Charter Schools: What Does the Science Say?

New data provide important insights on charter school student outcomes.

Posted Jun 29, 2020

There are few topics that generate more controversy among education stakeholders—and among politicians, policymakers, and researchers—than charter schools. Broadly, charter schools are publicly funded but administered and run by independent groups outside the normal state and local education authorities.[1] Nationally, this includes a pantheon of nonprofit and for-profit “independent groups,” and even some corporations, such as Kipp and Success Academy, that run charter schools in multiple states. The issue of whether public education funds should be paid to private nonprofit and profit entities has long been—and continues to be—a hotly debated ideological issue.[2]

Without getting drawn into this core and perhaps intractable disagreement, from a scientific perspective, it is perhaps even more important to ask the essential question of whether charter schools perform better, worse, or no different than regular public schools. Although this may seem to be a straightforward question, it is actually very difficult to address in a fair and unbiased manner.

A foundational principle in scientific discovery is the systematic experimental control of various factors in order to test causality.[3] For example, pharmaceutical companies are required to provide scientific studies with random assignment to treatment groups in order for the Federal Drug Administration to approve a new medication. In an experimental design, the new drug is given to patients who qualify using a lottery system, so that confounding factors, such as age, socioeconomic status, disease severity and so on, are "controlled" within the context of random assignment to ultimately determine whether the new drug is better than an older medication—or no drug at all. But there are many situations wherein controlled experimental studies are not possible.

As an example, consider epidemiological studies of smoking. In a tightly controlled experimental study, people could be prospectively assigned to smoking and non-smoking conditions. Of course, this would be highly unethical. Therefore, smoking must be scientifically examined within the context of studying—and comparing—people who choose to smoke to those who do not.

But this approach is “confounded” by the fact that there may be inherent underlying differences in people who decide to smoke. Please permit me to be crystal clear in saying that nonexperimental studies are not in any way “unscientific!” Rather, there are situations where experimental control via random assignment is not possible.

In these cases, questions such as the health consequences of smoking can still be studied scientifically, but it is much more difficult to control for confounding factors. Nonetheless, with regards to lung cancer and coronary disease, it is scientifically warranted to note that smoking dramatically increases the risk of these and other adverse health outcomes. Thus, scientific approaches can address questions—and generate credible answers—even when traditional experimental methods cannot be implemented.

And this is certainly the case with charter schools. A strictly experimental approach would include random assignment of a representative sample of students to charter schools or to regular schools with appropriate testing over time in order to generate reasonable comparisons for the outcomes of these two educational “treatment conditions.” But there are many factors that can impact who does—and who does not—attend charter schools so that conducting fair, unbiased comparisons can be problematic.

Consider as an example the student outcomes of highly selective private universities as compared to lower cost and less restrictive community colleges. If outcomes end up favoring the private institutions, one could credibly argue that this may, in large part, be attributable to the fact that these universities preselect and admit primarily high achieving students as compared to community colleges and that high tuition makes private college attendance biased towards wealthier families. 

A similar argument has been made regarding charter school outcomes. Some suggest that these schools use selective admission to siphon talented students and scarce education dollars away from regular schools and thereby bias outcome measures.[4] And there are many other factors that could account for better outcomes in charter schools if those indeed are the results. But can factors such as selection bias be controlled in order to get a "real" answer to the question of charter school outcomes relative to regular schools? More broadly, can "nonexperimental" scientific approaches be applied to this question?

For example, one possibility could be to compare outcomes using a “natural experiment” from charter schools that are in the same building alongside comparable regular schools, include students from the same district with similar demographic characteristics, with similar (or less) costs per student and to directly compare performance between identical grades.

A new book by Dr. Thomas Sowell at the Hoover Institution at Stanford University (Charter Schools and Their Enemies, Basic Books) provides multiple databased perspectives on charter school performance. In Charter Schools and Their Enemies, Professor Sowell describes outcomes from precisely these kind of “natural” experiments. The results are both compelling and credible.

As one representative example selected from many in the book, the 2017-2018 results from “Uncommon Schools Charter Schools” in Brooklyn, New York, are illuminating. These charter school students are housed in the same buildings at the same grade levels as the regular school students. Further, Dr. Sowell reported that “more than 90 percent of the students in both kinds of schools were either black or Hispanic,” and that a majority in both groups were classified as economically disadvantaged.[5]

The outcome is a bit stunning. In the Uncommon Schools Charter Schools condition, a majority of students received “proficient” or above scores in English for 15 of the 22 grade levels and “proficient” or above in 13 of the 18 grades in Mathematics. The results for traditional school condition—students in the same buildings and from similar demographic backgrounds—indicated that none of the 25 grades scored proficient in English, and in none of the grades did the majority of students score proficient or higher in Math. Dr. Sowell faithfully reported that in one school building, the regular education students did outperform the Uncommon Schools Charter Schools students in fifth grade for both Math and English. But, with that one exception, the charter school students clearly performer far better than those in regular classrooms, even when key confounds were taken into consideration.

The results from Uncommon Schools Charter Schools are not a lone exception or a cherry-picked “outlier.” Indeed, the book Charter Schools and Their Enemies provides many examples of similar results—some less dramatic—but all compelling nonetheless. Taken together, these data systematically indicate that students in charter schools have better outcomes as compared to highly similar students in regular schools. This appears to be true even when multiple confounding factors are taken into account to the greatest extent possible, short of conducting controlled experiments wherein attendance at a regular or charter school is mandated by random assignment.

The policy debates over how Charter schools are funded and administered will no doubt rage on unabated, and there are strongly held ideological arguments on both sides. But regardless of one’s philosophical stance for—or against—charter schools, these analyses and debates should be informed by and predicated upon the copious and credible data presented in Charter Schools and Their Enemies.


[1] 6/29/2020

[2] 6/29/2020

[3]Laudan L. Science and hypothesis: Historical essays on scientific methodology. Taylor & Francis; 1981.


[5]T. Sowell (2020). Charter schools and their enemies. Basic Books. pp. 33-41.