Abstract
Although substantial evidence from the United States suggests that more qualified teachers are disproportionately concentrated in the schools and classrooms of academically and socioeconomically advantaged children, it is not clear whether the problem of teacher sorting is global in scope. This study uses data from the 2013 Teaching and Learning International Survey to examine whether and how school- and classroom-level teacher distribution patterns vary across 32 education systems with diverse national contexts and education policies. We find that cross- and within-school teacher sorting is common in most countries but within-school sorting is more pronounced in higher income countries. We also identify several national policy variables that are significantly related to both cross-school and cross-classroom sorting of teachers.
Keywords
In a 2002 study of nearly all public school teachers in the state of New York, Lankford, Loeb, and Wyckoff found that low-income, minority, and lower achieving students had consistently less access to qualified teachers than their more advantaged and nonminority peers. Subsequent studies in the United States have found similarly inequitable teacher distribution patterns in several other states and across the nation (Bacolod, 2007; Clotfleter, Ladd, & Vigdor, 2005, 2006; Darling-Hammond, 2004; Goldhaber, Lavery, & Theobald, 2015). In a 2007 study using data from 46 countries participating in the 2003 Trends in International Mathematics and Science Study (TIMSS), Akiba, LeTendre, and Scribner added cross-national evidence suggesting that the problem of inequitable teacher distribution might be global in scope. Akiba et al. found that in many countries, students with low socioeconomic status (SES) had consistently less access to qualified teachers than their more advantaged peers. Confirming earlier evidence from the United States, Akiba et al. also found that the United States had one of the largest teacher quality “opportunity gaps” in the world.
Akiba et al.’s (2007) research identified national exceptions like South Korea, where low-SES students actually had greater access to qualified teachers. Kang and Hong (2008) attributed this “negative opportunity gap” to national education policies like centralized teacher assignment, mandatory rotation of teachers across schools, and incentives to teach disadvantaged students. Although many previous studies have attributed inequitable teacher distribution to teacher “sorting,” whereby teachers with stronger qualifications exercise their preferences for pleasant working and teaching conditions to self-select into the schools and classrooms of more advantaged and higher achieving students (e.g., Kalogrides, Loeb, & Béteille, 2013; Lankford, Loeb, & Wyckoff, 2002), the example of Korea suggests that national policy may be able to prevent or reduce sorting of teachers across schools (Jeong & Luschei, in press).
In this study, we shift the focus of teacher sorting patterns from the individual preferences of teachers to the impact of national context and policy. Certainly, teachers, like any other human beings, wish to live and work in more desirable locations, all else equal. But country-specific contextual or policy-related factors may mediate or dampen teachers’ sorting behavior to reduce inequities that occur when more qualified teachers choose to work in the schools and classrooms of higher achieving and more advantaged students. For example, education systems with more centralized systems of teacher hiring and assignment may have greater success in ensuring equal access to qualified teachers across schools, as may systems that offer incentives for teachers to work in difficult-to-staff schools (Luschei, Chudgar, & Rew, 2013).
Regardless of the intentions and impact of teacher assignment and incentive policies on the distribution of teachers across schools, these policies may have a much different impact on the distribution of teachers across classrooms within schools. Once a teacher arrives at a school, national policies related to teacher assignment may be superseded by school-level dynamics. Although most teacher sorting research has focused on cross-district or cross-school teacher sorting patterns, a few studies from the United States suggest that within-school sorting of teachers presents a major challenge to educational equity (Grissom, Kalogrides, & Loeb, 2015; Kalogrides et al., 2013). Yet to date, very little research has examined whether cross-classroom teacher sorting patterns vary across national educational and policy contexts.
In this study, we exploit cross-national variability in educational contexts and policies to explore the nature and correlates of teacher quality opportunity gaps across 32 diverse countries and education systems. We also explore how national educational context and policies interact with teacher characteristics to increase or reduce cross-school and cross-classroom inequities in disadvantaged children’s access to qualified teachers. We use nationally representative data from the 2013 application of the Organisation for Economic Co-operation and Development (OECD)’s 2013 Teaching and Learning International Survey (TALIS) to explore the following questions:
Across 32 education systems with diverse political, social, and economic contexts, are less qualified teachers more likely than other teachers to work in schools with greater concentrations of disadvantaged students?
Across these 32 education systems, are less qualified teachers more likely than other teachers to work in classrooms with greater concentrations of disadvantaged students?
