Education discussions today have a time horizon of three months.  In the fall, will we mix at-home with in-school? Does everybody have a digital hookup? Should we have police in schools? As important as these issues are, they have the unfortunate effect of pushing aside more fundamental issues that could have much greater impact. This twin crisis of Covid-19 and of societal recognition of deep-seated inequities must be directed toward essential school improvement.

The single most important way to ensure that everybody can participate in the modern economy is to ensure that they have the skills that are demanded by the economy.  If in September schools miraculously return to what they were in February, the current cohort of students would see their lifetime earnings reduced on average by 3 to 6 percent, based on estimates of pandemic-related learning losses and the value of a year of school.  Moreover, disadvantaged students would face an even larger loss, because their experiences over the past several months have been worse than those of more advantaged students.

The current protests across the country and across the world support a moral imperative for social re-design, but they have not touched on the deeper issues of how to deal with the underlying disparities that foster broken cities and dead end jobs. Sadly, the virus is pushing the schools to focus on ever smaller issues that will in the end not produce the change we need or the outcomes we desire.

Few of the current discussions suggest a way to ameliorate the lost learning. Many schools appear to be largely treating the fall return primarily as a logistical problem. How do we ensure appropriate social distancing in the schools, with partial in-person and partial internet instruction, with disparities in availability of devices? They have focused more on surface problems and less on treating the fundamental problems of delivering quality education to a very needy population of students.

Most return scenarios under discussion note but effectively ignore the gaping achievement gaps that will be faced in September. The digital divide has compounded existing differences in families’ ability to support and guide their children.

This situation is not amenable to any quick technological fix. It is amenable, however, to immediate policy improvements that proceed from where we are currently starting and lead us to better long run solutions.

There are two policy imperatives that stand out after this external disruption in the schools.

The shutdown since March has underscored that some teachers and some schools are better able to adapt.  It has taught us that some people are better at dispersed education, and some are better at in-person delivery. The simple lesson is that we cannot treat all teachers as equally able to deal with any new configuration of schools. A better approach would be to make sure that teachers are deployed in a manner that lines up their job placement with their skills.

But that does not go far enough. To ameliorate the unfortunate learning losses, we actually have to make schools better than they were. The only way we know how to do that is relying more on the best teachers and less on the ineffective teachers.

We also know that some kids will come out of this lock down period better positioned for the next grade than others. The second policy imperative for the fall is aligning instruction to meet each student where he or she is. Instead of teaching to the average preparation, based on prior grade in school, it will become apparent that we must focus on true capacity, where students are placed at the learning level appropriate for them.

Variation in the quality of teachers on the one hand and variation in the preparation and needs of students on the other have long been recognized. Covid-19 has exacerbated the prior conditions. Unfortunately, we just never previously felt compelled to make significant changes to deal with these problems.

Civil outrage over existing disparities may be the catalyst for true change by galvanizing the public to consider urgent new actions.  Perhaps now we can focus strategically on the immediate needs of our students, particularly those most vulnerable to the instability of the schools.

Current financial stress on schools along with the palpable harm to current students necessitate allocating resources differently. Doing it right now can lay the foundations for change that brings greater equity and opportunity, while doing it wrong will likely lead to more future protests because the disparities are real and big.

Eric Hanushek is the Paul and Jean Hanna Senior Fellow at the Hoover Institution of Stanford University.

Read more from Education Next on coronavirus and Covid-19.

Income inequality has soared in the United States over the past half century. Has educational inequality increased alongside, in lockstep?

Of course, say public intellectuals from across the political spectrum. As Richard Rothstein of the liberal Economic Policy Institute puts it: “Incomes have become more unequally distributed in the United States in the last generation, and this inequality contributes to the academic achievement gap.” Harvard political scientist Robert Putnam, citing research by Stanford sociologist Sean Reardon, says, “Rich Americans and poor Americans are living, learning, and raising children in increasingly separate and unequal worlds.” Another well-known political scientist, Charles Murray, argues that “the United States is stuck with a large and growing lower class that is able to care for itself only sporadically and inconsistently. . . . The new upper class has continued to prosper as the dollar value of the talents they bring to the economy has continued to grow.”

These analysts have good reason to express concern. National competitiveness is at stake, as education advocates have argued since the Soviet Sputnik launch inspired the National Defense Education Act of 1958. Economic productivity and growth are greater in countries where students perform better in math, reading, and science than in those that do not provide their youth the same opportunities to learn (see “Education and Economic Growth,” research, Spring 2008). And while some might see income inequality as the result of adult life choices about matters such as how hard to work or where to live, educational inequality seems unfair, because the economic status of a child is outside the child’s own control. It is an inequality of opportunity that runs counter to the American dream.

Despite the topic’s importance, surprisingly little scholarship has focused on long-term changes in the size of the achievement gap between students from higher and lower socioeconomic backgrounds. Our new research, presented here, attempts to fill this void, using data from four national assessments of student performance administered to representative samples of U.S. students over nearly five decades.

Contrary to recent perceptions, we find the opportunity gap—that is, the relationship between socioeconomic status and achievement—has not grown over the past 50 years. But neither has it closed. Instead, the gap between the haves and have-nots has persisted.

The stubborn endurance of achievement inequalities suggests the need to reconsider policies and practices aimed at shrinking the gap. Although policymakers have repeatedly tried to break the link between students’ learning and their socioeconomic background, these interventions thus far have been unable to dent the relationship between socioeconomic status and achievement. Perhaps it is time to consider alternatives.

