Schools and policymakers are mandating new anti-bias training for teachers in an attempt to improve racial attitudes. Decades of research have shown that teachers often give racially biased evaluations of student work and that biased evaluations can affect students’ future learning and course-taking decisions. However, less is known about what school leaders can do to correct this problem. Research does not show current forms of anti-bias training to be especially promising in changing behavior.

There is, though, a relatively straightforward, if often overlooked, way to diminish the impact of teachers’ racial biases in student evaluation: standardizing grading rubrics. To gauge the potential impact of a standardized rubric on grading bias, I conducted an experiment comparing how teachers graded two identical second-grade writing samples: one presented as the work of a Black student, and one as the work of a white student.

My experiment found that teachers gave the white student better marks across the board—with one exception. When teachers used a grading rubric with specific criteria, racial bias all but disappeared. When teachers evaluated student writing using a general grade-level scale, they were 4.7 percentage points more likely to consider the white child’s writing at or above grade level compared to the identical writing from a Black child. However, when teachers used a grading rubric with specific criteria, the grades were essentially the same.

The experiment also included a series of questions asking teachers about their background and their racial attitudes. In exploratory analyses examining bias by teachers’ own race, gender, and the racial makeup of the schools where they teach, I found larger bias in grading by white and female teachers, who were less likely to rate the Black child’s writing as being on grade level compared to the white child’s writing. However, I didn’t find any connection between my measures of teachers’ implicit and explicit racial attitudes and the differences in grading the Black and white student writing samples.

This experiment suggests that racial stereotypes can influence the scores teachers assign to student work. But stereotypes seem to have less influence on teachers’ evaluations when specific grading criteria are established in advance. New instructional practices and tools, such as standards-based grading rubrics and mastery-based grading with specific criteria, present potentially effective approaches to promoting racial equity in schools. Limiting opportunities for biased decisions may have more immediate impact on equitable student evaluation than current forms of anti-bias training.

Building a Grading-Bias Experiment

My experiment took the form of a web-based survey, including demographic questions, a two-part grading task, and a test to measure racial attitudes. I contracted with a private survey provider to recruit a multi-state sample of U.S. schoolteachers. Some 1,799 unique users responded to a survey invitation. Of those, 1,549 teachers working in preschool through 12th grade completed the main survey tasks and were compensated directly by the company for participating. Their responses form the basis of my analysis.

At the start of the survey, teachers were informed that the researcher was interested in learning how educators evaluate student writing. As a subject area, writing is well suited for a study of grading bias for two main reasons. Substantively, the subject area is of interest given that tools for evaluating student writing vary in their focus and specificity. Methodologically, the personal narrative lends itself well to signaling the author’s racial identity in a relatively subtle way. Overt statements of a student’s race in a grading experiment could arouse suspicion among research participants and affect their responses.

Respondents also answered questions about their gender, race, and number of years in the field, as well as the grade that they teach and the racial composition of their school. Overall, 69 percent were white and 54 percent taught in a predominantly white school. By comparison, about 79 percent of U.S. teachers are white and approximately 45 percent of all U.S. teachers work in schools that are less than 50 percent white, according to federal data from 2017.

Two Grading Tasks

Teachers were randomly assigned to receive one of two versions of a writing sample. The writing sample was presented in a child’s handwriting and was purportedly by a male student in the fall of second grade. It took the form of a brief personal narrative in response to a prompt to write about his weekend and mentioned spending time with his brother and a friend. The versions were identical in all but one aspect: each used different names for the brother to signal either a Black or a white student author (see Figure 1). The name choices came from a list of the most racially distinct names reported by Steven Levitt and Stephen J. Dubner in Freakonomics. In one version, the student author refers to his brother as “Dashawn,” signaling a Black author; in the other, his brother is called “Connor,” signaling a white author.

Teachers were first asked to rate the writing sample on a relative grade-level scale with seven options, from weak to strong performance: far below grade level, below grade level, and slightly below grade level; at grade level; or slightly above grade level, above grade level, and far above grade level. Performance criteria were not explicitly defined.

Then, they were asked to rate the writing sample again. This time, teachers were given a rubric with more clearly defined performance criteria for a personal narrative. The rubric included four possible ratings, from weak to strong: fails to recount an event, attempts to recount an event, recounts an event with some detail, or provides a well-elaborated recount of an event. The rubric appeared after the grade-level scale, on a separate screen and without the option to return to the earlier screen. This was designed to ensure that teachers’ ratings on the grade-level scale were not influenced by the rubric’s criteria.

Substantively, these evaluation measures differ in two important respects. The grade-level scale is general in the sense that it does not specify what dimensions the rater should consider, such as grammar, spelling, creativity, or organization. It also does not clearly specify the gradations among the seven possible ratings, or how a teacher should determine whether the writing is “slightly above grade level” versus “above grade level.” In contrast, the rubric specifies which domain teachers should evaluate—in this case, how well the writer recounts an event—and provides more specific descriptors to guide teachers in their rating choices along a four-point scale.

Assessing Racial Attitudes

After the grading exercise, the survey then attempted to collect data on respondents’ racial attitudes. This presented a challenge. On one hand, if questions or activities designed to reveal racial attitudes were administered before teachers see the writing sample, the act of completing the racial attitude measures could influence their grading. In particular, the experiment could produce “demand effects,” in which teachers adjust their ratings to match what they view as a socially desirable response, such as being particularly careful to show no racial bias. However, if respondents complete the racial attitude measures after viewing the writing sample, the writing sample may influence their racial attitude scores. I opted for the second option, viewing this as less damaging to the experiment overall.

To measure teachers’ implicit stereotypes of white and Black students, I adapted a traditional implicit association test to assess respondents’ associations between race and competence. These computer-based tests ask respondents to react quickly to ideas and images by assigning them to one of two categories, such as “good” and “bad.” In my test, participants identified photos of students’ faces as either “African American” or “European American” by pressing a right- or left-hand key on a computer keyboard. Then, those same keys were also used to assign words like intelligent, confident, disorganized, and unskilled to one of two categories: either “Competent” or “Incompetent.” The test combines the racial categories and competency categories in various combinations, with the right-hand key used for “European American” and “Incompetent” on one round, and for “African American” and “Incompetent” on the next round, for example.

Implicit association test scores are calculated by comparing response speeds by category. In this case, the relevant question is how long it takes a respondent to assign descriptive “Competence” or “Incompetence” words when those categories use the same keyboard stroke as “African American” or “European American.” An implicit association result would indicate a preference for African Americans over European Americans, for example, if a participant’s responses were faster when “African American” and “Competence” were assigned to the same key. In my study, just 675 teachers completed the full test and produced valid scores that are included in my analysis. The other half either abandoned the survey or responded to test items so quickly that valid scores could not be calculated.

Finally, teachers were asked to respond to traditional “feeling thermometer” questions in which they rated their feelings toward African Americans and European Americans. A 1-10 scale was shown with 1 representing “very cold” and 10 representing “very warm.” I created a measure of explicit bias by calculating the difference in the warmth of each individual teacher’s feelings about white and Black Americans. A positive score indicates a preference for white Americans and a negative score indicates a preference for Black Americans. Some 1,549 teachers completed the feeling thermometer questions.

Results by Grading Tool

To compare teachers’ grades using both grading tools, I sort grades into two groups based on whether their rating is above or below a cut-off point on each scale. On the vague grade-level scale, I consider how many teachers rate each sample as “at grade level” or above. On the rubric, I look at how many teachers rate the sample as “recounts an event with some detail” or better. However, results were robust to a variety of other analytic choices.

