How expensive is a college degree? Usually, the answer is based on what students pay in tuition and fees compared to what they earn after graduation. As a result, policymakers often promote enrollment and applaud growth in tech-heavy programs that tend to produce high-earning graduates, like engineering and computer science—especially given the explosive growth in college prices, which have doubled in the past 30 years.

But that thinking only focuses on one side of the equation: the student’s private return on investment, based on the labor-market value of a degree relative to what the student paid. We know very little about the economic cost of running an electrical engineering program compared to, say, a history department, or the resource consequences of steering more students into these fields.

To fill this gap, we examine department-level data on expenditures, outputs, and factors of production for undergraduate, graduate, and professional degree programs at nearly 600 four-year institutions across the United States from 2000 to 2017. Our analysis compares the instructional costs per student credit hour at more than 8,000 departments in 20 disciplines, including both in-person and online study.

We establish five new facts about college costs. First, we find substantial cost differences across fields of study. On the whole, costs are higher in fields where graduates earn more and in pre-professional programs. Second, most of these patterns can be explained by differences in class size and, to a lesser extent, differences in average faculty pay. However, some fields with highly paid faculty, like economics, offset this with large classes.

Third, cost differences have evolved over time. Programs in some fields, such as mechanical engineering, chemistry, physics, and nursing, have shown steep annual declines in spending, while others like fine and studio arts, history, and political science have grown more expensive each year. Fourth, these trends are explained in part by a growing number of adjunct faculty as well as changes in class size and faculty teaching loads. Fifth, online instruction is not a cost-saver. It is neither more nor less expensive than in-person classes.

Our results underscore the potential wedge between the social and private returns to higher education. That is, the social return to investment in high-earning fields may be lower than wage premiums suggest, because high-return fields also tend to be more costly to teach. This highlights the need for policymakers to consider the cost implications of changes in the mix of fields students study.

In addition, our work suggests that while differences in production technology enable some departments to take different approaches to cost management, from changing the mix of faculty to increasing class size, online instruction does not have a meaningful association with college costs, at least in its current form. Any one-discipline-fits-all approach to addressing cost escalation in higher education, including moving more classes online, is likely to be ineffective.

Cost Drivers on Campus

Scholars have long noted the tendency for college costs to grow faster than the broader economy. Some argue that this is inevitable because postsecondary education is inherently labor-intensive and therefore has not benefitted from the kinds of productivity-enhancing innovations that have driven down costs in other industries. Other potential explanations include the proclivity of colleges to maximize revenue in an effort to compete for prestige, school spending on student amenities, and the expansion of unnecessary administrative positions.

In all cases, programs produce a set of outputs, such as undergraduate instruction or research publications, using a large set of inputs, such as faculty of different types, classrooms, office space, technology, and laboratories. The relationship between these varies across fields of study. Some disciplines require intense interaction between students and faculty to produce a given level of instructional quality—think of a debate-driven course with long extemporaneous papers to grade—while others require laboratory sessions that rely on expensive equipment and supplies. By contrast, other fields of study may be able to take advantage of economies of scale and scope, such as those that deliver “101”-style general-education courses for the entire institution. In addition, departments with undergraduate and graduate programs can tap graduate students to serve as lower-cost instructors. Such differences affect class size, faculty mix, faculty teaching load, and non-personnel expenditures—all of which determine the cost per unit of instruction.

Data and Method

To compare instructional costs by field of study, we use department-level data from 2000 to 2017 from University of Delaware^,s Cost Study (also known as the National Study of Instructional Costs and Productivity). Instructional activity is measured by student credit hours, organized class sections, and faculty full-time equivalents. Student credit hours and class sections are disaggregated by instructor type, such as tenure track, supplemental, or teaching assistants, and by course level, such as lower-division undergraduate, upper-division undergraduate, and graduate. Finally, institutions report total direct expenditures for instruction, research, and public service and total undergraduate and graduate student credit hours for the entire academic year.

We work with direct instructional expenditures per student credit hour as our main measure of costs, which include salaries, benefits, and non-personnel expenses. In 2015, the study added a component to the survey to capture information about online instruction. These data contain information on online student credit hours by department at the undergraduate and graduate levels.

Institutional participation in the cost study is voluntary. Therefore, we assess how well our sample matched the broader universe of public and private nonprofit four-year institutions operating in the United States, based on the Integrated Postsecondary Education Data System. We estimate that our sample represents 23 percent of all degrees awarded between 1998 and 2015, including 32 percent of degrees at public institutions and 8 percent of degrees at private schools.

We focus on data collected from doctoral, master’s, and bachelor’s degree-granting institutions in the United States. Our analysis looks at 20 core fields of study, including the largest fields (collectively accounting for more than half of student credit hours) and fields that are particularly salient for institutional leaders and policymakers. Our final sample contains 43,819 institution-year observations from 8,221 departments in 20 disciplines at 594 institutions. The data for online programs beginning in 2015 include 3,358 unique departments from 20 disciplines at 238 institutions across 3 years.

Using these data, we construct variables that measure costs, outputs, and inputs. Our primary outcome of interest is direct instructional spending per student credit hour, which we construct by dividing annual instructional costs by annual student credit hours at the department level. We also calculate this ratio for the personnel expenditures portion of costs. Finally, to analyze the sources of differences in costs across programs, we calculate average class size based on fall credit hours, faculty per student, and faculty teaching load.

Differences by Degree Program

We find substantial variation across disciplines in average costs per student credit hour (see Figure 1). Costs range from about $434 per student credit hour in electrical engineering to $163 per hour in math. Our benchmark department, English, costs $199 per hour; the mean cost across the group of 20 fields is $228. Most social-science disciplines and philosophy are relatively less expensive, while science, technology, and pre-professional programs like nursing are more costly. This broad conclusion holds across all institutions—nursing is a more expensive program to operate at elite private research institutions and less selective public institutions alike.

To compare costs across fields of study, we use the English department as a benchmark and look at costs by discipline from 2015–17. First, we calculate the average direct instructional cost per student credit hour for English at each school. We then calculate the within-institution difference between direct instructional expenditures per student credit hour for each of the other 19 disciplines and the same measure for English. We do this for all institutions and disciplines in our sample and then compute averages for each field of study.

In many cases, we find higher instructional expenses in the fields that also produce higher-earning graduates. For example, at $434 per student credit hour, electrical engineering costs 90 percent more than English. In looking at research by Brad Hershbein and Melissa S. Kearney, we see that adults with electrical engineering degrees have substantially higher average salaries throughout their careers. One year after graduation, electrical engineering majors earn more than double the average salary paid to English majors, or $63,000 compared to $31,000. Some 15 years later, that difference is about 83 percent, or $106,000 compared to $58,000.

There are several exceptions to these overall patterns. Math costs about 25 percent less than English, yet adults with mathematics or statistics degrees still earn more: $45,000 one year after graduation, a difference of 45 percent, and $76,000 in year 16, a difference of 32 percent.

In addition, education and fine and studio arts are among the most costly programs to operate, yet their graduates are also among the lowest paid. Education costs about $291 per credit hour, about 45 percent more than English. But adults with degrees in elementary education earn less than English majors throughout most of their careers—in year 16, they earn $44,000 compared to $58,000, a difference of about 35 percent. Similarly, fine and studio arts instruction costs about $273 per credit hour, but arts majors earn about the same or slightly less than English majors, on average: both graduates earn $31,000 one year after graduation. Some 15 years later, arts majors earn $55,000 versus $58,000 for adults with English degrees.

We also look at the growth in the number of student credit hours in each field during the study period of 2000–2017 and find that, on the whole, enrollment is growing fastest in some of the more costly fields. The highest annual rates of growth are in nursing, at 5.4 percent, and mechanical engineering, at 4.9 percent. Meanwhile, we also see that student credit hours are declining in four fields: English, history, education, and fine and studio arts. If those higher costs were associated with challenges in dynamically adjusting inputs, such as instructors and course sections, we would expect that fast-growing fields would have lower costs than slow-growing or declining ones. In fact we see the opposite.

More costly fields also are more likely to have access to additional revenue sources than English departments. Almost all of the most expensive fields are typically housed in separate schools from English, such as colleges of engineering, business, and education. This permits them to generate additional revenue through differential tuition bills or fees, and through separate fundraising efforts from alumni or industry. In some cases, these fields also have access to dedicated state appropriations for instructional costs. For example, in Texas and North Carolina, programs in the sciences, engineering, and nursing are eligible for more public support than programs in the liberal arts and social sciences.

