Analyzing the Temporal Factors for Anxiety and Depression Symptoms with the Rashomon Perspective

TL;DR

The paper tackles the problem of interpretability in mental health analytics by arguing that a single best-performing model can mask alternative explanations for anxiety and depression risk. It operationalizes the Rashomon perspective through bootstrap-generated near-optimal models (the Rashomon set) and uses Partial Dependence Profiles with bootstrap confidence intervals to assess the stability of predictor effects across models. Applying this framework to a large PHQ-4 dataset (n ≈ 34k), the authors find robust demographic effects (age, sex, education) and clear temporal patterns, including diurnal and circaseptan fluctuations, in predicted anxiety/depression risk. The study demonstrates the value of multiplicity-aware interpretation for improving the robustness and generalizability of screening-based inferences in public mental health and highlights practical implications for population surveillance.

Abstract

This paper introduces a new modeling perspective in the public mental health domain to provide a robust interpretation of the relations between anxiety and depression, and the demographic and temporal factors. This perspective particularly leverages the Rashomon Effect, where multiple models exhibit similar predictive performance but rely on diverse internal structures. Instead of considering these multiple models, choosing a single best model risks masking alternative narratives embedded in the data. To address this, we employed this perspective in the interpretation of a large-scale psychological dataset, specifically focusing on the Patient Health Questionnaire-4. We use a random forest model combined with partial dependence profiles to rigorously assess the robustness and stability of predictive relationships across the resulting Rashomon set, which consists of multiple models that exhibit similar predictive performance. Our findings confirm that demographic variables \texttt{age}, \texttt{sex}, and \texttt{education} lead to consistent structural shifts in anxiety and depression risk. Crucially, we identify significant temporal effects: risk probability demonstrates clear diurnal and circaseptan fluctuations, peaking during early morning hours. This work demonstrates the necessity of moving beyond the best model to analyze the entire Rashomon set. Our results highlight that the observed variability, particularly due to circadian and circaseptan rhythms, must be meticulously considered for robust interpretation in psychological screening. We advocate for a multiplicity-aware approach to enhance the stability and generalizability of ML-based conclusions in mental health research.

Figures (2)

• Figure 1: The Grouped PDPs for GAD-2 Risk Across the Factors. They illustrate the model-predicted relationship between the plotted factors and the estimated probability of the GAD-2 score being greater than 3. The lines represent the predicted probability values, and the surrounding bands indicate the 95% confidence intervals. Non-overlapping confidence intervals between groups suggest differences in predicted risk.
• Figure 2: The Grouped PDPs for the PHQ-2 Risk Across the Factors. They illustrate the model-predicted relationship between the plotted factors and the estimated probability of the PHQ-2 score being greater than 3. The lines represent the predicted probability values, and the surrounding bands indicate the 95% confidence intervals. Non-overlapping confidence intervals between groups suggest differences in predicted risk.