Taking another look at intelligence and personality using an eye-tracking approach

The study explores the interplay between intelligence, personality, and behavioral processes during cognitive tasks. Intelligence, particularly fluid intelligence (the ability to solve novel problems), and personality traits, assessed using the Big Five model (Openness, Conscientiousness, Extraversion, Agreeableness, Neuroticism), are key factors influencing learning and cognitive performance. Previous research has established correlations between certain personality traits, especially Openness (positively) and Neuroticism (negatively), and intelligence test scores. However, these studies primarily relied on self-report measures and lacked a process-oriented perspective. This study aims to bridge this gap by incorporating eye-tracking technology to capture behavioral data during an intelligence test. Eye-tracking offers a rich dataset reflecting cognitive strategies and processes during task completion. The researchers hypothesize that gaze patterns will predict intelligence test performance and that certain personality traits, particularly Openness and Neuroticism, will correlate with specific gaze patterns. Moreover, they explore whether personality traits uniquely predict intelligence test performance beyond the variance explained by gaze patterns. The use of a timed intelligence test, in contrast to previous untimed studies, adds a crucial element of ecological validity, mimicking real-world testing situations. This comprehensive approach promises to provide a more nuanced understanding of the complex relationship between intelligence, personality, and cognitive processes.

Existing literature reveals a complex relationship between intelligence and personality. The Big Five personality model has been widely used to study this relationship, with Openness to Experience consistently showing the strongest positive correlation with fluid intelligence. Investment trait theory suggests that Openness influences the time and effort invested in intellectual pursuits, leading to better cognitive development. Conversely, Neuroticism is often negatively associated with intelligence due to factors like test anxiety. The correlations between the other Big Five traits (Extraversion, Agreeableness, Conscientiousness) and intelligence are generally weaker and less consistent. There is growing evidence that facet-level personality measures (more specific aspects of the Big Five traits) provide a more accurate picture of personality-intelligence associations compared to trait-level measures. Previous research using eye-tracking during intelligence tests has shown that gaze patterns can predict performance and reflect test-solving strategies, such as constructive matching (mentally constructing the answer before searching for it) and response elimination (systematically rejecting incorrect options). However, most prior eye-tracking studies used untimed tests, raising questions about the generalizability of findings to timed settings common in real-world assessments.

The study utilized data from the publicly available TüEyeQ dataset, which included eye movement data from 315 university students completing part of the Culture-Fair Intelligence Test (CFT 20-R), a measure of fluid intelligence. After data cleaning and exclusion of participants with poor eye-tracking data, the final sample comprised 182 adults (Mage = 23.32, SDage = 2.89, 71.82% women). The Big Five personality traits and their facets were assessed using the German version of the Revised NEO Personality Inventory (NEO PI-R). Eye-tracking data were preprocessed by dividing the stimulus into areas of interest (questions, targets, distractors, etc.) and normalizing scanpaths (gaze sequences) along their temporal axes. A variation of the Needleman-Wunsch algorithm was used to calculate pairwise similarity between scanpaths. Machine learning models, specifically elastic net regression, were employed for prediction. For RQ1 (personality predicting intelligence), elastic net with 10-fold cross-validation was used. For RQ2 (gaze predicting intelligence), a k-nearest-neighbor (kNN) approach with leave-one-participant-out cross-validation was used, predicting CFT scores based on scanpath similarity. RQ3 (personality predicting gaze patterns) also used the kNN approach with leave-one-participant-out cross-validation. RQ4 (identifying strategies) used an iterative method to identify the most representative scanpaths for each item, relating them to existing strategies like constructive matching and response elimination. Finally, for RQ5 (combining gaze and personality to predict intelligence), elastic net with 10-fold cross-validation was used, combining gaze-based predictions and personality traits (both trait and facet levels).

The machine learning models revealed several key findings: 1. **Personality and Intelligence:** Big Five personality traits explained 3.18% of the variance in CFT scores, with Openness and Agreeableness showing the strongest predictive power. Facet-level analysis revealed a larger explained variance (7.67%), with significant contributions from Ideas and Values (Openness), and Compliance and Trust (Agreeableness). 2. **Gaze Patterns and Intelligence:** Gaze patterns significantly predicted intelligence test performance, explaining 35.91% of the variance. This indicates that how participants visually explored the test items provided substantial information about their cognitive abilities. 3. **Gaze Patterns and Personality:** No significant relationship was found between overall Big Five traits and gaze patterns. However, at the facet level, Self-Consciousness (Neuroticism) and Assertiveness (Extraversion) showed some association with gaze patterns. 4. **Test-Solving Strategies:** Gaze patterns reflected, to some extent, the test-solving strategies of constructive matching and response elimination, with constructive matching associated with higher success rates. However, the timed nature of the test likely blurred the distinction between these strategies. 5. **Combined Prediction:** A combined model incorporating both gaze-based predictions and personality traits explained 37.50% of the variance in intelligence test scores. Personality traits made a significant, albeit small, unique contribution beyond the predictive power of gaze patterns. This was also observed in the facet-level analysis (38.02% variance explained).

The study's findings underscore the significant role of behavioral data (gaze patterns) in predicting intelligence test performance, outperforming traditional self-report personality measures. The substantial variance explained by gaze patterns highlights the importance of considering cognitive processes and strategies during task completion. The independent contributions of both personality traits (Openness, Agreeableness) and gaze patterns in predicting intelligence support a multi-faceted view of cognitive ability. The surprising contribution of Agreeableness might reflect the influence of cooperative attitudes and compliance with instructions during the test situation. The facet-level analyses provided more refined insights compared to trait-level analysis, emphasizing the need for more granular assessments of personality in this context. The lack of strong links between personality traits and gaze patterns might be due to the specific, constrained nature of the test situation, potentially limiting the expression of personality in gaze behavior. Future research could investigate the role of personality in more dynamic and ecologically valid contexts.

This study integrated eye-tracking data with traditional psychological measures to offer a more comprehensive understanding of intelligence and personality. Gaze patterns were significantly better predictors of intelligence test performance than personality traits, while personality traits, particularly Openness and Agreeableness, still contributed uniquely. Future research could explore the interplay between personality and gaze patterns in more naturalistic settings, using various intelligence tests and personality frameworks, and investigating whether specific gaze patterns exhibit trait-like stability across items and time. Furthermore, exploring the use of additional physiological measures (heart rate, skin conductance, etc.) along with eye-tracking could provide even richer insights into cognitive processes during intelligence testing.

The study's sample was comprised of university students, limiting the generalizability of findings to other populations. The use of a timed intelligence test might have influenced the expression of personality traits and the clarity of test-solving strategies. Further, the relatively modest sample size relative to the analytic complexity of the models, particularly the facet-level analysis, may introduce a risk of Type I errors. Replication studies with larger and more diverse samples are crucial to confirm the results.