Reveal Data Scientists and Partners Lead Conversations at Premier Statistics Conference
In May, Reveal Global Consulting shared some of its latest innovative research and client solutions at the American Association for Public Opinion Research (AAPOR) Annual Conference, which convenes leaders of public opinion and survey research to discuss advances in the field. This year’s theme, “Reshaping Democracy’s Oracle,” prompted thought-provoking conversations about the promises and perils that artificial intelligence (AI) and our information technology landscape have in changing political polling and survey research.
A few major points of discussion emerged, including how AI is transforming the way surveys are conducted, what risks accompany this change, and how restrictions can be implemented to manage this change safely. Other topics of discussion included the demographics of regular AI users, how useful synthetic respondents are, and strategies for reaching hard-to-count populations.
Taylor Wilson, VP of Applied Statistics & Data Science at Reveal, noted that “AAPOR is a great way for professionals in the survey space to connect and share ideas. This year’s conference tackled big questions around AI as the field looks to transform itself in the face of new technology.”
Two teams of Reveal data scientists were selected to present their research at the conference. The first, titled “AI Language Translation Evaluation: A Mixed-Methods Approach,” investigated the question, “How do individuals perceive and interpret AI-translated texts compared to professionally translated texts?” Reveal data scientist Nicole Cabrera, the first author of this study, was awarded a 2025 Burns “Bud” Roper Fellowship, a recognition awarded annually to early-career researchers in public opinion research. Nicky and her co-authors, Yezzi Angi Lee, Youlee Shin, Aruna Peri, and Taylor Wilson, presented findings that sparked engaging questions from the audience regarding cognitive interviewing methods and prompting strategies and the potential for expanding AI translation research into other languages and AI platforms beyond GPT-4o.
“AAPOR was a fun and enlightening experience,” Nicky reflected. “It was amazing to see how others approached complex challenges – from multilingual survey translation to large-scale data classification with AI. As a data scientist and a public health doctoral student, I found it both useful and insightful to see how rigorous quantitative methods are being used to address evolving questions in public opinion research.” You can read more about their methods and findings here.
In the second Reveal-led session, Yezzi Angi Lee presented “Quality Control for Autocoding: Enhancing Industry and Occupation Coding in the American Community Survey (ACS).” Her team developed a quality control framework to evaluate the performance of ACS’s existing autocoding program. They also compared the current autocoder’s performance against a new Large Language Model (LLM)-based approach with the goal of identifying where the LLM outperforms or falls short compared to the current autocoding program, to determine which types of open-ended responses are better handled by LLMs and semantic models. Yezzi was supported by fellow Reveal data scientists Nicole Cabrera, Jackson Chen, and Jiahui Xu, as well as U.S. Census Bureau researchers Julia Beckhusen, Lynda Laughlin, and Ana J. Montalvo. The presentation prompted a discussion around the challenges and limitations of automated classification and highlighted the importance of developing a robust quality control framework to ensure accuracy and reliability in autocoding. Reveal also presented information about the LLM autocoder for the ACS at the FedCASIC Conference in April. You can read more about that here.
Reflecting on the conference, Yezzi shared “The AAPOR conference was a valuable experience where I had the chance to learn from ongoing research and hear how others are approaching similar challenges. It was encouraging to see so many scholars and teams tackling the same issues we face, and I appreciated the opportunity to exchange ideas, share our work, and engage in meaningful discussions.”
One of the highlights of the conference was a moderated discussion with representatives from the University of Michigan and the Pew Research Center on how to survey hard-to-reach populations, like those who speak languages that are uncommon in the United States, people who do not have homes, and college students whose listed address is somewhere other than their school. Presenters shared innovative techniques they are using to connect with people who fall into these categories, resulting in greater representation of Americans in research that affects policymaking. Reveal engaged in a partnership this past semester with the University of Michigan to work with talented students from the School of Information to help design and test innovative optical character recognition (OCR) technologies and their applications to federal use cases. We will be showcasing this work at the Joint Statistical Meetings in August.