Assessing the effect of school days and absences on test score performance

North Carolina administrative panel, grades 3-5 (statewide, ~1M+ student-year records)

North Carolina public elementary school students

Separates two distinct time margins that the literature usually conflates. Length of the school calendar prior to the test is identified from a state policy that shifted the number of instructional days before the testing window. Absences are identified with student fixed effects, and additionally instrumented with local influenza incidence; family-year fixed effects are used as a robustness check. Excused and unexcused absences are separated, and controls for student disengagement are included.

The two margins are not interchangeable and the gap is large. Adding 10 school days to the calendar raises math scores by only 1.7% of a SD and reading by 0.8% of a SD. Removing 10 absences raises math by 5.5% of a SD and reading by 2.9% of a SD - roughly three times more per day than adding calendar days. Effects are concentrated among low-performing students. The per-absence math estimate (0.0055 SD) is an order of magnitude below the headline IV figure in Goodman 2014 and close to Goodman's own within-student estimate.

Better handled than most of this literature. Student fixed effects remove time-invariant heritable child traits; family-year fixed effects remove contemporaneous family-level shocks shared by siblings; the influenza instrument supplies plausibly exogenous variation in illness-driven absence. Residual worry: within-student, time-varying shocks (a parental separation, a new health problem) that both raise absence and depress achievement are not differenced out, and illness propensity is itself partly heritable, so the flu-IV compliers are not a random draw.