From the data
Does Higher School Spending Buy Higher Test Scores?
We put twenty years of federal finance filings next to federal test results for six to seven thousand districts at a time. The pattern is weaker than almost anyone expects, and what signal exists is mostly the price of teachers, not the number of them.
"They say spending money attracts better teachers, which you'd think would result in better test scores." It is one of the most common claims in any school budget debate, made in good faith by people on every side of it. We now hold enough federal data to examine each link of that chain separately: twenty years of district finance filings, test results and graduation rates for six to seven thousand districts at a time, teacher counts from the federal civil rights collection, and the poverty rate of each district's children.
Here is what that record shows — and, just as important, what it cannot show.
The headline correlation is close to zero
Across roughly 6,700 districts with at least 1,000 students in FY2018, the correlation between operating spending per pupil and the share of eighth-graders proficient in math is +0.04 — statistically indistinguishable from no relationship. Spending differences explain about two tenths of one percent of the variation in scores. The same is true of reading, of grade 4, and of graduation rates, where the correlation is −0.01.
The relationship is not even stable in direction. Run the same comparison in FY2009 and it is mildly positive (+0.13). Run it in FY2012 and it is negative (−0.17). A relationship that flips sign depending on the year you pick is not a relationship you can build a policy argument on, in either direction.
Poverty is the signal that never flips
In every year we tested, the correlation between a district's child poverty rate and its test scores runs between −0.25 and −0.48, and against graduation rates about −0.43. It is five to ten times stronger than the spending correlation and it never changes sign. Whatever else is true about school finance, where the poor children are is a far better predictor of published outcomes than where the money is.
This also contaminates the spending question. Federal and state formulas deliberately send more money to poorer districts, and poorer districts post lower scores for reasons that have nothing to do with how well the money was spent. Money is assigned to difficulty, which drags the raw spending-versus-scores number toward zero and below.
Money does buy staff — measurably
The first link in the chain holds. Districts that spend more per pupil employ more teachers per hundred students, at a correlation of +0.53 — by far the strongest spending relationship anywhere in this record. Money visibly becomes adults in buildings.
What it buys much less of is credentials. The correlation between spending and the share of teachers holding state certification is only +0.08, partly because certification is already near-universal in most districts, leaving little room to differ.
Better-credentialed staff track better outcomes — weakly
The second link also holds, faintly. Districts where more teachers are certified graduate more students (+0.12) and post higher math scores (+0.12). Faint is the honest word: certification is a blunt proxy for teacher quality, and these correlations are a quarter the size of the poverty effect running the other way.
Adjust for what a teacher costs, and the signal disappears
A dollar does not buy the same teacher everywhere. Using the federal Comparable Wage Index for Teachers — built from what college-educated non-educators earn in each district's labor market, so it reflects local costs rather than district choices — we divided each district's spending by its local wage level.
The small positive correlation with math scores (+0.04) fell to −0.001: exactly nothing. In plain terms, the modest raw pattern was mostly districts in expensive places both spending more dollars and posting somewhat higher scores. Once you compare what the money buys rather than what it costs, districts that buy more education inputs do not post detectably higher test scores in a same-year comparison.
Comparing like with like
The national numbers mix a remote Alaska district where per-pupil costs are unavoidably high with a large suburb enjoying every economy of scale. Split the comparison by the federal locale classification and the FY2018 spending–math correlation is −0.02 in rural districts, +0.09 in suburbs, and −0.22 in cities — the highest-spending city districts post lower scores, which is what heavy compensatory funding aimed at concentrated poverty looks like in a scatter plot. Within individual states, the median correlation is mildly negative (−0.13), and negative in 25 of 36 states with enough districts to test.
What this does not mean
It would be a serious misreading of everything above to conclude that money does not matter. These are comparisons between different districts in the same year. They cannot see what happens when the same district's funding changes, which is the comparison that actually answers the policy question. The credible causal research — studies of court-ordered finance reforms and funding shocks, most prominently Jackson, Johnson and Persico's work on reforms since the 1970s — finds that sustained spending increases do improve completed schooling and adult earnings, with the largest effects for low-income students. Graduation and later-life outcomes move more reliably in that literature than test scores do.
Both things are true at once, and the tension between them is the finding: money assigned to difficulty hides its own effects in a snapshot, and a snapshot is what a correlation is. What the cross-section supports saying is narrower but still worth saying — a district's per-pupil figure, on its own, tells you almost nothing about how its students are doing, and anyone ranking districts by spending as if it were a quality score is measuring geography and poverty, not schools.
Sources and method
Finance: U.S. Census Bureau, Annual Survey of School System Finances (F-33), FY2005–FY2024; operating spending excludes capital outlay and debt service. Test results: EDFacts state assessments (grades 4 and 8, share proficient or above), and EDFacts adjusted cohort graduation rates, both via the Urban Institute Education Data Portal. Results ED suppressed into wide ranges to protect small cohorts were excluded rather than reconstructed. Staffing: U.S. Department of Education, Civil Rights Data Collection, school-level counts summed to districts; full-time equivalents, no individual-level data. Poverty: Census Small Area Income and Poverty Estimates (ages 5–17). Wage adjustment: NCES Comparable Wage Index for Teachers. Correlations are Pearson r across districts with at least 1,000 students; because state tests differ, cross-state comparisons carry that caveat and the within-state figures are the stricter test. State proficiency standards also changed in 84 detected state-years, which we treat as breaks rather than trends. If you believe a figure here is wrong, the correction path is the same as for every page on this site: submit the record and we will re-run the numbers.