Overview
In 15 years of increasing average test scores, black-white and Hispanic-white student achievement gaps continue to close, and Asian students are pulling away from whites in both math and reading achievement. For the improving groups, these long-term trends may be a major educational success story.
In stark contrast, Hispanic and Asian students who are English language learners (ELL) are falling further behind white students in mathematics and reading achievement. And gaps between higher- and lower-income students persist, with some changes that vary by subject and grade. Meanwhile, the proportion of low-income students in U.S. schools has increased rapidly, as has the share of minority students in the student population. The chances of ending up in a high-poverty or high-minority school are highly determined by a student’s race/ethnicity and social class. For example, black and Hispanic students—even if they are not poor—are much more likely than white or Asian students to be in high-poverty schools.
These disparities represent a stubborn educational failure story. Attending a high-poverty school lowers math and reading achievement for students in all racial/ethnic groups and this negative effect has not diminished over time. And attending a school in which blacks and Hispanics make up more than 75 percent of the student body lowers achievement of black, Hispanic, and Asian students but does not affect white students (in some of the analyzed years it actually had a small positive influence on math test scores for whites).
These patterns of change (or lack of change) could have important implications for what is happening in American society in general and in U.S. schools in particular. Sustaining our democratic values and improving our education system call for a host of more coordinated and widespread education, economic, and housing policies—including policies to raise curricular standards, tackle insufficient funding for schools with a large share of low-income students, promote access to education resources from early childhood to college, improve dual language programs, provide economic support for families, and create more integrated schools and neighborhoods.
Executive summary
A founding ideal of American democracy is that merit, not accident of birth, should determine individuals’ income and social status. Schools have assumed a major role in judging key elements of merit among young people—namely, academic skills, hard work, self-discipline, and cooperative behavior. Schools do so mainly by evaluating students in a variety of subjects deemed important for success later in life. No one expects outcomes at the end of the schooling process to be the same for every student, since initial ability varies, and some young people are more disciplined and willing to work harder in school than others. Yet, when students’ inherent characteristics—such as race, gender, or parents’ economic and social capital—rather than their innate ability, hard work, and discipline systematically affect their school outcomes, this threatens democratic ideals.
These apparent contradictions between the ideals and reality of U.S. schools have led analysts over the last few decades to study and try to explain persistent gaps in student achievement. Particular attention has been given to the gap between blacks and Hispanics versus whites, across social-class groups, and by gender. Research has provided evidence that race and ethnicity continue to be important factors in explaining achievement differences. However, much of the black-white and Hispanic-white achievement gaps are accounted for by social-class differences. That is, in the United States, race and often ethnicity are closely intertwined with social class. Minority children, particularly African-Americans and Hispanics, are more likely to be poor than white children because of the ways that race and ethnicity shape opportunity and economic outcomes. Black and Hispanic children are also more likely than their white or Asian-American counterparts to live in low-income, racially segregated neighborhoods and to attend schools with high concentrations of low-income, nonwhite students.
Notwithstanding these troubling realities, achievement differences between blacks and whites and between Hispanics and whites have shrunk in recent decades. The bad news is that until recently gaps between the higher and lower social-class groups were increasing, particularly between children in the highest income group and everyone else (Reardon 2011; Reardon, Waldfogel, and Bassok 2016; Putnam 2015).
This paper advances the discussion of these issues by analyzing trends in the influence of race/ethnicity, social class, and gender on students’ academic performance in the United States. It focuses on trends for two different grade levels—eighth and fourth—and two different subjects—mathematics and reading—over the past decade. Trends in eighth-grade mathematics since the mid-1990s are also examined. This paper also explores the ways in which English language ability relates to Hispanics’ and Asian Americans’ academic performance over time (Nores and Barnett 2014). We use individual student microdata gathered from the National Assessment of Educational Progress (NAEP) to estimate the math and reading performance of students in the fourth and eighth grades from 2003 to 2013, and the math performance of eighth-graders from 1996 to 2013.
Our study has six objectives:
To describe changes in the racial characteristics and socioeconomic status (SES) of the student population, and in the composition of student bodies in U.S. schools over the past two decades in the periods 1996–2003 and 2003–2013
To describe the types of schools (high- and low-poverty, high and low concentrations of blacks plus Hispanics) that black, Hispanic, white, and Asian children attend and how these have changed over the past 10 and 20 years
To estimate changes in students’ achievement gaps by social class and race/ethnicity, including gaps for students designated as English language learners (ELLs), over the past 10 and 20 years
To estimate changes over the past decade in the influence of school composition—such as concentration of students by poverty, race, and ethnic status—on students’ achievement gaps by social class and race/ethnicity
To estimate whether and how much the trajectories of social class and race/ethnicity achievement gaps differed over the past 10 years for male and female students
To estimate whether and how much these trajectories differed over the past 10 years for lower-achieving students and higher-achieving students
Our unique approach, which uses individual student microdata gathered from NAEP over a substantial period of time (10 to 17 years, depending on the subject and grade), allows us to estimate changes in race/ethnic gaps, controlling for English-language learner designation, gender, and socioeconomic status. The approach also lets us estimate changes in socioeconomic gaps, controlling for race/ethnicity, gender, and ELL designation. Moreover, we can assess changes over time with regard to the sensitivity of race/ethnic and socioeconomic gaps to the inclusion of controls for school characteristics in terms of the proportion of poor and minority children in the student body. The percentage of students receiving free or reduced-price lunch (FRPL) is used as a proxy measure for the poor children in the student body.1 We characterize a school as high-poverty when more than 75 percent of its students are eligible for FRPL.