Across these 32 education systems, how do national educational context and policy relate to the distribution of teachers across schools and classrooms?
In addition to exploring whether cross-school and cross-classroom teacher sorting are global in nature, we also seek to draw policy implications for the United States and other education systems facing challenges to ensure access for disadvantaged children to qualified teachers. In the section that follows, we review international evidence related to teacher quality and opportunity gaps. We then describe our data and methods and present the results of our analysis. After discussing the implications of these results for policy, we conclude with a call for future cross-national investigation of the causes and implications of teacher sorting.
Teacher Sorting and Opportunity Gaps in Global Perspective
In addition to Lankford et al.’s 2002 study of teachers in New York State, which found systematic evidence of teacher sorting according to student race/ethnicity, income, and achievement, subsequent studies have found evidence of inequitable teacher distribution patterns in California (Darling-Hammond, 2004), North Carolina (Clotfelter et al., 2005, 2006), Washington State (Goldhaber et al., 2015), and nationally (Bacolod, 2007). Although these studies have generally examined teacher distribution across districts or schools, several studies have also found evidence of teacher sorting across classrooms within schools. Using data from all public school districts in the state of Washington, Goldhaber et al. (2015) examined cross-district, cross-school, and within-school patterns of teacher distribution. They found that teachers with greater experience, licensure exam scores, and value-added impact on student achievement are less likely than other teachers to work with students from underrepresented minority groups, lower income students, and lower achieving students. Although they found that cross-district sorting of both teachers and students tends to explain most of the teacher quality gap, cross-school sorting also strongly influences teacher quality gaps, especially in middle schools. Goldhaber and colleagues found less evidence of teacher sorting across classrooms within schools, but they did find that teachers with higher value-added scores are less likely than other teachers to work with seventh grade students with low prior achievement. According to the authors, this result suggests that “low-performing seventh graders may be disproportionately ‘tracked’ into classrooms with previously ineffective teachers” (p. 304).
In a study of the Miami-Dade County Public School District, Kalogrides et al. (2013) found that lower achieving students are more likely to have less experienced, minority, and female teachers, as well as teachers with degrees from less selective undergraduate institutions, relative to other students in the same school. In another study of Miami-Dade, Grissom et al. (2015) found that teachers with more experience in a given school are more likely to be assigned to classes with higher achieving students and students with fewer absences and suspensions. They also found that more experienced teachers are less likely to teach Black and low-income students.
Because differences in school location and conditions are controlled, cross-classroom teacher distribution strongly illustrates teachers’ preferences for student characteristics, which have been a focus of much of the teacher sorting literature in the United States. As Kalogrides et al. (2013) observe, “many studies find that teachers demonstrate preferences for teaching in schools with easier to serve student populations. When given the opportunity, more qualified and experienced teachers tend to choose schools with higher achieving students, fewer minority students, higher income students, and schools that are safer and experience fewer disciplinary problems” (p. 104). The teacher sorting “story” in the United States has focused on teachers and their preferences largely because teachers in the United States are assigned to schools locally and can bargain with school leaders or school district officials for positions. As a result, teachers with stronger qualifications or experience can often secure positions in schools with more pleasant working conditions. In contrast, teacher assignment in many nations—including Korea—takes place more centrally. Teachers in Korea are also required to rotate to new schools every 5 years, and teachers can receive incentives if they agree to work in difficult-to-staff schools (Kang & Hong, 2008). In this context, the influence of national policy may supersede teachers’ local enactment of preferences to ensure a more equitable distribution of teachers across schools, as evidenced by a wide disparity in teacher quality opportunity gaps between South Korea and the United States (Akiba et al., 2007).
Despite the apparent global ubiquity of teacher sorting, cross-national evidence suggests substantial variability in the degree and nature of teacher sorting across national contexts. Although Akiba et al. (2007) found that low-SES students had less access than high-SES students to qualified teachers in 24 countries, they found the reverse in 15 countries. There is also evidence that teacher sorting patterns differ across developing countries, as well as between developing countries and developed countries (Chudgar & Luschei, 2016; Luschei et al., 2013), suggesting that teacher sorting patterns reflect underlying social and educational contexts. For example, research on teacher distribution in Latin America—a region long plagued by social, economic, and educational inequality—finds strong teacher sorting patterns that place economically disadvantaged and rural children at a substantial disadvantage (Cox, 2010; Luschei, 2012; Luschei & Carnoy, 2010). In contrast, teacher sorting appears to be less pronounced in many Asian nations, which generally have lower levels of social and economic inequality (Chudgar & Luschei, 2016).