Teachers shown the “Dashawn” version of the writing sample are 4.7 percentage points less likely to rate it as being on grade-level or above compared to teachers shown the “Connor” version (see Figure 2). Some 35 percent of respondents rate the version written by a white student at grade-level or above compared with about 30 percent for the version written by a Black student. However, when those same teachers use a rubric with specific grading criteria, they give essentially identical ratings to the Black and white authors—about 37 percent rate both the “Dashawn” and “Connor” versions as “recounts an event with some detail” or better.

In exploratory analyses, I also investigate differences in grading based on the gender and race of the teacher. Prior research has found that teachers show preference for students with identities similar to their own. In particular, white teachers tend to have lower expectations for Black students than for similar white students (see “The Power of Teacher Expectations,” research, Winter 2018). I see evidence of this in my experiment as well, though only when teachers apply the vague grade-level scale. In all groups, when teachers use a specific grading rubric, estimates of bias are small and not significant.

Female teachers assessing a young boy’s writing sample show racial bias in their grading, but male teachers do not. Whereas females are 7 percentage points less likely to rate the “Dashawn” sample as being on grade level than the “Connor” sample, the difference for male teachers is small and not statistically significant (see Figure 3).

In looking at teachers by race, I find white teachers are approximately 8 percentage points less likely to rate the Black student’s writing as being at grade-level or above compared to the white student’s writing. By contrast, teachers of color do not show evidence of evaluation bias.

I also estimate grading bias among teachers who work in more and less racially diverse schools. As in my other analyses, bias is evident when teachers applied vague grade-level standards but not when they used specific criteria on a rubric. Teachers working in schools where no one particular race or ethnic group makes up a clear majority of enrolled students show the most bias in applying the grade-level scale. They are 13 percentage points less likely to rate the writing sample as on or above grade level if it was written by a Black student. There are no significant differences in the ratings assigned by teachers working in predominantly Black, Latinx, or white schools.

Some of these results raise important questions about student-teacher race and gender matching. Given that my teacher subgroup analyses are exploratory, and that this sample is not nationally representative (though it is national in scope), we cannot know whether these findings reflect patterns in the broader population. But they should inspire new hypotheses for further research. Female teachers showed racial bias in grading a Black male student compared to a white male student, whereas male teachers did not. Is a teacher less likely to exhibit racial bias against a student if the student shares the teacher’s gender?

Results by Racial Attitudes

I also look at the relationships between teachers’ grading bias and racial attitudes as measured by the implicit association test and explicit “warmth” questions.

Both tests showed attitudes that favor whites. The implicit attitudes test found that teachers had a significant association of white students as being more competent than Black students, by 41 percent of a standard deviation. The explicit “thermometer” questions measure showed a small and not significant preference for European Americans compared to Black Americans.

It stands to reason that teachers with higher measured levels of bias could show more bias on the grade-level evaluation measure. However, I find no relationship between teachers’ measured attitudes and levels of grading bias, either on the vague grade-level scale or the specific rubric. In no case does the magnitude of the bias differ significantly by teachers’ implicit or explicit racial attitudes.

There are several possible explanations for this. First, tests of implicit bias have limitations, and the implicit association test’s validity as a test of individual attitudes has been questioned. Second, it may be that my sample size was too small to detect the true influence of implicit attitudes on grading bias.

Indeed, this study did find some divergence in explicit and implicit attitudes. While the implicit attitudes test showed, on average, that teachers had a significant implicit association of white students as being more competent than Black students, the explicit measure showed a much smaller, and not significant, preference for whites. However, if in fact teachers were summoning explicit attitudes to override an initial implicit instinct to rate the “Dashawn” writing prompt lower on the grade-level scale, the experimentally observed grading bias suggests they were not entirely successful. Teachers may have been able to summon their explicit attitudes to dampen, but not entirely eliminate, the influence of implicit attitudes on their grading.

Implications

Scholars and decisionmakers have focused on two distinct approaches for mitigating the effects of negative implicit racial attitudes: training programs that aim to reduce people’s general implicit associations and efforts that engineer circumstances to reduce the impact that implicit stereotypes can have on a person’s behaviors or judgments. My study shows strong potential from the latter approach when it comes to teacher grading. When teachers use a rubric that orients grading decisions to a limited number of specific, demonstrable criteria, they show no bias in their grading decisions. When teachers are asked to rate student work along a vaguer spectrum of performance, based on meeting “grade-level” standards, their grading favored the white student.

These findings raise a number of questions for future research. There may be something unique about the elements of this experiment that contribute to my results, whether the rubric, the writing sample, or the combination of the two. I can’t rule out the possibility that the different sizes of each scale—seven points on the grade-level scale and four on the rubric—affected the amount of bias detected. We also should consider whether the impact of bias on grading could differ depending on the academic subject or the nature of the work being evaluated. The evaluation of student writing is likely more subjective than determining whether a student arrived at the correct answer to a math problem, for example. Additional experiments with other rating scales or other kinds of student work would be helpful.

The generalizability of these findings to the classroom is unknown. How might grading bias differ when teachers are grading their own students? Teachers’ bias regarding students that they personally know may differ from the bias we find in this experimental setting, though past research has found some evidence of teachers’ racial and gender-based bias being directed toward their own, familiar students. There may therefore be reason to recommend grading rubrics as a means to mitigate bias.

The present study does not offer direct evidence on whether rubrics would produce bias-reducing effects in school classrooms. It is possible that teachers hold strong student-specific biases that rubrics are less effective at overcoming. The findings in this study may be more generalizable to settings where raters are conducting anonymous reviews of essays in which the authors’ identities may be signaled through context clues, such as state writing exams or SAT and GRE scoring.

I also note that, in this study, teachers were presented with a rubric without any training or examples on how it is appropriately applied. While using the rubric reduced bias, the grading task was simple and did not have any time constraints, unlike the complex evaluations of student work that teachers make as part of their day-to-day jobs. Previous research has suggested that rubrics do not improve grading reliability unless teachers are trained in how to use them. And in general, any efforts to reform or standardize teachers’ classroom practices are less likely to succeed in the absence of aligned coaching and training.

Finally, policies that establish predetermined and clearly defined grading criteria may prove powerful in light of another finding from this study: that the overall bias was driven by white teachers and this bias seems to have been driven by an in-group preference among white teachers for white students. This finding aligns with calls to diversify the teaching force. Nationwide, 79 percent of teachers are white compared to 48 percent of students. Recent research has shown that although the share of teachers of color has grown in recent years, this growth has not kept pace with the increase in the share of students of color, which suggests an ongoing disadvantage for students of color. Insofar as this imbalance powers biased evaluations of student work, it may lead to a vicious cycle in which initial racially biased evaluations from a teacher cause lower future performance from students, which reinforces stereotypes held by teachers, which in turn leads to future bias in evaluations.

There are a great many ways in which racism, past and present, affects the educational opportunities of Black students. One proximate cause of inequality that school and district leaders have an opportunity to address is teachers’ racially biased evaluations of students. A relatively simple tool may help start to mitigate the effects of teachers’ racial biases on students.

David Quinn is assistant professor of education at the University of Southern California Rossier School of Education.

Imagine your ideal coworker or friend. She communicates well and is a good teammate. She’s in touch with her emotions but stays calm under pressure. She’s not a quitter. You’d probably describe her as hardworking, understanding, and flexible—the sort of person who helps solve big problems.

Research in economics, psychology, and sociology has found that, compared to people who are otherwise similar, those who demonstrate these sorts of mindsets and skills tend to have better outcomes in school and in life. Studies also show that contextual factors influence the degree to which people demonstrate these mindsets and skills. Supporting social-emotional development, such as by fostering experiences of belonging and promoting sharing and productive communication, has long been part of preschool and elementary school programs. Now, high schools increasingly are focused on social-emotional development, too.