Importantly, because we use individual student data from large-scale assessments for our analysis, we can identify those students assigned to the English language learner track. We can therefore separate Hispanic and Asian ELL students from their non-ELL ethnic counterparts and examine their distinct performance and trends. Finally, our approach enables us to show how estimates of race/ethnic achievement gaps are affected by the unequal share of race/ethnic groups across those states in which students have systematically performed better or worse on the NAEP.
The results of our analysis yield important insights into the changing nature of inequality in the U.S. education system.
We find that between the mid-1990s and 2013, the proportion of low-income students in U.S. schools—those eligible for free or reduced-price lunch (FRPL)—increased rapidly. By 2013, more than half of eighth-grade mathematics students were eligible for FRPL (52.1 percent), up from 35.1 percent in 2000. In addition, the proportion of Hispanic and Asian students increased, in contrast to a steady decline in the percentage of non-Hispanic white and black students.
As the overall proportion of low-income students (those eligible for FRPL) increased in U.S. schools, the percentage of all students attending high-poverty schools (those with more than 75 percent of students eligible for FRPL)2 rose substantially from 2003 to 2013. The proportion of black and Hispanic students in these high-poverty schools was much higher than for white or Asian students. By 2013, more than 40 percent of black and Hispanic students attended a high-poverty school (43.5 percent of blacks, 40.3 percent of Hispanic non-ELLs, and 55.8 percent of Hispanic ELLs, respectively). In contrast, only about 7 percent of white students (6.9 percent) attended such schools. At least one in five black and Hispanic students (20.7 percent of blacks, 15.1 percent of Hispanic non-ELLs, and 33.9 percent of Hispanic ELLs, respectively) who were not eligible for free or reduced-price lunch attended a high-poverty school compared with just 3.2 percent of non-eligible white students.
Asian non-ELLs generally attended schools that had even lower levels of poverty than those attended by white students, although poor Asian non-ELL students were much more likely to attend high-poverty schools than poor white students.
We confirm earlier studies showing that although the black/white test-score gap remains large, it has declined substantially in the past two decades.
The achievement gap between white students and Hispanic non-ELL students also closed substantially in this period. The achievement gap between white students and Asian non-ELLs greatly increased in favor of Asians. By 2013, Asian non-ELL students scored almost half a standard deviation (SD) higher than white students in math. Moreover, the gap between Asians and whites in math was even larger among higher-scoring students.
In stark contrast to the shrinking achievement gap between white students and Hispanic non-ELL students and the expanding achievement gap between Asian non-ELL students and whites, we find that Hispanic ELL and Asian ELL students are falling further behind white students in mathematics and reading achievement.3
Adjusting for the higher concentration of Asian non-English language learner students in California and Hawaii, two low-scoring states, reduces our estimates of Asian students’ scores compared to what they would have been had they lived in higher-scoring states. This “state effect” also tends to be true for our estimates of the scores of Hispanic non-ELL students (who are concentrated in California and the Southwest), and for the estimated scores of African Americans (who are concentrated in southern states, which generally score lower on the NAEP).
Attending a higher-poverty school had a negative influence on the math and reading achievement of students from all racial/ethnic groups in both fourth and eighth grades. This negative influence was smaller for Hispanic non-English language learners than for whites and blacks, and it was larger for Asians than for other groups. In contrast, attending a school with more than 75 percent black plus Hispanic students had a larger negative effect on black, Hispanic, and Asian students than on white students.
Attending the highest-poverty school (with a high proportion of poor students, meaning more than 75 percent of students eligible for FRPL) continues to have a strong negative impact on individual students across racial/ethnic groups, but that impact has not changed over the period 1996–2013. (Note that this is also true for schools with more than 50 percent of students who are FRPL-eligible.) We do not find clear evidence that either the black-white or the Hispanic-white achievement gap is increasing more among those students who attend higher- versus lower-poverty schools, or among those who attend schools with higher concentrations of black plus Hispanic students versus those students who do not attend such schools.
Our results are also inconclusive about changes in the achievement gap between higher- and lower-income students. We find that changes in the gap vary by subject and grade. Our data are limited to measuring the gap between students who are “not poor,” “somewhat poor,” and “very poor,” but not between students at the top of the income distribution and low- and middle-income students. The divisions used in this analysis are useful for testing differences in student achievement between middle-income and lower-income students, but not between the very highest-income students and those in the rest of the income distribution. It is at the very top of the income distribution (the top 10 percent) where analysts have found student achievement rising compared to everyone else.
In terms of gender differences in performance, the advantage of male students over females in mathematics decreased, as did female students’ advantage over their male peers in reading, when controlling for race/ethnicity and social class. The gender gaps are now small compared to race/ethnicity differences, but are still significant.
We argue that these patterns of change (or lack of change) have important implications for what is happening in U.S. schools and American society.
The decline in the gap between whites and Hispanic non-English language learners may help explain the sense among white workers in lower socioeconomic levels that Hispanics are increasingly competing for their jobs. Although the 2016 presidential campaign has put the focus on undocumented immigrants, the real issue may be that there are increasing numbers of second- and third-generation Hispanic Americans with achievement levels similar to those of whites when adjustments are made for socioeconomic background.
Among high-achieving students competing for places in elite universities, the major increase in Asian students’ achievement relative to whites’ (and everyone else’s), especially in mathematics, has probably increased the pressure on upper-middle-class white families to invest even more in their children’s tutoring and outside-of-school academic activities. The percentage of Asian students in the top-25 U.S. universities (as defined by U.S. News and World Report) reached 21 percent of the undergraduate student body in 2007, and has remained at that level. Over the same period, the percentage of whites in those universities fell from 48 percent to 43 percent. As Reardon (2011) argues, increasing inequality in incomes over the past three decades seemed to be a major driver of the widening achievement gap between pupils from the highest 10 percent-income families and everyone else (note that very recent research by Reardon, Waldfogel, and Bassok (2016) indicates that this trend may have been reversed in the last decade). However, another explanation could be the fact that higher-scoring Asians constitute an increasing proportion of high-income Americans, and (non-Asian) high-income Americans are increasingly forced to respond to the reality of academic competition from this group for admission to elite universities.