Implications of Teacher Sorting
Global evidence of teacher sorting has clear implications for educational equity. From the student perspective, sorting of more qualified teachers leads to achievement and resource inequities. In the first case, disproportionate concentrations of less qualified teachers working with disadvantaged students are likely to exacerbate well-documented racial and socioeconomic gaps in educational achievement (Darling-Hammond, 2006). Second, due to the universal application of salary schedules that reward teachers’ experience and training, more experienced and trained teachers command higher salaries (OECD, 2015). Since teacher salaries make up a very large fraction of most education budgets, schools, districts, and nations with disproportionate concentrations of more qualified teachers in the classrooms and schools of more advantaged students divert resources away from the students who need them the most (Chudgar & Luschei, 2016).
From the teacher perspective, teacher sorting introduces inequities for those without the experience or political and social capital to influence their teaching assignments. New teachers, who tend to be more sensitive to their working conditions (Hanushek, Kain, & Rivkin, 2004), may respond to what they perceive as unfair and more difficult assignments by leaving their current schools or exiting the profession entirely (Feng, 2010; Kalogrides et al., 2013). To the extent that these new teachers work with the least advantaged students, their early exit confronts these students with a double disadvantage of systematically lower access to experienced teachers and high degrees of teacher attrition.
What Causes (or Prevents) Teacher Sorting?
Although the consequences of teacher sorting are clear, the causes are less so. The U.S. literature has stressed the role of teachers’ preferences for working conditions, especially student composition. These studies often assume an environment in which teachers negotiate locally with schools or districts for teaching positions or transfers to schools with more pleasant working conditions. Within schools, teachers can accumulate experience and capital over time that they can exercise to influence the composition of classes and their own teaching assignments (Grissom et al., 2015; Kalogrides et al., 2013).
In approaching this question cross-nationally, we shift attention from teachers and their preferences to the influence of national education policy and context. This reorientation implies that teachers are not completely to blame for teacher sorting patterns; a number of contextual and policy factors may influence the degree to which teachers sort across classrooms and schools. Across much of Latin America and the Caribbean, for example, new teachers are assigned to schools through “concursos,” or competitions, in which the highest scoring teachers are able to choose schools with the most pleasant working conditions (Luschei & Carnoy, 2010; Luschei et al., 2013). One exception is Cuba, a highly centralized system that places a strong emphasis on equality of educational opportunity across rich and poor and urban and rural (Carnoy, 2007). Limited cross-national evidence also suggests that teacher distribution in Cuba is among the most equitable in Latin America and the Caribbean (Chudgar & Luschei, 2016; Luschei, 2012).
The case of Korea also suggests that a strong national commitment to educational equity may lead to the development of policies that promote equity in teacher distribution, such as uniformly high quality in the teacher labor force, mandatory teacher assignment, and incentives to work with disadvantaged student populations (Kang & Hong, 2008). Further, teacher assignment occurs centrally in Korea, at the level of the province or municipality. As a result, educational officials can assess teacher demand across schools and assign teachers according to the greatest need. In decentralized systems like the United States, it is much more difficult for education officials to deploy teachers across schools in a way that ensures equity (Jeong & Luschei, in press).
Finally, policies that influence school composition and teachers’ work may also exacerbate or reduce sorting of teachers across schools and classrooms. We identify four types of policies that could influence who teaches where and whom: (1) across- and within-school tracking or grouping of students, (2) performance incentives for teachers, (3) school accountability policies that make use of student achievement data, and (4) school autonomy in curriculum and instruction and resource allocation. To some extent, each type of policy either reduces or increases options and incentives for teachers to sort into schools or classrooms of higher achieving or more advantaged students.