Before drawing this conclusion, though, it is important to document the long-term trends in the connection between socioeconomic background and school achievement. Press coverage of the subject typically mentions only the most recent shifts in achievement levels and gaps. Our study broadens the perspective by making full use of nearly 50 years’ worth of historical data available from four intertemporally linked assessments of achievement in math, reading, and science administered to nationally representative samples of adolescent students born between 1954 and 2001. (By “intertemporally linked,” we mean that the test makers in each of these assessments design the tests to be comparable over time by doing things such as repeating some of the same questions across different waves.) These testing programs also collect information on students’ socioeconomic backgrounds, which we use to construct an index of socioeconomic status. We report changes in the gaps in performance between students from more- and less-advantaged backgrounds over the past half century.

We find that the socioeconomic achievement gap among the 1950s birth cohorts is very large—about 1.0 standard deviations between those in the top and bottom deciles of the socioeconomic distribution (the “90–10 gap”) and around 0.8 standard deviations between those in the top and bottom quartiles (the “75–25 gap”). These are very extensive disparities, as 1 standard deviation is approximately the difference in the average performance of students in 4th and 8th grades, or four years’ worth of learning. But though these inequalities are large, they have neither increased nor decreased significantly over the past 50 years.

It could be, however, that the picture is not as dismal as suggested. If overall changes in society, coupled with policy initiatives, have proportionately lifted all boats at the same rate, everybody might be better-off, even if gaps have not significantly changed. Using the same data as for the gap analysis, we find gains in average student performance of about 0.5 standard deviations for students at age 14, or roughly 0.1 standard deviations per decade. But, surprisingly, over the last quarter century, those gains disappear for students by age 17. In other words, there is no rising tide for students as they leave school for college and careers.

Prior Research

The effects of family background on student achievement are well-documented, but few studies track changes in the relationship between demographic characteristics and student performance over time. This scarcity of longitudinal analysis partly reflects measurement challenges.

Family background and achievement. There is little dispute that students’ performance in school is strongly affected by their family background. James Coleman and colleagues, in their seminal 1966 study, Equality of Educational Opportunity, found that parental education, income, and race are strongly associated with student achievement, while school resources such as per-pupil expenditures and class size are much less significant. Subsequent research has confirmed these early findings (see “How Family Background Influences Student Achievement,” features, Spring 2016).

A variety of mechanisms link socioeconomic status to achievement. For instance, children growing up in poorer households and communities are at greater risk of traumatic stress and other medical problems that can affect brain development. College-educated mothers speak more frequently to their infants, use a larger vocabulary with their toddlers, and are more likely to use parenting practices that respect the autonomy of a growing child. Higher-income families have access to more-enriching schooling environments, and they generally do not face the high rates of violent crime experienced by those in extremely impoverished communities. All these and other childhood or adolescent experiences contribute to profound socioeconomic disparities in academic achievement.

Trends in the socioeconomic achievement gap. Des- pite firm documentation of a strong connection between socioeconomic status and student achievement, only two studies provide information on trends in the opportunity gap over time. In an appendix table of a 1998 paper, Larry Hedges and Amy Nowell report the relationship between student performance and several background characteristics across six nationally representative surveys administered between 1965 and 1992. Among these variables, parental education has the strongest correlation with student achievement, and that connection endures over time. The correlation between achievement and family income in the six surveys is weaker and declines over time.

In a second investigation, published in 2011, Sean Reardon draws on data from 12 surveys that contain information on both student achievement and reports of parental income to estimate gaps in math and reading performance of students at the 90th and the 10th percentiles of the household income distribution. In contrast to Hedges and Nowell, he finds that the “income achievement gaps among children born in 2001 are roughly 75 percent larger than the estimated gaps among children born in the early 1940s.” For those born after 1974, children in families at the median income were falling farther behind those at the 90th percentile, leading Reardon to conclude that “The 90/50 gap appears to have grown faster than the 50/10 gap during the 1970s and 1980s.”

Reardon’s study and its conclusions have been widely cited by both academics and in the general media, and the idea that income-related achievement gaps have dramatically increased has become contemporary conventional wisdom. In a 2012 article, the New York Times asserts that “while the achievement gap between white and black students has narrowed significantly over the past few decades, the gap between rich and poor students has grown substantially during the same period.” Another Times piece quotes Reardon as saying, “The children of the rich increasingly do better in school, relative to the children of the poor. . . . This has always been true, but is much more true now than 40 years ago.”

Differences between the findings reported in the two studies may be owing to the focus of Hedges and Nowell on overall correlations between socioeconomic status and achievement, while Reardon discusses disparities between the extremes of the income distribution. They could also reflect the fact that Reardon’s analysis makes use of twice as many surveys as the earlier study, including data on more recent cohorts.

We, however, explore a third possibility—methodological limitations common to both studies. Both estimate trends from data collected by different surveys that are administered to students of varying ages and use disparate methods of estimating achievement levels and socioeconomic characteristics. As Federal Reserve economist Eric Nielsen points out, when “data sources have income and achievement measures that do not map easily across surveys, they add an additional layer of complexity and uncertainty to the analysis.” It is this uncertainty that we seek to mitigate by relying on surveys that allow for consistent, intertemporally linked measures of both student achievement and socioeconomic status.