The significant increase, in the 2003-2013 period, of students who attend high-poverty schools appears to have had a negative impact on the achievement gains of all groups of race/ethnic students, particularly whites, blacks, and Asians, in math as well as reading and in both the fourth and eighth grades. Concentration of low social class (and black and Hispanic students) is likely to be significantly reducing math and reading gains in U.S. schools across all states.
Finally, although English language learner designation is not an innate characteristic but one that can disappear as the student becomes proficient in the language, English language ability and usage may nevertheless reinforce race/ethnic and social-class identities and stigma. This, in turn, can make the ELL designation a “feature” that carries some of the same negative/positive aspects of academic expectations and treatment as does race/ethnicity and social class. This suggests further exploration may be needed: to see if the fact that Hispanic ELL and Asian ELL students are falling further behind white students is the result of changing rules for assigning students to ELL classes or a decline in the quality of teaching in ELL classes. It also suggests that we should pay greater attention to the widening achievement gap between students who are and are not on the English language learner track and that we should improve our understanding of the effectiveness of dual-language programs that serve minority students who need them to keep progressing in school.
Introduction
A founding ideal of American democracy is that merit, not accident of birth, should determine individuals’ income and social status. Schools have assumed a major role in judging key elements of merit among young people—namely, academic skills, hard work, self-discipline, and cooperative behavior. Schools do so mainly by evaluating students in a variety of subjects deemed important for success later in life. No one expects outcomes at the end of the schooling process to be the same for every student, since initial ability varies, and some young people are more disciplined and willing to work harder in school than others. Yet, when students’ inherent characteristics—such as race, gender, or parents’ economic and social capital—rather than their innate ability, hard work, and discipline systematically affect their school outcomes, this threatens democratic ideals.
Analysts have studied persistent gaps in U.S. student achievement—particularly between blacks and whites, Hispanics and whites, and different social-class groups—for many decades (Coleman, Campbell, and Hobson 1966; Jencks and Phillips 1998; Fryer and Levitt 2004, 2006; Rothstein 2005; Card and Rothstein 2007; Reardon and Galindo 2009; Reardon 2011; Reardon, Robinson-Cimpian, and Weathers 2015; see Musu-Gillette et al. 2016 for a recent review of education indicators by race/ethnicity). They have also examined achievement gaps between boys and girls (for a review, see Robinson and Lubienski 2011). Considerable evidence exists that race continues to be an important factor in explaining achievement differences. However, social-class differences account for much of the black-white and Hispanic-white achievement gap (Reardon, Robinson-Cimpian, and Weathers 2015). Disadvantaged minority children, such as African-Americans and Hispanics, are much more likely to be poor than are white children (DeNavas-Walt and Proctor 2015). Furthermore, there are new questions about whether race and social class interact with gender, resulting in a particularly deleterious effect on the academic performance of disadvantaged minority boys (Gregory, Skiba, and Noguera 2010), and whether school conditions have a greater effect on boys or girls (Autor et al. 2016).
Black and Hispanic children are also more likely than whites or Asian-Americans to live in low-income, racially segregated neighborhoods and to attend schools with high concentrations of low-income black and Hispanic students. Part of the achievement gap between race/ethnicity and social-class groups appears to result from social class and racial spatial segregation, and from the concentration of student populations by socioeconomic group and race/ethnicity in different schools (Coleman, Campbell, and Hobson 1966; Hanushek, Kain, and Rivkin 2002). An important issue is the way in which changes in U.S. demographics and racial/ethnic segregation in schools contribute to changes in the achievement gaps between whites and racial and ethnic minorities (Reardon and Yun 2001; Orfield et al. 2014).
The good news in the literature is that achievement differences between blacks and whites and between Hispanics and whites have apparently declined over time (Jencks and Phillips 1998; Reardon and Galindo 2009; Rothstein 2013; Reardon, Robinson-Cimpian, and Weathers 2015). The bad news is that until recently the achievement gap between higher- and lower-social class groups appeared to be increasing, particularly between the children in the highest-income group and everyone else (Reardon 2011; Putnam 2015). This may, however, have reversed somewhat in the first decade of the 21st century (Reardon, Waldfogel, and Bassok 2016).
This paper advances the discussion of these issues by analyzing trends in how race/ethnicity, social class, and gender relate to academic performance in U.S. schools. Our focus is on different grades and different subjects (mathematics and reading) over the past 10 years and on mathematics since the mid-1990s. We analyze changing achievement gaps between students of different race/ethnic identification, gender, social class, and English language-ability designation in the fourth and eighth grades over the past decade and a half, and how sensitive these gaps are to school composition in terms of the proportion of poor or minority peers. Many of these achievement gaps develop well before entry into school (Lee and Burkam 2002; García 2015) and, on average, continue—or get larger as students progress in school (because those who start out behind academically are more likely to attend schools with fewer resources, which may compound, instead of compensate for, initial disadvantages). Lower-income families are also less able and less likely to invest in academically enriching activities for their children outside of school. The student achievement scores we estimate in the fourth and eighth grade reflect these many influences.
We use individual student microdata from the National Assessment of Educational Progress (NAEP) to estimate the math and reading performance of students in fourth and eighth grade from 2003 to 2013, as well as students’ performance in eighth-grade mathematics only from 1996 to 2013.