First, cross- and within-school ability grouping practices, whereby schools track students of different perceived ability into separate schools or classrooms, are likely to increase differences in student composition across schools and classrooms and provide more choices for teachers to sort across, thereby exacerbating teacher sorting patterns (Goldhaber et al., 2015; Jeong & Luschei, in press; Oakes, 2005). However, cross-school tracking may concentrate certain types of students into specific schools, thereby making within-school student populations more homogeneous and decreasing cross-classroom teacher sorting. In contrast, within-school ability grouping is likely to increase teacher sorting across classrooms by making classrooms more heterogeneous and giving teachers more incentives to sort. We hypothesize that tracking of students across schools will increase cross-school teacher sorting but reduce cross-classroom sorting. In contrast, within-school ability grouping is unlikely to influence cross-school sorting but will exacerbate cross-classroom teacher sorting by concentrating populations of students with different needs or advantages into specific classrooms.
Second, performance incentives for teachers, both in terms of salaries and career advancement, may increase the stakes of student performance and encourage teachers with stronger qualifications and options to choose to work with higher achieving students (Jeong & Luschei, in press). Such performance incentives may also be linked to accountability policies that induce school leaders to assign their strongest teachers to classrooms that are more likely to be tested or to perform well on tests (e.g., Cohen-Vogel, 2011; Fuller & Ladd, 2013). As a result, we hypothesize that the presence of teacher performance incentives will exacerbate both cross-school and cross-classroom sorting, as teachers with greater qualifications seek to work with higher achieving students who are more likely to perform better on standardized tests.
Third, as with performance incentives, school accountability policies that track or publicly report student achievement data provide an additional incentive for teachers to select higher achieving students where possible. As a result, we hypothesize that the presence of accountability policies based on student achievement data will exacerbate teacher sorting patterns across schools and classrooms.
Finally, greater school-level autonomy in curriculum and instruction and resource allocation is likely to allow teachers greater options in how they use resources and run their classrooms (Ingersoll, 2006; Ingersoll & May, 2012), which could in turn influence the types of students that teachers choose to teach. At the same time, greater autonomy may also empower teachers concerned with issues like performance incentives and accountability policies, weakening their incentives to respond to such centrally driven mandates and reducing the stakes of student composition in schools and classrooms (e.g., Ingersoll, 1997; White, 1992). As a result, it is difficult to predict exactly how increased autonomy will relate to teacher sorting.
The key point is that while all of these policies have some type of intended consequence—often to encourage greater teacher efficiency or student performance—they may also result in unintended consequences in terms of their influence on teacher quality opportunity gaps. To examine whether this is the case, we turn to our analysis below.
Data and Methods
Data
We use data from the 2013 application of the OECD’s TALIS, which followed an earlier survey in 2008. In each of 32 participating countries or education systems, TALIS 2013 administered questionnaires to lower secondary school leaders and 20 teachers in 200 schools, reaching over 9,000 school leaders and over 170,000 teachers. Because the TALIS sample is based on the entire population of teachers within each education system, we can generalize our findings to teachers at the national level and make cross-national comparisons of teacher distribution. 1 The sampling of 20 teachers in each school allows us to compare teacher distribution across classrooms, which is not possible with other cross-national surveys like Programme for International Student Assessment (PISA) and TIMSS.
In this study, we limit our analysis to the TALIS sample of public schools, in which 8,078 school leaders and 138,395 teachers were surveyed. We use this sample for our descriptive analyses and analyses of cross-school teacher sorting. However, in calculating descriptive statistics, the teacher sample falls to 116,650 due to observations missing data on at least one of our teacher quality variables. Further, in our cross-classroom analyses of teacher sorting, we limit our sample to teachers stating that the target classroom they report on in the teacher questionnaire is representative of all the classrooms they teach. This exclusion—which helps to account for possible differences across classrooms resulting from within-school tracking of students—results in the loss of just under 15% of the public school teacher sample, from 116,650 to 98,625.
TALIS contains rich information on teacher and student characteristics at the school and classroom levels. Teachers report information related to their training, experience, and other qualifications, while school leaders and teachers are asked about the characteristics of their schools and students. The TALIS data include several measures of student disadvantage reported as percentages in the school or classroom: language minority status, economically disadvantaged students, and students with special needs. As a result, we can examine the distribution of teacher characteristics across various dimensions of student disadvantage.
We also use data from the OECD’s 2012 PISA to derive a set of country-specific education policy variables from each of the 32 education systems participating in TALIS 2013. PISA has been administered internationally to 15-year-old students in reading, mathematics, and science every 3 years since 2000. In the 2012 PISA, 65 countries or education systems participated, reaching 510,000 students. In addition to student tests, PISA administers questionnaires to school leaders and representatives of each education system. From these data we derive several national policy variables, including cross- and within-school ability grouping, teacher performance incentives, school accountability, and school autonomy. We use these variables to examine how teacher sorting and distribution differ across various national education system policies.
Variables
To examine differences in the characteristics of teachers across schools and classrooms, we use school- and classroom-level dependent variables from TALIS (see Table A1 in the Appendix for names and coding schemes of all TALIS variables). For the school-level analyses, we include proportions of language minority students, low-SES students, and students with special needs in each school. These percentages, which are reported by school principals, are grouped ordinally: 0% to 10%, 11% to 30%, 31% to 60%, and more than 60%. 2 For classroom-level analyses, we use classroom-level proportions of students with the same characteristics but reported by teachers. These percentages are reported in the same ordinal categories as school-level percentages.
To predict school and classroom characteristics, we use several teacher-related variables: total years of teaching experience, teacher education level, and self-reported efficacy in instruction. Teacher experience and education are important for two reasons: First, there is considerable evidence of a positive association between these variables and student achievement (Greenwald, Hedges, & Laine, 1996; Rockoff, 2004). Second, teacher education level and experience determine teacher salaries across most of the globe, so they represent substantial financial resources (OECD, 2015). Teacher self-efficacy, the TALIS measure that most closely aligns with teachers’ own sense of effectiveness, has also been positively linked with student achievement (e.g., Tschannen-Moran & Barr, 2004).
The teaching experience variable is a continuous measure of total years. The teacher education level variable is also continuous, representing the total years of education a teacher has received. 3 Teacher self-efficacy is a continuous index variable derived from teacher questionnaires, which ask teachers how well they can accomplish activities in the areas of classroom management, instruction, and student engagement (OECD, 2014b, p. 196). 4 Finally, we used principal component analysis to combine these three teacher variables into a fourth measure, a composite “overall teacher quality factor” (Lankford et al., 2002). 5 We use all four teacher measures in most analyses, with the exception of the analysis of variance (ANOVA) and robustness checks described in more detail below, for which we use only the overall teacher quality factor.
In addition to the TALIS data, we use several country-specific national context variables including per capita gross national income (GNI) measured in 2012 U.S. dollars (converted using purchasing power parity, or PPP) and income inequality as measured by the 2012 Gini index. The Gini index is continuous, ranging from 0 to 1. A higher number represents greater income inequality (Central Intelligence Agency, 2017). We also use per pupil educational expenditures expressed in 2012 U.S. dollars using PPP, from the OECD’s 2015 Education at a Glance report.
To examine our hypotheses related to the impact of national policy on teacher sorting patterns, we use four sets of education policy and system variables from the 2012 PISA: (1) cross- and within-school ability grouping, (2) teacher performance incentives for career advancement and financial bonus, (3) school accountability using student achievement data, and (4) school autonomy in curriculum and instruction and resource allocation (Table A2 in the Appendix). The cross-school grouping variable is continuous, indicating the country-level percentage of students in schools where principals reported that student records of academic performance and recommendations of feeder schools are always considered for admission. The within-school ability-grouping variable indicates the country-level percentage of students whose principals reported no ability grouping for any class in the school. The teacher career incentive variable indicates the country-level percentage of students in schools whose principals reported that appraisals of teachers lead directly to a change in the likelihood of career advancement. Similarly, the teacher monetary incentive variable indicates the country-level percentage of students in schools whose principals reported that appraisals of teachers lead directly to a financial bonus or another kind of monetary reward. The school accountability variables indicate the country-level percentages of students in schools that use achievement data in the following way: (1) posted publicly and (2) tracked over time by an administrative authority. Finally, school autonomy variables represent country-specific indices of school autonomy in both curriculum and instruction and resource allocation, with greater values representing more autonomy. 6
Empirical Model
For the analysis of teacher sorting, we use a model of teacher assignment into schools and classrooms with varying compositions of disadvantaged students as a function of teachers’ individual characteristics and country-specific contextual effects. Thus, we argue that teachers exercise influence on the schools or classes they teach, through sorting behavior based on their experience or other qualifications. This approach follows previous research that models school and classroom characteristics as a function of teacher characteristics (e.g., Grissom et al., 2015). The following equation describes the model of teacher sorting across schools:
where Sisc represents the school-level characteristics of students for teacher i in school s and country c, Tisc indicates the teacher’s individual characteristics aggregated to the school level, and Dc
At the classroom level, our empirical model is as follows:
where CLisc represents the classroom characteristics of students for teacher i in school s and country c, Tisc indicates the teacher’s individual characteristics, and Dc represents dummies for country-specific contextual effects. For cross-classroom analyses, we also add dummy variables for the subject matter of the teacher’s target classroom. This approach allows us to control for potential cross-subject differences in teacher sorting. Additionally, we run robustness checks to examine whether we find differences within subject groups. To avoid excessive loss of observations in this analysis, we combine reading with social science and math with science. However, we include subject-specific dummies to adjust for possible differences between the combined subjects.
As with the cross-school analysis, we use ordinal logits for the cross-classroom analysis of teacher sorting. For both cross-school and cross-classroom analyses, we also enter into our models a set of country-specific education policy and system variables (Ec
Results
We first examine descriptive statistics for the teacher workforce in public schools across the 32 countries and education systems that participated in TALIS 2013 (Table 1). Across the full sample, we find that average teacher education level and experience are nearly 16 years each. Whereas teacher education level ranges from 11 to 18 years, experience ranges from 0 to 58 years. Average self-efficacy is 12.69 and ranges from 2.949 to 15.870. The mean value of the overall teacher quality factor is .004, ranging from −3.049 to 6.069. 7
Teacher Qualification Attributes in 32 Countries and Education Systems
Note. Samples were weighted by the teacher sampling weight.
The measure of teacher education provided in TALIS is categorical based on the 1997 International Standard Classification of Education (ISCED) classification. We converted the variable into a continuous one indicating years of teacher education by exploiting the PISA 2012 technical report (p. 444). The report provides the table of mapping ISCED to years for each participating country.
The index of overall teacher quality factor was derived from the principal component factor (PCF) analysis using the above three teacher-level variables: years of teacher education, years of teaching experience in total, and self-efficacy in instruction. A likelihood-ratio test of independence against the saturated model indicates that the chi2 statistic is 342.18, significant at 0.1%. A factor score was calculated for each individual teacher from the PCF analysis.
Research Question 1: Teacher Sorting across Schools
Table 2 presents the ANOVA results of the overall teacher quality factor by education system, sorted according to per capita GNI. Column 2 provides mean squares between schools, or variation in the teacher quality factor due to differences across schools. Column 3 reports mean squares within schools or variation due to differences across individual teachers. The final column indicates the intraclass correlation (ICC), which represents the percentage of total variation in the teacher composite measure due to between-school variance. Mean squares between schools range from 0.781 in England (low cross-school inequality) to 3.912 in Mexico (high inequality), while within-school variance ranges from a low of 0.391 in Portugal (low cross-classroom inequality) to 1.453 in Serbia (high inequality). The ICC ranges from −0.002 in the Netherlands to 0.220 in Mexico. Among all participating systems, Mexico has the most variance in the overall teacher quality factor explained by cross-school differences (ICC = 0.22), followed by Chile (0.182) and Brazil (0.155). In the Netherlands, with an ICC of −0.002, within-school differences explain virtually all of the variance in the teacher quality composite, while cross-school differences explain essentially none of this variance. 8 Because an ICC lower than 0.5 indicates that within-school differences explain a greater percentage of variance in the overall teacher quality factor than cross-school differences, within-school teacher sorting is greater than cross-school sorting in all education systems in our sample.
Analysis of Variance of Overall Teacher Quality Factor by Country/System
The values for gross national income (GNI) per capita are in 2012 U.S. dollars converted using purchasing power parities (PPPs).
Mean squares between schools were calculated by dividing sum of squares of grand means minus school means by the number of schools in the country.
Mean squares within schools were calculated by dividing the sum of squares of individual values minus school means by the number of teachers.
The intraclass coefficient of correlation is calculated as
p < .05. **p < .01.
In Figure 1, we plot countries according to mean squares between schools (x axis) and within schools (y axis) of the overall teacher quality factor. Further, we divide the countries into quadrants according to between- and within-school variance in the composite teacher measure by using each country’s mean squares average. We find that most countries are in either the high-between and high-within quadrant (particularly Croatia, Mexico, Serbia, and Sweden) indicating high inequality along both dimensions of teacher sorting, or the low-between, low-within quadrant (including the United States, England, and Canada), suggesting less sorting. One notable exception is Singapore, which has a high degree of within-school variance but relatively low cross-school variance. Only two countries fall in the high-between and low-within quadrant: Brazil and Israel.