The most common instruments used to measure social-emotional development are student surveys, in which adolescents report their experiences, behaviors, and attitudes related to school. Can these surveys reveal which high schools best support social-emotional development? And does attending such a school improve students’ long-term outcomes?

We examine results from a detailed annual survey about social-emotional development and school climate administered to students in the Chicago Public Schools. Through value-added analysis, we identify individual high schools’ impacts on 9th-grade students’ social-emotional development and test scores. We then trace the effects of attending a school that excels along each of these dimensions on short-term outcomes, such as absenteeism and school-based arrests, as well as on longer-term outcomes, like high-school graduation and college enrollment. Our focus on 9th grade is intentional, because it is a critical transition year of schooling, when young adolescents are most vulnerable to becoming off-track for high-school graduation due to accumulating an insufficient number of credits.

Our analysis confirms that some schools are better at supporting students’ social-emotional development than others. But these effects are not all the same. School effects cluster in two domains, social well-being and work habits, and some schools are better at one than at the other. Schools that promote social well-being have larger effects on students’ attendance and behavioral infractions, while those that improve work habits have larger effects on academic performance.

We also calculate each school’s value-added to student test scores and then look to see how well these measures predict student success. Compared to test-score value-added, social-emotional value-added is far more predictive of the behaviors that support student success, such as having fewer absences and being on-track to graduate. And it is more predictive of positive longer-run outcomes as well, such as graduating from high school and enrolling in a four-year college.

These results show that students’ own assessments of their social well-being and work habits provide valuable information about their development. They also show that these surveys can be used alongside traditional indicators like test scores to provide a more complete picture of how schools prepare students for the future. This analysis represents an important early step toward understanding how schools influence the social-emotional development of adolescents, how that can be measured, and how this can be useful for policy.

Surveying Social-Emotional Development

Chicago Public Schools is a large urban school district with 133 public high schools, including neighborhood, charter, vocational, and magnet schools. About 86 percent of students are from families with economic disadvantage. Forty-two percent of students are Black, and 44 percent are Latinx.

Since 2010–11, students in grades 6–12 have participated in an annual survey about their experiences previously known as My Voice, My School and now called the 5Essentials survey. The survey includes 21 questions designed to measure students’ social-emotional development, including their interpersonal skills, level of school connectedness, academic engagement, grit, and study habits. Students register their level of agreement on a numerical scale with statements like, “I’m good at working with other students,” “I don’t give up easily,” and “People here notice when I’m good at something” (see Figure 1). These questions assess students’ beliefs about themselves and their environments, both of which can influence learning.

The most widely used theoretical framework for social-emotional learning, developed by the Collaborative for Academic, Social, and Emotional Learning, divides habits, mindsets, and skills into five domains. They are self-management, responsible decision-making, social-awareness, self-awareness, and relationship skills. Based on how students respond in our survey data cluster, however, we group questions into just two summary indexes. Questions about interpersonal relationships and school connections are sorted into a “social well-being” index, and questions regarding academic effort, academic engagement, and grit form a “work habits” index.

Our analysis uses administrative, test-score, and survey data for 157,630 students. We look mainly at cohorts of first-time 9th-grade students who attended high school between 2011 and 2017, which includes 55,560 students who are now old enough to have attended college. Some 78 percent of students invited to take the survey responded.

We look at shorter-term and longer-term outcomes for each student. In the short term, outcomes include how many times they were absent, how many disciplinary incidents made them eligible for suspension, and whether they earned at least five full-year course credits and no more than one F for a semester of work in a core course in their first year of high school. This “freshmen on track” measure is used by the district and is a more accurate predictor of graduation than test scores or demographics. In 2018–19, 89 percent of 9th-grade students in Chicago Public Schools were considered “on track.”

Another key outcome is whether a student has ever been arrested for activities conducted on school grounds, during off-campus school activities, or due to a referral by a school official. In Chicago, roughly 20 percent of juvenile arrests in 2010 were school based, so these arrests have important, long-term implications. During our sample period, 4 percent of Chicago public high-school students had a school-based arrest, including 5.3 percent of males and 7.9 percent of Black males.

In looking at longer-term outcomes, we consider high-school completion and college enrollment based on district and National Student Clearinghouse data. Overall, about 79 percent of first-time 9th graders in our sample went on to graduate high school, and about 53 percent enrolled in college within two years of their expected graduation date.

Estimating School Effects

Our analysis involves two key steps. First, we identify which schools add the most value to students’ social-emotional development and test scores. Then, we estimate the effects of attending those schools.

Our value-added model seeks to isolate the causal effects of individual schools on students’ test scores, social well-being, and work habits. We compare the test scores and survey responses of students at each school to similar students elsewhere at the end of 9th grade. To determine which students are similar, we look at a range of data gathered at the end of each student’s 8th-grade year: test scores, survey responses, course grades, discipline incidents, attendance, and demographics. The demographic characteristics we consider include students’ gender and ethnicity, the socioeconomic status of their Census block, and whether they qualify for free- or reduced-price school lunch. We calculate a school’s value-added by determining how much it increases students’ test scores and social-emotional development relative to the observed changes for similar students at other schools.

We then quantify the effects of attending a school with a value-added score in each domain—social well-being, work habits, and test scores—that is one standard deviation higher than the average school in our study. This is roughly equivalent to attending a school in the 85th percentile of performance in that domain rather than the average school.

We estimate these school effects two different ways. First, we simply ask whether students do better when they attend a school with high value-added in each separate domain. We then look at the predicted gains from attending a school that has high value-added in all three domains. This second approach reveals whether knowing a school’s value-added to social-emotional development provides additional information, over what is already evident from value-added to test scores, about how well the school supports student success.

Impacts on Social Well-Being

First, we look at how the high-school students attend affects their self-reported levels of social well-being while in 9th grade. When students attend high schools that demonstrated high value-added to social well-being in other school years, students’ reports of their own social well-being increase by 9 percent of a standard deviation compared to students attending the average school. Those students are more likely to agree that they are noticed when they are good at something, can end arguments among others, and are included in activities. This result provides compelling evidence that schools can, and do, influence students’ self-reported social well-being.

Schools with high value-added to work habits also improve students’ self-reported social well-being. In this case, the increase is 6.2 percent of a standard deviation compared to students in the average school. The effect on social well-being of attending a school with high test-score value-added is also positive but smaller, at 3.8 percent of a standard deviation.

What happens when we consider different aspects of school performance in combination with one another? An increase of one standard deviation across all three dimensions of performance leads to a 9.1 percent increase in students’ reports of their social well-being, almost identical to the gain from an increase in value-added to social well-being alone. In other words, the measure of value-added to social well-being captures virtually all of the detectable variation in school impacts on self-reported social well-being. Compared to test-score value-added, the inclusion of the two measures of social-emotional development more than doubles our ability to predict a school’s effects on social well-being.

Impacts on Work Habits

We then turn to school impacts on students’ self-reported work habits. When students attend a school with a track record of high work-habits value-added, their own work habits in 9th grade improve by 6.4 percent of a standard deviation. These students are more likely to agree that they try to do their best, study even when a subject doesn’t interest them, and finish what they start.

As with social well-being, we investigate the extent to which other aspects of a school’s performance predict positive effects on students’ persistence and hard work. A school with high value-added to social well-being has effects of 6 percent of a standard deviation, while a school with higher test-score value-added has an impact of 3.3 percent of a standard deviation.

In looking at these aspects of school performance in combination, we find that an increase of one standard deviation across all three dimensions increases work habits by 6.7 percent of a standard deviation. A school’s track record in developing students’ work habits is the best predictor of its success in boosting effort and grit among current students.

