A Machine Learning Approach to Assess Differential Item Functioning of PISA 2018 ICT Engagement Questionnaire

Item Functioning of PISA 2018 ICT Engagement Questionnaire

Authors

  • Gozde Sirganci Southern Methodist University

Abstract

This study aimed to investigate the different item functioning (DIF) of the Programme for International Student Assessment (PISA) 2018 information, communication, and technology (ICT) questionnaire items based on country, gender, Economic, social, and cultural status (ESCS) variables. The sample included 29,277 15-year-old students from Eastern Europe and Central Asia (EECA) countries. The study employed the generalized partial credit model with lasso penalization, a machine learning approach, to evaluate DIF. The findings showed that two out of 21 items exhibited DIF based on country, gender, and ESCS overall. There were seven out of 21 items that showed DIF in favor of males and six out of 21 items that showed DIF in favor of females. According to ESCS, only three out of 21 items displayed DIF. All items exhibited DIF based on countries. According to the GPCMlasso coefficient, when reference group Bulgaria and focus group Georgia, Croatia, Kazakhstan, and Turkey are compared, 28%, 71%, 33%, and 33% of all items hold DIF, respectively. In pairwise comparisons, the most DIF-prone items were found between Bulgaria and Croatia, while the fewest were between Bulgaria and Georgia.

Downloads

Published

2023-09-01