Between- versus within-school variation of overall teacher quality factor by country/system.
Figure 2 allows us to examine the relationship between national income (GNI) and the ICC. In general, we find that higher income countries have a lower value of the ICC, suggesting less teacher sorting across schools and more sorting within schools, according to the composite teacher measure. This pattern is largely driven by the three Latin American countries in the sample—Brazil, Chile, and Mexico—as they all have high ICCs but relatively low GNI per capita. This pattern is not surprising, given the overall high degrees of both income inequality and educational inequality across Latin America (Cox, 2010).
Intraclass correlation of overall teacher quality factor between schools GNI per capita.
Although the ANOVA results reported in Table 2 and Figures 1 and 2 provide important evidence regarding the distribution of the teacher composite variable, they do not indicate toward whom this teacher measure is skewed. To explore whether teacher characteristics are skewed toward certain types of students, we estimate ordered logit regressions of school-level student characteristics on teacher characteristics (Table 3). Coefficients indicate the likelihood that teachers with certain characteristics work in schools with greater percentages of a given student characteristic. Across four teacher quality measures and three measures of student disadvantage, we find only one statistically significant result: Teachers with greater educational attainment are less likely to work in schools with greater concentrations of low-SES students, a clearly inequitable result.
Teacher Distribution Across Schools Within Countries by Student Characteristics
Note. Ordered logit regressions of student characteristics in school were run on the school-averaged values of teacher characteristics including overall teacher quality factor, years of teaching experience in total, years of teacher education, and self-efficacy in instruction, respectively. Country dummies were entered in order to adjust for possible country-specific differences. The dependent variables are classified into four categories: 0%~10%, 11%~30%, 31%~60%, and more than 60%. Schools were weighted with the sampling weight. Robust z statistics are in parentheses based on the standard errors adjusted for clustering of countries.
p < .05. **p < .01.
To summarize results for Research Question 1, we find that teacher sorting appears to occur across the sample, although within-school inequality in the overall teacher quality factor is greater than cross-school inequality in all countries. In terms of the relative influence of cross-school versus within-school variance, the influence of cross-school inequality appears to be higher in lower income countries (especially in the three Latin American countries in our sample), while the importance of cross-classroom inequality increases with national income. We find little evidence of teacher sorting according to school-level student composition, with the exception of disproportionately more educated teachers working in schools with higher SES students.
Research Question 2: Teacher Sorting Across Classrooms
Although cross-school teacher sorting raises equity concerns, the classroom-level, student-teacher match represents the most concrete measure of whether disadvantaged students have equal access to qualified teachers. However, it is important to note that in contrast to the ANOVAs reported above, in which we compared cross-school teacher distribution to within-school distribution, the cross-classroom analysis for Research Questions 2 and 3 estimates relationships between teacher and student characteristics across classrooms within countries, not within schools. Consequently, we cannot use these results to compare cross-school to within-school patterns. 9 Instead, our approach for Research Questions 2 and 3 allows us to compare cross-classroom teacher sorting among countries with different sets of education policies. As we describe below, our analysis for Research Question 3 also allows us to examine whether national policies are significantly related to cross-school and cross-classroom sorting patterns.
Table 4 reports the results of ordered logit regressions of classroom-level student characteristics on teacher characteristics, controlling for teacher subject matter to account for differences in teacher sorting across subjects or departments. 10 We find inequities in teacher distribution across nearly all of the measures of student disadvantage and teacher qualifications. Teachers with higher values of the overall teacher quality factor and more experienced teachers are less likely to teach in classrooms with more low-SES students and students with special needs. Additionally, teachers with more years of education are less likely to teach in classes with greater proportions of language-minority students. In contrast to these equity-dampening results, teachers with higher self-efficacy are more likely to teach classes with greater concentrations of low-SES students.
Teacher Distribution Across Classrooms Within Subjects and Countries by Student Characteristics
Note. Ordered logit regressions of student characteristics in class are run on the individual teacher characteristics including overall teacher quality factor, years of teaching experience in total, years of teacher education, and self-efficacy in instruction. Country dummies were entered in order to adjust for possible country-specific differences. The dependent variables are classified into four categories: 0%~10%, 11%~30%, 31%~60%, and more than 60%. Teachers are weighted with the sampling weight. Robust z statistics in parentheses were calculated based on the standard errors adjusted for clustering of countries.
p < .05. **p < .01.
