How closely do large language model scores from parent interviews match self-reported parenting?

Large language models (LLMs) offer a potential approach to inferring standardized questionnaire responses from qualitative interviews, but their utility for assessing parenting remains unclear. We evaluated six generative LLMs using zero-shot prompting to infer responses to the Multidimensional Assessment of Parenting Scale (MAPS) from interviews with 43 parents. Models inferred responses to 35 MAPS items from two introductory questions about family, parenting roles, and the meaning of caregiving. Performance was evaluated through exact item-response agreement, correlations between inferred and self-reported parenting scores, systematic score differences, and internal consistency. Across models, Fisher-averaged correlations between inferred and self-reported MAPS subscale scores ranged from .320 to .434, with no consistently superior model. Correlations were strongest and most consistent for Warmth, whereas Physical Control showed little correspondence and substantial floor restriction. Across parent–item pairs, 35.5%–44.7% of inferred responses exactly matched parents’ selected response categories. All models assigned systematically lower response categories than parents and rarely selected the highest category; score underestimation was particularly evident for Positive Parenting. Model-derived subscales generally showed internal consistency, although higher internal consistency did not consistently correspond to stronger correlations with parent-reported scores. Exploratory analyses indicated modest similarity in the associations of inferred and self-reported parenting scores with other psychological constructs. Preliminary comparisons also suggested that shorter interview segments substantially reduced computational requirements without clear evidence of poorer performance. These findings support further evaluation of LLM-based parenting assessment while highlighting limitations in reproducing individual responses, matching score levels, and capturing some parenting dimensions.

Authors

Publication Details

Journal
PsyArXiv (OSF Preprints)
Published
2026-11-18
DOI
https://doi.org/10.31234/osf.io/47nvm
Primary Topic
Psychometric Methodologies and Testing
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

How closely do large language model scores from parent interviews match self-reported parenting?

Violeta J. Rodriguez, Qimin Liu, Aryan Ramchandra Kapadia, Anudeepa Ramachandiran
3 links
PsyArXiv (OSF Preprints)
Psychometric Methodologies and Testing
preprint

How closely do large language model scores from parent interviews match self-reported parenting?

Violeta J. Rodriguez, Qimin Liu, Aryan Ramchandra Kapadia, Anudeepa Ramachandiran
preprint en

Abstract

Large language models (LLMs) offer a potential approach to inferring standardized questionnaire responses from qualitative interviews, but their utility for assessing parenting remains unclear. We evaluated six generative LLMs using zero-shot prompting to infer responses to the Multidimensional Assessment of Parenting Scale (MAPS) from interviews with 43 parents. Models inferred responses to 35 MAPS items from two introductory questions about family, parenting roles, and the meaning of caregiving. Performance was evaluated through exact item-response agreement, correlations between inferred and self-reported parenting scores, systematic score differences, and internal consistency. Across models, Fisher-averaged correlations between inferred and self-reported MAPS subscale scores ranged from .320 to .434, with no consistently superior model. Correlations were strongest and most consistent for Warmth, whereas Physical Control showed little correspondence and substantial floor restriction. Across parent–item pairs, 35.5%–44.7% of inferred responses exactly matched parents’ selected response categories. All models assigned systematically lower response categories than parents and rarely selected the highest category; score underestimation was particularly evident for Positive Parenting. Model-derived subscales generally showed internal consistency, although higher internal consistency did not consistently correspond to stronger correlations with parent-reported scores. Exploratory analyses indicated modest similarity in the associations of inferred and self-reported parenting scores with other psychological constructs. Preliminary comparisons also suggested that shorter interview segments substantially reduced computational requirements without clear evidence of poorer performance. These findings support further evaluation of LLM-based parenting assessment while highlighting limitations in reproducing individual responses, matching score levels, and capturing some parenting dimensions.

PsyArXiv (OSF Preprints)
Psychometric Methodologies and Testing
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon