The Relationship Between Student Attendance and Achievement, Pre- and Post-COVID

North Carolina statewide administrative data, pre- and post-pandemic cohorts

North Carolina public school students

First-differences models on administrative data. First-differencing removes time-invariant student characteristics but leaves every time-varying family and health shock that jointly moves absence and achievement in the same year - which is the dominant confound for absence, not a nuisance. No instrument, no exogenous shock, no within-family comparison. Also reports peer-absenteeism associations, which are subject to the same objection at the classroom level.

In 2023-24, each day absent was associated with a 0.0057 SD decline in math achievement. The authors' own policy arithmetic is the most useful number in the paper and is strikingly deflationary - returning absence rates to 2018-19 norms would improve achievement by 0.017-0.025 SD, accounting for only 13-19% of the recovery needed. They also find the absence-achievement association was WEAKER in the immediate post-COVID years, which is hard to reconcile with a stable instructional dose-response and easy to reconcile with the association being driven by who is absent rather than by how much instruction is lost.

High. This is a within-student association with no exogenous variation in absence. Absence in the post-COVID period tracked household illness, family disruption, disengagement and parental follow-through, all heritable or family-transmitted and all independently predictive of achievement. The instability of the coefficient across periods is itself evidence that it is not measuring a fixed causal parameter.