Introduction and executive summary
Inequalities in education outcomes such as test scores or degree attainment have been at the center of education policy debates for decades. Indeed, the first seminal national report on the state of U.S. education—the 1966 Coleman Report—examined some of these inequalities 50 years ago. Since then, researchers have examined performance gaps by income level and race or ethnicity in depth, as well as inequalities in educational attainment (degrees earned, etc.), employment opportunities, earnings, and even health status and overall well-being—all of which can be seen, partly, as long-lasting consequences of earlier education gaps (Altonji and Blank 1999; Cutler and Lleras-Muney 2010; Duncan and Murnane 2011a; Jencks and Phillips 1998; Magnuson and Waldfogel 2008; Morsy and Rothstein 2015; Rothstein 2004; Schultz 1980).
This study seeks to broaden the debate by examining the education gaps that exist even before children enter formal schooling in kindergarten, and showing that the gaps extend to noncognitive skills, which are also critical for adulthood outcomes (Heckman 2008; Heckman & Kautz 2012). Regarding the analysis of early education gaps, this paper is modeled on Lee and Burkam’s 2002 monograph Inequality at the Starting Gate: Social Background Differences in Achievement as Children Begin School, which found that cognitive gaps between children of different socioeconomic backgrounds and races and ethnicities were both sizeable and statistically significant at school entry in kindergarten.1 This is important for policymakers because, if unaddressed, there is the potential that gaps persist over time and compound. Such early-in-life inequalities point to the need for substantial interventions to reduce them, including early educational interventions, to ensure that children arrive in kindergarten ready to learn and for compensatory policies to support these children throughout the school years (from kindergarten through 12th grade). Moreover, the social and economic disadvantages that generate these gaps should be addressed directly and eliminated through social and economic policies, not just education policies (Morsy and Rothstein 2015; Putman 2015; Rothstein 2004).
Related feature
View summary
Get the summary and infographics
Visit Early Education Gaps by Social Class and Race for a summary of this report’s major findings, and infographics on skill gaps in the first year of school.
Newly available data on kindergartners in the 2010–2011 school year allow us to examine the presence of education gaps for a recent cohort of children at their educational “starting gate,” their kindergarten year. Given the different conditions in which, relative to earlier cohorts, today’s young children have spent their early years, we might expect that gaps among groups in the recent cohort would be different. Presumably, today’s kindergartners would have benefited from a decade in which parents, practitioners, policymakers, and researchers actively sought new ways to boost young children’s educational experiences (Kagan and Kauerz 2012; Pianta, Cox, and Snow 2007). On the one hand, the newest generation of students potentially started school in a much better shape than the earlier cohort, as they were exposed, on average, to several welfare and education interventions designed to improve their school readiness and skills (such as expanded prekindergarten school, among others). On the other hand, students in 1998 entered school in years of prosperity, while the economy and context for this more recent group of children and their families has been characterized by economic stagnation and high rates of unemployment (Mishel et al. 2012). In addition to these differences, there have been demographic shifts as the proportions of low-income, immigrant, and minority individuals increased.2
Whether and how these dynamics have changed education inequalities is the focus of this study. Using recent data from a younger cohort of kindergarten students—the National Center for Education Statistics’ Early Childhood Longitudinal Study, Kindergarten class of 2010–2011 (hereafter, ECLS-K 2010–2011 NCES), this paper delineates an updated picture of education inequalities among our youngest children in school. We produce a comprehensive analysis of gaps in both cognitive and noncognitive skills among this cohort of children.3 We conclude with a discussion of the research and policy implications of these findings.
Following are the major findings of this report:
Inequalities based on socioeconomic status (SES) are very significant. Cognitive and noncognitive skills are least developed among those with the lowest socioeconomic status and sharply increase as one ascends the socioeconomic ladder, as these examples show:
The relative advantage of a child in the top fifth of the SES distribution (referred to in this report as “high SES”) relative to a child in the bottom fifth (“low SES”) is of 0.8 standard deviations in reading and math, and 0.4 standard deviations in persistence in completing tasks.
Middle-socioeconomic-status children have a relative disadvantage with respect to children in the top SES fifth of about 0.4 standard deviations in the cognitive skills, and almost 0.25 standard deviations in persistence in completing tasks.
There are statistically significant education inequalities by race and ethnicity before accounting for the circumstances in which children live (i.e., their social class). After these factors are taken into consideration, race-based gaps shrink (and even vanish, in some cases). Importantly, this supports other evidence that education gaps are driven by socioeconomic differences (i.e., racial gaps reflect that racial minorities have lower socioeconomic status).
For cognitive skills such as reading, when controlling for socioeconomic differences, the only group that shows a highly significant inequality compared with whites is the Hispanic ELL (English Language Learner) group. Black children’s disadvantage relative to whites is marginally significant, and small. Hispanic non-ELL children are statistically equivalent to white children, while Asian children are ahead of white children by 0.4 standard deviations. In math and some of the executive function skills, these gaps are larger.
Children’s reported levels of noncognitive skills differ significantly depending on the race and ethnicity of the parent and whether parents or teachers are doing the assessments. For example, Hispanic ELL or Asian parents’ assessments of their children’s approaches to learning are lower than white parents’ (about 0.2 standard deviations below), but teachers’ assessments of these two groups’ approaches to learning do not confer any relative advantage to white children (gaps are about 0.1 standard deviations, but statistically insignificant). The opposite can be seen among black parents’ and teachers’ assessments of black children: black parents’ assessments of their children’s approaches to learning are equal to white parents’ assessments, but teachers confer on black children an average disadvantage of 0.2 standard deviations relative to whites in this skill.
