Association for Public Policy Analysis & Management

This paper uses the National Assessment of Educational Progress (NAEP) to measure learning during the school year. I identify learning using random variation in assessment dates, within and across years of testing, in a nationally representative sample of 40,000 schools. Baseline learning in 4th grade is double that in 8th grade, when measured in test standard deviations: in math, 4th grade students gain 0.80 SD, and 8th grade students gain 0.40 SD, per additional year of instruction; in reading, 4th grade students gain 0.50 SD, and 8th grade students gain 0.25 SD. The estimates are precise (t=10 to 20), robust across specifications, and stable across years of NAEP testing. Question-level evidence shows that differences across grades and subjects are not due to response patterns, question difficulty, or reliability. Instead, learning rates decline from 4th to 8th grade partly because tests are cumulative (as in Cascio and Staiger, 2012), and partly because the taught curriculum is progressively less aligned with the tested curriculum.