Beyond the crucial philosophical role that equality of opportunity plays in the American identity, why is it important to study school achievement? Is there a significant relationship between achievement and economic and social outcomes? The answer is both obvious and complex. With regard to the obvious, higher test scores are associated with a greater probability of completing high school, and attending and completing a four-year college. Higher levels of school attainment, in turn, are strongly related to improved life outcomes, including but not limited to higher earnings (Belfield and Levin 2007; Alexander, Entwisle, and Oslon 2007). For example, interventions such as increasing the academic activities of low-income youth in summer could have a large enough effect on their test scores to significantly increase the probability that they will attend a four-year college (Alexander, Entwisle, and Oslon 2007). Yet, with regard to the complex, when we account for individuals’ level of education, the effect of student achievement (as measured by test scores) on economic and social outcomes is much smaller than generally assumed (Bowles and Gintis 1975; Murnane, Willett, and Levy 1995; Castex and Dechter 2014; Balart, Oosterveen, and Webbink 2015). Murnane, Willett, and Levy (1995), and Bowles, Gintis, and Osbourne (2001) estimate that, controlling for other factors—including social class—a very large increase of one standard deviation of a test score (about 34 percentage points on a 100 point scale) is associated with (only) a 9 to 10 percent increase in wages.
With this in mind, we consider that achievement differences between groups do have economic and social meaning; they give us insights into how well our school systems and society are adapting to demographic, social, and political changes. In other words, we have a system where higher scores produce better outcomes but we have a labor market that seems to reward higher test scores much less than the political rhetoric would have us believe.
Our study has six objectives:
To describe changes in the racial characteristics and socioeconomic status (SES) of the student population, and in the composition of student bodies in U.S. schools over the past two decades in the periods 1996–2003 and 2003–2013
To describe the types of schools (high- and low-poverty, high and low concentrations of blacks plus Hispanics) that black, Hispanic, white, and Asian children attend and how these have changed over the past 10 and 20 years
To estimate changes in students’ achievement gaps by social class and race/ethnicity—including gaps for students designated as English language learners (ELLs)—over the past 10 and 20 years
To estimate changes over the past decade in the influence of school composition—such as concentration of students by poverty, race, and ethnic status—on students’ social class and race/ethnicity achievement gaps
To estimate whether and how much the trajectories of social class and race/ethnicity achievement gaps differed over the past 10 years for male and female students
To estimate whether and how much these trajectories differed over the past 10 years for lower-achieving students and higher-achieving students
Our unique approach using NAEP individual microdata over a 10- to 17-year period allows us to estimate changes in race/ethnicity effects, controlling for SES, the designation of English language learner (ELL), and gender, and changes in SES effects, controlling for race/ethnicity, ELL, and gender.4 In addition, we estimate changes in race/ethnic achievement gaps, controlling for state achievement differences (see Carnoy, García, and Khavenson 2015 for a discussion). Since Hispanics and particularly Asians are more concentrated in certain states that may perform lower, on average, than others, controlling for state fixed effects may influence estimates of ethnic achievement differences. We also explore how interactions between racial/ethnic and socioeconomic traits of U.S. students and the characteristics of the schools these students attend relate to student outcomes. We repeat these procedures for male and female students separately to test whether gender differences are important in regard to how race and social class influence student performance. Finally, we estimate tercile regressions to assess whether changes in race and class achievement gaps vary across student achievement levels.
The results yield important insights into the changing nature of inequality in the U.S. education system. We confirm earlier studies showing that although the black-white test score gap remains large, it is gradually declining (Hedges and Nowell 1999; Magnuson, Rosenbaum, and Waldfogel 2008). The gap in test scores between Hispanics and whites also continues to decline for non-English language learners (ELL) students, as does the negative female-male gap in math and the positive female-male gap in reading.
Because we use individual student data for our analysis, we can identify whether students have been assigned to the English language learner track or not. Thus, we can address the role that language status plays in the trajectory of ethnicity achievement gaps (Nores and Barnett 2014; Nores and García 2014). We find large and somewhat increasing gaps in mathematics and reading achievement between Hispanic and Asian ELL students and other groups, including whites and non-ELL Hispanics and Asians. This is in stark contrast to the achievement gap between white students and non-ELL Hispanic students, which decreased substantially from 2003 to 2013, and the achievement gap between whites and Asian non-ELLs, which increased substantially during this period.
It is important to note that English language designation is not an innate characteristic, but one that can change as the student becomes proficient in English. Therefore, it does not have the same meaning as race, ethnicity, gender, and some elements of social class. Nevertheless, English language ability and usage can reinforce race/ethnic and social class identities and stigma, which can make ELL designation a student “feature” that carries some of the same negative/positive aspects of academic expectations and treatment as race/ethnicity and social class. This suggests that we should be paying much more attention to the “language gap” in U.S. education for the about 9 percent of ELL students in fourth grade (of which 76 percent are Hispanic students and 10 percent Asian students) and the about 5 percent in eighth grade (of which 70 percent are Hispanics and 13 percent Asians).5 Similarly, it is important to improve our understanding of the effectiveness of dual-language programs that serve minority students who need them to continue their progress in school.
We are not able to confirm that the achievement gap is unequivocally increasing between students from high- and low-social class families; changes in the gap vary by subject and grade. However, our data are limited to measuring the gap between students who are “not poor,” “somewhat poor,” and “very poor.”6 These are reasonable measures for testing differences in student achievement between middle- and lower-income students, but not between the very highest-income students and students at the rest of the income-distribution levels. It is only at the very top of the income distribution (top 10 percent) where analysts have found student achievement rising compared to everyone else (Reardon 2011).