Method

We draw data from four testing programs: two that are part of the National Assessment of Educational Progress (NAEP)—the Long-Term Trend and Main NAEP; the Trends in International Mathematics and Science Study (TIMSS); and the Program for International Student Assessment (PISA). (See sidebar for details.) We include all tests administered to students age 14 or thereabouts and at age 17. (For convenience, we identify all those tested at ages 13 to 15 as “14 years old.”) All told, we compile observations of achievement levels and gaps from 46 tests in math, 40 in reading, and 12 in science, or a total of 98 intertemporally linked tests over a 47-year period. Across this time span, achievement data are available for 2,737,583 students.

To measure these students’ socioeconomic status, we use indicators of parental education and home possessions as reported by students to construct an index similar to one designed by PISA. The choice of indicators is determined by the fact that all four assessments collect information on family background directly from students themselves. Young people are thought to be aware of their parents’ level of educational attainment but to have only an imperfect knowledge of their parents’ earned income. As a classic study investigating this question puts it, income is “a matter of speculation for many students and thus inaccurately reported.” For this reason, the surveys collect economic information by asking students about household items, such as the number of durable goods and educational items present in the home. Students are likely not to have seen their family’s tax return. These same students, though, are well aware of whether they sleep in their own bedroom or share one. They also know whether their home includes a dishwasher or a computer. Our analysis thus differs from Reardon’s study, which excludes assessments that do not ask students or their parents a direct question about household income.

We use our constructed index to estimate two disparities for each test: 1) the difference in achievement between the highest and lowest deciles of the socioeconomic distribution (the 90–10 gap) and 2) the difference between the highest and lowest quartiles (the 75–25 gap). We then fit simple quadratic trend lines through these points in order to document how, if at all, the magnitude of these disparities has changed over time.

Achievement Gaps

As can be seen in Figure 1, the disparities in achievement between students from the highest and lowest socioeconomic status groups are strikingly persistent throughout the time period. The socioeconomic achievement divide hardly wavers over this half century. In the 1954 birth cohort, the achievement gap between the average of those in the top and bottom deciles of the socioeconomic distribution stood at slightly less than 1.2 standard deviations. For those born in 2001, the gap is only slightly less—about 1.05 standard deviations. That is, the most-disadvantaged students have made the same gains in achievement over the decades as those realized by the most-advantaged students.

The disparity between students in the top and bottom quartiles of the socioeconomic distribution was about 0.9 standard deviations for the 1954 birth cohort. This 75–25 gap falls slightly during the next two decades, settling at barely below 0.8 for the cohort born in 2001.

Trends are similar for math and reading separately. The gap in math achievement, particularly for the 90–10 comparison, shows a little movement over the period—narrowing in the early years but returning to a position below the initial level in recent decades. The 75–25 math gap narrows slightly over time. In reading, the pattern appears essentially flat for the entire period.

To see whether an alternative measure of socioeconomic status yields similar results, we estimate the gap between students who are eligible for the federal school-lunch program and those who are not, as reported on the Main NAEP, the one assessment that contains this information. The federal program provides free lunch to extremely poor students from households below the poverty line, while a reduced-price lunch is available to moderately poor students with somewhat higher incomes (1.85 times the poverty line). The gap between the extremely poor students and other students in the 1982 birth cohort is a sizable 0.73 standard deviations (Figure 2). When the extremely poor are combined with the moderately poor, the gap for this cohort is nearly as large. Over the next 20 years, the gap between the extremely poor and students from families above the eligibility line narrows by just 0.02 standard deviations, while the gap between ineligible students and all those eligible for participation in the program widens by 0.01. In sum, this alternative measure of the achievement gap between students from higher and lower socioeconomic backgrounds also shows only minuscule change over the course of the past two decades.

Figure 2 also shows the white-black achievement gap. While this is not accurately thought of as a socioeconomic gap because of the improvements in black incomes, it represents another potential dimension of continuing societal disparities. As Figure 2 shows, there is a sizable shrinking of the racial gap in the early period but little change across the last two decades.

Some have hypothesized that the lack of success in diminishing the size of the socioeconomic gap is due to changes in the racial and ethnic composition of the school population. It is true that the ethnic makeup of the school-age population has changed dramatically over the past half century, with the share that is white declining from about 75 percent to 55 percent. However, these changes do not seem to have materially affected trends in performance gaps. The 90–10 socioeconomic achievement gap among white students born in 1954 was one standard deviation. By the middle of the period, the divide had declined by about 0.2 standard deviations, but it then rose again by a commensurate amount. Trends for the 75–25 socioeconomic achievement gap among whites are much the same, confirming that changes in the ethnic composition of student cohorts do not account for the unwavering divide between the haves and have-nots.

In sum, our results confirm Reardon’s finding of large gaps in academic performance between students at the extremes of the socioeconomic distribution. The average 90–10 income achievement gap across the surveys suggested by the Reardon analysis is very similar to the 90–10 socioeconomic achievement gap we identify. We are, however, unable to replicate Reardon’s finding that achievement differentials have risen by as much as 75 percent over the past 50 years. His results may be a function of a reliance upon cross-sectional studies that use disparate methods for collecting both income and achievement information. Whatever the reason, the trends estimated in his analysis differ markedly from the gaps we observe by using a uniform measure of socioeconomic status and data from intertemporally linked surveys administered to students of the same age.

Rising Tides?

