Demand-Sensitive Time Allocation in Digital Reading: Evidence from 25 Education Systems

Abstract

Digital assessments record item-visit durations and revisiting. These traces can reveal responses to task difficulty, but achievement associations do not establish stable person-level measures. We linked digitalPIRLS 2021 process, response, and achievement data across 25 education systems, obtaining 121,908 students with usable process records and 116,206 complete cases for the primary achievement models. Item difficulty was estimated from other students in five groups, excluding each target student’s own responses. We measured how each student’s first-visit time and revisiting changed with difficulty. More difficult items elicited longer first visits in all 25 systems, whereas the average difficulty effect on revisiting was weak. A one-standard-deviation increase in time adaptation was associated with 19.92 overall-reading points, compared with 3.19 points for revisit adaptation. Effect magnitudes varied substantially across systems. Repeated analyses of separate item halves showed limited stability for time adaptation (0.317) and moderate stability for revisit adaptation (0.529), after correction for test length. Time adaptation estimated from four fifths of the items remained associated with performance on the remaining fifth in all 25 systems. These findings support time adaptation as a task-responsive behavioral indicator within digitalPIRLS. They do not establish a highly reliable trait score or identify a single cognitive mechanism.