It is important to note that unadjusted skills gaps (not controlling for socioeconomic status, family characteristics, and other variables) by race relative to white children are highly statistically significant, especially for black and Hispanic ELL children. Those gaps, and not the adjusted gaps, indicate the degree of disadvantage with which black and Hispanic ELL children start school in reality. However, the adjusted results show that it is the factors that are highly correlated with race that drive the racial gaps (Ladd 2012; Rothstein 2004). If instead of race we could control for all that race mediates for, on average, adjusted gaps by race would shrink and/or become statistically insignificant. This implies that if we actually lived in a race-neutral economy these unadjusted racial gaps would be absent.
Analysis of education gaps by gender at the starting gate lead to two conclusions: that any preexisting cognitive gap between girls and boys when they enter school is very small (with a slight relative advantage of girls in reading and a slight relative advantage of boys in math—0.07 standard deviations in each case); and that girls’ noncognitive skills are noticeably superior to boys, as rated by teachers and parents alike (but with teachers’ assessments manifesting more pronounced differences between boys and girls).
Our analytic approach confirms that the following must be considered when designing policies and programs:
We need to be more discerning when looking at children’s needs by subgroup. To effectively identify the performance and needs of groups of children that are highly heterogeneous in themselves analyses must first group them by common underlying characteristics and put them into more homogenous subgroups. Such targeted analysis is especially essential in light of minorities’ increasing share of the U.S. population, their heterogeneity, and the concentration of disadvantages many face. For example, among Hispanics, focusing on subgroups of ELL or non-ELL children can help us better understand their performance relative to other groups and their different needs.
We need to look holistically at what matters for children’s development, in terms of the outcomes (cognitive skills and noncognitive skills) and agents involved in the process (children, teachers, and parents). Having this broader understanding will shed light on the real opportunities that children have been given—and the needs that they have—as they cross the school’s starting gate.
Our findings have important implications for policy:
The existence of significant education inequalities at the starting gate poses a strong challenge to education policy. Programs and policies must account for the fact that schools and teachers serve students who do not start school on equal terms. Many students haven’t participated in preschool education and care, nor have they engaged in equal amounts of developmental and play time with adults. Not only are children unequally prepared to learn when they enter school, but, as research shows, their chances of attending unequally resourced schools are high, as they are much more likely to share school with children who face the same circumstances (Adamson and Darling-Hammond 2012; Baker and Corcoran 2012; García and Weiss 2014; Rothstein 2014). In short education policies must grapple with the relative disadvantages that many children face—disadvantages that are concentrated and compounded, and accumulate over time.
Whether a child is faring better or worse than her peers is largely determined by her socioeconomic status. The high poverty levels among the 2010–2011 kindergarten class (a quarter of children live in poverty) and significant income disparities by race and ethnicity (close to two thirds of black and Hispanic ELL children live in poverty) call for critical policy attention to the effect of these inequalities on outcomes—on the real opportunities given to children. (In this data set, poverty is defined as having a household income at or below 200 percent of the federal poverty line.)
Education policy alone is unlikely to suffice. Because severe education inequalities develop before children reach school, addressing these inequalities cannot be left to education policy alone. Interventions need to include wider economic and social policies to tackle the socioeconomic disadvantages that constrict opportunities before children even reach the school starting gate. These broader policies would include strategies to make poor children less poor (including employment, criminal justice, immigration, health, and housing policies); early educational interventions and programs to boost parents’ capacity to provide educational opportunities at home; and compensatory policies integrated throughout school (from kindergarten through 12th grade) to offset children’s disadvantages at school entry.
This study first reviews the existing literature analyzing education inequalities. The technical details of the analysis are explained in the “Dataset and methodology” section and appendix A. In the fourth section, we describe the characteristics of the 2010–2011 kindergarten class, in terms of race, socioeconomic status, and the other determinants of gaps that are examined in the study. In the results section, we discuss current socioeconomic-based and race-based inequalities in cognitive and noncognitive skills of students at the beginning of their school life. The report concludes with a review of results and a discussion of the implications of the findings for both future research and policy.
Literature review
Research on achievement gaps can be grouped under three main topics: existence and persistence of gaps, mechanisms driving the gaps, and consequences of the gaps for subsequent learning and development.4 Major contributions in these three different areas are described below. This section also includes an explanation of the contribution of this paper to the broader literature and provides a justification of why a more comprehensive understanding of the gaps (one that includes assessment of cognitive and noncognitive gaps) could help advance more successful strategies to close them.
Existence and persistence of gaps
In the United States, the study of education inequalities has been largely associated with the study of education outcomes of whites relative to outcomes of minority groups, especially African Americans. The precedent for this research was the Equality of Educational Opportunity study requested by the Civil Rights Act of 1964 and conducted by James Coleman and colleagues (see Coleman et al. 1966), which assessed the differences between the resources or inputs available to minority students attending public schools and those available to white children and discussed the consequences of those differences in terms of outcomes.5 In response to extensive research demonstrating that the race-education performance link is not a direct association but rather an indirect relationship strongly mediated by income and other factors related to income, recent studies have focused more on income-based gaps, and less on those based on race or ethnicity (Duncan and Murnane 2014; Duncan and Magnuson 2011). Research on gaps in noncognitive skills between ethnic or economic groups is still scarce (Grissmer and Eiseman 2008; Nores and García 2014; Rothstein 2004).
Whether race- or income-based, multiple studies have documented substantial and persistent differences in performance among population subgroups. These works tend to agree that gaps originate early in life, persist over time, and are measurable throughout a person’s lifetime.6 On average, a black student’s academic score is about 75 percent the score of a white student, and the difference is visible among children as young as three or four years old (Jencks and Phillips 1998). Indeed, one of the main findings in Lee and Burkam’s 2002 report is that children from different racial and ethnic groups begin school on very unequal terms. For children starting kindergarten in 1998, math achievement was 21 percent lower for blacks than for whites, and 19 percent lower for Hispanics than for whites. These sizable race-based gaps are described in-depth in recent volumes edited by Magnuson and Waldfogel (2008) and Duncan and Murnane (2011a).