Impacts on Test Scores

How do these novel measures of school performance predict effects on student test scores? Intriguingly, the two measures of value-added to social-emotional development are nearly as good predictors of impacts on test scores as a school’s test-score value-added. Attending a school with strong test-score value-added increases 9th-grade test scores by 6.8 percent of a standard deviation (see Figure 2). Both social-emotional value-added measures have similar effects on test scores when considered on their own: 6 percent of a standard deviation for value added to social well-being and 5.7 percent for work habits value-added.

Considering these value-added estimates in combination reveals a remarkable result: measures of social-emotional value-added substantially improve our ability to predict a school’s impacts on test scores. Relative to using the test-score value-added measure alone, adding the two social-emotional value-added measures increases the share of the variation in students’ test scores that we can explain based on the school they attend by 47 percent. This stands in stark contrast to the pattern for social well-being and work habits, for which the vast majority of a school’s effect is captured by the estimates of the school’s value-added within those domains. In other words, schools that raise test scores don’t necessarily focus on academic achievement alone; fostering social-emotional development may be foundational for academic success.

This alone is revealing. But from a policy perspective, the key question is whether a school’s effectiveness in supporting students’ social-emotional development has implications for their success over the long haul. We explore this below.

Long-Term Impacts

At schools with higher value-added to social-emotional development, students are more likely to go on to graduate high school and enroll in a four-year college (see Figure 3). Both the social well-being and work habits value-added measures are stronger predictors of long-term school impacts than test score value-added.

An increase of one standard deviation in the test-score value-added of the high-school students attend in 9th grade increases their likelihood of graduating from high school by about 1.2 percentage points. The impact is larger for increases in the school’s value-added to social-emotional development, at 1.6 percentage points for both social well-being and work habits value-added. Attending a school that has high value-added in all three domains increases the likelihood of graduating by 1.9 percentage points, a substantial jump compared to the gains from attending a high school with high test-score value-added alone.

We then turn to students’ college-going and look at whether students enroll in any college within two years of their expected high-school graduation date. We see the same pattern: social-emotional value-added measures are more predictive than test-score value-added for college attendance, as well. An increase of one standard deviation in test-score value-added increases college-going by 1.7 percentage points. For value-added to social well-being, the impact is 1.7 percentage points, and for work-habits value-added, it’s 2 percentage points. In considering all three value-added measures together, the predicted gain in student college-going is 2.3 percentage points. As with high-school completion, social-emotional value-added estimates predict more of the differences across schools in college-going rates than do estimates of value-added to test scores.

To delve deeper into the college results, we explore impacts on enrollment in both two-year and four-year schools. We find no effects of our value-added measures on enrollment in two-year schools, but we find large effects on four-year college-going. An increase of one standard deviation in test-score value-added boosts four-year college-going by 2.3 percentage points. The increase for value-added to social well-being is larger at 2.9 percentage points, and that for work-habits value-added is largest at 3.2 percentage points. Looking at all three value-added measures in combination, the increase in four-year college-going is 3.6 percentage points.

Potential Mechanisms

When more students report strong connections to school, healthy relationships, and the habits that support hard work, how does that play out in terms of their behavior? Does their high-school experience change in ways that could explain the positive longer-term outcomes we’ve just documented? We estimate the effects of high schools’ social-emotional value-added on three key metrics of student success in 9th grade: being on-track to graduate in four years, attendance, and the number of disciplinary incidents (see Figure 4).

Schools with high social-emotional value-added have larger impacts on students’ on-track status than schools with high test-score value-added. A school that is one standard deviation higher in test-score value-added improves on-track rates by 1.9 percentage points. The gains from attending a school with high value-added to social well-being or work habits are 1.9 and 2.1 percentage points, respectively. A school that has high value-added across all three dimensions improves on-track rates by as much as 2.5 percentage points. This indicates that much of what high schools do to keep students on track to graduate is not captured by impacts on standardized tests. And it tells us that school impacts on self-reported survey measures capture more of a school’s impact on staying on track than impacts on test scores.

School attendance in 9th grade is another important predictor of high-school graduation. All three value-added measures predict absence rates, but value-added to social well-being, which reflects the strength of students’ connections to school and relationships with their peers, tells us more about how often a student will miss school than the other two. An increase of one standard deviation in a school’s test-score value-added reduces absences by 0.9 days in 9th grade. By comparison, an increase of one standard deviation in value-added to social well-being reduces absences by 1.3 days—roughly an 8.6 percent reduction compared to the average numbers of days a student is absent. This larger impact of value-added to social well-being is consistent with earlier findings that students who feel a greater sense of belonging are more likely to attend school.

Finally, we examine impacts on the number of disciplinary incidents in 9th grade and school-based arrests throughout high school. Both value-added to social well-being and test-score value-added predict fewer incidents, while work-habits value-added does not. An increase of one standard deviation in value-added to social well-being reduces the number of incidents by about 1 percentage point compared to 0.8 percentage points for test score value-added. Social-emotional value-added also has a greater effect and more predictive power than test scores when looking at school-based arrests. High value-added to social well-being reduces the likelihood of an arrest by 0.7 percentage points and high work-hard value-added leads to a reduction of 0.8 percentage points. Increasing test-score value-added reduces the likelihood of an arrest by 0.6 percentage points. Using all three value-added measures, the effect of attending a school that is stronger along all three dimensions is 0.9 percentage points. This represents a decrease of about 21 percent compared to the probability of a school-based arrest across all students.

Expanding the Definition of a “Good” School

The high-school years are formative, and attending a high-performing high school can build a foundation for success in adulthood. But what do we mean by high-performing? What are our metrics for success? Our study provides fresh answers to these enduring questions.

We find that some high schools are better than others at helping students develop healthy social lives, community connections, and the skills and habits that promote hard work and grit. We also find that students who attend such a school are more likely to experience positive outcomes in school and after graduation, from being more likely to attend a four-year college to having less interaction with the criminal-justice system. We focus our analysis on students in 9th grade. This transition year is an important window of opportunity to establish strong ties to school.

This is the first broad effort to validate measures of school impacts on social-emotional development that are based on self-report surveys, and our evidence shows that these estimates are arguably causal. Our finding that school impacts on social-emotional growth have larger effects on short- and long-term outcomes than schools’ impacts on test scores has important implications for how policymakers measure school quality.

We also show the potential for surveys to identify high-performing schools using more diverse indicators of success, at least in a low-stakes environment. Surveys reveal that schools that raise test scores are not always those that improve students’ social-emotional development, and vice versa. These results suggest that school quality is multidimensional and show that value-added estimates of impacts on social-emotional growth predict impacts on longer-term outcomes that are not captured by measures of a school’s value-added to test scores.

We conclude with a central question: which school practices improve social-emotional development? While much work remains to be done, our analysis represents an important early step toward a fuller picture of how schools influence student success.

C. Kirabo Jackson is the Abraham Harris Professor of Education and Social Policy at Northwestern University, where Sebastián Kiguel is a PhD student. Shanette C. Porter is director of research and a senior fellow at the Mindset Scholars Network. John Q. Easton is a senior fellow at the UChicago Consortium on School Research, where Alyssa Blanchard is a research analyst. This article is adapted from a study titled “School Effects on Socio-emotional Development, School-Based Arrests, and Educational Attainment,” forthcoming from American Economic Review: Insights.

During the 2020 presidential primaries, Democratic candidates proposed forgiving $640 billion in outstanding federal student-loan debt. Proponents of loan forgiveness argue that this debt places overwhelming burdens on today’s young graduates—and college dropouts—and that lightening their load would help both the borrowers and the economy overall. Yet these proposals have raised questions about who would benefit most and whether student-debt relief is efficient. Is loan forgiveness for all a good idea?

