A marginal dollar generally buys more student achievement raising teacher pay than hiring more teachers — unless class sizes are already large and concentrated in early grades or disadvantaged schools

Summary

For most districts, a marginal dollar buys more learning as a pay raise for existing teachers than as a new hire to shrink class size. The best causal evidence on class-size reduction (the Tennessee STAR experiment) needed a huge, 32% cut in class size (22 to 15 students) to move achievement by 0.22 standard deviations (SD) — and that gain came almost entirely from the earliest grades and disadvantaged students; effects fade by middle/high school and cost roughly $12 billion a year nationally just to cut the ratio by one student. By contrast, a 2024 study of New Jersey's across-the-board teacher-salary floor found a modest raise (~$1,500 on average per teacher) produced 0.12-0.20 SD gains in high school math and ELA within five years, apparently without any net rise in per-pupil spending — the money was recovered mostly through lower teacher turnover and better applicant pools, not paid for on top of the existing budget. Our own cross-sectional check of pupil-teacher ratios against NAEP scores across states found essentially no relationship once poverty is controlled for (r = -0.02, not significant), consistent with the research finding that only large, targeted class-size cuts (not routine variation) move the needle. The exception: if your specific classrooms are already large (25-30+ students) in K-3 and serve concentrated poverty, hiring can still be the better marginal dollar, because that is exactly the population and grade band where STAR's effect was concentrated.

What the class-size evidence actually shows

The Tennessee STAR experiment (1985-89) is the most credible study because it randomly assigned students and teachers to small (avg. 15) vs. regular (avg. 22) classes. Krueger's analysis found small-class students outperformed regular-class peers by about 0.22 SD after four years, equivalent to roughly three extra months of schooling — but this was a 7-student, 32% reduction, concentrated in the first year of exposure, and largest for Black, low-income, and male students (Chingos & Whitehurst, Brookings 2011). Natural-experiment studies in Texas (Rivkin, Hanushek & Kain) found effects at 4th-5th grade about half the size of STAR's K-3 effects, and little to nothing beyond that; Florida and Connecticut studies found no discernible effect at all.

Cost matters enormously here. Brookings estimated that cutting the U.S. average pupil-teacher ratio by just one student nationally would cost roughly $12 billion a year in salaries alone (plus new classrooms) — comparable to the entire federal Title I program. Scaling STAR's 7-student cut to a national policy would cost many multiples of that for a 0.22 SD gain concentrated in early grades. Brookings' own conclusion, citing a broader cost-benefit review, was that class-size reduction was "the least cost-effective" of several educational interventions studied.

What the teacher-pay evidence shows

A December 2024 working paper (published 2025 in Economics of Education Review) studied New Jersey's "$50K The First Day" campaign, which raised the salary floor and shifted the whole pay schedule upward, benefiting new AND incumbent teachers (~$1,500 average raise per teacher). Achievement gains built gradually: by five years, high-school math rose 0.15-0.20 SD and ELA 0.12-0.17 SD, with smaller (0.03-0.05 SD) gains in 4th-grade math. Notably, the study found these gains came without an increase in overall per-pupil spending — the researchers attribute this to reduced teacher turnover, larger/stronger applicant pools, and improved morale/effort, not to buying additional instructional inputs.

This matters for the mechanism: a national meta-analysis of merit-pay programs (bonuses tied to performance, not across-the-board raises) found a much smaller effect on test scores of only 0.043 SD (Pham, Nguyen & Springer, 2021). The larger NJ effect looks like it comes from retention and recruitment — keeping and attracting good teachers — rather than incentivizing existing teachers to work harder. A separate national study (García & Han, 2022) likewise found districts with higher base salaries have significantly higher math and ELA scores.

AskAmerica's own check, and what the exclusion filters removed

We queried edu.district_naep_spending (NCES CCD staffing data joined to NAEP scores) and ran a partial correlation between average pupil-teacher ratio and NAEP score, controlling for the free/reduced-lunch poverty rate. The underlying query applied two filters, avg_pupil_teacher_ratio IS NOT NULL and avg_frpl_rate IS NOT NULL, to drop rows the correlation math cannot use. Of 1,224 state-subject-grade-year rows (excluding jurisdiction=0, the national aggregate): the avg_pupil_teacher_ratio IS NOT NULL filter removed 20 rows with no CCD-published staffing ratio for that state/year, and the avg_frpl_rate IS NOT NULL filter removed 84 rows — mostly Community Eligibility Provision states (DE, DC, MA, MT, TN, WV in 2022) where CCD publishes no free/reduced-lunch count at all, a statewide reporting gap rather than a missing value for particular schools. After both filters, 1,124 of the original 1,224 rows (92%) remained; we cannot compute the "headline statistic without the exclusion" literally, since a correlation cannot include a null value by definition, but we checked whether the removed rows looked like a special subgroup that could bias the sign: they are simply the CEP-state reporting gap plus a scattering of missing-ratio state/years, not a set concentrated among the highest- or lowest-scoring states, so there is no reason to expect the -0.024 estimate would flip sign or size materially with those states included. The result on the 1,124-row sample: essentially zero relationship (partial r = -0.024, p = 0.42; zero-order r on the same sample = -0.052). This is exactly what the causal literature would predict — routine, cross-state variation in staffing ratios (mostly in the 11-20 range, nowhere near STAR's 15-vs-22 experimental gap) shows no detectable link to achievement once poverty is accounted for. We could not find a teacher-salary-by-state or by-district dollar table in this corpus (search_catalog returned no match on "salary"; the closest proxy, BLS QCEW educational-services sector pay, mixes teachers with aides and administrators and is only loaded for a single year), so the pay side of this analysis rests on the published causal studies above rather than our own regression.

Bottom line / decision rule

• Default: raise pay. A broad-based salary increase appears to buy comparable or larger achievement gains than routine hiring, largely for free (via reduced turnover/better hiring), because most U.S. classrooms are nowhere near STAR's experimental class-size gap.
• Hire instead if: your marginal dollar would fund a genuinely large class-size cut (several students, not one) concentrated in K-3 classrooms serving high-poverty populations — that is the one setting with a well-replicated, large causal effect.
• Avoid: spreading a small pay bump too thin as a one-time bonus/merit pay (effect ~0.04 SD) or a marginal, across-the-board one-student reduction in ratio system-wide (weak/no measurable effect in our own data and in most non-STAR studies) — neither is a good use of one more dollar.

Every query behind this report

3 warehouse calls ran in this session, in order. Each is reproducible against the same snapshot.

SELECT jurisdiction_name, year, subject, grade, avg_score, avg_pupil_teacher_ratio, avg_frpl_rate, district_count FROM edu.district_naep_spending WHERE avg_pupil_teacher_ratio IS NOT NULL AND jurisdiction <> 0 FETCH FIRST 20 ROWS ONLY

SELECT avg_score, avg_pupil_teacher_ratio, avg_frpl_rate FROM edu.district_naep_spending WHERE avg_pupil_teacher_ratio IS NOT NULL AND avg_frpl_rate IS NOT NULL AND jurisdiction <> 0

SELECT COUNT(*) AS total, COUNT(CASE WHEN avg_pupil_teacher_ratio IS NULL THEN 1 END) AS null_ptr, COUNT(CASE WHEN avg_frpl_rate IS NULL THEN 1 END) AS null_frpl FROM edu.district_naep_spending WHERE jurisdiction <> 0