Idiographic Personality Gaussian Process for Psychological Assessment

TL;DR

The paper addresses the challenge of balancing shared population structure with individual-level deviations in dynamic, longitudinal ordinal psychometric data. It introduces IPGP, a multi-task Gaussian process coregionalization framework that models grouped batteries with an ordinal likelihood and unit-specific deviations, using stochastic variational inference for scalable learning. IPGP enables both accurate prediction of responses and discovery of individualized taxonomies, supported by theory-testing via Bayes factors. Empirically, it demonstrates superior performance in simulations, cross-sectional Big Five validation, and a longitudinal ESM study, offering a principled path toward personalized psychological diagnosis and treatment.

Abstract

We develop a novel measurement framework based on a Gaussian process coregionalization model to address a long-lasting debate in psychometrics: whether psychological features like personality share a common structure across the population, vary uniquely for individuals, or some combination. We propose the idiographic personality Gaussian process (IPGP) framework, an intermediate model that accommodates both shared trait structure across a population and "idiographic" deviations for individuals. IPGP leverages the Gaussian process coregionalization model to handle the grouped nature of battery responses, but adjusted to non-Gaussian ordinal data. We further exploit stochastic variational inference for efficient latent factor estimation required for idiographic modeling at scale. Using synthetic and real data, we show that IPGP improves both prediction of actual responses and estimation of individualized factor structures relative to existing benchmarks. In a third study, we show that IPGP also identifies unique clusters of personality taxonomies in real-world data, displaying great potential in advancing individualized approaches to psychological diagnosis and treatment.

Figures (5)

• Figure 1: Proposed ipgp model for inferring latent factors and factor loadings from dynamic ordinal data. Input ordinal observations across channels are modeled as ordinal transformations of latent dynamic Gaussian processes with individualized rbf kernels and loading matrices.
• Figure 2: Illustration of raw correlation matrix (left) and our estimated Big Five loading matrix (right). Both correlation matrices displace a block pattern, where estimated interpersonal variation show strong correlation between questions within the same factor of the Big Five personalities and weak correlation across different factors. Besides, questions corresponding negative emotionality show minor negative correlation with those corresponding to extraversion and conscientiousness, suggesting trait-by-trait interaction effects.
• Figure 3: Predictive accuracy and log lik of ipgp and ipgp-nom for the forecasting task and leave-one-trait-out cross-validation task.
• Figure 4: Four residual correlations as identified by our k-mean clustering. Each heatmap displays the trait-level residual correlation averaged across corresponding batteries for one cluster, with darker red and blue indicating larger positive and negative deviations. For instance, agreeableness (A) is more correlated to extraversion (E) than the population profile in the first profile, but less correlated to openness (O) in the second profile. Moreover, these two directions of deviations are even exacerbated in the third and fourth profiles.
• Figure 5: Estimated correlations of selective individuals for the identified four profiles in the longitudinal study.