Beth Akers is a senior fellow at the Manhattan Institute for Policy Research, a center-right think tank. Sandy Baum is a nonresident senior fellow for the Center on Education Data and Policy at the Urban Institute, which leans toward the center-left. In this forum, both authors argue against universal loan forgiveness.

In 2011, when the Occupy Wall Street movement called the nation’s attention to the wealth-and-income gaps between the top 1 percent of the population and everyone else, activists began to promote the idea of forgiving student-loan debt. Those in the Occupy Student Debt campaign argued that all current education debt should be eliminated immediately. They asserted that policies such as limiting loan payments to an affordable share of income were “micro-cosmetic,” and that creditors needed to free debtors from their “bondage.”

At the time, only a small minority of people subscribed to the idea, but recently it has gone mainstream, with Democratic presidential candidates Elizabeth Warren and Bernie Sanders proposing broad student-debt forgiveness policies. To help families cope with financial pressures during the Covid-19 crisis, the Democratic Party platform calls for up to $10,000 in student-debt relief per borrower. Longer-term provisions in the platform include forgiving all debt on undergraduate tuition loans for those who earn under $125,000 and who attended public institutions. That benefit would also apply to those who hold tuition debt from attending historically Black private colleges and universities.

Democrats included a student-debt relief provision in their proposals for the Covid-19 rescue package. Ultimately, the Coronavirus Aid, Relief, and Economic Security Act of March 2020 suspended loan payments and waived interest for six months but did not include debt forgiveness. The payment waiver now extends to the end of the year.

Proponents of large-scale erasure of education debt characterize the idea as progressive, in part because such a policy, which would benefit relatively affluent people, might be financed (as Bernie Sanders proposed) by people who are even better off. Truly progressive policies, though, provide disproportionate benefits to households in the lower reaches of the income distribution. They are designed to diminish the gaps between the haves and the have-nots.

The realities of student debt in our country make it clear that proposals to eliminate these obligations do not meet the criteria for progressive policies. Households in the upper half of the income distribution hold more student debt than those in the lower half. The highest-income quartile of households owes about one-third of that debt; the lowest-income quartile owes about 12 percent. People who don’t go to college don’t have student debt. They have lower incomes and more constrained job opportunities than others.

There are some people who borrowed and either didn’t complete their programs or never saw the anticipated earnings payoffs to the credentials they did earn. These individuals make up a large share of the low-income adults who do hold student debt. The circumstances of these borrowers explain why the government has developed an income-driven repayment system for federal student loans. The system is far from perfect, but it does not require payments until a borrower’s income exceeds 150 percent of the poverty level and then generally requires payments equal to 10 percent of the borrower’s income beyond that level. Those whose incomes never support affordable repayment of their debts will see their remaining balances forgiven after 20 years (or 10 years for those with public-service jobs and 25 years for those with graduate school debt).

Just 7 percent of borrowers owe more than $100,000 in student loans. This small share of borrowers owes more than one-third of the outstanding balances. Doctors and lawyers and MBAs have lots of debt, but they also tend to have high incomes. About 40 percent of federal student loans go to graduate students each year. There are strict limits on how much undergraduate students can borrow from the federal government—$31,000 total for those who are dependent on their parents and $57,500 for those who are older, married, or otherwise independent of their parents. Graduate students, though, can borrow virtually unlimited amounts.

More than one-third of borrowers owe less than $10,000. They hold just 5 percent of the outstanding student debt. Many of them are the borrowers who struggle most to pay back their loans because their limited skills restrict their job opportunities.

In short, forgiving all student debt would deliver a big windfall to a few people: those who can afford to pay. Virtually all of those with the largest debts have bachelor’s degrees, and most have advanced degrees. That is not a progressive policy.

The CARES Act provided for one-time relief payments of up to $1,200 to individuals making no more than $99,000 annually. The idea of sending checks to everyone did not survive—there is an income limit. Maybe there should not be an income limit. Maybe the checks should be much bigger. But would anyone explicitly propose sending checks only to those who went to college? This would be shocking even absent the reality that highly educated workers are more likely than others to be able to work remotely. Many of the restaurant workers, taxi drivers, retail clerks, and maintenance staff who have lost their incomes did not go to college and do not have student loans. If they do have loans, they may well not have been required to make payments even before the implementation of the waiver and might eventually have their debts forgiven under existing policies.

The call to relieve each borrower of up to $10,000 in debt would be akin to sending a check in that amount only to those with outstanding student loans. Quite a few people in addition to those who never went to college would be left out under such a policy: Borrowers who have just finished repaying their loans, for instance, and students who worked long hours to avoid borrowing. Imagine college classmates from similar families who borrowed similar amounts. Student A decided to work hard to pay off all his debt before following his dream to try to make it as a musician. Student B decided to travel around the world and postpone paying her loans. Now, under loan forgiveness, the taxpayers will repay Student B’s loans, but Student A, who paid back every dime on his own, will receive no such benefit.

What about borrowers who put their student-loan payments on their credit cards to avoid default? They’d be out of luck. What about those Americans who have debt from medical procedures? From utility bills? From payday loans? Or fines that accumulate when debts go unpaid?

Aside from all of these inequities, one-time elimination of student debt makes little sense if future students will continue borrowing similar amounts. Some students might even feel encouraged to borrow more in the hope that those debts, too, will be forgiven. Many advocates hope that college will become tuition free, solving this problem. But the reality is that “free” college will not eliminate borrowing for college. Public colleges are already essentially tuition free for a large share of low-income students, because Pell Grants and state grants cover those charges—but many of those students still borrow to cover living expenses. In fact, students who pay no tuition graduate with almost as much debt as those who do pay tuition.

We should forgive some student debt, such as that carried by students who borrowed for education that did not pay off or who were defrauded by their schools. We already have separate policies to deal with those issues—policies that should be simplified, improved, and carried out.

Universal forgiveness would benefit many students from relatively affluent families who attended expensive private colleges. It would also be a gift to those who borrowed for graduate school. The Congressional Budget Office recently examined the potential cost of the existing income-driven repayment plans designed to protect borrowers from unaffordable debt payments. The study found that 20 percent of those in repayment are graduate borrowers. These borrowers owe half of the funds that are now in repayment. So, half of the benefit of forgiving that debt would go to people who went to graduate school.

Wiping out the student debt of borrowers who took these loans to invest in themselves and who are reaping the benefits of their education is not a progressive policy. Most of these individuals will have increased earnings potential and a wide range of opportunities throughout their lives that would not otherwise have been available to them. The federal government is right to provide the loans that create these opportunities. Eliminating the federal student-loan program or restricting its ability to serve students who have not yet proven themselves would erode opportunities for upward mobility. The government should continue to offer student loans while ensuring that students can’t use those loans at very poorly performing institutions and that borrowers don’t have to make payments that would deprive them of the ability to meet basic needs.

The economic crisis wrought by the pandemic has highlighted the sad reality that too many Americans were living on the edge even before the virus hit. Some of the people now facing the most serious struggles do have student debt, and they need a lot of support—not only so they can keep up with their education debt but also, more urgently, so they can pay rent, have enough to eat, and provide for their children. The majority of student debt, however, is owed by people who are in better circumstances than most Americans.

Student-debt relief should be a targeted policy that is part of a truly progressive agenda—not a special-interest subsidy that disproportionately helps a segment of the relatively privileged.

The popular-media coverage of student loans would have you believe that a generation of young workers is being crushed by unaffordable student-loan debt they can’t escape. Indeed, when presidential candidate Senator Elizabeth Warren last year put forth a proposal to cancel $640 billion dollars in education debt, the plan met with popular approval. According to a May 2019 poll by Politico/Morning Consult, 56 percent of voters supported Warren’s proposal, while just 27 percent opposed it.