Along with a review of tentative explanations for the evolution of the gaps, Rothstein (2013) describes changes in the black-white gap since the 1970s. His study highlights a reduction of the gap among these groups, driven by increases in educational achievement among black students that are greater than increases of their white peers in the 1970s and in the early 2000s. In part, the relative convergence was explained by a relative improvement of black parents’ educational attainment and subsequent occupational status and income levels, and by reductions in family size (fewer children per family) over those decades (Grissmer et al. 1994; Rothstein 2013). Barton and Coley (2010) agree with the assessment, and highlight that the convergence stalled at the end of the 1980s. They review the impact of school-level policies (such as reductions in class size) or other public policy stimuli (such incentivizing neighborhood desegregation, etc.), which also partially contributed to the narrowing of the gap during the decades when this positive phenomenon occurred.7
At the same time, as noted above, education gaps by income (or, more broadly, socioeconomic status, which includes income and other indicators of education attainment, occupation status, or wealth or possessions) are increasingly noticeable. In Lee and Burkam’s study, cognitive achievement of children in the highest socioeconomic group is 60 percent higher than that of children in the lowest socioeconomic group, as measured by test scores (and cognitive skills are much less closely related to race/ethnicity after accounting for socioeconomic status). In terms of the evolution of the income gaps, a review of research on trends in education gaps by income gradient offers two complementary views. The intergenerational mobility approach (Reardon 2011) suggests that the academic achievement gap between children at the 90th and at the 10th percentiles of the income distribution increased in recent decades; Reardon estimated that the gap was between 30 and 40 percent larger among children born in 2001 than among children born in 1975.
Another perspective consists of studying intragenerational education inequalities, or how the gaps evolve over time for the same cohort of students. A recent study on performance gaps (see Nores and García 2014) examines student performance from kindergarten to 8th grade. The study divides Hispanic students into different subgroups depending on their knowledge of English and their immigration status, and finds that cognitive inequalities between white and some Hispanics subgroups—especially non–English language speaking Hispanic children—in reading and math achievement at the beginning of kindergarten significantly shrank over the school years (the opposite was true for Hispanic-immigrant children). With respect to noncognitive performance, the gaps are smaller overall and diminish over time, which suggests a small relative advantage of Hispanics versus whites in skills such as approaches to learning, internalizing and externalizing behavioral problems, and self-control (all reported by teachers). Besides the methodological contribution, this paper illustrates the importance of understanding the heterogeneity within certain groups (such as Hispanic children by their knowledge of English, for instance), in order to better disentangle which groups are relatively lagging behind and consequently, to better identify policies to address obstacles and needs (Waldfogel 2001).8
Causes or mechanisms driving gaps
In light of these sizeable achievement gaps, researchers and policymakers have concentrated their efforts on identifying the mechanisms that generate gaps at such early stages. As already mentioned, one of the factors most strongly correlated with achievement disparities among different groups of students is a child’s social class or socioeconomic status. As the empirical research shows, socioeconomic status affects achievement gaps in two (compounded) ways. First is the well-documented direct association between education outcomes and individual economic (dis)advantage, whereby low-SES status is associated with lower academic performance.9 Second is the indirect link between SES and outcomes through the statistically significant associations between economic (dis)advantage and multiple factors also related to education results (Ladd 2012; Rothstein 2004; Coley and Baker 2013). These factors include the environment in which a child grows up (neighborhood factors and family characteristics), a child’s participation in early childhood programs, the quality of those programs, and even the type and quantity of instructional and motivational activities that parents engage in with their children and that affect child development and school readiness. All of these associations are significant, and all help better explain the link between education inequalities and economic inequalities.
Indeed, a child’s early environment is one of the fundamental drivers of race- and SES-associated education gaps. Importantly, this attention to conditions in which children live has, for the most part, disproved the misleading theory that innate or genetic factors partly explain the gaps (or account for any unexplained part of them).10 Two important contributions in this area are Shonkoff and Phillips (2000) and related research, and Wilson (1978), and related research.11
Shonkoff and Phillips’ book From Neurons to Neighborhoods made widely accessible the explanations of how the environment influences human development, and how neurobiology research could contribute to documenting these relationships.12 The authors emphasize that “every aspect of early human development … is affected by the environments and experiences that are encountered in a cumulative fashion, beginning in the prenatal period and extending throughout the early childhood years” (Shonkoff and Phillips 2000, 6).
The book illustrates, in particular, how disparities in infants’ and toddlers’ experiences in out-of-the-home settings translate into large gaps in school readiness. For instance, less affluent parents have less access to information about the importance of children’s interactions with adults, less economic capacity to buy stimulating toys, and less time to go to museums (Phillips 2011). Moreover, these early disparities compound differences in children’s health and well-being at birth. As Janet Currie and her colleagues have documented, low-income mothers’ lack of access to health care during pregnancy, as well as other influences of their environment, increase their babies’ health risks (Currie and Goodman 2010; Currie and Almond 2011). In fact, Currie’s findings indicate that these health disparities at birth already predict some of the subsequent large education gaps (Currie 2011; 2009).