Variety in Cost Contributors Across Fields of Study

We next investigate the factors behind these cost differences, looking at faculty salaries, class sizes, faculty workload—defined as the number of organized class sections divided by the number of full-time equivalent faculty—and non-personnel costs like equipment and supplies.

Instructional cost differences across fields can mostly be explained by large differences in class size across disciplines and, to a lesser extent, differences in average faculty pay. Teaching loads and other non-personnel expenditures explain relatively little.

Instructional Style

Although each field is slightly different, a few general patterns emerge. Economics, political science, accounting, and business have high faculty salaries that are mostly offset by large classes. Engineering and nursing are more expensive than English as a result of higher faculty salaries and lower teaching loads without commensurately larger classes. Workload and non-personnel expenses are important for some of the sciences with laboratory components—namely, biology and chemistry—but otherwise explain relatively little of the observed cost differences.

Consider economics, which is approximately 8 percent less expensive than English. Economics faculty are more highly paid than English professors. Postsecondary English instructors earn about 54 percent less than economics professors, with mean annual wages of $80,340 a year compared to $123,720. Thus if all cost drivers other than average pay were equalized between the two fields, economics would be more expensive. On the other hand, economics classes tend to be much larger than English classes, so class-size differences would make economics instruction less expensive. The faculty workload is a little lighter in economics than in English, so if that were the only driver of cost differences, economics would be about 3 percent more expensive.

Putting these findings together, we see that economics departments are able to field classes that are large enough to more than offset the higher salary and slightly lower workload of economics faculty, resulting in slightly lower average costs than English. Yet economics graduates earn substantially more: one year after graduation, the average salary is $48,000, or about 55 percent more than English majors. Fifteen years later, that difference grows to about 83 percent, or $106,000 for economics majors compared to $58,000 for adults with English degrees.

Mechanical engineering, a fast-growing field that is 62 percent more expensive than English, provides a counterexample. Engineering instructors have mean annual wages of $114,130, about 43 percent more than English instructors, as well as lower teaching loads than English faculty. As a result, the average difference in faculty pay across these two fields contributes substantially to the overall cost difference. Unlike economics, however, classes are only modestly larger in mechanical engineering than in English. Class-size differences are not large enough to offset the higher salary and lower teaching load, and thus mechanical engineering remains much more expensive than English. Research shows wide differences in earnings for mechanical engineering graduates compared to adults with English degrees: $60,000 one year after graduation, a difference of about 95 percent, and $104,000 some 15 years later, a difference of about 80 percent.

Faculty Type and Class Size

A department’s faculty salaries are driven by the mix of tenure-track and adjunct instructors as well as the salary for that particular discipline. We find quite a bit of variation in the share of tenure-track faculty by field (see Figure 2). In nursing, nearly 60 percent of instructors are not on the tenure track, while in mechanical and electrical engineering, three quarters of faculty are in tenure-track roles. English, communications, and math also have relatively low shares of tenure-track faculty. Thus, greater use of tenure-track faculty, which is more expensive, is one explanation for higher personnel costs in engineering, economics, and the sciences. Across the board, we see declines in the share of tenure-track faculty during the study period, with departments relying more on lower-cost adjunct instructors.

The second key cost driver, beyond faculty salaries, is class size. Differences in class size are a function of the mix of course types offered, such as lower-level undergraduate, upper-level undergraduate, and graduate, as well as the average class size for those courses. Lower-level classes tend to be larger and therefore less costly, whereas upper-level and graduate courses have smaller class sizes and are therefore more expensive.

We find substantial differences in the mix of course types offered, with relatively fewer lower-division courses in professional fields like education and business, and many lower-division courses in mathematics and science disciplines like physics and chemistry. Fields with relatively little undergraduate instruction, like engineering and nursing, tend to be more expensive. Looking over the study period, we find average class size conditional on level of course is fairly steady for most fields in the social sciences and humanities, in contrast to marked increases in undergraduate class sizes in engineering and biology.

Trends Over Time

We consider the average annual change in instructional costs for each of our 20 fields of study by looking at pieces attributable to changes in faculty salaries, class sizes, faculty workload, and other expenses. Across many fields, changes in faculty salaries and class sizes account for the bulk of the changes in instructional costs between 2000 and 2017. For instance, mechanical engineering saw a 2.10 percent reduction in cost each year, which is more than fully explained by an increase in class size. Costs for accounting rose by 0.64 percent annually, driven by faculty salary growth of 1.43 percent that outpaced increases in workload and class size. Some fields saw notable changes in faculty workload: education, English, and history all saw reductions in faculty workload over this period, which increased costs, while chemistry experienced a large increase. Only for nursing did changes in non-personnel expenditures increase costs. We also find appreciable declines in expenditures in a few tech-related fields, such as physics and computer science, perhaps reflecting falling costs for technology or lab supplies.

No Real Savings from Online Instruction

We then turn our attention to online instruction, which has commanded sustained interest from policymakers and institutional leaders as a possible strategy for containing college costs and expanding postsecondary access. We look at data from 2015–17 and find substantial variation in the prevalence of online instruction in the 20 disciplines we study. In undergraduate coursework, the share of online credits ranges from essentially zero in engineering to 13 percent of credits in nursing programs (see Figure 3). The average share of online credits is 6 percent, and 51 percent of the programs in our sample have no online enrollment at all. However, some graduate programs have considerable shares of online credits. For example, online coursework accounts for about one quarter of graduate education programs and one third of graduate nursing programs.

We investigate the potential cost savings of online classes and find a negligible association between online credits and instructional costs. Our estimates imply that adoption of any online coursework is associated with a cost increase of 0.4 percent, but an increase of 10 percentage points in online coursework is associated with a cost decrease of 1.4 percent. Neither of these is statistically significant. We view these estimates as small, especially given the hype about the cost-saving potential of online instruction.

A common hypothesis holds that online coursework can cut labor costs because such classes can be larger and require less face-to-face instruction. However, there is debate about the appropriate size for online courses relative to traditional in-person ones, with some institutions imposing lower enrollment caps for online courses than in-person study. We find some evidence that an increase in the share of undergraduate coursework completed online is related to lower salary costs. But estimates for the other cost drivers suggest that any short-run cost savings on salaries are offset by smaller class sizes and an uptick in non-personnel expenditures.

We note two caveats here: first, this analysis looks only at three years of data and thus cannot illuminate long-run cost changes that might emerge from the sustained adoption of online instruction; and second, we do not observe costs shared across departments, such as capital costs or costs for technology support. We also note the open question of whether the quality of online instruction is comparable to that of in-person classes, especially for less-prepared students. For example, a study by Stanford University’s Eric Bettinger and colleagues looked at a non-selective, for-profit school and found that students earned lower grades and were less likely to persist in school when they completed their coursework online. Some fields may find online education a more useful tool than others in lowering costs without compromising quality. Better understanding this potential is a productive path for future research.

Making Decisions at the Department Level

Over the past 17 years, average instructional costs per credit hour have increased only modestly. However, this relatively flat trend in average costs obscures variation in such cost trends by field of study. We find steep declines in spending in some science and technology fields, due to larger classes and increases in faculty workloads. Other fields, such as nursing, also saw declining costs that reflect a shift in the composition of faculty, with greater reliance on non-tenure-track staff. Yet other fields, such as business and accounting, have experienced escalating costs driven by rapid growth in faculty salaries. For all its promise, online education, arguably the highest-profile change to the delivery of higher education over this time period, is not associated with short-run cost savings.

Public debate about college costs usually focuses on differences between institutions. But the wide variety in costs by field of study should be part of that conversation too. It has important implications for institutional leaders facing decisions
such as differential tuition pricing and for government leaders considering programs to support more students to study engineering or nursing, for example.

Institutions have little control over the prevailing market wages for faculty, but changes in faculty workload, class size, and mix of course types across disciplines show some of the ways that costs might be kept in check. However, changes along these margins are also likely to shape research productivity and the capacity for public service. Thus, changes aimed at reducing instructional costs must balance potential effects on other valued activities of academic departments.