We might feel differently about these persistent achievement gaps if we found that all achievement was rising and thus suggesting improved economic futures for all. To place the achievement gaps in context, we describe changes in the average level of achievement among students at age 14 and age 17 for students born between 1954 and 2001. Figure 3 shows a significant upward trend in the average achievement level for all adolescent students of approximately 0.3 standard deviations over the course of the past half century, or approximately 0.06 per decade. This trend differs by the age of the student, however. Students at age 14 show an overall increase of about 0.43 standard deviations, or approximately 0.08 per decade, but gains among students at age 17 amount to only about 0.10 standard deviations, or 0.02 per decade. Further, we see no improvement in the performance of older students after the 1970 birth cohort.

Trends in average levels of achievement do differ in magnitude by subject, but the overall patterns are quite similar. In math, the younger adolescents register average gains of 0.9 standard deviations, while the older ones show a shift upward of only 0.25. At both ages, the reading gains are less. The trend among younger adolescents amounts to just 0.20 standard deviations over the half century and, among older ones, the trend is flat, showing no upward trend at all.

The differences in trend lines for students at different ages presents a puzzle for which we have no easy answer. Even setting aside the oldest students in our data, we see that the average improvement in test performance among 13- and 14-year-olds who take the NAEP tests and the TIMSS is larger than that registered by 15-year-olds on the PISA tests. This may reflect differences in test design, or it may suggest that the fade-out in gains begins in the early years of high school. The lack of a positive trend among 17-year-olds for the past quarter century also suggests that high schools do not build upon gains achieved earlier, a signal, perhaps, that the high school has become a troubled institution. In any event, there is no sign of a rising tide that lifts all boats at age 17 when these students are going into further schooling or into the labor force.

Importantly, the age anomaly that we see in the trends in achievement levels is not found in the performance gaps. Constant social gaps are found across all age groups.

Discussion

The achievement gap between haves and have-nots in the U.S. remains as large as it was in 1966, when James Coleman wrote his landmark report and the nation launched a “war on poverty” that made compensatory education its centerpiece. That gap has not widened, as some have suggested. But neither has it closed.

The question remains: why has the gap remained constant? The tempting answer is that nothing significant enough has happened to alter its size. But this would ignore a wide variety of factors that have shifted over the years. It is more likely that some changes within families and within schools have worked to close the socioeconomic achievement gap while other changes have widened it, with these factors largely offsetting one another.

Families. In terms of family background, there is the widening differential in household income that motivated Reardon’s work. Socioeconomic differences in the age of the mother at the birth of the child have also increased in the past 50 years. The incidence of single-parent households has increased and is likewise concentrated at the lower end of the socioeconomic spectrum. Each would tend to exacerbate socioeconomic achievement gaps.

But these negative factors could be offset by other, countervailing demographic changes. Most importantly, differences among children in their parents’ level of educational attainment have narrowed as overall education levels have climbed. So have differences in the number of siblings in the household. Both factors are important determinants of student achievement. The balance among all these factors may well have left the family contribution to the achievement gap at much the same level today as it was for cohorts born in the 1950s.

Schools. Similarly, there may be opposing forces within the educational system that have offset one another. On the one side, over the past 50 years, the federal government has enacted compensatory education programs for school-age children and the Head Start program for students at ages three and four. Brown v. Board of Education and the Civil Rights Act of 1964 accelerated school desegregation, particularly in the South. The Individuals with Disabilities Education Act funded school services for students with disabilities, a group disproportionately composed of children from low-income families. States systematically changed their funding of local schools, often in response to court orders, leading to more equal funding between rich and poor school districts. Overall school funding increased dramatically on a per-student basis, quadrupling in real dollars between 1960 and 2015. And finally, states have introduced measures holding schools accountable for student performance, as required by the 2002 No Child Left Behind Act. Accountability mandates were disproportionately directed toward schools serving low-income students. Each is aimed at closing gaps.

On the other hand, the quality of the teaching force—a centrally important factor affecting student achievement—may well have declined over the course of the past several decades. Women have greater access to opportunities outside the field of teaching. Teachers’ performance on standardized tests has slipped, along with other indicators of selectivity. Teacher salaries have declined relative to those earned by other four-year college-degree holders and are currently low relative to comparable workers in other occupations (see “Do Smarter Teachers Make Smarter Students?” features, Spring 2019).

These changes affecting the quality of the teaching force are likely to have had a disproportionately adverse effect on disadvantaged students. Collective-bargaining agreements and state laws have granted more-experienced teachers seniority rights, leaving disadvantaged students to be taught by less-effective novices.

In other words, a growing disparity in teacher quality across the social divide may have offset the impacts of policies designed to work in the opposite direction.

Conclusion

Two surprises emerge from this analysis of long-term trends in student-achievement levels and gaps across the socioeconomic distribution. First, gaps in achievement between the haves and have-nots are mostly unchanged over the past half century. Second, steady gains in student achievement at the 8th-grade level have not translated into gains at the end of high school.

Because cognitive skills as measured by standardized achievement tests are a strong predictor of future income and economic well-being, the unwavering achievement gap across the socioeconomic spectrum sends a discouraging signal about the possibilities of improved intergenerational social mobility. Perhaps more disturbing, programs to improve the education of disadvantaged students, while perhaps offsetting a decline in the quality of teachers serving such students, have done little to close achievement gaps. These steadfast disparities suggest the need to reconsider the current direction of national education policy.

Two areas for further exploration seem especially critical. First, researchers have uniformly found that teacher effectiveness is a predominant factor affecting school quality. While there has been ample commentary on teacher recruitment and compensation policies, few programs and policies at scale have directly focused on enhancing teacher quality, particularly for disadvantaged students. Second, the achievement gains realized by students at age 14 fade away by age 17, yet policymakers have left high schools—like the achievement gap itself—in many ways untouched.

