Addressing Chronic Absenteeism in Schools: A Meta-Analysis of Evidence-Based Interventions

22 random and nonrandom controlled intervention studies, published 2000-2018

Pre-K through grade 12 US public school students

Pools randomised AND non-randomised controlled designs, so it inherits the confounding of the weaker half of what it aggregates. Interventions coded into three families - behavioural, family-school partnership, and academic. The outcome pooled is ATTENDANCE ONLY; this meta-analysis does not estimate effects on achievement.

Between-group weighted g = 0.25 (95% CI 0.14-0.36) on attendance outcomes. Within-group g = 1.04 (95% CI 0.32-1.76), which is a textbook illustration of why pre-post designs are worthless here - a fourfold inflation over the controlled estimate. By intervention family, all three produce small effects (g = 0.09-0.26), and family-school partnerships have a CI including zero. The authors' own summary is the honest one: most practices to improve attendance are understudied, produce small effects, or both.

Medium-high. The pooled set mixes randomised with non-randomised controlled designs, and self-selection into voluntary attendance programmes is exactly the channel by which motivated families sort into treatment. The within-group g = 1.04 is almost certainly regression to the mean - students are enrolled because they were unusually absent, and unusually absent years regress. Also note that a g = 0.25 effect on a behavioural outcome carries no implication for achievement.