Our findings highlight the broad differences in costs and cost contributors among disciplines. We see a strong need for additional research that sheds light on the effects of instructor types, class sizes, and online classes on field-specific outcomes, including measures of quality such as student performance and success after college completion. For example, it may be that the adoption of online instruction reduces average instructional costs without impinging on quality in math but compromises student performance in chemistry.

Resource allocation decisions have strong effects on learning, instructional quality, and student outcomes, and these effects are likely to differ by field. Further research should explore these differences by discipline to help policymakers and institutional leaders work to reduce spiraling college costs while maintaining the quality of education that students strive to acquire.

Steven W. Hemelt is associate professor at the University of North Carolina at Chapel Hill. Kevin M. Stange is associate professor at the University of Michigan. Fernando Furquim is Director of Institutional Effectiveness at Minneapolis College. Andrew Simon is a postdoctoral scholar at the University of Chicago. John E. Sawyer is professor at the University of Delaware.

Recent changes to the National School Lunch Program have allowed some schools to offer free meals to all students, regardless of family income. This program expansion, known as the Community Eligibility Provision, applies to schools serving communities with low-income rates that exceed a federally designated threshold. Now, spurred in part by the impact of the Covid pandemic on students, President Joe Biden has proposed lowering this threshold and increasing the generosity of subsidies to Community Eligibility Provision schools. Some policymakers and advocates are pushing for even further expansion: offering free breakfast, snack, lunch, and dinner to every student, regardless of family income; having schools provide boxed lunches or groceries when school meals are unavailable; or sending pandemic-style payments to families to defray meal expenses during summers, school vacations, or weekends.

While the new proposals vary, the fundamental idea of making school lunch free for all students makes a lot of sense. A robust body of evidence points to positive effects of free lunch for all on a range of student outcomes, including test scores, nutrition, and disciplinary suspensions, with no persuasive evidence of negative unintended consequences for students or school budgets. In a 2021 national study, Krista Ruffini analyzed data from the Stanford Education Data Archive and found that math performance increases by 0.02 standard deviations in districts with the largest shares of students in Community Eligibility Provision schools (though effects on reading scores were inconsistent and statistically insignificant). Our own work finds slightly larger effect sizes for both math and English language arts (for both poor and non-poor children) using student-level administrative data from New York City. Susan M. Gross and colleagues surveyed 427 students in eight schools that met the Community Eligibility threshold. They found that food insecurity was higher among students attending schools that chose not to participate in the Community Eligibility Provision than among students at schools that did adopt the program. As for behavioral effects, Nora Gordon and Krista Ruffini studied national school-level suspension data from the U.S. Department of Education Office for Civil Rights and found that schoolwide free meals reduced suspensions among white male elementary-school students by about 17 percent, with statistically insignificant but substantively meaningful reductions among other subgroups of elementary-school students. Previous research has also consistently found that adopting the Community Eligibility Provision increases student participation in school-lunch programs.

A somewhat thinner literature documents small positive effects of free school breakfast, with lower student participation and cost. In 2014, for example, Diane Schanzenbach and Mary Zaki used U.S. Department of Agriculture experimental data to estimate impacts of universal free breakfast and breakfast-in-the-classroom programs, finding small increases in meal-program participation but little evidence that students increased their overall daily food consumption. (They found improved health only among students attending urban, high-poverty schools, and improved behavior only among minority students.) In a 2013 study, Jacob Leos-Urbel and colleagues used student-level data from New York City to explore impacts of universal free breakfast, finding increased participation for poor and non-poor students but limited evidence of impacts on academic outcomes overall (though there were some positive effects on attendance among poor Black students).

As for dinner or snacks, the research is thin, but the logistical challenges and resource requirements suggest that the costs of providing these meals would be high and that student participation in a dinner program would be limited.

Current Programs

Under the traditional rules of the school lunch program, meals are offered free to students from families with income under 130 percent of federal poverty line, at a reduced price to those with family income under 185 percent, and at full price to those with family income exceeding 185 percent. The federal government reimburses school districts based on the number of meals served, with rates depending on the proportions of meals served to students eligible for free, reduced-price, or full-price meals. Under the Community Eligibility Provision, schools (or groups of schools) can adopt “universal free meals”—extending free meals to all children, regardless of income—if the school’s share of students participating in SNAP or other means-tested programs exceeds 40 percent, with federal reimbursement determined by the share of students participating in these programs.

While reimbursement rates, eligibility criteria, regulations, and financing are set at the federal level, schools and districts have considerable discretion in program adoption and implementation. This broad latitude leads to wide variation in program specifics. Schools vary in meals served (breakfast, lunch, or snacks), menus (hot or cold options, fresh or prepackaged cuisine), schedule (service hours, school year only or summer too) and dining location (cafeteria or classrooms). Federal guidelines for school meals ensure they meet certain nutrition standards, so for many students school meals may be more healthful than meals brought from home or purchased outside school.

At the same time, parents and students can decide whether they want to participate. Students can bring brown-bag meals and forgo the school lunch. Some eat breakfast at home, and others skip it. Many participate on some days and not on others. Indeed, participation is far from complete, even among poor students and even when it is free. To some extent, participation decisions reflect family preferences and resources and the quality, cost, and accessibility of outside options (say, fast food restaurants nearby). But participation also reflects the quality and appeal of school meals and the social and physical context. Cramped cafeterias, long service lines, awkward schedules, student perceptions of “low quality” meals “for poor kids,” and stigma all dampen participation among poor and non-poor students alike. Universal free meals are likely to reduce stigma and boost participation—again, among both poor and non-poor—although participation is rarely “universal.”

Prior to the pandemic, more than 30 million children received school lunch and 15 million had breakfast daily in almost 100,000 schools and other institutions nationwide, and roughly 30,000 schools had adopted the Community Eligibility Provision. Afterschool snacks were offered at roughly 25,000 sites and the Summer Food Service Program was in place at 47,000 locations, reaching a much smaller population.

Expanding Access to Free Breakfast and Lunch

The Biden administration proposes two changes to expand universal free meals under the Community Eligibility Provision: 1) lowering the poverty threshold for eligibility, thereby increasing the number of schools that qualify and 2) increasing reimbursements for school meals, making it more affordable for schools to participate. There are several reasons to believe these changes will benefit America’s children.

Perhaps the most compelling reason to expand the program is the persuasive evidence that it works. Studies have shown that the Community Eligibility Provision and other universal free-meal programs increase student participation in school meals, which research indicates are nutritionally better than the meals students might otherwise consume. In 2015, for instance, Michelle Caruso and Karen Cullen studied lunches brought from home by 242 elementary-school and 95 middle-school students in 12 schools in Houston, Texas. The researchers found higher sodium content and fewer servings of fruits, vegetables, whole grains, and fluid milk in these lunches than required under National School Lunch Program standards. As for academic outcomes, recent research shows schoolwide free-meal programs improve student performance on standardized tests, reduce suspensions, and may improve attendance. As might be expected, studies show significant improvements among kids for whom school meals would not have otherwise been free—driven, perhaps, by impacts on students whose family income only
modestly exceeds the threshold (lower-middle-income families). Studies also show improvement among the poor students who would qualify for free meals anyway—perhaps owing to a reduction in the stigma associated with participating in school lunch, changes in menu options, or easier program administration.

At the same time, there is little evidence of unintended negative consequences—no increases in child obesity and no explosion or meaningful increases in school food-program deficits (indeed, our 2021 working paper showed that deficits shrank, on average, in New York State). And free meals for all may even have some unintended positive consequences—a 2021 working paper by Jessie Handbury and Sarah Moshary suggests that a school’s adoption of the Community Eligibility Provision may even lower food prices in nearby grocery stores!

To be sure, federal outlays have increased, but by a small amount per pupil, and the bang for the buck in academic outcomes is larger than that of many alternative education reforms, including class-size reduction or increases in teacher salaries. Altogether, annual federal spending on school-meal programs of roughly $390 per pupil ($18.4 billion total) is 3 percent of total educational spending, which amounts to roughly $13,000 per pupil.

Will the proposed changes yield similar benefits? The design of Biden’s reforms is promising: lowering the eligibility threshold extends the option to more schools, but doesn’t require universal free meals. If schools decide wisely, those opting in will benefit, and those that will not benefit will opt out. Increasing the subsidy’s generosity makes it easier—and more attractive—for schools to opt in, offsetting lost revenue from school-lunch fees or higher costs. To be sure, participation is likely to remain incomplete. While the adoption of Community Eligibility has been increasing, nearly one-third of the nearly 50,000 schools eligible for the program have so far declined to join, citing reasons such as high rates of lunch purchases among “full-pay” students, constraints on physical and human capacity, and concerns about financial implications, among others.