Eric A. Hanushek is the Paul and Jean Hanna Senior Fellow at the Hoover Institution of Stanford University. Paul E. Peterson is senior editor of Education Next; professor of government and director of Harvard’s Program on Education Policy and Governance; and a senior fellow at the Hoover Institution. Laura M. Talpey is a research associate at the Stanford Institute for Economic Policy Research. Ludger Woessmann is professor of economics at the University of Munich.

---

Data Sources

We use surveys from four testing programs to investigate achievement gaps and levels over time. These surveys use consistent data-collection procedures to trace the achievement of representative samples of U.S. adolescents over time. They also collect information about the cultural and economic resources of the students’ families using student reports of their parents’ education and of a wide variety of durable material and educational possessions in the home. Each data set comprises student-level data that we aggregate by demographic group.

Long-Term Trend National Assessment of Educational Progress (LTT-NAEP)

The LTT-NAEP dates back to 1971 and assesses students age 9, 13, and 17. Data are available for math in select years from 1978–2008 and for reading from 1971–2008. We create a panel of math and reading scores for students age 13 and 17, beginning with the 1954 birth cohort, who turned 17 in 1971. LTT-NAEP is the only source of information for cohorts born between 1954 and 1976. In a typical year, approximately 17,000 students participate.

Main National Assessment of Educational Progress (Main NAEP)

The Main NAEP started in 1990 and assesses students in grades 4, 8, and 12 every two to four years. We create a panel of math and reading scores for 8th graders from 1990–2013. The Main NAEP is aligned to school curricula and designed to provide results for representative samples of students in the United States as a whole and for each participating state. For each test administration, the Main NAEP 8th-grade sample is over 150,000 observations.

Trends in International Mathematics and Science Study (TIMSS)

TIMSS, administered by the International Association for Evaluation of Educational Achievement (IEA), is the current version of an international survey that originated as an exploratory study of mathematics conducted across 12 countries in the 1960s. The tests are curriculum-based and developed by an IEA-directed international committee. Beginning with the 1981 birth cohort (tested in 1995), the TIMSS tests have been designed to generate scores that are comparable over time. We use the TIMSS 8th-grade math and science tests beginning with this cohort by compiling national data files from 2003, 2007, and 2011, and international data files from 1995, 1999, and 2015. The only difference between the national and international data is that the latter do not contain an indicator of race or ethnicity. For this reason, our estimates of the black-white achievement gap for TIMSS are only available for 2003, 2007, and 2011. The U.S. TIMSS 8th-grade sample includes roughly 10,000 students for each administration of the test.

Program for International Student Assessment (PISA)

PISA, administered by the Organization for Economic Co-operation and Development, began in 2000 and assesses students’ math, reading, and science literacy at age 15 every three years. Its assessments are designed to measure practical applications of knowledge. The United States has participated in every wave of the test, though results are not available for reading for the 1991 birth cohort. We use national PISA data, available every three years from 2000 to 2015. PISA does not collect information on race or ethnicity, so these tests are not used in our analysis of the black-white achievement gap. The U.S. PISA sample includes over 5,000 students for each administration of the test.

Student achievement varies widely across developed countries, but the source of these differences is not well understood. One obvious candidate, and a major focus of research and policy discussions both in the United States and abroad, is teacher quality.

Research and common sense tell us good teachers can have a tremendous impact on their students’ learning. But what, exactly, makes some teachers more effective than others? Some analysts have pointed to teachers’ own scholastic performance as a key predictor, citing as examples teacher-recruitment practices in countries where students do unusually well on international tests. One oft-cited statistic notes that high-scoring Singapore, Finland, and Korea recruit their teacher corps exclusively from the top third of their academic cohorts in college; by contrast, in the U.S., just 23 percent of new teachers come from the top third of their graduating class.

Can we provide systematic evidence that teachers’ cognitive skills matter for student achievement? Do smarter teachers make for smarter students? And if so, how might we recruit teachers with stronger cognitive skills in the U.S.?

To investigate these questions, we look at whether differences in the cognitive skills of teachers can help explain differences in student performance across developed countries. We consider data from the Organization for Economic Cooperation and Development (OECD), an association of 36 largely developed countries that has assessed nationally representative samples of both adults and students in reading and math. We use these data to estimate the effects of teacher cognitive skills on student achievement across 31 OECD countries.

We find that teachers’ cognitive skills differ widely among nations—and that these differences matter greatly for students’ success in school. An increase of one standard deviation in teacher cognitive skills is associated with an increase of 10 to 15 percent of a standard deviation in student performance. This implies that as much as one quarter of the gaps in average student performance across the countries in our study would be closed if each of them were to raise their teachers’ cognitive skills to the level of those in the highest-ranked country, Finland.

We also investigate two explanations for why teachers in some countries are smarter than in others: differences in job opportunities for women and in teachers’ salaries compared to those of other professions. We find that teachers have lower cognitive skills, on average, in countries with greater non-teaching job opportunities for women in high-skill occupations and where teaching pays relatively less than other professions. These findings have clear implications for policy debates here in the U.S., where teachers earn some 20 percent less than comparable college graduates.

The importance of teacher quality

While many factors influence student success, the most convincing research has focused on differences in learning gains made by students assigned to different teachers. Studies of teachers’ contributions to student reading and math achievement consistently find variations in “value-added” that far exceed the impact of any other school-based factor.

