Impact of AI-Generated Customized Quizzes on Learner Sentiment: A 6-Week Crossover in Grades 3–6 (Multan, Pakistan)

Abstract

Compares AI-customized vs. generic quizzes with 20 students using an AB/BA crossover design. Measures weekly affect (PANAS-C + Smileyometer) alongside accuracy and time-on-task, analyzed via mixed-effects models. Investigates whether personalization of question content meaningfully shifts how students feel about quizzes — not only how they perform on them.

Research Question

Does the use of AI-generated, student-specific quiz content produce a measurable shift in learner affect — as reported through validated affective instruments — compared to a generic quiz administered to the same cohort in alternating weeks?

Study Design

• AB/BA crossover: students alternate between AI-customized quizzes (Condition A) and generic class-wide quizzes (Condition B) across six weeks
• 20 students enrolled across Grades 3–6 at a single site in Multan, Pakistan
• Each student serves as their own control, eliminating between-subject variance
• Weekly data collection ensures sufficient within-subject observations for mixed-effects modeling

Measures

• Affect (primary): PANAS-C (child-adapted Positive and Negative Affect Schedule) + Smileyometer — administered weekly after each quiz session
• Performance (secondary): Quiz accuracy score and time-on-task (minutes to completion)
• Analysis: Linear mixed-effects models with student as a random effect, condition and week as fixed effects

Expected Contribution

Most EdTech research focuses on performance outcomes. This study isolates the emotional and motivational dimension — arguing that if AI personalization improves how students feel about assessments, the downstream effects on engagement and long-term learning may be more durable than accuracy gains alone.