Might it be better to devote school-lunch funds exclusively to low-income students? Lowering the threshold means an increasing share of students in newly eligible schools will be “non-poor,” and more middle- and high- income families will benefit. Despite the intuitive appeal of targeting, however, universality seems to benefit poor students more than the traditional means-tested programs and helps other needy students at the same time.

Should Schools Offer Dinner Too?

While the success of the free breakfast and lunch programs might suggest schools should offer dinner as well, logistical and practical challenges and likely lower demand make for a weak case. Since dinner-at-school would probably be offered some hours after the end of the school day, it could require keeping school buildings open longer, extending work hours for cafeteria workers or adding a second shift, and forgoing competing civic and educational uses of school buildings. Further, demand is likely to be lower for a meal not adjacent to the school day, as students participate in activities elsewhere and many people value family dinnertime. The farther food delivery gets from the core instructional day (vis-a-vis time of day, week, or year), the less efficient such meal service is likely to be. Dinner-at-school may be “a meal too far.”

Similar considerations make providing summer meals and weekend meals less compelling. Expanding SNAP or other federal anti-poverty or anti-hunger programs may be a more effective way of supporting child nutrition outside of school hours. Still, there are circumstances under which extending the food program does make sense—in high-poverty schools, for example—and pilot programs such as those implemented years ago for the school-lunch program, breakfast program, and the Community Eligibility Provision could serve to test the merit of these possible expansions of the program.

Let Them Eat Lunch (and Breakfast)

The Biden administration proposes expanding universal free lunch and breakfast through a lower eligibility threshold for adopting the Community Eligibility Provision and through a more generous reimbursement rate. These proposals are backed by research on the effects of the school-lunch program, and the availability of free meals for all promises to provide much-needed help to students working to overcome challenges posed by the pandemic. The comparative advantage of schools providing meals for students during school hours, or just before or after, appears quite high. School meals offered beyond these hours, however, are not likely to draw robust participation, may stigmatize partakers, and could strain resource-strapped schools and districts already juggling a panoply of responsibilities.

Before the pandemic, more than half of American public-school students were eligible for free or reduced-price school lunch. Senator Bernie Sanders of Vermont has proposed expanding the program to provide free breakfast, lunch, and dinner to every American public-school student. President Joe Biden is pushing a more modest reform: lowering the threshold for “community eligibility” for free or reduced-price lunch to cover another projected 9.7 million students—though this number may prove an underestimate.

There is a strong case for having the government provide food to children whose parents can’t afford to feed them adequately, but that’s not the question at hand. The question is whether the government should feed children whose parents can afford it. Conservatives have traditionally argued “no” from the perspective of fiscal responsibility. Progressives counter that universal school lunch would reduce paperwork burdens, yielding administrative efficiency gains. Another claim is that universal free lunch will fight the stigma and taunting kids who get free lunch reportedly experience. If everyone were eligible for the program, the argument goes, the lower-income kids wouldn’t get “lunch-shamed.”

Suppose, for the sake of argument, that universal free lunch would indeed mitigate bullying and that kids wouldn’t just find other pretexts to pick on each other.

Bullying, though, is hardly the only moral question that’s involved. An international perspective may help clarify some of the issues. Schools in Switzerland, by and large, do not provide lunch. Rather, students break for two hours midday and generally walk back home to be fed by their parents. The sight of young children walking through the streets by themselves is unremarkable in high-trust Swiss society. In low-trust American society, by contrast, it can provoke calls to the police.

Swiss parents like having their children come home for lunch, because the thought of turning them over to a government institution all day is abhorrent to them. Many Americans, in contrast, find it hard to imagine a non-pandemic world in which a parent would be expected to be at home during the workday to serve lunch. Which attitude is more conducive to a flourishing society? The economists who compiled the 2021 World Happiness Report ranked Switzerland third and the United States 18th on citizen happiness.

The Swiss have it righter, I suspect. My judgment is based less on the word of economists than that of my mother. She frequently told me that she took great joy in preparing my breakfast and lunch every day. That struck me as a natural and beautiful thing. Parents have a primal drive to provide food for their children. But parents are also sensitive and responsive to the social pressures their children face. If kids apply stigma to behaviors that go against norms, then universal free lunch could generate a stigma against kids bringing brown-bag lunches, discouraging parental food preparation.

Would that really be good for parents or for children? I’m reminded again of my mother. In cultures and religious traditions throughout the world, it is an age-old custom to give thanks to God before or after meals, recognizing that eating is partly a sacramental act. When my mother was a teacher in the Cleveland, Ohio, public schools, she was unnerved by the entitled attitude her elementary students took to the free school breakfasts consumed in her classroom (not to mention by the massive food waste). So she asked her students, before they ate, to say in unison: “Thank you, State of Ohio.”

She thought this was better than no expression of gratitude. She was probably right, but I found it unsettling. As a high school student, I couldn’t articulate why, but today I can. The children had to contemplate the state as provider, rather than reflecting on how the love and labor of parents brought food to their plate. That experience shapes a child’s moral worldview, with human consequences that evade econometric analysis. Since the government, not the family, is already providing the education, the food may seem like a minor detail. But as the religions recognize, it carries significant meaning.

Progressives eager to expand school lunches, breakfasts, and dinners may be disappointed to discover that even after all the heavily touted efforts to make school lunches more local and nutritious, what gets served in school cafeterias remains heavily influenced by Big Agriculture and its lobbyists. Expect more mystery meat and french fries, not free-range arugula or heirloom citrus. The same progressives who blame meat consumption for global warming now want to serve more factory-farmed hamburgers, bacon, and sausage in school cafeterias. Even the vaunted liberal commitment to “diversity, equity, and inclusion” is forgotten when it comes to children who may want to eat kosher or halal, prefer to avoid meat entirely, or just have allergies or food sensitivities. A salad bar or cereal may be available, but school cafeteria menus largely serve majority tastes.

And to what end? A literature review published in 2020 in the American Journal of Public Health examined the effects of the Community Eligibility Provision that Biden seeks to expand. The review, by Johns Hopkins scholar Amelie Hecht and co-authors, noted that, of five studies on universal breakfast, “3 found no change in test scores and 2 found some improvements.” Among studies focusing specifically on Community Eligibility for lunch, “2 detected improvement in test scores for some subjects and age groups and the third detected no change.” The authors note that the positive effects were “relatively small” and “similar in magnitude to those seen when families receive other forms of income support, such as the earned income tax credit.” Why not, then, just allocate the additional money directly to parents, the way that the government already does with the Supplemental Nutrition Assistance Program and did during the pandemic when schools were closed with the Pandemic Electronic Benefit Transfer Program?

Perhaps the strongest argument for spending additional money on the federal school-lunch system is that school lunches are, reportedly, much healthier than grocery-store bought food. A 2021 study in JAMA Network Open by Junxiu Liu and co-authors found that, after the Obama administration’s school-lunch overhaul, according to the American Heart Association’s diet index, “diet quality for foods from schools improved significantly. . . . By 2017–2018, food consumed at schools had the highest quality, followed by food from grocery stores, other sources, worksites, and restaurants. . . . Findings were similar for [the USDA’s] Healthy Eating Index.” That study relied on survey respondents’ self-reports of what they ate, rather than on an external observation. School lunches are nutritionally different from other foods, but whether they are better or worse depends on whether the USDA and the American Heart Association are right or wrong in their dietary assumptions. The USDA’s food pyramid is subject to all kinds of political pressure, and expert advice on what to eat is subject to change. It wasn’t that long ago that the experts were telling us to eat margarine instead of butter, then reversing course to warn of the dangers of trans fats. Or those same experts were telling us to eat pasta instead of fats, then discovering that too many carbohydrates were bad for us. When the menu planning is nationally centralized, or organized to meet federal standards rather than a family’s, erroneous nutrition choices are magnified and imposed on tens of millions of children.

The old saying goes: “There ain’t no such thing as a free lunch.” It’s true, but not only from a budgetary perspective. Making school lunch universally available would also come at a high moral, social, and potentially health-related price.

