Deep Beats, Deep Thoughts? Predicting General Cognitive Ability from Natural Music-Listening Behavior

TL;DR

This study explores how music-listening behavior can predict general cognitive ability using real-world smartphone data.

Contribution

It introduces a novel approach to assessing cognitive ability through natural, non-achievement-related behavior.

Findings

Random forest models showed small but reliable associations between music-listening behavior and cognitive ability.

Lyrics-based preferences were the most informative feature for predicting cognitive ability.

Audio characteristics contributed little to the predictive power of the models.

Abstract

Music is more than just entertainment. It is a complex auditory stimulus that engages various cognitive processing systems. Accordingly, natural music-listening patterns may reveal insights into individual differences in general cognitive ability (GCA). In this study (N = 185), we used real-world smartphone-based music-listening records collected over five months to explore this question. We quantified participants’ listening habits (e.g., listening durations) and music preferences based on audio characteristics (e.g., tempo, mode) and lyrical characteristics (e.g., positive emotion words, affiliation words) of the songs they had listened to. These strictly behavioral features were used to predict GCA scores using linear LASSO regression and nonlinear random forest models. Out-of-sample cross-validation indicated modest predictive performance, with only the random forest model detecting small but reliable associations between music-listening behavior and GCA. Interpretable machine learning analyses showed that lyrics-based preferences were the most informative feature group, followed by listening habits, whereas audio characteristics contributed little predictive value. We discuss how these findings offer initial evidence that cognitive ability may be reflected, albeit subtly, in micro-patterns of everyday, non-achievement-related behavior, and outline conceptual and methodological challenges for future work using digital behavioral data to complement traditional cognitive assessment.

Figures