These studies are unhelpful in explaining international differences in student achievement, however: they focus primarily on the U.S. and have not identified correlates of teacher value-added that can be measured consistently across countries. Such differences are a major concern for the United States, where policymakers are searching for strategies to shore up the country’s economic competitiveness. American students score rather unimpressively on the OECD’s Program for International Student Assessment (PISA), which measures high-school students’ skills in math, reading, and science every three years. On the most recent PISA math assessment in 2015, for example, American teenagers ranked 40th, well below most major Asian and European countries.

Importantly, research conducted within the U.S. and in other settings has shown that common measures of teacher qualifications such as advanced degrees, experience levels, and professional preparation are not consistently related to classroom effectiveness. The story differs for research on teacher cognitive skills and salaries, however, in ways that motivate our analysis in this article.

Prior studies of teacher cognitive skills, largely from within the U.S., provide some evidence of positive impacts on student achievement. These studies have relied on small and idiosyncratic data sets, and their results are not entirely uniform. Nonetheless, compared to alternative measures of teacher quality, test scores are most consistently related to student outcomes.

The relevant evidence on teacher salaries is different. While studies conducted within specific countries tend to find that salaries are unrelated to effectiveness, the limited available cross-country evidence suggests that students perform better where teachers are better paid. These divergent results suggest that salary levels may have important ramifications for the quality of the overall pool of potential teachers—even if the distribution of salaries within a country is not a good index of effectiveness.

Measuring teacher cognitive skills

To measure teacher cognitive skills, we use data from the OECD’s Program for the International Assessment of Adult Competencies (PIAAC) survey in 2012 and 2015, which tested the literacy and numeracy skills of more than 215,000 randomly selected adults age 16–65 in 33 countries. We focus on the 6,402 test-takers in 31 countries (those where we also have information on student achievement) who reported their occupation as “primary school teacher,” “secondary school teacher,” or “other teacher.” The number of teachers tested ranges from 106 in Chile to 834 in Canada, with 207 per country on average. We use the median literacy and numeracy scores of the teachers tested in each country as our measures of teachers’ cognitive skills.

These data reveal vast differences in teacher cognitive skills across countries. Figure 1 compares median teacher numeracy and literacy skills in each country to the skills of all employed adults in different educational groups within Canada, the country with the largest PIAAC sample. Teachers in Turkey and Chile score well below Canadian adults with only a vocational post-secondary degree, while teachers in Italy, Russia, and Israel perform at the level of vocationally educated Canadians. At the other end of the spectrum, the skills of teachers in Japan and Finland are higher than those of Canadians with a master’s or doctoral degree. Teachers in the Netherlands, Norway, and Sweden have skill levels similar to Canadians with a bachelor’s degree.

Teachers in the United States perform worse than the average teacher sample-wide in numeracy, with a median score of 284 points out of a possible 500, compared to the sample-wide average of 292 points. In literacy, they perform slightly better than average, with a median score of 301 points compared to the sample-wide average of 295 points. While teacher literacy skills are higher than numeracy skills in some countries (including the U.S.), the reverse is true in others—a pattern we will return to below when examining the consequences of differences in teacher skills across subjects.

These differences in teacher cognitive skills reflect both where teachers are drawn from within each country’s skill distribution and where a country’s overall cognitive-skill level falls in the world distribution. While median teacher cognitive skills are close to the median skills of college graduates in most countries, teachers perform better than the median college graduate in countries like Finland, Singapore, Ireland, and Chile, and perform worse than the median college graduate in others, such as Austria, Denmark, the Slovak Republic, and Poland.

However, teachers from relatively lower parts of the distribution may have greater cognitive skills than their peers abroad if their home country’s skill level is higher overall. For example, math teachers in Chile and Finland are drawn from similarly high levels of their overall distributions of college graduates, yet Chilean teachers score at the bottom of all 31countries while Finnish teachers score at the top. Teachers from the Slovak Republic are drawn from the lowest point in the country skill distribution of the 31 countries, yet have teachers in the middle of the international skill distribution of teachers.

Measuring student achievement

Our data on student achievement come from PISA in 2009 and 2012, which tested the math and reading skills of more than a half million 15-year-old students in nationally representative samples in more than 60 countries, including 31 of the countries participating in PIACC. We use those two PISA cycles because participating students would have been taught by the teacher cohorts tested in 2012 and 2015 in PIAAC.

Student performance in math and reading also varies widely across the countries in our sample, with especially pronounced differences in math. Students in top-performing Singapore scored 70 points above the sample-wide average of 498—the equivalent of nearly two school years. U.S. students scored well below that average at 484. In reading, Singapore students again earned the highest score of 534 compared to 445 for Chile, the lowest-scoring country in our sample. The U.S. score of 498 was not statistically different from the average of 497—leaving American students roughly one school year behind students in Singapore.

Linking teacher and student skills

How do these differences in student achievement relate to international differences in teachers’ cognitive skills? We use two different approaches to address this question. In both, we measure teacher cognitive skills only at the country level, in part because our data do not let us link students to their actual teachers but also to avoid bias due to factors such as parents choosing better schools and teachers for their children.

We first examine the association between median teacher cognitive skills and individual students’ performance across the 31 countries in our sample. That is, we ask whether students perform better on the PISA math and reading tests when their country’s teachers have stronger numeracy and literacy skills, respectively. In making these comparisons, we control for a wide range of other factors that could influence students’ performance. These factors include the skill levels of all adults age 25–65 as measured by PIACC; student characteristics such as age, gender, and migrant status; family background characteristics such as parents’ education levels and the number of books in the home; school characteristics such as enrollment size and instructional time in math and reading; and the average per-pupil spending and school starting age in each country. Importantly, we also control for a measure of the cognitive skills of the parents of PISA test-takers, as estimated based on the relationship between demographic characteristics and cognitive skills among parents tested in PIACC.

