[Paper Review] Hierarchical Latent Space Item Response Model for Analyzing Mental Health Vulnerability of Elementary School Students in South Korea

Soyeon Park, Seoyoung Shin+3 more|arXiv (Cornell University)|Mar 14, 2026

Mental Health Research Topics0 citations

TL;DR

The paper introduces HLSIRM, a Bayesian hierarchical latent space IRT model with inner-product respondent-item interactions to identify school- and student-level mental health vulnerability patterns among Korean elementary students. It applies the model to 2,210 students across 35 Incheon schools to derive school-specific vulnerability domains and targeted intervention insights.

ABSTRACT

Mental health difficulties among elementary school students represent a growing public health concern in South Korea, yet analytical tools for identifying school-specific vulnerability patterns from item response data remain limited. We propose the hierarchical latent space item response model (HLSIRM), which adds hierarchical respondent effects and an inner-product latent interaction for signed respondent-item associations, yielding a unified interaction map that separates school, individual main effects from school/individual-item interactions. We apply HLSIRM to mental health vulnerability data from 2,210 elementary school students across 35 schools in Incheon, South Korea. Clustering item vectors by directional similarity identifies four empirically derived vulnerability domains. School-level analysis reveals that the absence of counseling experience is the primary vulnerability domain aligned with most school vectors, while stress, depression, and smartphone dependency concentrate in specific schools. Within-school analysis demonstrates how individual student positions in the interaction map translate into targeted intervention strategies that address school-specific needs.

Motivation & Objective

• Motivate the need to identify school-specific mental health vulnerability patterns among elementary students in Korea.
• Develop and formulate the hierarchical latent space item response model (HLSIRM) that combines hierarchical respondent effects with inner-product respondent–item interactions.
• Enable separation of school/individual main effects from school/individual–item interaction effects within a unified interaction map.
• Apply HLSIRM to Incheon data to uncover empirical vulnerability domains and inform targeted interventions.

Proposed method

• Propose a Bayesian HLSIRM with logit-prior response model incorporating student intercepts, a fixed item intercept, and an inner-product interaction term between student and item latent vectors.
• Use a hierarchical structure where student parameters are drawn from school-level distributions to capture between-school differences and within-school heterogeneity.
• Adopt conjugate priors for tractable Bayesian inference and implement MCMC with joint parameter updates and Procrustes alignment to address rotational invariance.
• Fix the interaction map dimension to D=2 for visualization and interpretability, while maintaining measurement invariance by keeping item parameters non-hierarchical across schools.

Experimental results

Research questions

• RQ1 How do school-level and student-level effects interact to produce mental health vulnerability patterns across different schools?
• RQ2 Can the inner-product latent space capture meaningful positive, negative, and neutral respondent–item associations in a mental health context?
• RQ3 What are the empirical vulnerability domains across items that emerge when accounting for hierarchical structure and interactions?
• RQ4 How can school-specific interaction patterns inform targeted, school-level intervention strategies?

Key findings

• The model identifies four empirically derived vulnerability domains via clustering of item vectors by directional similarity.
• School-level analysis shows the absence of counseling experience as a primary vulnerability domain aligned with most school vectors.
• Certain domains such as stress, depression, and smartphone dependency concentrate in specific schools.
• Within-school analysis links student positions in the interaction map to targeted intervention needs addressing school-specific vulnerabilities.
• The approach preserves measurement invariance of item parameters while capturing between-school heterogeneity through a hierarchical latent space.