The second contribution to the acceptance of the early environment as a fundamental driver of race- and SES-based education gaps is a set of studies, beginning with those by William Julius Wilson, whose scope also goes beyond the limits of school walls. These studies note that children of certain minorities are more likely to live in concentrated poverty (Wilson 1978, 1987; Jargowsky 2013; Orfield 2013; 1978; Rothstein 2004), and to do so over prolonged periods of time (Sharkey 2013). Deprived neighborhoods mean deficient learning environments, since growing up in a poor or violent neighborhood limits a child’s access to role models, exposes him or her to pollutants in the air and soil, leads to consistently high levels of stress, is associated with lower-quality schooling opportunities, and limits his or her economic opportunities (Sharkey 2013). As well, as recently shown, accumulation of problems in neighborhoods translates into stronger prevalence of disadvantage around minority children in those neighborhoods’ schools (such as the proportion of children eligible for free or reduced lunch, the proportion of children not living with their two parents, etc.) (García and Weiss 2014). As a result of all these circumstances, students in highly segregated schools, who are less prepared on average in the fall, make lower relative gains by spring than students in nonsegregated schools (García and Weiss 2014).
Disparities in access to preschool education are widely seen as another major driver of education gaps. Preschool has been identified as one of the most important contributors to school readiness and education success (Magnuson et al. 2004; Cabell et al. 2011; Barnett and Belfield 2006; Barnett 2011; Diamond et al. 2013; Duncan and Magnuson 2013; Heckman 2004; 2000). Studies find that early childhood education increases a child’s exposure to learning and provides opportunities to develop his or her social interaction skills with peers and adults. Because all students benefit from early childhood education, but wealthier children are more likely to attain it, lack of universal access can be expected to widen gaps. Indeed, if participation in preschool, the ability to benefit from it, and/or the quality of preschool programs differed by ethnic group (or, more likely, by socioeconomic status), the effects could just be exacerbating other existing differences (Barnett and Yarosz 2007; Pianta et al. 2009; Bridges et al. 2004; Bartik 2011; Kagan 2009).
Parents’ efforts to promote their children’s development constitute another important contribution to student development and school readiness (Hart and Risley 1995; Belfield and García 2013; Phillips 2011; Brooks-Gunn and Markman 2005). Simple adult-to-child interactions during playtime during the first three years of life improve the child’s vocabulary and have been found to drive other educational outcomes (Hart and Risley 1995). Reading to children and other parenting practices likewise contribute to children’s learning and development (Barbarin et al. 2010). And parenting styles supportive of children’s autonomy have been positively associated with executive function skills, such as working memory and impulse control, at later ages (Bernier, Carlson, and Whipple 2010). As is true of unequal access to high-quality preschool education, low-income parents have less ability to afford leisure and educational time with their children relative to their more economically advantaged counterparts, further increasing gaps by income and racial status (Van Voorhis et al. 2013; Waldfogel 2006; Rothstein 2004; Phillips 2011; Brooks-Gunn and Markman 2005). Disparities in monetary investment in children’s education also contribute to gaps; spending on education-enhancing activities by parents in the top income fifth has nearly tripled since the 1970s (from $3,500 in 1972 to $8,900 in 2006), while spending by parents in the bottom income fifth has remained low and more stable ($800 in 1972 and $1,300 in 2006) (Duncan and Murnane 2011b). 13
A natural next question, then, is whether these early disparities are compensated for, or compounded by, the U.S. formal education system. There is some descriptive information that suggests that low-SES children begin kindergarten in lower-quality elementary schools than more advantaged children—whether measured by level of student achievement, school resources, teacher qualifications, positive attitudes toward learning, neighborhood characteristics, or school type (i.e., private or public school) (Adamson and Darling-Hammond 2012). Low-SES students are more likely to fall behind their more advantaged peers during the summer breaks (Peterson 2013). Minority and/or low-SES children are also normally in schools in which the proportion of poor children is high (García and Weiss 2014). Whether this is a cause or a consequence of historical segregation, housing segregation, economic segregation, or any other reason is not clear, but this factor could be highly likely to alter the schooling and economic opportunities of a child (Lee and Burkam 2002; Orfield 2013; Rothstein 2014).
Consequences of the gaps for later learning and development
As described above, research emphasizes that early skills gaps, both cognitive and noncognitive, translate into differences in students’ subsequent learning and development (Duncan et al. 2007; Duncan and Magnuson 2011). And early investments in education strongly predict adolescent and adult development (Heckman 2008; Heckman and Kautz 2012; Cunha and Heckman 2007). Children with stronger skills at school entry are on a more favorable pathway toward academic success than are students with weaker initial skills. For instance, students with higher levels of behavioral skills learn more in school than peers whose attitudinal skills are lower (Jennings and DiPrete 2010).14 In general, as Heckman asserted, “skills beget skills,” meaning that creating basic, foundational knowledge makes it easier to acquire skills in the future (Heckman 2008).
Conversely, children who fail to acquire this early foundational knowledge may experience some permanent loss of opportunities to achieve to their full potential. Indeed, scholars have documented a correlation between lack of kindergarten readiness and not reading well at third grade, which is a key point at which failing to read well greatly reduces a child’s odds of completing high school (Fiester 2010; Hernandez 2011).15
Why do we study noncognitive skills at the starting gate?
Noncognitive skills, which include skills such as persistence, respect for others, academic confidence, teamwork, interpersonal relationships, and creativity, are central to this analysis for a number of reasons.16 These skills directly affect the productivity of a person (as a student, worker, and citizen) and also alter the productivity relationships between factors and educational outcomes (for instance, the effect of some teaching styles on learning can differ among children depending on their socioemotional skills). First, noncognitive skills help nurture children’s learning. As noted above, children whose behavioral skills are high learn more than children with weak behavioral skills (Jennings and DiPrete 2010); and noncognitive skills help explain achievement gaps between black and white students at young ages (Grissmer and Eiseman 2008). Although the empirical evidence on how these traits predict later cognitive and noncognitive performance is still relatively scarce, research shows a positive and reciprocal relationship between noncognitive and cognitive abilities (García 2013), and of “self-productivity” and “dynamic complementarities” between investments in the two types of skills (Cunha and Heckman 2007).17 These relationships suggest that boosting cognitive productivity may not be possible without paying attention to noncognitive skills, and that increased attention to noncognitive skills in education policy can thus increase children’s opportunities and pathways to develop (García 2014). More broadly, as are cognitive skills, noncognitive skills are a component of a person’s development and life potential. Consequently, knowing whether there are gaps in these skills at earlier stages would reveal important mechanisms behind inequalities among children of different characteristics.