Despite these adjustments, however, the relationship between teacher skills and student achievement we estimate in our first approach could still reflect differences between countries that are more difficult to measure. Countries that emphasize the importance of a good education, for example, may have both teachers with high cognitive skills and parents who do more to support their children’s education in the home.

To address this concern, our second approach exploits the fact that both students and teachers were tested in two subjects and asks whether differences in teacher cognitive skills between numeracy and literacy are systematically related to differences in student performance between math and reading. In other words, do students perform relatively better in math (compared to reading) in countries where teachers have relatively higher numeracy skills? By focusing on comparisons within the same country, this approach eliminates the influence of any differences across countries that affect student achievement similarly in both subjects.

The only lingering concern is the possibility of unmeasured differences between countries that are subject-specific. For example, the results of our second approach would be biased if parents in societies where teachers have strong numeracy skills place more value on supporting their children in math than in reading. We provide evidence that such factors are unlikely to be important below.

Cross-country comparisons

Our first approach reveals a strong relationship between teacher cognitive skills and student achievement across countries. We estimate that increasing teacher numeracy skills by one standard deviation increases student performance by nearly 15 percent of a standard deviation on the PISA math test. Our estimate of the effect of increasing teacher literacy skills on students’ reading performance is slightly smaller, at 10 percent of a standard deviation, but the difference in magnitude across the two subjects is not statistically significant. Further analysis shows that the impact of teacher skills is somewhat larger for girls than for boys and for low-income students compared to wealthier students, particularly in reading.

As expected, a country’s cognitive skill level of all adults (age 25–65) is also strongly correlated with student performance. However, when controlling for teacher cognitive skills, the estimates for adult skills substantially decrease in size and lose statistical significance. In other words, the relationship between teacher cognitive skills and student performance is not driven by overall skill levels in the country; it is what teachers know that matters.

How much does it matter? To gauge the magnitude of our estimates, we simulate the improvements in student performance if each country brought its teachers up to the cognitive-skill level of Finnish teachers, the highest-skilled teachers in our sample. In math, U.S. students would improve by roughly one third of a standard deviation, and students in lowest-ranked Turkey and Chile would improve by more than half of a standard deviation. Overall, we estimate that bringing teachers in each country to the Finnish level would reduce the dispersion of country-level PISA scores by about one quarter, reducing the standard deviation of scores from 29 to 22 points in math and from 22 to 16 points in reading.

Improvements of this size may not be realistic in the short run for many countries, however. To match the cognitive skills of Finnish teachers, Turkey would have to draw its median teacher from the 97th percentile of the college numeracy distribution instead of the 53rd percentile. The U.S. would need to recruit its median math teacher from the 74th percentile instead of the current 47th percentile, and its median reading teacher from the 71st percentile instead of the 51st.

It is also important to note that the teacher-skill estimates do not capture the effect of a single school year with a talented teacher but rather reflect the cumulative effect of teacher cognitive skills on student performance through age 15. Thus, these projections are long-run impacts that presume that the quality of students’ teachers across the first 10 grades would improve to the level of Finland.

Within-country comparisons

While teacher cognitive skills are significantly related to student performance in both math and reading, it remains possible that this association is driven by unobserved differences across countries. Therefore, we now exploit only within-country variation to provide even more rigorous evidence on the effect of teacher cognitive skills on student performance.

The overall story is easy to see in a simple graphic: differences in teacher cognitive skills between numeracy and literacy are systematically related to differences in student performance between the same two subjects (see Figure 2). Remarkably, the magnitude of the relationship is very similar to that observed in the cross-country analysis: an increase of teacher cognitive skills of one standard deviation is estimated to improve student achievement by 11 percent of a standard deviation.

Our confidence in this result is strengthened by various “placebo” tests, all of which indicate that our estimates reflect the impact of teacher cognitive skills and not just those of the broader society.

In the first placebo test, we replace teacher cognitive skills with the cognitive-skill level of workers in 14 other occupations, including managers, scientists and engineers, health professionals, business professionals, clerks, sales workers, and service workers. If our results for teachers were really just the result of countries differentially valuing numeracy or literacy, we would expect to find an “effect” of these other workers’ skills on student test scores as well. Instead, we find there is no occupation other than teaching whose skill level is systematically related to student performance.

In a second placebo test, we replace teacher skills by the skill level of a randomly chosen sample of adults matched by age, gender, and educational attainment to the teacher sample in each country. We then draw 100 samples of matched “teacher twins” and compare their estimated student impacts. None of the 100 samples of teacher twins produces larger impacts than teachers.

We also investigate cross-subject effects, that is, the effect of teachers’ numeracy skills on student reading performance and the effect of teacher literacy skills on student math performance. If it is subject-matter skills that are important, as we have assumed in our within-country analysis, then teacher skills in one subject should be only weakly related—if at all—to student performance in the other subject.

Consistent with this logic, we find that teacher numeracy skills have a substantially larger association with student math performance than with reading performance. Similarly, teacher literacy skills are more important for student reading performance than for math performance. The most convincing evidence comes from simultaneously including teacher numeracy and literacy skills. Here, teacher skills in either subject affect only student performance in the same subject.