Further, we show that as the overall proportion of poor students in schools increased from 2003 to 2013, the percentage of both black and Hispanic students in high-poverty schools rose substantially. We also show that Hispanic students are as or more likely than white students to attend high-poverty schools (García and Weiss 2014). We find inconsistent evidence that the achievement gaps for blacks and for Hispanics in high-poverty schools are increasing. However, we do find that among black and Hispanic students, the achievement gap are increasing between those blacks and Hispanics who attend schools with a high concentration of black plus Hispanic students versus those who do not.
We suggest that the increase in Hispanic and Asian non-ELLs in the U.S. school population, combined with the declining achievement gap of Hispanic non-ELLs and the increasing Asian non-ELL gap, may help explain broader social phenomena. One is the seeming increase in opposition to Hispanic immigration among less-educated whites, which some speculate is rooted in shifts from an industrial to a service economy. But these anti-immigration sentiments could conceivably be fueled by the rising school performance and labor competitiveness of non-ELL second- and third-generation Hispanics, as well as their growing numbers. Another phenomenon is increased pressure on some groups to invest more in raising their children’s test scores. The pressure may be associated with the growing gap between whites and Asian non-ELLs, and it is probably greatest on higher-income—and higher-scoring—whites “competing” in the battle for places in elite colleges and high-income jobs. This pressure to increase test scores may help explain why the gaps in test scores among different social classes were expanding until recently (Reardon 2011; Reardon, Waldfogel, and Bassok 2016).
The paper is divided as follows: In the next section, we present our estimation strategy. The section after that presents the results, and the final section discusses the results and draws conclusions.
How do we estimate achievement gaps by race/ethnicity and social class over time?
The United States’ national assessment test, the National Assessment of Educational Progress (NAEP), is the main data source for trends in mathematics and reading achievement for individual students with different characteristics, for schools with different student populations, and, since 1992, for sufficiently consistent samples of students and schools within states. NAEP assessments are administered uniformly using the same set of test booklets across the nation, and the assessment stays essentially the same from year to year and includes only carefully documented changes. This permits the NAEP to provide a clear picture of student academic progress over time. We use the “Main NAEP” microdata rather than the “Long-Term Trend NAEP” because the Main NAEP provides data on students’ eligibility for free or reduced-price lunch dating back to 1996 and on individual states, is grade specific, and was used more often in the shorter period of interest to us (1996–2013). The NAEP is applied to students in specific grades (fourth, eighth, and twelfth) to obtain a stratified random sample of schools in each state.7 We focus on the fourth- and eighth-grade results in mathematics and reading in the period 2003–2013, when all states were required to participate in the NAEP, and extend our analysis back to 1996 for eighth- grade mathematics.8
We use ordinary least squares to estimate a series of step-wise models of student achievement in mathematics and reading in fourth and eighth grade as a function of (1) gender; (2) race/ethnicity; (3) whether the student is designated an English language learner; (4) parental education; (5) eligibility for free or reduced-price lunch; (6) whether a student is enrolled in an individual education plan (special education); (7) the percentage of students eligible for FRPL in the school the student attends; and (8) the total percent of black plus Hispanic students in the school the student attends. We also estimate OLS regressions that include interactions of student race/ethnicity and eligibility for free or reduced-price lunch with the percentage of students eligible for free lunch and the racial composition of the school the student attends.
We know that average test scores can differ among states for reasons that are unrelated to individual characteristics of students or the demographic composition of schools in the state. For example, some states have performed considerably better on the NAEP than others, even when adjusting for student and school demographics (Carnoy, García, and Khavenson 2015). States may have more- or less-effective educational systems that can affect all groups’ test scores positively or negatively. If minority or lower-SES groups are not randomly distributed across states, this could influence relative test scores over time in ways that are not related to either ethnicity or social class. Since accounting for state differences can therefore be important to estimating SES and race/ethnicity effects on achievement, we include an estimate that adds state fixed effects.9
The complete model for each year, subject, and two grades of the NAEP test can be represented as follows:
Aij = a + bXi + FRPLj + gMj + dXi * FRPLj + hXi * Mj + eij (1a)
Aijs = a + bXi + FRPLj + gMj + dXi * FRPLj + hXi * Mj + lStates + eijs (1b)
Where Aij is achievement on NAEP mathematics or reading tests of student i in the fourth or eighth grade in school j;10 Xi is a vector of student characteristics; FRPLj is the percent of students eligible for FRPL in school j; Mj is the percent of black plus Hispanic students in school j; Xi*FRPLj and Xi*Mj are interactions of student characteristics with school composition variables; and eij is an error term. In equation 1b, Aijs is achievement on NAEP mathematics or reading tests of student i in the fourth or eighth grade in school j in state s. States are state dummies.
As mentioned above, equations 1a and 1b are estimated following a step-wise procedure. Model I includes only student characteristics as independent variables; Model II (“without state fixed effects”) estimates the b’s, including controls for school FRPL and minority composition; Model II (“with state fixed effects”) estimates the b’s, including controls for school FRPL and minority composition plus state fixed effects; and Model III is the complete model, including interaction variables without state fixed effects.
In this analysis, the parameters of interest are the b’s, d’s, and h’s for each of the years of the NAEP from 1996 to 2013. These can be used to trace the trajectories of relationships between students’ race and ethnicity, SES, and the interactions of these individual characteristics with school composition. We posit that the changes in these parameters of interest represent an approximate estimate of the changing minority/white and poor/not poor student achievement gap over time, as well as a measure of how changes in school composition over time may influence the achievement gaps of particular groups of students. To test for heterogeneity of our parameter estimates by gender and student performance level, we also estimate equation (1a, without the state fixed effects) for male and female students separately and for terciles of student achievement.