Moreover, the examination of noncognitive skills available in the ECLS-K: 2010–2011 study provides an additional, and potentially useful, insight—that the ratings of these skills come from parents’ and teachers’ subjective assessments, and, as such, reflect those adults’ individual characteristics and own social and cultural norms as well as the children’s observed abilities and behaviors. Deeper understanding of why parents and teachers rate children differently (through studying the influence of biases and stereotypes), and whether these influences affect children’s development, could help us to better understand and address educational inequalities at the starting gate and throughout the school years.
Dataset and methodology
The analysis developed in this study is based on data from the Early Childhood Longitudinal Study, Kindergarten Class of 2010–2011 (ECLS-K: 2010–2011), sponsored by the National Center for Education Statistics (Institute of Education Sciences, U.S. Department of Education). This study will follow a nationally representative sample of children starting in their kindergarten year, through their elementary school years.18 It provides information on multiple dimensions of children’s development, early learning, and progress in school, as well as information on children’s families and on teachers’ and parents’ perceptions of children’s skills and behaviors. The tracking of students over time is one of the study’s most valuable features, as is the availability of two ECLS-K studies (ECLS-K: 1998–99 and ECLS-K: 2010–2011), which will allow for cross-comparisons “of two nationally representative kindergarten classes experiencing different policy, educational, and demographic environments” (Tourangeau et al. 2013).19
Both the outcome variables and the individual level characteristics (control variables) that are used in the analysis are described below.
Variables—Outcomes
For the current analysis, we focus on measurements of the child’s cognitive and noncognitive skills at the beginning of the school year (assessments were conducted from August through mid-December 2010).
The definitions that follow summarize and paraphrase the information reported by Tourangeau et al. (2013), which can be consulted for more details. See also Appendix B for a more detailed explanation of the variables used in this analysis.
Cognitive skills and executive function
These cognitive skills and executive function skills20 are assessed with instruments that measure the child’s:
Reading skills: print familiarity, letter recognition, beginning and ending sounds, rhyming words, word recognition, vocabulary knowledge, and reading comprehension.
Math skills: conceptual knowledge, procedural knowledge, and problem solving; number sense, properties, and operations; measurement; geometry and spatial sense; data analysis, statistics, and probability; and patterns, algebra, and functions.
Cognitive flexibility: ability to sort a series of picture cards according to different rules, and response time at this task.
Working memory: ability to repeat increasingly long strings of orally presented numbers in reverse order.
Principal noncognitive skills
We use the term “principal” to identify a set of noncognitive skills that are measured by both the ECLS-K 1998–1999 and 2010–2011 surveys, and that have been relatively extensively used in research. We distinguish these “principal noncognitive skills” from “other noncognitive skills” described later, which, while not available in the public data from the kindergarten class of 1998–1999, are nevertheless important noncognitive skills to measure.
Teachers are asked to assess the child’s:
Self-control: ability to control behavior by respecting the property rights of others, controlling temper, accepting peer ideas for group activities, and responding appropriately to pressure from peers.
Approaches to learning: organizational skills (keeps belongings organized); curiosity (is eager to learn new things); independence (works independently); adaptability (easily adapts to changes in routine); persistence in completing tasks; focus (ability to pay attention); and ability to follow classroom rules.
Internalizing problems: degree of internalizing behavioral problems as measured by the frequency with which the child shows anxiety, loneliness, low self-esteem, and sadness.
Externalizing problems: degree of externalizing behavioral problems as measured by the frequency with which a child argues, fights, gets angry, acts impulsively, and disturbs ongoing activities.
Parents are asked to assess the child’s:
Self-control: ability to control behavior by refraining from fighting, arguing, throwing tantrums, and getting angry.
Approaches to learning: persistence (keeps working at something until finished); curiosity (shows interest in a variety of things); focus (concentrates on a task and ignores distractions); helpfulness (helps with chores); intellectual curiosity (is eager to learn new things); and creativity (in work and play).
Social interaction (with peers and adults): ease in joining in play, ability to make and keep friends, and capacity to positively interact with peers (e.g., by comforting or helping).
Other noncognitive skills
The ECLS-K 2010–2011 includes a range of interesting measures not available in the ECLS-K 1998–1999 public data and our analysis encompasses many of these measures. Other skills reported by teachers and covered in our study include the child’s interpersonal relationships (based on items describing the child’s skill in forming and maintaining friendships; getting along with people who are different; comforting or helping other children; expressing feelings, ideas, and opinions in positive ways; and showing sensitivity to the feelings of others); closeness to the teacher (based on items that measure the affection, warmth, and open communication that the teacher experiences with the student); eagerness to learn, persistence in completing tasks, and attention (whether the child “pays attention well”).
Other skills reported by parents and covered in our study include the child’s persistence (ability to work until finished); eagerness to learn new things; and level of creativity in work or play.