Popular support notwithstanding, widespread student-loan forgiveness is a bad idea—not because such a program would be too costly or because it would undermine the social compact for individual responsibility. Rather, it’s a bad idea because the problem it’s designed to address isn’t the one that needs fixing.

The picture of the issue painted by the media is distorted. Why? First of all, the typical student borrower has but a modest monthly payment to make relative to his or her income. That’s because college degrees, on average, pay big dividends in the form of higher wages over the course of one’s working career. The typical college graduate with debt will have borrowed $28,500 in pursuit of a bachelor’s degree. The borrower can repay that amount with monthly installments of $181 on a standard, 20-year repayment plan.

Second, the media narrative generally ignores the fact that the federal student-loan program, which accounts for more than 90 percent of the outstanding student-loan balances in the country, has since 2009 allowed borrowers to reduce their monthly payments to an amount that’s pegged to their current income. According to a recent report from the Congressional Budget Office, almost half of borrowers are currently making reduced payments on an income-based repayment plan.

So, for the most part, student loans aren’t unaffordable. Nor are they inescapable. Pundits and politicians often mention the fact that such loans aren’t easily dischargeable in bankruptcy—but, in fact, the federal loan program includes protections to prevent borrowers from reaching the brink of bankruptcy. When federal student-loan debt remains unaffordable for a long time, that is, when the borrower’s investment in education has failed to yield returns in the form of a well-paying job, the debt is automatically forgiven. That process takes 20 years for those who work in the private sector (or don’t work at all) and 10 years for borrowers who work in public service. Either period is a long time to have a large debt hanging over your head, but, as noted above, borrowers needn’t be making unaffordable monthly payments during that time, since all are eligible to make reduced, income-based monthly payments. Those with very little or no income don’t have to make payments at all.

The problem, generally, with wide-scale student-loan forgiveness is that it would be layered onto a system that already does a decent job of helping out those who need help the most. Any expansion of eligibility for education-loan forgiveness would almost necessarily bring about a regressive change in the allocation of resources—increasing the proportion of aid being delivered to already well-off borrowers.

For example, a recent report from the often-left-leaning Brookings Institution analyzed the distribution of benefits that would result from the loan-forgiveness plan that Warren proposed. They found that the benefits would disproportionately accrue to higher-income households. The bottom 60 percent of households would receive only 34 percent of the benefits.

Those who struggle the most to pay back education loans are borrowers who don’t complete their degree programs. Often, these individuals have not acquired the skills or credentials to secure the higher-paying jobs that can make loan repayment affordable. Even with the safety nets that are now woven into the federal student-loan program, many in this group face financial distress. Instead of lessening the cost of college for those who’ve “made it” by offering broad loan forgiveness, policymakers should consider tailoring solutions to those who most need help. One possibility would be to allow students who are normally eligible to receive federal Pell Grants during each semester of enrollment to collect more of these funds in their initial semesters. Shifting more of the grant aid to the first year or two of college would mean that students could potentially accrue less debt early on. This relatively modest change in the timing of Pell Grants could go a long way toward reducing the financial distress among student borrowers. The highest rate of default on student loans is among borrowers with less than $5,000 in debt, who are often those who started college but did not finish.

Advocates of blanket student-loan forgiveness sometimes concede the points I’ve made here but continue to defend the policy on the grounds that it would be “good for the economy.” There is some sense in that. Sending taxpayer dollars out to those who are likely to spend them does have the potential to stimulate the economy (as long as the effect isn’t offset by others anticipating their taxes will go up and reducing consumption accordingly). But if the main goal is stimulating the economy rather than supporting borrower welfare, there are other policy options that would more effectively achieve that end. Even throwing dollar bills out of a helicopter might work.

Under the CARES Act, Congress temporarily excused all borrowers from making payments on their federal student loans, which was a reasonable step. Even though it likely delivered poorly targeted benefits, disproportionately aiding borrowers who didn’t lose their jobs, it ensured that workers who did lose income wouldn’t face negative repercussions from missing a payment; nor would they need to hustle to sign up for an income-based repayment plan that would excuse them from making payments.

With that benefit set to expire at the end of 2020, the policy conversation has progressed beyond stopgap measures to a more comprehensive overhaul of the system. Senator Lamar Alexander, chair of the Senate education committee, introduced a proposal that is largely consistent with the ideals I’ve laid out above. Rather than offering widespread loan forgiveness, he has proposed a streamlining of the existing safety-net provisions, which will more effectively ensure that those with unaffordable debt won’t have to pay. Alexander’s proposal offers borrowers two options: a standard 10-year plan and an income-based plan. Under the latter, borrowers with no income would be excused from making payments. When a borrower starts earning income, the payment would be capped at 10 percent of the person’s income that surpasses 150 percent of the federal poverty line. Higher-income earners would not be eligible for this benefit.

Unlike some Democratic plans that call for a one-time pardon from debt for all, Alexander’s plan ensures a system that will work to ease the economic burden of student debt for troubled borrowers today but will also ensure student-loan affordability during future downturns, and even when individuals face personal financial crises.

It’s easy to appreciate why widespread student-debt relief is an appealing proposal. Those of us with some years behind us empathize with today’s young adults who are coming of age in a difficult economy. We want them to have every opportunity to succeed, but borrowing for education actually creates opportunities for most people. Loan forgiveness should be reserved for those who truly need it.

In mid-February 2020, Antonio Roca’s team at the education-management company Academica started to get some disturbing reports from the firm’s Milan branch. It appeared that the virus ravaging Wuhan, China, had made its way to Northern Italy. This menace wasn’t going to stay contained for long. Could it make it to the States? The team should probably prepare.

Roca is the managing director for the virtual education division of Academica, a large U.S.-based education service provider. The company manages 200 brick-and-mortar charter schools in 11 states, serving some 125,000 students. Via online instruction, it serves an additional 20,000 students in 11 countries, including Italy. It is one of the for-profit charter-school companies that left-leaning education activists have set their sights on.

The 2020 Democratic Party platform promises a ban on all federal funding for for-profit charter schools, explaining that “education is a public good and should not be saddled with a private profit motive.” In May 2020, more than 200 activists, including Diane Ravitch, Jonathan Kozol, Danny Glover, and Michael Moore, signed an open letter to presidential candidate Joe Biden calling for an outright prohibition on such schools.

A look at Academica’s response to the Covid-19 crisis might temper some of that distrust.

In early March, the company surveyed parents of its on-campus students, hoping to identify potential problems with student Internet access and hurdles that the firm would need to overcome to move instruction online. Staff developed both hard-copy and digital resources (in English, Spanish, and French) and distributed thousands of devices to students on campus before schools had to close.

On Friday, March 13, the first group of Academica schools closed in South Florida. The company trained 5,000 teachers via videoconferencing. The videoconferencing tools had already been vetted for security via the firm’s Colegia platform, a central hub of educational applications, content, and communications that Academica had created to ensure continuity of live instruction. Luckily, many of its schools were slated to be on spring break the following week, buying the staff time to train the rest of the teachers.

On Monday, March 16, the first tranche of Academica schools reopened online, with the rest reopening as they came back from spring break. For the remainder of the school year, Academica offered at least four hours of remote live instruction per day. Students reported online at the usual school start time, wearing their school uniforms. Academica created professional-learning communities for teachers to work together to navigate the obstacles that emerged during the pandemic. It developed online tools that offered confidential meeting rooms for pullout services and direct instruction for students with special needs. It continuously surveyed parents and tracked attendance every day. In late April, the company was seeing 94 percent attendance, a better rate than in its traditional brick-and-mortar schools.

