More per-pupil spending has not tracked with higher NAEP scores across states in the last decade — but this cross-state snapshot cannot answer the causal question the credible quasi-experimental literature does

Summary

Over the last available decade of comparable state data (2016-2024), the states that raised per-pupil spending fastest were no more likely to see NAEP scores rise (or fall less) than states that raised spending the least — the state-level correlation is essentially zero (r = -0.06, p = 0.66, n = 51). National NAEP grade-8 scores fell almost everywhere over this window (composite math+reading average score down from 275.5 in 2013 to 264.9 in 2024), a decline dominated by the pandemic-era drop after 2019, and that decline swamps any state-level spending signal in this simple before/after comparison. This does NOT mean money doesn't matter: it means a raw cross-state snapshot is the wrong instrument to answer a causal question, because states that raise spending fastest are not a random sample — they differ in cost of living, student need, and other confounds, and every state was hit by the same pandemic shock regardless of its spending trajectory. The credible peer-reviewed literature, which uses court-ordered and legislated school-finance reforms as a source of spending variation unrelated to a state's own achievement trend, generally DOES find a positive, causal effect of increased spending on achievement — concentrated in low-income districts. Both things are true at once: our observational comparison finds no relationship, and the quasi-experimental literature finds a real positive effect that this kind of comparison is not designed to detect.

Step 1 — What the peer-reviewed literature already establishes

Three findings recur across independent lines of research:

• Lafortune, Rothstein & Schanzenbach (2018, AEJ: Applied Economics) use the staggered timing of post-1990 court-ordered and legislative school-finance reforms as an event-study design. Reform events caused sharp, sustained increases in spending, concentrated in low-income districts, and NAEP achievement in those districts rose over the following years — the effect phases in gradually and is described as large.
• C. Kirabo Jackson and coauthors (NBER w28517, and Jackson, Johnson & Persico 2016) similarly find that school-finance-reform-driven spending increases raise test scores, educational attainment, and adult earnings, with larger effects for low-income students; conversely, recession-driven spending cuts lowered test scores and college attendance.
• Jackson & Mackevicius (2024 meta-analysis, published in the American Economic Journal: Applied Economics) pool many well-identified (quasi-experimental) US spending studies and estimate that a policy sustaining a $1,000/pupil spending increase for four years raises test scores by about 0.03 standard deviations and college-going by roughly 2.8 percentage points — a real but modest effect, not detectable in a two-point cross-state snapshot.

The common thread: these studies identify effects using spending changes plausibly UNRELATED to a state's pre-existing achievement trend (a court order, a funding-formula rewrite) — not a raw comparison of whichever states happened to spend more.

Step 2 — What we computed directly from the government data available here, and why it disagrees with the literature's framing rather than contradicting its finding

Data used. Census Bureau F-33 Annual Survey of School System Finances (edu.f33_district_finance), aggregated from district to state level, weighted by enrollment, for the two finance years actually loaded in this warehouse: FY2016 and FY2024 (the table's declared window is FY2023-2024 only; FY2016 was recoverable via a workaround join through edu.ccd_districts because — a data-quality defect we logged via report_issue — every FY2016 row has a NULL state_fips column in the finance table itself). NAEP grade-8 math+reading composite scores (edu.naep_state_trends) for the two available years bracketing that window, 2017 and 2024. FY2016 dollars were converted to real 2024 dollars using the BLS CPI-U (deflator 1.307, verified via adjust_inflation).

Result. Mean real per-pupil current spending rose from about $15,900 (2016, in 2024$) to about $17,900 (2024) — a 13.4% real increase, with every state raising nominal spending and most rising in real terms (range: -8.2% in Alaska to +39.5% in New Mexico). Over the same span, the average NAEP grade-8 composite score FELL by about 9.0 points nationally, and fell in nearly every state regardless of how much its spending grew. Regressing the state-level change in NAEP score on the state-level change in real per-pupil spending (verified with askengine's ols_regression tool) gives a coefficient of -0.12 NAEP points per $1,000 of real per-pupil spending increase, standard error 0.28, p = 0.66 — statistically indistinguishable from zero, and if anything the wrong sign. A leave-one-state-out check (askengine sensitivity_analysis) confirms this is not one outlier state driving a null: the coefficient stays negative and insignificant across every possible single-state exclusion (range -0.28 to 0.00, no sign flip, no p-value crossing 0.05).

Why this doesn't refute the literature. Two confounds this simple design cannot separate: (1) a nationwide, pandemic-driven score decline hit essentially every state between 2019 and 2024 regardless of its spending trajectory, so the outcome variance in this window is dominated by a common shock, not by cross-state spending differences; and (2) spending increases here are NOT randomly assigned or reform-driven — states raise spending for many reasons (rising costs, enrollment shifts, political cycles) that are themselves correlated with a state's baseline trajectory, so this is an observational comparison, not a natural experiment. The quasi-experimental studies above are designed precisely to strip out both confounds; this state cross-section is not.

Step 3 — Data-quality note

edu.f33_district_finance's declared coverage window is FY2023-2024 only, and its FY2016 rows (the only pre-2023 year loaded, with FY2017-2022 an interior gap) all carry a NULL state_fips — a genuine defect, logged to the issue tracker in this session, that would silently prevent any state-level aggregation of that year without the ccd_districts workaround join used here. Given this, the two-finance-year window (2016, 2024) is the widest usable spending comparison this warehouse currently supports; a true year-by-year 10-year panel is not available from this table as loaded.

What This Report Does Not Answer

• Cover roughly the last ten years: edu.f33_district_finance's only usable pre-2023 year is FY2016 (an interior gap covers FY2017-2022, and the table's declared window is FY2023-2024 only) — the widest true finance-to-NAEP comparison available in this warehouse is 2016-2024 (8 years), not a full 10-year year-by-year panel. Reported as a fiscal-year data-availability limit, and the literature review (Step 1) supplements this with findings from studies covering longer historical windows.