For the purpose of the analysis, all variables are standardized to have a mean of zero and standard deviation of one. 21
Variables—Child and family characteristics (education inputs)
Variables describing the children and their families are used as controls in our estimates. These variables include:
Race/ethnicity: The groups of interest in the paper are white, black, Hispanic, Asian, or other. Hispanic children are divided into two groups, depending on whether the language spoken at home is English or not. This decomposition is first described and utilized by Nores and Barnett (2014) and Nores and García (2014).22
Socioeconomic status (SES): SES is based on five different components, including parents’ (or guardians’) educational attainment, occupational prestige score, and household income (see more details in Tourangeau et al. 2013, 7-56 to 7-60). We divide the variable into quintiles or fifths, where “low SES” indicates the first or bottom quintile; “middle-low SES” indicates the second quintile; “middle SES” is the third socioeconomic quintile; “high-middle SES” indicates the fourth SES quintile; and “high SES” represents the top quintile.
Other individual and family characteristics: These other characteristics include the child’s gender, age, disability status, immigrant status, ELL status (whether the child is an English Language Learner versus whether she speaks English), number of siblings, and whether the child lives with both parents.
Prekindergarten care and parenting experiences: This variable indicates whether the child was cared for in a center-based setting during the year prior to the kindergarten year23 and engaged in enrichment activities with parents (as measured by a composite that captures early literacy practices, leisure activities, other rules, and routines24).
Following the traditional framework to estimate gaps, we use the economic approach in which education outcomes—cognitive and noncognitive skills—are explained as a function of education inputs, including an indicator for each of the population groups of interest, whether by race/ethnicity (white, black, Hispanic ELL, Hispanic non-ELL, Asian, or other) or socioeconomic status (low SES, middle-low SES, middle SES, high-middle SES; and high SES) (García 2013; Nores and García 2014; Todd and Wolpin 2003).
In order to estimate education gaps for the 2010–2011 kindergarten cohort, we follow a parsimonious strategy with three models for each of the two sets of estimated gaps: by race and by SES.25 The baseline model shows unadjusted skills gaps by race/ethnicity and language, or by SES, reflecting absolute performance gaps, and model 2 incorporates only controls for SES, or race/ethnicity. The final model provides adjusted race/ethnicity-based and SES-based gaps for both cognitive and noncognitive skills after controlling for other individual and family characteristics, and early educational practices such as pre-K experience and parental activities with children.
In order to control for across-school differences and to account for potential selection of students into schools, we use a schools-fixed-effects model, in which the estimated gaps are within-school gaps.26 This approach controls for biases that may arise due to selection processes; for instance, certain types of students are more likely to attend certain schools, which in turn is also associated with their outcomes. In the absence of longitudinal information for a child (or in absence of individual fixed effects), this strategy has been utilized to account for such selection both for cognitive and noncognitive outcomes (Neidell and Waldfogel 2010; Nores and García 2014).27
The specifications are shown in Appendix A. Coefficients of interest for race- and ethnicity-based gaps are represented by β1 to β4, and represent the unadjusted (Model 1R) to fully adjusted (Model 3R) skills gaps by race/ethnicity (R) and language (L). For the socioeconomic-based gaps, the coefficients of interest are δ1 to δ4, obtained under Models 1S (unadjusted socioeconomic gaps) to Model 3S (fully adjusted socioeconomic gaps).
Analytic sample
While the ECLS-K: 2010–2011 is designed to provide a nationally representative sample of the 2010–2011 U.S. kindergarten population, the study has experienced the problem of lack of responses (i.e., missing data) on some variables of interest. As such, the analytic samples do not fully represent the intended population of interest.28
In order to select the analytic samples used in the study, we proceed as follows. The descriptive analyses in the next section are based on the maximum number of per-child responses for each variable, and provide an updated description of the student population at kindergarten entry as of 2010. Analytic samples supporting the results shown in the “Gaps at the starting gate” section vary as a function of the complete response in the predictors (the control variables, e.g., race, ethnicity, SES, etc.) minus the missing responses of each skill. A more detailed discussion of how missing data affects the different outcomes and the predictors is included in Appendix B.
Child-level weights are used in all results in the following sections (see more details in Appendix B).
A description of the kindergarten class of 2010–2011
Our analysis starts with a description of the characteristics of the members of the 2010–2011 kindergarten class, with a focus on several demographic dimensions that are relevant to assessing educational performance. In particular, we describe the characteristics of the slightly over four million children in the 2010–2011 class by their race/ethnicity and socioeconomic status. We also examine parents’ characteristics, including their investment in their children’s prekindergarten care/schooling, and activities aimed at promoting their children’s development.
Who is entering kindergarten?
Table C1 in Appendix C shows the characteristics of the kindergarten class of 2010–2011. The first set of variables indicates that white students represent just over half of the group, while black students are about 14 percent (13.7 percent). One of every four kindergarteners is Hispanic (and, among the respondents for the immigration question, almost one of every five students is a Hispanic English language learner (ELL). Four percent (4.4 percent) of the children are Asian, and the remaining 5.5 percent are classified in the “other races/ethnicities” group.
With respect to children’s families, we highlight the fact that almost one-third of kindergarten entrants live in a family that does not include two parents (31.8 percent). The vast majority of the children speak English at home (84.7 percent), and three-fourths (74 percent) are native born with native parents. Of particular importance for our analysis of achievement gaps, as reported in Table C1 (second panel), one of every four children (25.5 percent) lives in poverty.29
What did parents do to boost their children’s development before entering kindergarten?
As discussed above, research makes clear the importance of providing all children with a high quality preschool education (Gormley, Phillips, and Gayer 2008; Barnett 2013; 2010). While providing such support has not yet become the norm, economic and employment realities dictate that most young children receive some type of paid care outside the home. Indeed, more than half of the students in the cohort have received some center-based pre-K care (55.1 percent), and nearly four-fifths of parents made some nonparental care arrangements during the year prior to kindergarten (79.3 percent).
