School spending, charters, and what school-level choices actually move outcomes

Verdict

Two questions hide inside "does money/school choice matter": does more money help? (a little, expensively, mostly for the poor) and what should a school buy? (the bundle — which is cheap).

Money. The causal record overturned the old "money never matters" null: court-ordered finance reforms moved attainment, wages, and poverty for poor children (+0.46 years education, +9.6% wages for low-income cohorts) in a low-spending era. But the priced average is sobering: $1,000/pupil sustained 4 years buys ~0.032 SD — per test-score dollar, roughly 5–10x worse than high-dosage tutoring — and the heterogeneity is huge: many funded policies bought nothing. The average is not your effect; allocation is everything.

Schools. The lottery record is the sharpest intake-dependence result in education: Boston No-Excuses charters add ~0.4 SD/yr of middle-school math — the largest replicated causal school effects in US data, and in-district pilot schools with the same autonomy but not the model add ~0 — while the national 36-school lottery finds the average charter is a precise null, and the same design applied statewide in Massachusetts makes nonurban charters serving advantaged students negative (−0.16 SD/yr vs +0.34 urban). The charter label buys nothing; the model does, and only against weak counterfactual schools. (The sector-wide CREDO 41-region matching study points the same direction with far weaker identification — coverage, not causation.) Five practices explain ~45% of school effectiveness — high-dosage tutoring, extended time, data-driven instruction, frequent teacher feedback, high expectations — while the things districts actually buy with marginal dollars (class size, spending, credentials) explain ~zero. And the bundle transplants: Fryer's Houston experiment delivered charter-sized math gains (+0.15–0.18 SD/yr) inside ordinary district schools at $355/student (elementary, IRR ~28%) to $1,837 (secondary, IRR ~13%).

The recurring pattern: math moves 2–5x more than reading in every study in this cluster — school structure buys computational/procedural skills readily, reading comprehension barely (consistent with the knowledge-based view of comprehension).

Hereditarian-lens assessment

Risk: low for the verdicts (lotteries and reform-timing designs). This cluster is where naive observational reasoning goes worst: raw charter-vs-district gaps are dominated by selection on (heritable) family traits; the lottery evidence shows urban gains are real treatment — larger for lower-baseline students, not peer-driven — but substantially recovery from bad counterfactual schools. For an advantaged intake, the grade-A evidence for intensive No-Excuses schooling is absent-to-negative: the model treats environmental deprivation, and where there's little to treat, its costs show. Match model to intake.

Practical guidance

• Budget by bundle, not level: tutoring > coaching/evaluation > extended time > data systems with decision rules. Skip what doesn't correlate with effectiveness even among charters: marginal class-size cuts, credential pay, facilities beyond sufficiency.
• For a disadvantaged intake: the full-intensity bundle, unapologetically.
• For an advantaged intake: don't import No-Excuses intensity wholesale — there's no evidence base for it there; buy the still-universal pieces (tutoring for whoever's behind, teacher selection, coherent curriculum).
• Expect structural levers to move math and attainment, not reading comprehension — and nothing to move g.