Creating a smarter teacher workforce

International differences in teacher cognitive skills reflect both where teachers are drawn from in each country’s skill distribution as well as the overall skill level of each country’s population—and policies to improve teacher skills could in theory focus on either of these dimensions. While improving the cognitive skills of the entire population is a valuable goal, it has been widely discussed elsewhere. In contrast, the determinants of where teachers are drawn from the overall skill distribution of a country’s population has received little attention. Our international data enable us to investigate how external forces and policy choices affect the part of the overall skill distribution from which countries recruit their teachers.

We examine two major factors. First, how has teaching been affected by competition from other occupations that demand high skills? In most countries, women historically have been segregated into a constrained set of occupations, one of which is teaching, and teaching remains a female-dominated occupation worldwide. Across the 23 countries used in the analysis below (where we exclude Turkey and all post-Communist countries due to their distinctive labor-market histories), more than two thirds (69 percent) of teachers are female, ranging from 59 percent in Japan to 79 percent in Austria. At the same time, women previously were much more concentrated in teaching than they are today, and the extent of this change varies across countries.

To compare women’s access to high-skill occupations across countries and over time, we compute the proportion of female teachers relative to the number of females in all high-skill occupations in three cohorts of adults defined by their birth years. We define what counts as a high-skill occupation empirically for each country, based on the average years of schooling among males working in each job category.

For both numeracy and literacy, we find that teacher cognitive skills are higher in countries where more females work in teaching relative to other high-skill occupations. The size of these relationships is quite substantial. For example, across all 23 countries in the sample, the share of women in high-skill occupations who are teachers decreases from 29 percent in the oldest age cohort (born in years 1946–1960) to 22 percent in the youngest cohort (born 1976–1990), reflecting the increased availability of alternative job opportunities for women over time. Our results imply that this change is associated with a decline in teacher numeracy skills of one quarter of a standard deviation. Another benchmark comes from comparing the share of women in high-skill occupations who are teachers across countries, which ranges from 18 percent in the U.S. to 38 percent in Singapore. Our estimates suggest that if the employment choices of U.S. women were as constrained as those in Singapore, the numeracy skills of U.S. teachers would be nearly three quarters of a standard deviation higher, lifting them to just above the international average.

While these results shed light on one important explanation for differences in teacher cognitive skills, restricting job opportunities for women is hardly an appealing strategy to improve teacher quality. Our second analysis, therefore, focuses on the impact of teacher pay on teachers’ cognitive skills.

To investigate the salary-skills relationship across countries, we first estimate whether teachers are paid a positive or negative wage premium compared to other college graduates with the same gender, work experience, and literacy and numeracy skills. We find a wide range of wage premiums, ranging from a positive 45 percent in Ireland to a negative 22 percent in the United States and Sweden (see Figure 3). This means that American teachers are paid 22 percent less than comparably experienced and skilled college graduates doing other jobs.

We then assess how these pay premiums relate to the position of teachers in a country’s skill distribution, and we find that countries that pay teachers more also tend to draw their teachers from higher parts of the college skill distribution. In terms of magnitude, a higher teacher wage premium of 10 percentage points is associated with an increase in teacher skills of about one tenth of a standard deviation for a given level of college graduates’ skills. Those pay choices appear to carry through to student performance in the classroom (see Figure 4). In countries where teachers are better paid, students achieve higher levels.

Implications

Our findings have broad application for American policymakers aiming to build a better teaching workforce. Prior research conducted within the U.S. has highlighted the importance of teacher quality for student achievement. But while such work provides useful information about relative learning gains among current teachers, it does not indicate what would be possible if the teacher corps were drawn from a different pool of candidates. Rather than assessing the relative talent of our current workforce, our study of teachers in 31 countries suggests what might be possible if the pool of potential teachers in the U.S. resembled those in the most successful education systems in the world.

To be sure, our work does not speak definitively to the sources of individual teacher talent. But we find that differences in teacher cognitive skills across countries are strongly associated with international differences in student performance. An increase in teacher cognitive skills of one standard deviation is associated with an increase in student performance of as much as 15 percent of a standard deviation in the PISA test.

Our international data also allow us to investigate how external forces and policy choices affect the skills of the teaching force and ultimately, student outcomes. We find that cross-country differences in women’s access to high-skill occupations and in wage premiums paid to teachers (given their gender, work experience, and cognitive skills) are directly related to teacher cognitive skills in a country. Teachers’ wage premiums are also highly correlated with student achievement across countries.

These results speak to the potential value of increasing teacher pay but must be interpreted with care. In particular, we have not provided causal estimates of how the quality of teachers would change if teacher salaries in the U.S. were raised. Increasing teacher salaries would undoubtedly expand the pool of potential teachers and help to reduce teacher turnover. Our evidence does not, however, indicate that more talented teachers would be hired out of the enlarged pool, nor does it indicate that the teachers induced to stay in the profession would be the most effective. Thus, while making it clear that a more skilled teaching force is generally found in countries with higher relative salaries, policymakers will need to do more than raise teacher pay across the board to ensure positive results. They must ensure that higher salaries go to more effective teachers.

Eric A. Hanushek is the Paul and Jean Hanna Senior Fellow at the Hoover Institution of Stanford University. Marc Piopiunik is an economist at the ifo Center for the Economics of Education at the CESifo Group. Simon Wiederhold is professor at the Catholic University, Eichstätt-Ingolstadt. This article is based on “The Value of Smarter Teachers: International Evidence on Teacher Cognitive Skills and Student Performance,” which is forthcoming in the Journal of Human Resources.