We caution that these are not causal estimates. In the case of estimating the parameters of students’ characteristics, we are primarily concerned with adjusting for a number of variables that could influence race/ethnic and SES achievement gaps in order to understand how the parameters mentioned above influence student performance, and how they have changed over time.
Results
Two major trends characterized the student composition of U.S. elementary and secondary public schools between the mid-1990s and 2013. First, the proportion of non-Hispanic white and black students declined. This trend contrasted with the steady increase in the proportion of Hispanic and Asian students. Further, even with the decline in the proportion of black students, minority students of black or Hispanic origin increased greatly, from 30.0 percent to 40.5 percent.11 The second major trend was the rapid increase in the proportion of low-income students (those eligible for free or reduced-price lunch). In 2013, students eligible for FRPL represented 52.0 percent of all public school students, up from 38.3 percent in 2000.12
The shares of students who fall into various racial/ethnic and FRPL categories are somewhat different in the NAEP fourth- and eighth-grade samples than in the K-12 public education system, due mainly to demographic trends that increase the proportion of Hispanics in lower grades (K–3). The proportion of whites in the NAEP samples also tends to be higher: In addition to sampling students in public schools, the NAEP samples students in private schools, where students are much more likely to be white and less likely to be classified eligible for FRPL. Thus, in the NAEP eighth-grade math sample, white students in 2013 were about 54.2 percent of the total; in the fourth-grade math sample, 52.8 percent.13 Nevertheless, the trends for the NAEP eighth- and fourth-grade samples over time are similar to those trends in the public school system as a whole, in terms of both racial composition and proportion of students eligible for FRPL (see Tables 1 and 2). Because of the rapid increase of Hispanics in the younger population, the proportion of Hispanic students is higher in the fourth than in the eighth grade.
Table 1
Table 2
Segregation by class and race
We found that largely because of the increases in the percent of students eligible for free or reduced-price lunch, the proportion of fourth-graders in schools with more than 50 percent of students eligible for FRPL also increased—from 41.4 percent in 2003 to 54.4 percent in 2013, and in eighth grade, from 32.0 percent to 48.3 percent. The proportion of fourth-graders attending schools that were more than 25 percent minority also increased, from 48.2 to 58.3 percent and, in eighth grade, from 44.4 to 57.1 percent.14
The divisions across race/ethnicity, socioeconomic status (as measured by the degree of poverty), and language learner status groups described below (see Tables 3a, 3b, 3c, and 4) highlight two characteristics of class divisions in U.S. schools. The first characteristic is that a much higher fraction of black and Hispanic students attend high-poverty schools than white or Asian students. The second characteristic is that black and Hispanic students are much more likely to attend high-poverty schools even when they are not poor; i.e., black and Hispanic students who are not poor are still more likely to attend schools that have large proportions of poor students than are white or Asian students (generally, see details below).
Poor students, and black and Hispanic students, were much more likely to attend a school with a high percentage of students eligible for FRPL, and to attend a school with a high percentage of black and Hispanic students (see Tables 3a, 3b, and 3c, and Table 4, year 2013 panels). For example, in 2013 (Table 3a), only 6.1 percent of the economically more advantaged eighth-grade students in the NAEP math sample in 2013 (that is, those who were not eligible for FRPL) attended a school where more than 75 percent of students were FRPL eligible. Only 25.8 percent of the more advantaged group attended a school where more than 50 percent of students were eligible for free or reduced-price lunch. A much larger proportion of students who were not eligible for FRPL (39.4 percent) attended a school with 25 percent or less students eligible for FRPL. At the other extreme, 39.5 percent of those students eligible for free lunch attended a school where more than 75 percent of students were eligible for free or reduced-price lunch, and 71.2 percent of students eligible for free lunch attended a school where more than half the students were FRPL-eligible. Thus, a high proportion of poor students attend schools with other poor students, and a high proportion of students who are not poor attend schools with relatively few poor students.
When we look at students by race/ethnicity and language status (Table 3b), we see that over 40 percent of black and Hispanic eighth-grade math students in 2013 attended a school with more than 75 percent FRPL-eligible students (43.5 percent of black students, 40.3 percent of Hispanic non-English language learners, and 55.8 percent of Hispanic ELLs attended such a school). In contrast, only 12.0 percent of Asian non-ELLs and 29.8 percent of Asian ELLs attended a school where more than 75 percent of students were eligible for free or reduced-price lunch. An even lower 6.9 percent of white students attended such a high-poverty school.
Further, a very low 3.2 percent of advantaged white students (those ineligible for FRPL) attended a high-poverty school (a school where over 75 percent of students were FRPL eligible) in 2013 (see Table 3c); and one of six (16.0 percent) poor white students (those eligible for free lunch) attended a high-poverty school. In contrast, among advantaged black students, 20.7 percent attended a high-poverty school, while more than one-half (52.5 percent) of poor black students attended a high-poverty school. That is, poor black students were three times as likely to attend a high-poverty school as poor white students, and non-poor (or advantaged) blacks were more than six times as likely to attend a high-poverty school as non-poor whites.
Advantaged Hispanic non-ELL students were less likely than advantaged black students to attend a high-poverty school, but much more likely than advantaged white students to attend high-poverty schools (15.1 percent of Hispanic non-ELL students were in high-poverty schools). Also, a slightly lower proportion of poor Hispanic non-ELLs (51.1 percent) attended a high-poverty school than did black students (52.5 percent), and both these shares were far greater than whites’ proportion (16.0 percent). However, relative to white students of similar poverty levels, a much higher proportion of both advantaged and poor Hispanic ELL students attended a high-poverty school: 33.9 percent of non-poor and 59.1 of poor Hispanic ELLs attended high-poverty schools. Among non-poor students, Hispanic ELL students were twice as likely as Hispanic non-ELLs, and 10 times as likely as whites, to attend high-poverty schools.
