigate changes in the engagement behavior and the nature of support from human counselors—both of which are essential for seekers’ mental well-being and counseling success. We observe human counselors’ disengagement and their downgrading support in the forum. Concerningly, the disengagement became more pronounced in tougher cases, such as seekers expressing suicidal ideation. Furthermore, the observed effects extended to services that were not directly exposed to the chatbot, indicating a spillover effect. Overall, our study provides important implications for AI integration in mental healthcare, offering actionable insights for multiple stakeholders.

One world, one opinion? The superstar effect in LLM responses

arXiv (Cornell University) · 2024-12-13

As large language models (LLMs) are shaping the way information is shared and accessed online, their opinions have the potential to influence a wide audience. This study examines who the LLMs view as the most prominent figures across various fields, using prompts in ten different languages to explore the influence of linguistic diversity. Our findings reveal low diversity in responses, with a small number of figures dominating recognition across languages (also known as the "superstar effect"). These results highlight the risk of narrowing global knowledge representation when LLMs retrieve subjective information.

Fighting Misinformation on Social Media: An Empirical Investigation of the Impact of Prominence Reduction Policies

Production and Operations Management · 2024-09-09 · 4 citations

Misinformation has dire implications for both public welfare and the operational aims of user-generated content platforms. As a result, platforms have adopted various content moderation policies aimed at decreasing the volume and impact of misinformation. However, implementing new platform policies runs the risk of decreasing user contribution and alienating core users, and results regarding the efficacy of such policies are mixed. Herein, we empirically assess a prominence reduction policy applied to a problematic group that is high in misinformation. The goal of this policy is to reduce the visibility of misinformation on the platform (rather than deleting misinformation or banning users). The results show that while prominence reduction diminishes misinformation dissemination in the focal group, this method also results in a spillover of misinformation to topically related spaces. This spillover is short-lived and driven primarily by a small set of problematic users. As misinformation is not contagious, we find that this spillover to external groups diminishes over time. Finally, prominence reduction is found to have no impact on non-misinformation contribution on the studied platform. The findings of this study have important implications for platform operations and provide useful recommendations for managers regarding effective ways to reduce the spread of misinformation.

Evaluating LLMs for Gender Disparities in Notable Persons

arXiv (Cornell University) · 2024 · 2 citations

• Political Science
• Political Science
• Psychology

This study examines the use of Large Language Models (LLMs) for retrieving factual information, addressing concerns over their propensity to produce factually incorrect "hallucinated" responses or to altogether decline to even answer prompt at all. Specifically, it investigates the presence of gender-based biases in LLMs' responses to factual inquiries. This paper takes a multi-pronged approach to evaluating GPT models by evaluating fairness across multiple dimensions of recall, hallucinations and declinations. Our findings reveal discernible gender disparities in the responses generated by GPT-3.5. While advancements in GPT-4 have led to improvements in performance, they have not fully eradicated these gender disparities, notably in instances where responses are declined. The study further explores the origins of these disparities by examining the influence of gender associations in prompts and the homogeneity in the responses.