Private tutoring chains raise U.S. test scores but widen racial gaps, study finds

An NBER working paper tracking chains including Kumon, Sylvan Learning and Mathnasium finds higher district-wide math and reading scores after a first center opens, but the math gains are concentrated among White and Asian students

New research finds the arrival of private tutoring chains is associated with higher district math and reading scores, but some of the strongest math gains are concentrated among White and Asian students

The arrival of a private tutoring chain in a U.S. school district raises average math and reading scores, but can also widen racial achievement gaps because access to tutoring is heavily concentrated among higher-income, White and Asian families, new research finds.

The August 2026 National Bureau of Economic Research working paper, Kumon Up: Private Tutoring Centers and Student Achievement, was written by Boston University researchers Joshua Goodman and Samuel Nitkin.

Using 25 years of business location data alongside household spending and school district test scores, the researchers estimate that the opening of a district’s first chain tutoring center increased average math scores by 0.029 standard deviations and reading scores by 0.016 standard deviations.

The math gains were not evenly distributed. White and Asian students recorded a combined 0.040 standard deviation increase after a chain entered their district. The estimated effect for Black, Hispanic and Native American students was 0.018 standard deviations and was not statistically significant.

Within the same districts, the White-Black math achievement gap increased by 0.037 standard deviations and the White-Hispanic gap by 0.020 standard deviations.

The paper is an NBER working paper circulated for discussion and has not been peer-reviewed.

Score gains came with wider racial gaps

Private tutoring chains have expanded considerably across the U.S. The researchers identified roughly 500 chain tutoring centers in 1997, rising to more than 4,000 at their peak in the mid-2010s. Across more than 7,000 unique locations in the study’s main period, Kumon accounted for 38%, Sylvan Learning for 25%, and Mathnasium and Huntington Learning for 10% each.

To test whether those openings changed student outcomes rather than simply appearing in already high-performing communities, Goodman and Nitkin linked center locations to standardized math and reading scores from the Stanford Education Data Archive.

Their main analysis covers grades 3 to 8 between 2009 and 2019. Around 402 school districts received their first chain tutoring center during that period and were compared with 6,959 districts that had no chain center between 1997 and 2019.

Those groups did not start from the same position. Districts attracting tutoring chains were already larger, higher-income, higher-scoring, more suburban and had a higher proportion of Asian students.

The researchers therefore used a staggered difference-in-differences approach that tracked how scores changed before and after a first center opened. They found no clear pre-existing trend in scores before entry, and the results remained broadly stable after controls were added for enrollment, racial composition, unemployment and median income.

With those controls, the estimated effect rose slightly to 0.034 standard deviations in math and 0.021 in reading.

The racial pattern was clearest in math. The researchers found the widening White-Black achievement gap was larger in more racially segregated districts, increasing by 0.047 standard deviations compared with 0.021 in less segregated districts, where the latter estimate was not statistically significant.

For the White-Hispanic gap, the increase was 0.028 standard deviations in more segregated districts, while researchers detected no comparable effect in less segregated districts.

One possible explanation examined in the paper is spillover inside schools. If students receiving private tutoring need less instructional support or help classmates through peer learning, some of the benefit could extend to other students at their school.

Because private tutoring is disproportionately used by White and Asian families, those spillovers could also remain concentrated in more segregated schools.

The evidence is consistent with that explanation, but the study does not establish that spillovers are the mechanism causing the pattern.

Nearly half of tutoring users are in the top income decile

The inequality in outcomes closely follows who is paying for tutoring in the first place. Using Consumer Expenditure Survey data from 2009 to 2019, the researchers found that only 1.5% of households with school-age children reported tutoring spending in a typical month.

Among families that did pay for tutoring, average expenditure was $272 a month, equivalent to around $2,500 over a nine-month school year.

Usage rose steeply with income. Some 4.6% of households in the top 10% of the income distribution reported tutoring expenditure, compared with 1.4% in the next 40% and 0.5% among households in the bottom half.

Top-decile households accounted for 46% of all families reporting tutoring spending.

There were also substantial racial differences. Monthly tutoring expenditure was reported by 4.5% of Asian households, compared with 1.7% of White households, 1.2% of Black households and 0.7% of Hispanic households.

Much of the White, Black and Hispanic gap was explained by income. Within comparable income groups, their tutoring rates were much closer.

Asian families remained the exception. Even among households in the top income decile, nearly 9% of Asian households reported tutoring spending, substantially above the other groups.

The researchers also found that greater availability of chain centers in a local area was associated with higher tutoring expenditure among White, Asian and above-median-income households, but not among the other groups examined.

That helps explain why an intervention paid for outside the public school system can shift district-wide test scores without benefiting students evenly.

Goodman made that point when announcing the research on LinkedIn, arguing that private tutoring is also changing the test-score measures used to judge schools and districts.

Chain centers performed differently from other tutoring models

The findings do not extend to every form of private tutoring. When the researchers repeated their analysis for independent tutoring centers operating fewer than 10 locations, they found no statistically significant overall effects on math or reading scores.

The same was true for chains providing tutors in students’ homes. That leaves the positive results concentrated among brick-and-mortar tutoring chains.

The paper examines several features shared by the largest operators. Kumon, Sylvan Learning, Mathnasium and Huntington Learning all use diagnostic assessments and individualized learning plans, while their franchise structures provide standardized curricula, instructor requirements and some level of quality oversight.

Their teaching models can still look quite different. Kumon relies heavily on self-directed worksheets and daily practice, while Sylvan uses more direct instruction in small groups. Mathnasium focuses exclusively on math and gives students individualized learning pathways.

The typical student appears to spend around two to three hours each week either at a center or completing work assigned by it.

Researchers estimate that a first chain center induces around 3% of students in a district to take up math tutoring. Because the measured test-score gain is averaged across the entire district, including the large majority who never attend, the implied effect for students actually receiving tutoring is considerably larger.

Those estimates require assumptions about how long learning gains persist, however, and the authors caution that they cannot separate lasting learning from other factors such as centers gradually increasing enrollment.

The apparent cost advantage also needs context. The researchers estimate private tutoring associated with a first center costs around $75 when spread across every student in the district. They compare that with previous research finding that an additional $1,000 in annual public school spending sustained for four years produces a similar district-level test-score increase.

But those are not equivalent investments. Tutoring costs are borne by families, reach only a small share of students and are concentrated among groups already more likely to access private educational support.

The strongest finding is therefore not simply that private tutoring works.

Chain tutoring centers were associated with measurable improvements in district test scores, particularly math, while the families most able and likely to buy that additional instruction received a disproportionate share of the benefit.

The result was higher average achievement alongside wider racial gaps in the same school districts.