When the Center on Reinventing Public Education tracked a set of 82 public school districts last spring, it found that 27 of them—one third—did “not set consistent expectations for teachers to provide meaningful remote instruction” during the closures, and 13 of the 27 did “not require teachers to give feedback on student work.” Meanwhile, Academica was kicking into overdrive.

Why?

“We’re sensitive to our customers,” Roca said. “If we don’t respond, they’ll talk with their feet.”

When Is a School a For-Profit?

According to data from the National Alliance for Public Charter Schools, approximately 13 percent of the nation’s more than 7,000 charter schools are operated by for-profit education management organizations or education service providers. About twice that number of schools are run by nonprofit charter management organizations, and five times as many are freestanding charters that operate independently. Over the past decade, the number of schools operated by nonprofit organizations has grown more than twice as much as the number operated by for-profit organizations (see Figure 1).

But is a school managed by a for-profit organization a for-profit school?

“Most people don’t understand what ‘for-profit’ means” in the context of charter schools, said Brian Britton, CEO of the for-profit education management organization National Heritage Academies. “We are partnering with nonprofit school boards.”

Kevin Chavous, president of academics, policy, and schools for K12 Inc., agrees. “We don’t actually hold any charters,” he said. Chavous likens K12, a for-profit that provides online schooling and curricula, to other companies with which schools typically do business, asserting, “schools have vendors for curriculum or roof repair. We are a vendor.”

Roca of Academica puts it this way: “Although Academica itself is a for-profit company, the CMO networks and schools who hire Academica are nonprofit entities. Together, we are referred to as hybrid organizations.”

In every state except Arizona (and, until recently, California), for-profit charter operators are not allowed to receive government funding or hold a school’s charter. However, nonprofit charter boards can choose to contract with a for-profit management organization instead of contracting with a nonprofit operator or hiring a traditional management team to run the school.

Rudyard Ceres, a board member of Brooklyn Excelsior Charter School, described one benefit of partnering with National Heritage Academies: “As a result of NHA’s scale, they are able to spot patterns and see what is working across a wide range of schools,” he said, “which helps us to troubleshoot in real time. Further, the model allows us to rely on NHA’s business expertise, which permits our educators and staff to focus on what matters most: helping our students reach their fullest potential.”

Charter boards can, and do, fire their management organizations and contract with new ones. Last year, for instance, Indianapolis Public Schools chose to end its contract with Charter Schools USA, which had been managing three of the district’s charter schools since 2011. The district has chosen a nonprofit charter management organization, Christel House International, to operate one of the schools and is considering whether to manage the other two schools itself or contract with another management organization.

Schools of all types contract with vendors. They use for-profit companies to provide food services or perform maintenance on the building. They hire security guards and buy textbooks from for-profit publishers. They contract for professional development and office supplies. For-profit companies provide back-office support and run learning-management systems. They provide assessments and data analytics. The list goes on.

Some schools contract with vendors to manage and operate their schools. It is these, and only these, that are described as “for-profit” schools. But why draw the line at school management and operation?

Conversely, why would a nonprofit charter school that has its own management team but contracts out its curriculum and textbooks, its learning-management system, its food services, its back-office accounting and human resources, and its substitute teachers be considered nonprofit?

Examine the vendor contracts of traditional public school districts, and you will see lines blurring there as well.

Maybe the contrast drawn between for-profit education management organizations and non-profit charter management organizations is a distinction without a difference.

Transparency, Scale, and Skin in the Game

Debates about the role of for-profit operators go back decades. The first issue of this journal, in 2001, featured a forum with contributors John Chubb, then of the for-profit Edison Schools, and Henry Levin of Teachers College headlined “The Profit Motive: Will it benefit kids?”

At the time, private businesses had only recently begun running public schools. Increasingly, as Chubb wrote, commercial firms were “seeking to actively manage entire public schools—hiring and firing; supervision, evaluation, and compensation; professional development; curriculum, instruction, and assessment; educational technology; plant management—everything.”

To many people, the idea was anathema. The mission of schools, after all, was educating the young, not making a buck. Chubb’s essay, though, laid out arguments in favor of the concept. First, private companies, in contrast to the nation’s patchwork of (at that time) 15,000 school districts, could consolidate administrative functions to serve multiple schools and employ economies of scale to drive down the prices of goods and services. Second, the capital-raising ability of for-profit companies, along with the profit motive itself, could spur greater investment in research and development.

And finally, because private companies would not be hampered by collective bargaining agreements or bureaucratic red tape, they had the potential to create more nimble organizations that could better meet the needs of students.

Henry Levin countered by describing the potential hazards of for-profit school management and operation. He doubted that businesses could compete profitably with government-subsidized public schools and nonprofit private schools. Most privates, he pointed out, set tuition levels below their costs and supplemented tuition dollars with vigorous fundraising.

Furthermore, the for-profit sector had yet to innovate significantly or distinguish its approaches from those of traditional schools. Although for-profit companies were experimenting with larger class sizes, using more junior teachers or contract employees, and starting to introduce technology, their schools looked largely the same as public and nonprofit schools.

Levin also warned of malign effects that could develop from the profit motive and companies’ efforts to attract “clients.” For-profit schools might be tempted to focus narrowly on maximizing test scores, a goal that could lead to pernicious practices such as recruiting only students likely to perform well on tests, narrowing the curriculum, and selecting, rewarding, and punishing teachers based mainly on their ability to improve test scores.

Over the years, the key points in this debate have changed little. When I asked leaders in the for-profit realm to name the advantages of their structure, these familiar themes returned.

Brian Britton of National Heritage Academies asserted that his company’s for-profit status allows it to scale up much more efficiently than a nonprofit organization could.

The primary reason? “We don’t have to fundraise,” Britton said.

Consider two charter operators, one for-profit and one nonprofit, both of which want to start a new school. Leaders at the nonprofit will have to raise philanthropic capital, up to millions of dollars, and have the capacity to continue raising money until enough state funding comes their way through student enrollment. Whether the founders acquire and renovate an existing building or purchase land and construct a new one, developing a school facility will be a major undertaking that eats up time and energy.

Businesses, by contrast, can borrow money more easily and can deploy capital faster. Britton says that without the capital he was able to access, he wouldn’t have been able to renovate and expand the former Catholic school on the East Side of Detroit that now houses the Detroit Enterprise Academy. According to data from the state of Michigan, that school now has higher rates of student proficiency than schools with similar student characteristics, faster rates of student growth, better teacher retention, and fewer mid-year transfers, all with lower staff-to-student ratios than comparable schools. It would appear that, because National Heritage Academies had access to capital, 739 K–8 students in Detroit have access to a better education.

Beyond the ability to grow more nimbly, for-profits can deploy further advantages. Private businesses can leverage economies of scale in ways that most nonprofits can’t. As Kevin Chavous of K12 puts it, with anything from shipping computers to designing and deploying curriculum, “we are able to create value-add services at a much better price point than if people had to put them together themselves.” This agglomeration effect can help schools operate much more efficiently.

For-profit management companies also face a higher level of accountability than traditional public and nonprofit schools generally do. Both Britton and Chavous argued that their operations were more transparent and more accountable than nonprofit operators.

“There is no transparency like in a for-profit, publicly traded company,” Chavous said. He would know. He served as a member of the Washington, D.C., city council for more than a decade, and also as a nonprofit leader, before entering the for-profit sector.

Private companies have the additional accountability that comes with having skin in the game. As Britton puts it, National Heritage Academies has an “extra dose of heightened accountability” because “the pressure of operating as a for-profit imposes a significant amount of discipline.” If a school fails, it loses money, and investors punish it.

“If we run great schools, that is major shareholder value,” Chavous said. “If people feel we run bad schools, that is bad for business.”