A fuller version of this essay is available as an AEI report, ”The Case Against Universal Free Lunch.”

There are more children in school worldwide today than at any other time in history, pandemic-related disruptions notwithstanding. In 2010, the average adult had completed 7.6 years of school, more than double the 3.2 years completed by the average adult in 1950. These estimates, based on our review of data from 164 countries around the world, appear to represent a substantial achievement toward human progress. In particular, they would seem to indicate the potential for dramatic economic growth in the developing world, which accounts for much of the enrollment difference.

But does more schooling necessarily lead to economic progress? Development experts have long argued that an expansion in school enrollments would equip the next generation with knowledge and skills to enhance their economies and promote a better quality of life. But a new database of detailed student-achievement data suggests that greater school enrollments have been followed by little to no growth in learning in most parts of the world.

We constructed the Harmonized Learning Outcomes database by linking regional assessments of core academic subjects in less-developed countries to international achievement tests commonly administered in the wealthier parts of the world. This database of 164 countries allows us to compare student performance in regions that are typically excluded from international comparisons, such as sub-Saharan Africa and Latin America, to that of more affluent students in Europe and North America. Our data cover 98 percent of the global population and track student-learning outcomes from 2000 to 2017.

We find that while rates of school enrollment topped 90 percent in every region by 2010, learning outcomes are low and have barely budged over the past couple decades. Students are in school, but they learn very little.

Wealthy countries like the United States are not immune to what international aid organizations have termed the global “learning crisis.” In North America, where school enrollment rates have been high for decades, our database shows that student performance on standardized tests in reading, math, and science barely budged between 2000 and 2015. Meanwhile, countries that have made dramatic progress in achieving high rates of schooling, such as Brazil and Ghana, still have very low rates of student learning.

Expansion in access to school may provide more opportunities to learn and may have other benefits as well, but schooling does not guarantee learning, and the relationship is not one to one. To understand the potential benefits of the global growth of education to economies around the world, we cannot assume that schooling leads to learning. Measuring learning directly is essential.

An Expanded Inventory of Student Learning

In recent decades, international assessments such as the Programme for International Student Assessment and Trends in International Mathematics and Science Study have measured student learning in a steadily growing number of countries. In 2015, students in 71 countries took the PISA, and students in 65 countries took the TIMSS.

These psychometrically designed, standardized assessments of cognitive skills enable global comparisons of learning across countries and over time. However, these tests exclude many low- and middle-income countries, which makes it difficult to analyze learning patterns in the developing countries that may have the most to gain from improving education quality.

To bridge this gap in learning outcomes data, we built the new Harmonized Learning Outcomes database, which includes measures of student learning that cover 98 percent of the global population. The database expands the number of countries surveyed to 164, two-thirds of which are developing countries, by linking international assessments to their regional counterparts. For example, rather than the PISA or TIMSS, most students in sub-Saharan Africa take exams administered by the Southern African Consortium for Monitoring Education Quality, and most students in Latin America take tests from the Laboratorio Latinoamericano de Evaluación de la Calidad de la Educación. Altogether, our database includes seven assessment regimes: three international tests that are global in scope, three regional tests, and the Early Grade Reading Assessment. The latter provides at least one data point in the past 10 years for an additional 48 countries, including large developing economies such as Bangladesh, Nigeria, and Pakistan. Each test covers between 10 and 72 countries.

To create comparable measures, we convert regional test scores to an international scale within subjects, which include mathematics, reading, and science, and within levels of schooling, such as primary or secondary. This is possible when one or more countries participated in both a regional and an international test in adjacent years during our target period of 2000 to 2017. We then apply this conversion to countries that participate in a regional test but not an international test to produce comparable scores, which we call Harmonized Learning Outcomes. On this scale, a score of 625 is considered high performance. The scores of individual countries vary; for example, in the latest available year, the United States has a score of 545, Mexico has a score of 453, and Uganda has a score of 391.

Schooling Isn’t Learning

The scores show that progress in learning has been limited over the past two decades, even where schooling opportunities have expanded. Looking across regions, we find 72 countries with data for primary school that include at least two data points for both enrollment and learning. First, we consider enrollment rates based on estimates from Jong-Wah Lee and Hanol Lee, which show a clear pattern of high and increasing school enrollment through 2010. Then we compare that to trends in student learning based on estimates from our database, which show little to no progress in every region of the world (see Figure 1).

For example, in the Middle East and North Africa region, enrollment rates for primary education increase to 99 percent by 2010. However, from 2000 to 2015, learning levels remain at around 380—not far above the low-performance benchmark of 300. In looking at individual countries, we find that many developing countries have achieved substantial levels of schooling but have not yet realized high rates of learning. Ghana has 11.6 years of expected schooling, yet its learning score of only 229 does not meet the low-performance benchmark. Brazil has reached 11.7 years of expected schooling, yet achieves a learning score of just 426—far below the high-performance benchmark of 625.

These findings reinforce messages in a 2018 report by the World Bank, which similarly highlighted the acute gap between schooling and learning in developing countries. For example, in Kenya, Tanzania, and Uganda, three-quarters of the students in grade 3, who are typically eight or nine years old, cannot read a basic sentence such as “the name of the dog is Puppy.” And in rural India and Uganda, more than 80 percent of grade 2 students cannot solve a subtraction problem with numbers greater than 10, such as 46-17.

It is possible that in regions such as sub-Saharan Africa, as enrollment grows substantially higher, and new, lower-performing students participate in learning assessments, average scores decrease owing to a selection effect. In particular, strong enrollment growth since 2000 points to this possibility. These enrollment increases followed prominent declarations, such as “Education For All,” as well as the introduction of compulsory schooling laws and the elimination of school fees in many developing countries. However, we observe slow learning progress even in regions where enrollment levels are relatively constant and high, such as North America and the Caribbean. This suggests that there is more to these trends than selection.

We also calculate average learning levels in reading, math, and science in primary school in all 164 countries represented in our database. Average learning levels in high-income countries far outpace those in developing economies (see Figure 2). In looking across regions, we find that North America has the highest average learning levels, followed by Europe and Central Asia. Meanwhile, sub-Saharan Africa and South Asia have the lowest average learning levels.

However, the performances of individual countries in these regions varies, and some countries buck regional trends. For example, in sub-Saharan Africa, countries such as Kenya and Tanzania have higher average levels of learning, on par with many countries in Latin America. Within Latin America, a few higher-performing countries such as Chile are on par with learning levels in Europe. Most countries in the Middle East perform similarly or worse than those in Latin America, while some Asian countries, such as Singapore, outperform North America and Europe.

Learning, More Than Schooling, Is Linked with Economic Growth

These differences have broad implications for countries’ economies that are not yet fully captured in leading measures of global development. For example, the Human Development Index published by the United Nations, which creates a single country-level measure of well-being based on average life expectancy, education, and standard of living, uses numbers of years of schooling as its proxy for human capital. While access to school correlates with multiple measures of national well-being, prior research has shown that the relationship between years of schooling and economic growth is weak. By contrast, measures that use learning to define human capital are more strongly associated with economic growth (see “Education and Economic Growth,” research, spring 2008).

The Harmonized Learning Outcome data allow us to document these patterns on a global scale (see Figure 3). When we use students’ test scores to define a country’s human capital, the scores account for between one-fifth to around half of differences in income levels among countries. That is nearly twice as much as when those estimates are based on the amount of schooling the average student in a country receives. Learning has a stronger and more positive relationship with economic growth than schooling.

 

However, greater incomes do not inevitably produce a more educated and skilled society. We looked at per-capita income and average levels of learning for all of the countries in our database and found broad variation in countries’ performance relative to what their incomes would predict (see Figure 4). For example, the United States, when compared to all other countries, does as well as its per-capita income level predicts. However, it underperforms some countries with similar income levels, like Singapore. We also find countries that are much less developed but do just as well, such as Poland and Vietnam, along with other rich countries, such as several in the Middle East, that drastically underperform the achievement level expected given the per-capita income. This may be because their economies rely heavily on natural resources, which requires a lesser investment in education, rather than on highly skilled labor, which requires high levels of human capital. South Africa and Kenya provide another notable contrast, with Kenya outperforming its much richer regional neighbor.