On the other hand, advantaged Asian non-ELL students were more likely than advantaged whites to attend very low-poverty schools (schools where 10 percent or less of students were eligible for free or reduced-price lunch). The share of non-poor Asian non-ELL students attending very low-poverty schools was 27.4 percent compared with 19.0 percent of non-poor whites in such schools, although poor Asian non-ELL students were much more likely than poor white students to attend a high-poverty school (28.5 percent versus 16.0 percent).
Not surprisingly, U.S. schools are also racially segregated. This is particularly important because, as we show below, the proportion of blacks and Hispanics in the schools these students attend is negatively correlated with their individual achievement. In 2013 (see Table 4), a white eighth-grader (in the math sample) was 73.9 percent likely to attend a school with less than 25 percent black or Hispanic students. Yet, a black student or a Hispanic non-ELL student were, respectively, 13.8 percent and 14.8 percent likely to attend a school with less than 25 percent black or Hispanic students. In addition, black and Hispanic non-English language learners were about 43 percent likely to attend a school with 75 percent or more black or Hispanic students (42.8 percent and 43.5 percent, respectively). The figures for Hispanic ELL students were 9.0 percent in low-minority schools and 55.5 percent in high-minority schools. Asian ELL students had only a 13.0 percent likelihood of attending a school with more than 75 percent black or Hispanic students, and this percentage was even lower for Asian non-English language learner students (8.4 percent).
Changes in segregation by class and race
Our results show that in the first decade of the 21st century, there was a large increase in the percentage of students eligible for free or reduced-price lunch (our measure of student poverty). Largely because of this, the total percentage of students in schools with more than 75 percent poor students increased from 2003 to 2013. We also find that, following what had started in previous decades, there was a large increase in the proportion of Hispanic students, which raised the total percentage of schools with large shares of minority students. This to a certain degree expanded the concentration of blacks and Hispanics in schools with high concentrations of FRPL-eligible students, since these black and Hispanic students were also more likely to be poor than the average student.
The percentage of students eligible for free or reduced-price lunch increased from 2003 to 2013, from 39.6 percent to 52.0 percent in eighth-grade (math sample, see Table 1), and in fourth grade, from 47.0 percent to 56.0 percent (math sample, see Table 2). The total percentage of eighth-graders in schools with more than 75 percent FRPL students, for example, increased between 1996 and 2013, from 15.2 percent to 21.6 percent; all of that increase occurred after 2003.15, 16 The percentage of free lunch eligible students—the poorest students—attending schools with more than 75 percent FRPL students also increased in this period, but entirely before 2003. The main increases in the percentage of those attending schools with high percentages of FRPL students occurred for those less-poor students (eligible for reduced-price lunch) (Table 3a).
Table 3a
At the same time, the proportion of students attending a high-poverty school increased more overall for blacks and Hispanics than for whites and Asians (Table 3b). Table 3c shows changes over time in the proportion of students attending low- and high-poverty schools (as measured by the percentage of FRPL-eligible students in the schools’ student body) by race/ethnic group and individual students’ own level of poverty (as measured by eligibility for FRPL). It was not the proportion of the poorest blacks and Hispanics (those eligible for free lunch) attending a high-poverty school that increased; rather, the increase was highest among less-poor blacks and Hispanics (those eligible for a reduced-price lunch). Although less-poor Hispanics constitute a much smaller group than those eligible for free lunch, it is possible that for this less-poor group of blacks and Hispanics, the negative effect of being in a high-poverty school might be greater. We test this proposition in the analysis below.
Table 3b
Table 3c
We also find that as the percentage of black and Hispanic students increased from 1996 to 2013, the likelihood that students of all ethnic groups would attend schools with a high fraction of black and Hispanic students also increased. In percentage-point terms the increase was modest for white eighth-grade students (from 5.2 percent to 8.6 percent) attending a school with more than 50 percent blacks and Hispanics, but greater for blacks (from 51.8 percent to 64.2 percent), and for Hispanics (from 64.2 percent to 76.5 percent for ELLs, and from 60.3 percent to 66.1 percent for non-ELLs). The proportion of Asians attending schools with more than 50 percent Hispanics or blacks also increased (from 25.7 percent to 37.5 percent for ELL students, and from 19.3 percent to 23.1 percent for non-ELLs (Table 4).
Thus, Table 4 shows some evidence of a greater concentration of blacks and Hispanics in schools with high concentrations of black and Hispanic students, particularly between 1996 and 2003. Equally important, during this entire period, the differences in the racial/ethnic composition of schools that whites attend and that blacks and Hispanics attend remained vastly different. Even Asian students attended schools that were likely to have a higher fraction of black and Hispanic students than those attended by whites.
Table 4
Changes in race/ethnicity, gender, English language-ability designation, and social-class achievement gaps in reading and math, 2003–2013 and in eighth-grade math, 1996–2013
Our main findings on changes in student achievement during this period are that the black-white and the non-ELL Hispanic-white achievement gaps fell in the late 1990s and the first decade of the 2000s, while the non-ELL Asian-white gap (in favor of Asians) increased substantially. This was not the case for Hispanic English language learners and Asian English language learners, as the large negative gap between white students and both groups increased during this period. We also find that the social-class achievement gap between students from poor and non-poor families decreased in the 1990s, but then increased somewhat in the 2000s. These trends were generally the same for both fourth- and eighth-graders.