In addition to early education arrangements, parents undertake a multitude of activities with their children, from reading to children to ensuring playing time, which also contribute to children’s development. ECLS-K includes an extensive set of questions about the frequency with which parents engage in different activities. According to the descriptive findings (available upon request), the majority of children are read to and/or told stories on a daily basis (52.0 and 39.4 percent respectively). In addition, parents sing songs with their children (45.3 percent) and practice reading (61.3 percent) and writing (50.7 percent) with their children, and children read picture books or read outside of school daily (36.9 percent). The majority of parents also play games (41.6 percent) and sports (37.4 percent) with their children with a high frequency (three to six times a week) and 36.9 percent have children help with chores. Parents report that they engage their children less frequently (once or twice per week) in talking about nature (49.2 percent), building things (42.5 percent), or doing art projects (36.7 percent).
Given the high correlation among the activities, for the empirical analysis, we construct an index that captures the joint variance of all these activities (see Appendix B). The index reflects the frequency with which parents engage in a range of educational and leisure activities with their children.
Characteristics of the kindergartners by ethnic and socioeconomic backgrounds
Understanding the education gaps we are studying requires, first, analyzing some of these inequalities with respect to inputs. Table C2 shows the descriptive statistics of such inputs by racial/ethnic group and by socioeconomic status.
Over half (52 percent) of white children are in the two highest socioeconomic quintiles (high-middle or high), while only 8.9 percent fall into the lowest SES quintile. A similar pattern is true among Asian kindergartners: 59.9 percent are in the highest two quintiles, and 11.8 percent are in the lowest. For black and especially for Hispanic children, however, the situation is reversed. Over half (56.8 percent) of black children and over two-thirds (66.6 percent) of Hispanic children are in the two lowest quintiles, and fewer than one in 10 of either group are in the highest SES quintile (8.3 percent of black children and 6.8 percent of Hispanic children). Another angle through which to see these numbers is the proportion of children who live in poverty by race/ethnicity: 13.1 percent of white children, 17.3 percent of Asian children, and nearly half of black children (45.5 percent) and Hispanic children (46.3 percent). Among Hispanics, 30.5 percent of non-ELL, and 62.5 percent of ELL children live in poverty. Among all racial/ethnic groups, the Hispanic ELL group has the largest share living in poverty.
Other disparities are also clear along racial/ethnic lines. Almost two-thirds of black children (64.5 percent) do not live with two parents, compared with 9.6 percent of Asian children. Both Asian and Hispanic children are more likely to speak a language other than English at home (54.5 percent and 47.5 percent respectively), versus white and black children (1.8 percent and 4.0 percent respectively). And Asian children are the most likely to have received center-based pre-K care (61.7 percent), while Hispanic children—especially ELL-Hispanic—are among the least likely to have participated in center-based care (46.5 and 41.3 percent respectively).
Regarding the disparities by socioeconomic status (shown in the bottom half of the table), all statistics consistently confirm the correlation between socioeconomic status and obstacles to educational development (selected control variables are shown in table). Low-SES students are more likely than their higher SES peers to not speak English, to not live with two parents, to be immigrants, to not have participated in center-based pre-K care activities in the previous year, and to have a lower index of early literacy practices at home. Among children in the low SES group, half (50.4 percent) are Hispanic, 23.1 percent are white, 19.6 percent are black, and 2.5 percent are Asian.
Gaps at the starting gate: Results from the econometric approach
This section includes the results of the analysis estimating education (or more specifically, cognitive and noncognitive skills) gaps at the school starting gate. Results are presented for different socioeconomic groups (see specific results in Appendix D, figures D1 to D6 and tables D1 to D6; Appendix F, tables F1 to F6; and Table 1) and for different racial/ethnic groups (see specific results in Appendix E, figures E1 to E6 and tables E1 to E6; Appendix F, tables F1 to F6; and Table 2). The section ends by highlighting some other relationships between outcomes and inputs of interest (see results in Appendix F, tables F1 to F6; and Table 3).
The sizes of the real education gaps between groups of U.S. kindergartners are revealed by the unadjusted gaps estimates. As the findings illustrate, skills gaps by SES in both cognitive and noncognitive dimensions are sizeable by the time children enter kindergarten. In other words, if we compare the achievement of each of the SES groups relative to the lowest SES children (the reference group in the analyses), we see that gaps exist between all groups, as average scores increase for each step up the SES distribution. Low-income children come to the starting gate well behind their more affluent peers, those in the four SES groups above them. And if we compare their position with each of the four SES groups, that gap widens steeply for each step up the SES distribution.While the skills gaps by socioeconomic status shrink slightly when adjusted for controls such as race, other individual and family characteristics, and pre-K care arrangements and parenting activities, substantial inequalities remain, and this is true for all the skills analyzed. Conversely, the unadjusted gaps by race/ethnicity are, in many cases, statistically significant, but they shrink—and even disappear for some groups—after the inclusion of the different covariates that identify the children’s socioeconomic background (i.e., they are very sensitive to the inclusion of SES covariates). Although skills gaps for black children and a subgroup of Hispanic children (the ELL Hispanic group) also diminish when adjusted for the covariates, the analysis by race/ethnicity for these two groups points out their relative disadvantage, compared with white children.
Gaps based on socioeconomic status
When children in the bottom socioeconomic quintile (low SES) are compared with children in the other four quintiles (low-middle, middle, high-middle and high SES), we find no educational outcome for which a sizeable gap does not exist under the unadjusted to fully adjusted models. All gaps and gradients are sizeable, and virtually all are statistically significant (with just a few exceptions for average performance between students in the low-middle and middle SES groups compared with the poorest children). While strong and persistent, though, all of the gaps narrow slightly with the addition of controls, which implies that gaps can be narrowed to some extent by using compensatory policies in favor of children (such as preschool and parental engagement activities) and by providing support associated with the family circumstances that most contribute to the reduction of the unadjusted gaps (economic support, knowledge of English, immigration status, etc.).