Focus on Student Learning

This analysis gives us both good and bad news to share. The good news: the world is on track to achieve universal primary-school enrollment by 2030. The bad news: there is little progress in learning outcomes in most countries. Our results show the learning crisis is worldwide.

Now is the time to take action. Data from high-income countries, where standardized tests that measure student learning are regularly administered, show that estimates of economic growth based on learning outcomes are far more predictive than those based on enrollment. Those estimates have often not been available on a globally comparable scale for middle- and low-income countries.

We must shift our focus to measuring and advancing learning as a priority goal around the world, just as we have measured and improved access to schooling for the past 50 years. The world largely succeeded in greatly expanding schooling, in part because it was measured carefully and consistently. Learning must receive the same attention and care.

Our global database and analysis can point the way. With the expansion of standardized testing and new tools to make clear comparisons across assessments administered around the world, we can measure learning more accurately and more completely than in the past. Countries need to provide students with more than just access to school. Around the world, citizens need the educational experiences that promote higher levels of learning to ensure global prosperity in the twenty-first century.

Noam Angrist is a fellow at the University of Oxford, cofounder of Young 1ove, and a consultant with the World Bank, where Harry Patrinos is practice manager. Simeon Djankov and Pinelopi Goldberg, both formerly of the World Bank, are senior fellows at the Peterson Institute for International Economics. Goldberg is also the Elihu Professor of Economics at Yale University. This article is adapted from a paper originally published in Nature in March 2021, Measuring human capital using global learning data.

Bill Oberndorf has committed his resources to expanding opportunities for children from disadvantaged backgrounds. He chairs the American Federation for Children, which funds scholarships for low-income students to attend private schools and supports pro-school-choice political candidates at the state level. In 2021, Oberndorf was named the Simon-DeVos Philanthropist of the Year by the Philanthropy Roundtable, a national association of philanthropists. Education Next senior editor Paul Peterson recently spoke with Oberndorf about the state of school choice in America.

Paul Peterson: Why did you de-cide to focus much of your philanthropy on helping disadvantaged children attend private school?

Bill Oberndorf: I felt extremely fortunate that I was able to attend a wonderful private school in Cleveland, and only because my grandparents set aside and saved money for the education of my brothers and me. I felt that every kid who wants to work hard in school, whose parents want something better for them, should have access to the kind of education that best fits the needs of that child. I feel that this is the civil-rights issue of our time.

The idea of private-school choice through government-funded vouchers was proposed by Milton Friedman in the 1950s. Seventy years later, we have only a few such programs in this country. Why has it been so difficult to build public support for this idea?

I remember talking to Milton Friedman about this shortly before he died. He said, “Well, we’re just about right on schedule. It takes decades for ideas to take root before they really can flourish.” So Milton was not deterred. The opposition has come from the teachers unions, which are such a powerful force and funding source for the Democratic Party that this has created major obstacles along the way.

But the good news is that now there are private-school choice programs in 22 states. And 45 states plus D.C. have charter-school programs.

Yes, but in recent years it seemed like progress was stalling out. In 2016 in Massachusetts, for example, a ballot initiative to expand charter schools was defeated, even though charter schools in Massachusetts seemed to be doing very well. There were also divisions within the school-choice movement, and the energy seemed to be disappearing. How were you assessing the state of school choice at that time?

The charter-school movement had scaled up to around 3 million students enrolled, and suddenly, for the first time, that sector was feeling the kind of union opposition that the private-school choice movement had felt all along. This did create a lull, but since then, some important things have happened that have helped change the overall trajectory of the advocacy and implementation of private-school and charter-school choice.

What’s happened, of course, is the Covid-19 pandemic, and the shutting down of district schools across the country, with private schools remaining open in many places. Do you think that’s critical to what seems to be a turning of public opinion today?

Yes. The tide went out because of Covid, and many people who never had been touched by the impact of union power suddenly felt that impact. The other factor was that because so much remote instruction was going on, parents actually saw for the first time the quality of the teaching in their kids’ classrooms, and they didn’t like what they saw. This was a real eye-opener, and it has caused the acceptance and popularity of education choice to skyrocket.

Over the last school year, a lot of people moved away from the standard district-run school, either to the private sector, to charter schools, or to homeschooling, which has exploded. Is this people voting with their feet against what was happening during the pandemic?

Absolutely. And at the American Federation for Children, which is now the largest school-choice organization in the country, we start with funding state legislative races and directly backing candidates. You referenced the ballot initiative losing in Massachusetts. There has never been a ballot initiative that’s passed, because it’s too easy to knock them off. Instead, we look at states where we feel that, over a three- to five-year period, we can change the legislative composition to be favorable to choice and where we can help elect a governor who is receptive to signing such legislation.

In 2020 we backed 390 state legislators and won 337 of those seats, concentrated in 13 states. And what resulted in 2021 was the passage of legislation funding 150,000 new private-school seats, at about $6,000 dollars apiece—almost $900 million of government money. And, as you mentioned, there were also increases in homeschooling and in charter enrollment. This shift is having a big political influence too, in how people vote once they see how their children are benefiting from these programs.

What do you see as the main driver here?

I think it’s the culmination of a lot of frustration that parents have had over the years—and particularly the kind of parents we try to help, low-income parents, and this is changing how they are voting.

Governor Doug Ducey from Arizona told me he got 44 percent of the Hispanic vote the last time he ran. He said, “That’s only because of this issue of school choice. That’s the only reason I got that kind of percentage.” And when Ron DeSantis’s opponent, Tallahassee mayor Andrew Gillum, said, “We’re going to end the school-choice programs in Florida,” DeSantis ended up getting 18 percent of the Black female vote in the gubernatorial election. That was 70,000 votes, and he won by 30,000 votes. So this is changing outcomes, with people who are simply tired of seeing what’s happening to their children, who are subject to sending their kids to schools that none of us would ever let our kids go to.

Are political leaders talking to one another from state to state? Is this what’s moving the conversation?

Yes, I think school choice is finally gaining traction in a way we’ve never seen before. And in this next election cycle, the federation will have 550 different state legislative races to invest in if we are able to raise the funds to do so. Governors understand the implications of school choice, and politicians of color are understanding it is good for their constituents. So we don’t view this as a Republican issue or a Democratic issue. About 20 percent of the money we give to candidates every year goes to Democrats. We’d like it to be a lot higher than that.

And I think that, with what happened in the wake of George Floyd’s murder and the riots in the summer of 2020, you cannot have a real conversation about systemic racism if you do not talk about K–12 and the outcomes for these kids. It is what’s holding back students of color in this country. It’s an inconvenient truth. If we do not talk about this, I don’t think we will be able to make substantial progress moving forward.

This is an edited excerpt from an Education Exchange podcast, which can be heard at educationnext.org.

For years, media attention to charter schools has focused on the horse race: which schools are better, charter schools or district schools?

What if one were to tweak this question and ask instead: which type of school shows greater capacity for improvement, and what can educators and policymakers learn from the answer?

For some time, research has indicated that charter schools, on average, provide a superior education to students living in poverty, Black students, and Hispanic students. Now, research also shows charter schools are improving at a faster rate than district schools.

For our most disadvantaged students, charter schools are not only out in front, but they are also widening their lead.

That is great news for the children enrolled in charter schools, but no consolation to those who are not. To accelerate the achievement of all children in all types of schools, it may help to take a closer look at why one group of public schools (charter) is improving faster than another (district).

The answer is twofold:

• The combination of choice and flexibility provides charter schools with the incentive and the ability to implement practices that lead to better results.
• The charter sector has taken decisive actions based on those results, closing low-performing schools and replicating those that are succeeding.

These two factors work in tandem and reinforce each other to drive improvement; one without the other would not likely produce the same level of progress.

States began enacting charter-school laws 30 years ago, in part to create a “laboratory” for learning about effective innovation and improvement that could be transferred to other public schools. Three decades in, that knowledge is available and, if we do learn from it and apply it throughout public education, it can be used to accelerate learning for all children.

Performance Data

M. Danish Shakeel and Paul Peterson recently published research examining the changes in student performance at charter and district schools on the National Assessment of Educational Progress from 2005 to 2017 (see “Charter Schools Show Steeper Upward Trend in Student Achievement than District Schools,” research, Winter 2021).

