Abstract
INTRODUCTION: This study investigates the confounding effects of reverse-coded items on the measurement of confidence in mathematics using data from the 2019 Trends in International Mathematics and Science Study (TIMSS). METHODS: The sample came from the Saudi Arabian cohort of 8th graders in 2019 involving 4,515 students. Through mixture modeling, two subgroups responding in similar ways to reverse-coded items were identified representing approximately 9% of the sample. RESULTS: Their response to positively valenced and negatively valenced items showed inconsistency and the observed unexpected response patterns were further verified using Lz*, U3, and the number of Guttman errors person fit indicators. Psychometric analyses on the full sample and the truncated sample after deleting the aberrant responders indicated significant improvements in both internal consistency reliability and factorial validity. DISCUSSION: It was concluded that reverse-coded items contribute to systematic measurement error that is associated with distorted item level parameters that compromised the scale's reliability and validity. The study underscores the need for reconsideration of reverse-coded items in survey design, particularly in contexts involving younger populations and low-achieving students.