Figure A shows the trends in the average eighth-grade NAEP mathematics scores of whites, blacks, Hispanics (ELL and non-ELL), and Asians (ELL and non-ELL), from 1996 to 2015, without controls for any other student or school characteristics/variables. The scores of Hispanic English language learners and blacks were much lower than those of whites. The scores of Asian ELLs and Hispanic non-ELLs were similar to one another and closer to but still below those of whites, while Asian non-ELLs’ scores were higher. In 2003, the white-black test score gap was -35.4 scale score points, equivalent to about -1 standard deviation (SD) (not shown in figure),17 the white-Hispanic ELL-gap was -51.2 points (about -1.4 SD), and the white-Hispanic non-ELL gap was -22.0 (about -0.6 SD). The white-Asian ELL gap was -26.6 points (about -0.8 SD), while the white-Asian non-ELL gap was 7.9 points (about 0.2 SD). For Asian non-ELLs, Hispanic non-ELLs, and blacks, groups, the NAEP scores relative to whites increased steadily from 2003 to 2013. For ELL Asian and Hispanic children, there was essentially no catch-up relative to whites. Yet, even in 2015, the black-white gap remained high, at -0.9 SD, and the white-Hispanic non-ELL gap was large at -0.4 SD, as was the white-Asian ELL gap, at -0.7 SD. The largest gap remained the white-Hispanic ELL gap, at about -1.3 SD, while the white-Asian non-ELL gap expanded to a 0.5 SD difference. It is important to remember that these racial/ethnic achievement gaps are not adjusted for any social-class differences or changes in social-class differences among groups.
Figure A
Table 5 shows the results of our regression estimates of race/ethnic coefficients in our main base year, 2003, controlling for students’ social class and other individual student characteristics.18 Black eighth-graders scored 0.71 standard deviations lower than their white counterparts in mathematics and 0.55 SDs lower in reading. Hispanic non-ELL students did better than black students (0.30 SD lower than whites in math and 0.27 SD lower in reading). Asian non-ELL students performed at the same level as white students in reading but higher in mathematics (0.17 SD). When compared with the unadjusted math gaps in Figure A, these adjusted math gaps suggest that about 20 percent to 25 percent of the unadjusted white-black and white-Asian non-ELL math gap, and about 50 percent of the white-Hispanic non-ELL math gap, in 2003 are explained by the student’s social class, gender, and special education designation, not by race/ethnicity.19
The one-fourth of Hispanic students classified as ELLs in eighth grade (math sample) in 2003 scored much lower, about one standard deviation below whites in both math and reading. Asian English language learners scored higher than blacks in math (0.59 SD lower than whites) but much lower in reading (0.92 SD lower than whites). Asian non-ELL students did not score significantly higher than whites in reading but scored 0.17 SD higher in math.20 These race/ethnic differences are large considering that we controlled for free or reduced-price lunch eligibility and parents’ education. As discussed earlier, these differences reflect a complex interaction among socioeconomic, “cultural,” language, and school factors. Noteworthy is the major role of language and the interaction between schooling and language (ELL designation) in school achievement. Whether Hispanic or Asian, English language learners scored lower in both math and reading, and the results are similar for fourth grade, where English language learner designation is more common.
Controlling for student race/ethnicity, gender, and whether a student was in special education, eighth-grade students eligible for free lunch scored 0.46 standard deviations lower in math in 2003 and 0.41 SD lower in reading than students not eligible. For students eligible for reduced-price lunch, the gap was about one-half that in both subjects. The gaps were larger for students eligible for free or reduced-price lunch in fourth-grade. As was the case with race/ethnicity, students’ poverty status was closely associated with their academic achievement. When race/ethnicity and poverty are combined, the effect is enormous. In 2003, black students eligible for free lunch (in poverty) scored about 1.2 SDs lower in eighth-grade math and about 1 SD lower in eighth-grade reading than white students not eligible for FRPL. The gap was even larger for poor Hispanic students designated ELLs.
Table 5
How did these gaps change over the decade of the 2000s, a period marked by an increasing proportion of fourth- and eighth-graders who are poor and Hispanic, and a period in which all ethnic groups—particularly Hispanics—trend toward attending schools with higher concentrations of low-income and minority (Hispanic plus black) students?
The patterns over time of black-white, Hispanic-white, and Asian-white achievement gaps for eighth-grade mathematics and reading scores are shown in Tables 6a and 6b. Model I estimates the race/ethnicity achievement gaps adjusting for student’s gender, whether a student is in an individualized education plan (special education), parents’ education, and whether a student is eligible for free or reduced-price lunch. Model II estimates the achievement gaps for the Model I variables, plus the percentage of students eligible for FRPL and the percentage of black and Hispanic students in the school each individual student attends. Model III adds the interactions of individual student FRPL eligibility with overall school FRPL, student race/ethnicity with school FRPL, and student race/ethnicity with the percentage of Hispanics and blacks in the school attended by the student. In all models for math performance, we offer estimates with and without state fixed effects.
The results show that in all three estimated models, the adjusted black-white achievement gap and the achievement gap between whites and Hispanic non-ELLs in eighth grade decreased from 2003 to 2013, and so did the black-white reading gap, though the decline was much smaller proportionately. For blacks, the math gap closes steadily over the 10 years, but for Hispanic non-English language learners, almost all the change in the math gap occurred from 2007 to 2013. One reason that the white-Hispanic non-ELL gap declined is that, across states, requirements for reassignment from ELL to non-ELL might have become more stringent over time. In that case, the Hispanic non-ELL group would have become more “exclusive” and, therefore, the smaller achievement gap may represent merely a change of membership in the group. However, if the requirements had changed, the gap between whites and Hispanics designated ELLs would have also decreased, as a result of improved test-taking capacity of the English language le