@inproceedings{beak-etal-2026-evaluating,
title = "Evaluating Bias in {LLM}-Synthetic Data across Majority and Minority Groups on General and Stigmatized Topics: The Case of Multicultural and Monocultural Adolescents in {S}outh {K}orea",
author = "Beak, Jiyun and
Han, Myokyung and
Kim, Lanu",
editor = "Atwell, Katherine and
Borah, Angana and
Dementieva, Daryna and
Kreiss, Elisa and
Kotonya, Neema and
Liu, Jiarui and
Olson, Liz and
Wan, Ruyuan and
Zhao, Jieyu",
booktitle = "Proceedings of the Fifth Workshop on {NLP} for Positive Impact ({NLP}4{PI})",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.nlp4pi-1.15/",
pages = "161--174",
ISBN = "979-8-89176-495-8",
abstract = "Large Language Models (LLMs) are increasingly used to replicate human responses in survey data, yet it remains unclear whether they perform equally well for social minorities and majorities. Using a unique survey conducted in South Korea that provided identical questionnaires to adolescents from multicultural and monocultural families, we examine group-level differences in the accuracy of LLM-generated synthetic data. We construct personas that mirror individual-level demographic profiles and prompt two LLMs, llama3:latest and gpt-4o, to answer 77 survey questions across 11 question domains. We find marginal differences in LLM performance between minority and majority groups. Instead, replication performance varies by whether question prompts contain socially stigmatized words and by which LLM model we use. These findings highlight both the potential and the limitations of using LLMs to generate synthetic survey data targeting social minorities."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="beak-etal-2026-evaluating">
<titleInfo>
<title>Evaluating Bias in LLM-Synthetic Data across Majority and Minority Groups on General and Stigmatized Topics: The Case of Multicultural and Monocultural Adolescents in South Korea</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jiyun</namePart>
<namePart type="family">Beak</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Myokyung</namePart>
<namePart type="family">Han</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lanu</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Fifth Workshop on NLP for Positive Impact (NLP4PI)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Katherine</namePart>
<namePart type="family">Atwell</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Angana</namePart>
<namePart type="family">Borah</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Daryna</namePart>
<namePart type="family">Dementieva</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Elisa</namePart>
<namePart type="family">Kreiss</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Neema</namePart>
<namePart type="family">Kotonya</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiarui</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Liz</namePart>
<namePart type="family">Olson</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruyuan</namePart>
<namePart type="family">Wan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jieyu</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-495-8</identifier>
</relatedItem>
<abstract>Large Language Models (LLMs) are increasingly used to replicate human responses in survey data, yet it remains unclear whether they perform equally well for social minorities and majorities. Using a unique survey conducted in South Korea that provided identical questionnaires to adolescents from multicultural and monocultural families, we examine group-level differences in the accuracy of LLM-generated synthetic data. We construct personas that mirror individual-level demographic profiles and prompt two LLMs, llama3:latest and gpt-4o, to answer 77 survey questions across 11 question domains. We find marginal differences in LLM performance between minority and majority groups. Instead, replication performance varies by whether question prompts contain socially stigmatized words and by which LLM model we use. These findings highlight both the potential and the limitations of using LLMs to generate synthetic survey data targeting social minorities.</abstract>
<identifier type="citekey">beak-etal-2026-evaluating</identifier>
<location>
<url>https://aclanthology.org/2026.nlp4pi-1.15/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>161</start>
<end>174</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Evaluating Bias in LLM-Synthetic Data across Majority and Minority Groups on General and Stigmatized Topics: The Case of Multicultural and Monocultural Adolescents in South Korea
%A Beak, Jiyun
%A Han, Myokyung
%A Kim, Lanu
%Y Atwell, Katherine
%Y Borah, Angana
%Y Dementieva, Daryna
%Y Kreiss, Elisa
%Y Kotonya, Neema
%Y Liu, Jiarui
%Y Olson, Liz
%Y Wan, Ruyuan
%Y Zhao, Jieyu
%S Proceedings of the Fifth Workshop on NLP for Positive Impact (NLP4PI)
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-495-8
%F beak-etal-2026-evaluating
%X Large Language Models (LLMs) are increasingly used to replicate human responses in survey data, yet it remains unclear whether they perform equally well for social minorities and majorities. Using a unique survey conducted in South Korea that provided identical questionnaires to adolescents from multicultural and monocultural families, we examine group-level differences in the accuracy of LLM-generated synthetic data. We construct personas that mirror individual-level demographic profiles and prompt two LLMs, llama3:latest and gpt-4o, to answer 77 survey questions across 11 question domains. We find marginal differences in LLM performance between minority and majority groups. Instead, replication performance varies by whether question prompts contain socially stigmatized words and by which LLM model we use. These findings highlight both the potential and the limitations of using LLMs to generate synthetic survey data targeting social minorities.
%U https://aclanthology.org/2026.nlp4pi-1.15/
%P 161-174
Markdown (Informal)
[Evaluating Bias in LLM-Synthetic Data across Majority and Minority Groups on General and Stigmatized Topics: The Case of Multicultural and Monocultural Adolescents in South Korea](https://aclanthology.org/2026.nlp4pi-1.15/) (Beak et al., NLP4PI 2026)
ACL