T\xfcEyeQ, a rich IQ test performance data set with eye movement, educational and socio-demographic information

Enkelejda Kasneci,Gjergji Kasneci,Peter Gerjets,Maike Tibus,Tobias Appel,Ulrich Trautwein,Johannes Haug,Franz Wortha

doi:10.1038/s41597-021-00938-3

Enkelejda Kasneci, Gjergji Kasneci + Show 6 more

Open Access

https://doi.org/10.1038/s41597-021-00938-3

Copy DOI

Abstract

We present the TüEyeQ data set - to the best of our knowledge - the most comprehensive data set generated on a culture fair intelligence test (CFT 20-R), i.e., an IQ Test, consisting of 56 single tasks, taken by 315 individuals aged between 18 and 30 years. In addition to socio-demographic and educational information, the data set also includes the eye movements of the individuals while taking the IQ test. Along with distributional information we also highlight the potential for predictive analysis on the TüEyeQ data set and report the most important covariates for predicting the performance of a participant on a given task along with their influence on the prediction.

Highlights

Background & SummaryFor many decades, research in various fields has been devoted to the question of what constitutes human intelligence[1,2], the ways in which it develops in the course of our life[3], and how it can be positively influenced[4]
To support the research community and the work at the intersection of psychological and educational sciences and artificial intelligence, we provide the TüEyeQ data set
We collected a comprehensive data set from 315 university students performing a culture fair intelligence test (CFT 20-R)

Summary

Background & Summary

Research in various fields has been devoted to the question of what constitutes human intelligence[1,2], the ways in which it develops in the course of our life[3], and how it can be positively influenced[4]. Along with the performance data, we provide socio-demographic and educational background information on the students as well as carefully annotated eye movement data of all participants during task solving. We believe that this data set will boost the research in various fields and will contribute to highly interesting research questions:. From a data science perspective the data set provides valuable means to analyse performance bias with respect to background information on the participants, such as education, training, viewing behaviour, gender and many more. From the AI perspective the question of whether machine learning algorithms can learn to reason as humans and whether it is possible to develop an AI system that correctly solves such a test is among current challenges in AI research (e.g.11–13)

Methods

Findings

Code availability