Socioeconomic-based gaps in cognitive skills
Overall, our results—showing significant socioeconomic-based gaps in cognitive skills—confirm what multiple other research analyses (e.g., Reardon 2011) have found: that students’ levels of readiness and development are closely associated with their parents’ socioeconomic status. Unadjusted differences in cognitive domains indicate that each move up a socioeconomic quintile in the SES distribution is associated with approximately a quarter of a standard deviation (sd) improvement in performance in both math and reading, with students in the top quintile (the high SES group) scoring nearly a full standard deviation above students in the bottom quintile (the low SES group).30 While the gaps shrink when controls are included, all of the adjusted differences remain statistically and educationally significant. Fully adjusted gradients show that moving each subsequent quintile up in the SES distribution improves performance by about 0.2 standard deviations (with a minimum of 0.15 sd and a maximum of 0.23 sd), in both math and reading, and the gaps between the lowest and highest SES groups still surpass three-fourths of a standard deviation in the two cognitive skills.31
One interesting finding is that, when all controls are included, the coefficients associated with math and reading performance of the two lowest SES quintiles narrow more, proportionately, than do the coefficients associated with the two highest SES quintiles. In other words, adding controls such as family composition and early education practices has a bigger influence on gaps at the low-SES versus the high-SES end of the distribution. This may indicate that education supports that children receive outside their homes and/or parenting enrichment activities are particularly beneficial for low- to middle-income children, since higher-income parents likely provide such supports from their own resources. This suggests that increasing low-income children’s access to educational activities that can complement the attention and stimulation received within their homes could substantially reduce their relative disadvantage in reading and math skills.
While SES-based gaps for a set of executive function indicators—cognitive flexibility and working memory—are also substantial, a closer look offers some interesting findings that contrast with the trends for math and reading. Controlling for race/ethnicity (Model 2S) significantly decreases the gaps, whereas the decrease was smaller in math and reading. There is also a small additional shrinking effect from adding child and family characteristics, pre-K schooling, and the various parenting educational activities (Model 3S), but this additional decrease is much smaller than it was for reading and math. The adjusted advantages for children across the socioeconomic distribution compared with children in the low-SES quintile in the cognitive flexibility skill are between 0.10 and 0.25 standard deviations. A similar gradient, but steeper, is also observed for the working memory skill (adjusted gaps relative to children in the low SES quintile are between 0.10 and 0.51 standard deviations).
Socioeconomic-based gaps in noncognitive skills
Estimates of SES-based gaps in noncognitive skills, as reported by parents and teachers, reveal two important trends. First, both parent- and teacher-based assessments reveal gaps or socioeconomic gradients. Related to this main finding, as we will see, the steepness of the gradients is not uniform, and depends both on the skills and on whether parents or teachers are providing the assessments. In terms of sensitivity to the controls, gaps do not always shrink when race/ethnicity controls are added (comparing unadjusted with adjusted-by-race gaps). Second, the size of the gaps in comparable constructs (e.g., persistence in completing tasks) differs depending on whether parents or teachers are doing the rating. In other words, parents sometimes rate their child as having better behaviors than the child’s teacher rates the child, and vice versa.
According to the estimates, children’s noncognitive skills as rated by teachers clearly differ by socioeconomic status, for all the skills studied. The steepest gradients are found in teachers’ assessments of children’s approaches to learning, eagerness to learn, persistence, and attention. In the approaches to learning category, fully adjusted scores for children rise from 0.13 standard deviations for children in the low-middle SES quintile to 0.51 standard deviations for children in the high SES quintile (compared with children in the poorest quintile). In the eagerness to learn category, the range from low-middle to high quintiles is from 0.09 to 0.41 standard deviations. In the persistence category, the range from low-middle to high quintiles is from 0.12 to 0.42 standard deviations. In the attention category, the range from low-middle to high quintiles is from 0.10 to 0.44 standard deviations.
Compared with the steep ladders just described, gradients are not as uniformly steep for internalizing behavioral problems, externalizing behavioral problems, and closeness to teachers, especially when comparing children in the two lowest SES quintiles. In each of these categories, under the fully adjusted model, there is no statistically significant difference between teacher’s ratings of the poorest children versus those of children in the low-middle SES. However, the gaps become significant when comparing the poorest children with children in the top two SES quintiles (these gaps range from 0.15 to 0.23 standard deviations). The gaps are also marginally statistically significant for middle-SES children relative to low-SES children in the internalizing behavioral problems and teacher closeness skills.
Finally, although SES-based gaps in noncognitive skills as rated by teachers are narrowed by the controls included in the fully adjusted model, intermediate adjustments show some trends that are in clear contrast with cognitive skills gaps. For example, controlling for race alone (model 2 in the tables) increases, rather than reduces, the perceived disadvantage of low-SES children relative to all the other children. The only exceptions are estimated gaps in closeness to the teacher, in which race controls decrease the gap, and interpersonal relationships, where the gap essentially remains the same). Moreover, when controlling for race/ethnicity and certain other family and child characteristics (not shown in tables), the gradients become fixed; they are not responsive to either prekindergarten attendance or parental enrichment activities. This is in contrast to parents’ reports of noncognitive skills.
Unfortunately, we found no research evidence explaining the key trends in teacher-reported gaps discussed above: why teachers rate noncognitive skills higher for higher-SES children and why gaps increase rather than decrease after adjusting by race. Clearly, it is important that further research explores why this occurs and the potential implications of these gaps for educational performance as children progress in school.
Turning to children’s noncognitive skills as rated by parents, we also see clear differences by socioeconomic status. The gradients or slopes are markedly steep for most measures of noncognitive skills. This is particularly true for approaches to learning, self-control, and persistence in completing tasks. In these three areas, the fully adjusted gaps relat