This study examined the impact of two questionnaire characteristics, scale position and questionnaire length, on the prevalence of random responders in the TIMSS 2015 eighth-grade student questionnaire. While there was no support for an absolute effect of questionnaire length, we did find a positive effect for scale position, with an increase of 5% in random responding over the course of the questionnaire (in both the shorter and the longer version). However, scale character turned out to be an unexpected but more important determinant. Scales about students' confidence in mathematics or science showed an increase of 9% in random responding, which is double the impact of scale position. Potential mechanisms underlying the confidence case and general implications of the results for questionnaire design are discussed.