Controlling for differences in students’ background characteristics, they found that student cohorts in the charter sector made greater gains than did those in the district sector (see Figure 1). “The difference in the trends in the two sectors amounts to nearly an additional half-year’s worth of learning,” the authors wrote. “The biggest gains are for African Americans and for students of low socioeconomic status attending charter schools.”

In 2013, Stanford University’s Center for Research on Education Outcomes, known as CREDO, reached a similar conclusion related to research the center had done four years earlier. “When compared to the 2009 results, the 2013 findings indicate overall improvement in learning gains for students at charter schools relative to their traditional public school peers in both reading and math,” the center reported.

Here too, the differences were most pronounced for low-income students:

“Compared to the learning gains of TPS [traditional public school] students in poverty, charter students in poverty learn significantly more in math,” the report said. “Moreover, this difference in performance has widened.” In 2009, charter students in poverty had an advantage of about 7 more days of learning in math each year than their TPS peers. In 2013, the edge was 22 additional days.

Patrick Baude and colleagues found similar results in a study of Texas charter-school performance from 2001 to 2011. “Charter school mathematics and reading value-added increased substantially relative to traditional public schools,” the researchers wrote. “This improvement is notable because there is evidence that traditional public schools were also improving on average.”

What explains the difference in these improvement rates? And what can policymakers and K–12 educators learn from this information?

The Role of Choice

Throughout the year, the principals and boards of charter schools focus on one particular set of data: the enrollment numbers for the coming school year. In the winter and spring, they look at the number of applicants and the grades to which they are applying. If demand is low, they are compelled to find ways to attract more students. In the summer, after lotteries have occurred, they project how many students will show up when school opens. In the fall, they compare actual enrollment and attendance to earlier projections.

Of course, principals at district schools also pay attention to enrollment, but not as often or in the same ways. For charters, the issue of enrollment spells constant pressure to improve.

That’s because a charter school’s enrollment has an immediate and significant impact on the school’s budget and the services it can provide. A school expecting 500 students that enrolls 490 may lose funding for that year that’s roughly equivalent to a full-time teaching position.

A district school that experiences the same enrollment shortfall would likely experience no impact at all. The district will shield that school from the revenue loss for that year and perhaps for years to come. (Those funds must come from somewhere, of course, and they come at the expense of other schools that are not losing enrollment.)

In a high-performing charter school, the incentive to achieve enrollment projections creates an organizational mentality focused on continuous improvement in every sphere: academics, culture, extracurriculars, teachers’ job satisfaction, communication with parents, and more. While some charter schools (often those that are struggling) spend big money on marketing campaigns, high-performing schools know that the most powerful marketing is parent word-of-mouth. If a school is delivering for students and families, others will learn about it and apply. If it is not delivering, people will hear about that, too.

The Power of Flexibility

Choice drives the quest for improvement in the charter sector, but choice itself does not improve teaching and learning. Rather, it is flexibility that enables charter schools to improve in ways that are less available to district schools.

One way that charters have tapped into their flexibility is by lengthening the school day and school year. Using data from the 2007–08 school year, the National Center on Time and Learning observed, “Charters, as opposed to traditional public schools, are more likely to extend their school year, offer longer days, and operate a year-round school calendar.” That year, the typical charter-school day was nearly 15 minutes longer than a day at its traditional public-school counterpart. While 23.5 percent of charter schools reported a longer school year than the conventional 180 days, 16.7 percent of traditional public schools did so. Ten percent of charter schools offered a significantly longer year of 187 days or more.

Fast forward to the National Center on Time and Learning’s 2012 report “Mapping the Field: A Report on Expanded-Time Schools in America.” Noting that the first serious proposal for expanding school time had appeared nearly 30 years earlier in the landmark report A Nation at Risk, the center’s publication underscored that “no movement ensued on the part of traditional public schools to break from the conventional calendar and/or schedule. The one notable exception to this adherence to school-time norms came from the emerging group of independent public schools known as charter schools.” Charter founders, the center’s report observed, had “crafted their schools—which had been established to be deliberately unlike the conventional—on a platform of a longer school day and/or year.”

Charter schools, which made up just 5 percent of public schools nationwide at the time of the center’s study, constituted 60 percent of all expanded-time schools.

The center also observed that it was much more common for start-up schools to adopt an expanded-time schedule than it was for an existing school to convert to the model. Nearly three-quarters (72 percent) of the expanded-time schools they identified were start-ups.

In 2015, the center’s findings on charter schools and time were even more pronounced. “The average charter school day has grown markedly over the last decade, with particular growth in the upper quartile,” the center noted in a review of research and practice.

Research indicates that more time in school generally leads to improved learning outcomes. Citing a meta-analysis of 15 studies, the National Center on Time and Learning found that additional time in school “can have a meaningfully positive impact on student proficiency and, indeed, upon a child’s entire educational experience. Such enhancement can be especially consequential for economically disadvantaged students. . . . For these millions of students, more time in school can be a path to equity.”

Of course, if extra time in school is to have this positive effect, a strong academic program is essential. As in all schools, academic excellence in charter schools is dependent on strong teachers. Research suggests that the workforce in charter schools differs from that of district schools in several important ways.

In 2012, roughly midway through the 2005–2017 timeframe studied by Shakeel and Peterson, the National Conference of State Legislatures reported that teaching looked different in charter schools in several areas:

Demographics. “Charter school teachers are more diverse; there are almost twice as many black and Hispanic teachers in these schools.” Further, “some data indicate charter school teachers are more likely to have graduated from a competitive or selective college.”

Licensure. “Fourteen states [out of 41 states with charter-school laws at that time] require only a certain percentage of charter teachers in each school to be licensed, varying between 30 percent and 90 percent. Four states and the District of Columbia have no requirement for licensure.”

Turnover. “Involuntary attrition is significantly higher in charter schools due to the lack of barriers to teacher dismissal and to a school’s possible instability.”

Collective Bargaining. “Twenty states and the District of Columbia exempt charter schools from collective bargaining agreements and only Iowa holds all charter schools to all existing school district collective bargaining agreements.”

While these teacher variables—demographics, licensure, turnover, and collective bargaining—could explain charter schools’ performance relative to district schools at any given point in time, they do not speak directly to the faster rate of charter-school improvement over time. A 2020 study by Matthew Steinberg and Haisheng Yang does. They first review prior evidence indicating that charter-school teachers improve with experience at a faster clip than district-school teachers. In their own study of Pennsylvania schools, they find that this is particularly the case for charter schools that are part of charter management organizations, whose teachers “improve more rapidly than teachers in its traditional public schools or standalone charters.”

Decisive Action

Imagine a city with 100 schools where, every year, the three or four lowest-performing schools in the city close and a handful of new schools open. The quality of the new additions ranges from weak to excellent, but in the aggregate, they are average. Over time, replacing the three or four lowest-performing schools with average schools will lead to improvement. Replacing them with above-average schools would lead to even faster improvement.

This scenario has not happened often among district schools, despite bold public policies like the No Child Left Behind Act and Race to the Top. With few exceptions, districts have resisted closing schools, even those that have persistently failed to educate children satisfactorily.

The charter sector, though, has embraced this scenario, annually closing 3 to 4 percent of its lowest performers for years. Over time, the sector has opened not only average schools, but a greater number of excellent schools—those run by charter management organizations.

According to the National Alliance for Public Charter Schools, from 2005–06 to 2017–18, the charter sector closed between 3.1 percent and 3.7 percent of its schools every year but two, with an average of 185 closures per year (see Figure 2).

Throughout that time, I led the National Association of Charter School Authorizers, which advocated for charter-school quality and accountability. In 2012, in our “One Million Lives” campaign, we called on public officials and authorizers to close a thousand low-performing charter schools by 2017 and to open two thousand new, high-quality charter schools. The goal was ambitious, since the sector had never closed 200 schools per year even once. Yet the campaign was widely embraced by the charter community, including advocates in states with many charter schools like Texas, California, Arizona, and Ohio.

In 2013, the Texas Charter School Association successfully advocated for the passage of a state law that raised performance standards for charter schools and provided for the closing of schools that failed to meet those standards for three successive years. The legislation also raised the cap on the number of charter schools allowed in the state and streamlined the renewal and replication process for successful schools. The Texas reform law embodied the charter philosophy: growing the number of high-quality schools and closing those that persistently failed to deliver for kids. During the two years leading up to the law’s passage, two charter schools had closed in Texas. In the two years following, 20 charters closed.