Leaders at K12 have had firsthand experience with this principle. In early September 2013, the company’s stock price was near its all-time high, at over $36 per share. But on September 17, investor and education-reform advocate Whitney Tilson presented a 110-slide PowerPoint presentation to the Value Investing Congress in New York City, detailing what he saw as the educational shortcomings of K12 and making his case for shorting the stock. He argued that “K12’s aggressive student recruitment has led to dismal academic results by students and sky-high dropout rates, in some cases more than 50 percent annually.” By the end of October, the stock’s price had fallen to half where it was before Tilson’s presentation. It would drop further, though less dramatically, until January 2016, when it bottomed out at just over $8 per share. Over the next three years, the stock climbed back up to its “pre-Tilson” price, only to fall again to about half of its peak.

K12 is not the only for-profit entity the market has disciplined. For years, Edison Schools (now EdisonLearning) was synonymous with for-profit education. Founded in 1992, it grew to manage 136 schools in 23 states. Founder Chris Whittle argued that his company could run schools for less money than public school districts did, and with better results. It couldn’t, and the market punished the firm. While the stock price rose to $36 per share in early 2001, by late 2002, it was trading at 14 cents. It was eventually sold and taken private in November 2003 at a share price of $1.76. It has since moved away from trying to manage schools to providing supplemental services like a more traditional school vendor.

The market’s harsh discipline affects the decisions that these enterprises undertake. School leaders have learned that unsustainable growth is poison for share value. So, when it comes to opening new schools or scaling up operations, the likelihood of success plays a huge role in the decision-making process. As Britton of National Heritage Academies puts it, “We don’t want to grow unless we can deliver what we’ve promised.”

Twenty years into the for-profit experiment, it is clear that skeptics were right to express concern about some of the recruiting and management practices that for-profit operators might adopt. Clearly, some of them did try to expand too quickly, and others engaged in practices that, while cutting costs, did not lead to higher student achievement. Yet these behaviors were not allowed to run amok—they were punished by the feedback loop of consumer sentiment and stock price. When schools behaved in ways that were detrimental to children and families, researchers and gadflies made this known, and investors took note.

Today, leaders of for-profit education management companies stress that they have learned from their predecessors and don’t want to repeat mistakes.

One hazard that critics of for-profits warned about did pose problems—but in traditional public schools and nonprofit charter schools. Accountability in the form of hiring, rewarding, and punishing teachers based on their ability to affect student test scores became a major policy initiative. Education advocates and parents often accused schools of narrowing the curriculum and “teaching to the test,” and some school districts made headlines for cheating on standardized tests. More informal gaming, whether by focusing on students who were scoring close to proficiency cutoff points on tests, or by moving strong teachers from non-tested to tested grades, would emerge in the subsequent years. The profit motive, as traditionally understood, had nothing to do with it.

What the Research Says

Perhaps research can indicate whether the experiment in for-profit education has been helpful or harmful.

In 2017, Stanford’s Center for Research on Education Outcomes released a study of charter management organizations across the country. It included a breakout of data on for-profit and nonprofit charter operators. The study found that students who attended schools managed by a nonprofit charter management organization outperformed demographically matched students in traditional public schools by 0.02 standard deviations in terms of annual growth in both math and reading, a difference that the researchers equate to about 12 additional days of learning. Those who attended for-profit schools scored lower than matched traditional public school students by 0.02 standard deviations in math but outperformed them by 0.01 standard deviations in reading, a finding that did not meet tests of statistical significance.

The Stanford study results on the three education management organizations mentioned in this article were mixed. The researchers found that Academica had a statistically significant positive impact on student reading scores, but no impact in math. K12 showed statistically significant negative impacts in both math and reading. National Heritage Academies showed statistically significant positive impacts in both math and reading.

In 2018, Susan Dynarksi and colleagues from the University of Michigan published a paper with the National Bureau of Economic Research that looked specifically at the performance of National Heritage Academies schools. The researchers compared the test scores of admissions-lottery winners and losers to estimate the charter schools’ effects on student achievement. They found significant positive effects in math, and smaller effects, which were not statistically significant, on other outcomes. They also found that the largest positive results, in contrast to findings in much of the literature on charter schools, were concentrated in wealthier, nonurban areas. The researchers surveyed school administrators at National Heritage Academies to try to tease out what the schools did differently. Administrators cited a number of the organization’s practices as potential differentiators: substantial parental engagement, ability grouping, a “no excuses” philosophy, mentoring of teachers by principals, and teacher bonuses.

In sum, the student-performance differences between for-profit schools and the traditional public schools their students would likely have attended neither convict nor exonerate for-profit management companies. There is a range of performance among providers; on standardized tests, their students perform about the same, on average, as students in the schools they are ostensibly attempting to replace.

Online charter schools, many of which are for-profit, have attracted their share of attention as well. In 2015, a team of researchers from Mathematica Public Policy Research, the Center on Reinventing Public Education, and Stanford published an in-depth examination of online charter schools. The headline finding was substantial negative effects for students in online schools, who lost the equivalent of 72 days of learning in reading and 180 days (a full school year) of learning in math.

One caveat applies to these results, though, and it’s related to the policy environment in which online schools operate. The Mathematica team found through surveys of students attending online schools that “student-driven, independent study” was the dominant mode of learning, with many online charters having high student-teacher ratios and offering limited contact time between students and teachers. The Center on Reinventing Public Education team noted that most state laws did not allow online schools to screen applicants to identify those most likely to thrive in an independent-learning environment (for example, students who are self-motivated and who have significant parental support). What’s more, in most states, school were paid based on enrollment rather than outcomes (for example, course completion).

The Stanford research design matched online students with non-online students to compare them on a set of observable characteristics. Online students, though, may vary from others in important but unobservable ways. For instance, they may be dealing with issues that cause them to need online schooling—issues that don’t show up in their demographic profile but that negatively affect their test scores.

That said, it is far more likely that online schools simply enrolled students who were not a good fit for the model. That is a problem of both regulation and incentives, and it could be addressed by changing the policy environment in which these schools operate.

It appears that, as in many realms of education, the quality of for-profit operators varies widely. Some for-profit schools are innovating and attempting new and different pedagogical strategies that are better meeting the needs of students, and some aren’t. Of course, one could say much the same about nonprofit charter schools and traditional district schools.

Going Forward

The pursuit of profit can lead people to take actions that reap good results for students, but it can also lead to pernicious practices. How can policymakers encourage the former and rein in the latter? They can hold all schools accountable, whether they are for-profit charters, nonprofit charters, or traditional public schools.

Giving parents the opportunity to opt out of schools that are not working for their children and opt in to schools that might do better is the most effective way to keep all of these forces in check. Empowering the people who are the “consumers” of education is the best tool in our toolbox to ensure that they are served well.

This is called a market. And, as for-profit operators have learned, it is a harsh disciplinarian.

But while the stock market in which for-profits operate is a true market, K–12 education in its current form is far from a free market. Prices are fixed. The largest player is the government. Gatekeepers restrict supply. Regulations stack hundreds of pages tall. The fastest-growing sector of alternative school operators is subsidized by tax-advantaged philanthropic giving.

This could help explain why for-profit education management organizations have remained a minor player in the charter-schooling landscape, which is itself a small player in the education system writ large. In 2007, according to data from the National Alliance for Public Charter Schools, education management organizations managed about 10 percent of all charter schools. By the 2016–17 school year, that share had only increased to 13 percent.

If policymakers want to get the most out of for-profit operators (with the added benefit of getting the most out of government-run and nonprofit schools, too), they will need to support a genuinely competitive market that allows all families to choose learning environments that work best for their children.

Michael Q. McShane is director of national research at EdChoice.