California’s state charter-school association also publicly pushed for high standards and the closure of charter schools that persistently failed to deliver results. Beginning in 2011, the association annually identified charter schools it recommended for nonrenewal. “We have too many persistently underperforming charters, and we need to come up with constructive suggestions to make sure there is sufficient accountability in the movement,” said Jed Wallace, president of the association.

In Arizona, DeAnna Rowe became the executive director of the Arizona State Board for Charter Schools in 2007. With over 500 schools throughout the state, Rowe said in an interview that the board had until then taken a “let a thousand flowers bloom” approach. “It was the right strategy at the time to launch and grow school choice for Arizona families, but at some point, you need to weed the garden.” The board improved its application process, a step that led to stronger start-ups and fewer closures. It also created a school-evaluation framework using guidance from the National Association of Charter School Authorizers, and used the framework to give greater latitude to well-performing schools and create improvement plans for lesser performers. The documentation that was created along the way “provided stronger evidence to close schools if it was necessary,” Rowe said.

Ohio focused on accountability and transparency. In 2015, the state enacted a law that required authorizers to be evaluated and certified by the state, made it easier to close failing charter schools, and prevented closed charter schools from gaming the system by transferring to another authorizer. Another much-needed weeding process followed. In the three years that followed, the number of charter schools in the state declined 14 percent, from 373 schools to 322. A 2020 study by Stéphane Lavertu subsequently found that students in grades 4 to 8 in Ohio’s brick-and-mortar charter schools made significant gains on state math and English Language Arts exams when compared to district students of similar backgrounds. Consistent with prior research, Black students made particularly strong progress. With accountability measures in place, the state has more recently turned its attention to supporting the replication of high-performing charter schools, allocating up to $1,750 per pupil for the creation of schools serving high-poverty communities.

By 2017, the goal of the One Million Lives Campaign was achieved, with a total of 1,080 failing schools shut down.

Smart Replication

The charter sector’s willingness to shutter poorly performing schools is matched by its commitment to replicating schools that excel, best illustrated by the work of the Charter School Growth Fund. The fund has invested more than $420 million in about 250 charter-school networks since 2010. Those cash infusions have helped open more than 625 new schools, and the charter-network segment of the sector has grown to serve 517,000 students in 2020–21 from about 140,000 in 2010.

Kevin Hall, chief executive officer of the growth fund, noted that “it wasn’t clear in the 2005–2010 timeframe that this idea would work at all. If you looked at school districts, you wouldn’t say that growing makes sense. There was not very much evidence.”

Since then, the number of CMO schools, the number of students served, and the quality of those schools have all increased. While freestanding charter schools still comprise the majority of charters and serve the most students, the proportion of charter schools that are part of a CMO nearly tripled (to 29 percent from 11 percent) between 2007 and 2019, and the proportion of charter-school students they enroll has more than tripled (to 30 percent from 9 percent), according to data from the National Alliance for Public Charter Schools (see Figure 3).

Rapid growth in the CMO segment has contributed to accelerated improvement in the charter sector overall, because CMO schools, on average, are delivering strong results. In 2017, CREDO studied academic performance by school-management type and concluded that “on the whole, . . . attending a charter school that is part of a larger network of schools is associated with improved educational outcomes for students” and that “research work has shown steady and consistent, even if gradual, improvement in charter school network performance.” CREDO also noted that nonprofit operators notched “significantly higher student academic gains” than did for-profits.

In reading, students attending a freestanding charter school were found to experience the equivalent of an additional 6 days of growth per school year, relative to traditional public schools, while students in CMO schools (nonprofit and for-profit), experienced an additional 17 days of growth.

Yet, 15 years ago, it was not at all clear that expanding and replicating charter schools would lead to high-quality outcomes. “In K–12 as a whole, scale does not necessarily translate into being better,” Kevin Hall said.

So why did it happen? The answer lies in “smart replication.”

“We sort of obsess on School One,” said Hall, “and then, ‘Is School Two as good or better than School One?’” What’s more, copying successful methods and approaches is not enough on its own. School operators “have to know why,” he said. “Why are they getting good results? What are they doing? Then there is a virtuous cycle. Can they attract talent, build their own talent? Do they codify what they’re doing so they can get better? All of those things happen in our
best performers.”

Hall and Ebony Lee, a partner at the Charter School Growth Fund, emphasize four key factors for charter-school success: talent, high expectations for students and the school team, high levels of support, and a forward-looking focus on what happens with students after they graduate.

Some large and influential charter authorizers, including the State University of New York, have also supported smart replication. Susie Miller Carello, executive director of the SUNY Charter School Institute, said her organization has tripled the number of schools under its umbrella over the past 10 years, keeping its focus on accountability and devoting time to learning about successful approaches to scaling.

“We went from ‘one good school at a time’ to ‘one good school as a proof point’ and being willing to support the replication of that school,” Carello said. “We talked with venture capitalists about how they determine if there is a good company they want to take on. We talked about the markers of being able to scale. You have a good program; can you convey it with fidelity to the next one or the next three? So now we give multiple charters at a time. We are venture bureaucrats.”

The charter sector’s approach to accountability and replication has had its critics, including some in the charter community itself. Some believe that authorizers have overemphasized standardized tests that define student success too narrowly and inhibit truly innovative educational models. There has been a backlash against “no excuses” models that produce high test scores but often rely on strict disciplinary systems in doing so. Companies that run virtual charter schools doubled down on this argument, maintaining that parent demand, not test scores, is the only valid measure of school quality.

Others have faulted wealthy donors for fixating on growing a relatively small number of charter networks that are disproportionately led by white founders from elite universities and from outside the communities their schools serve. Civic leaders in more than one city focused on recruiting brand-name national networks to their city rather than supporting local educators. Nonprofit organizations, variously referred to as “harbormasters” and “quarterbacks,” were launched with the purpose of saturating the market in cities picked by philanthropists with charter networks that were also selected by philanthropists.

In recent years, after the period studied by Shakeel and Peterson, the charter community has reassessed its approach on these fronts, supporting a broader definition of school quality and investing in new schools that emerge from the communities they serve. The Charter School Growth Fund has been a leader on both fronts. Still, it is worth noting that the fast pace of improvement captured by Shakeel and Peterson predates these changes. Indeed, Baude’s Texas study specifically noted the positive results from schools that focused on test scores: “Our evidence suggests that the increasing share of charter schools adhering to a No Excuses philosophy contributes to observed improvements in the sector.”

While it remains to be seen whether the new, evolved charter sector will deliver the same level of results as the old, “the whole charter premise is working,” Kevin Hall said. “High performers are replicating and, methodically, low performers are closing. It’s not perfect, but over time, this is what is happening.”

* * *

The system has its flaws. Charter performance remains weak in some states, and some schools cream-skim students. Cases of financial malfeasance are still too common, and almost all virtual charter schools have delivered substandard results. Most charter-school advocates recognize these problems and are pushing for improvement, as they have done for years.

In the summer of 2005, the newly established National Alliance for Public Charter Schools, led by its founding president Nelson Smith, convened leading charter advocates from across the country at a conference on Mackinac Island in Michigan. There, the alliance released a task-force statement that read:

If chartering is to thrive, and to play a central role in delivering public education, we must elevate quality to the highest priority. We must look inward at our schools, our authorizers, our state associations, and our own beliefs and habits of mind, so that nothing—nothing—gets in the way of pursuing higher student achievement.

For the next 12 years, the period studied by Shakeel and Peterson, the charter-school community heeded this call. As a result, charter-school performance improved because of choice, flexibility, and the sector’s commitment to taking decisive action based on results. During this time, under a Republican president and a Democrat, in red states and blue, these ideas were the dominant themes in all public education. The district sector often resisted them, the charter sector often embraced them, and charter schools showed the faster improvement.

More recently, though, some of these ideas, such as no-excuses models and the closure of failing schools, have been falling out of favor. Indeed, some former advocates of these concepts have turned their attention to other strategies. Public officials, education advocates, and educators of all stripes would do well to remember the lessons learned from research on charter schools: students receive a better education when we provide families with choices, when schools have the flexibility to implement proven practices, and when our system of public education opens more schools with a track record of strong results while closing those that persistently fail.

Greg Richmond is the superintendent of schools for the Archdiocese of Chicago and the founder and former chief executive of the National Association of Charter School Authorizers.