Although the negative direction of these coefficients signals inequities in teacher distribution, it is difficult to assess their magnitude. Unlike linear models, it is not easy to interpret the coefficient estimates of nonlinear models, in particular models with multilayered outcomes like ordered logits. One meaningful way to interpret these results is to determine how a marginal change in one regressor is related to a change in the distribution of the outcome variable. However, it is difficult to do so because marginal changes in nonlinear models are a highly nonlinear function of all the coefficients in the model. More importantly, related to the reporting of interaction terms, it is not possible to separately estimate the marginal effect of any interaction term from the marginal effects of the component terms. An alternative approach is to interpret the ordinal response model in terms of odds ratios for cumulative probabilities (Long, 1997). To illustrate an example from Model 2 of Table 4, the coefficient of years of teaching experience as it relates to the percentage of students in a classroom with low SES can be transformed to the odds ratio of having 0% to 10% versus 11% and more low-SES students in class. Specifically, for a 1-year increase in teaching experience, the odds of having 0%–10% low-SES students in class are 1.011 times greater, holding all other variables constant. Or teachers with 10 additional years of experience have 10.6% greater odds of having 0%–10% low-SES students in their classrooms. 11
In Table 5, we estimate the same ordered logit regression model as in Table 4, Model 1 (using the overall teacher quality factor as a teacher-level regressor) for subsamples of reading/social sciences teachers combined and math/science teachers combined. We use this approach as a robustness check to examine whether subject-matter differences influence the results presented above. We find that all coefficients in Table 5 are negative but only two are statistically significant. For reading/social science teachers, we find that the overall teacher quality factor is negatively related to greater concentrations of students with special needs. For math/science teachers, we find a statistically significant negative relationship for low-SES students. In other words, results of the analysis using subject-matter subsamples are quite similar to the results of the full sample using subject-matter controls. However, the loss of sample size due to separating the sample by subject area appears to have reduced statistical significance.
Robustness of the Results: Teacher Distribution Across Classes by Subject
Note. For each subsample specific to the subject, we repeated the ordered logit regression analysis reported in the first model of Table 4. Teachers are weighted with the sampling weight. Robust z statistics in parentheses were calculated from the standard errors adjusted for clustering of schools.
p < .05. **p < .01.
We conducted two additional robustness checks to address a potential problem related to the TALIS country sampling design. In 2013, eight countries participating in TALIS also participated in the TALIS-PISA Link program, which allowed for analysis of PISA data using TALIS variables. In these countries (Australia, Finland, Latvia, Mexico, Portugal, Romania, Singapore, and Spain), the TALIS sample was drawn to be representative at the student level rather than the teacher level. This sampling difference may influence the results in ways that are difficult to predict. In the cross-classroom analysis reported here, we ran the same model but excluding these countries. Similarly, to address the issue that some countries chose to survey teachers in upper secondary schools in addition to lower secondary schools, we ran an additional analysis omitting 10 countries in which TALIS was administered in upper secondary schools (Abu Dhabi [United Arab Emirates, or UAE], Australia, Denmark, Finland, Iceland, Italy, Mexico, Norway, Poland, Singapore, and UAE). In these two additional analyses, our results did not differ substantially from those reported here. Finally, our cross-country analyses include country-specific fixed effects to address differences in sampling design across countries. As a result, we believe that our results are robust to sampling differences across countries. 12
To summarize results for Research Question 2, while cross-classroom teacher sorting does appear to be global in scope (at least across the 32 education systems in our sample), the nature of this sorting differs across teacher qualification variables. Teachers with higher overall qualification measures and teachers with more experience are less likely to work in the classrooms of low-SES students and students with special needs. The negative relationship of the overall teacher quality factor with the proportion of special needs students holds across a subsample of reading/social science teachers, while the negative relationship with low-SES students holds for the subsample of math/science teachers. Further, across the full sample, teachers with more years of education are less likely to work with language-minority students. From the perspective of disadvantaged students, teacher sorting applies to all students in some way or another, with each group facing statistically significant inequities in at least one of the four teacher qualification measures.
Research Question 3: Relationship Between National Context and the Distribution of Teachers
Although our results to this point demonstrate equity challenges across countries, they do not suggest context- or policy-related explanations or solutions to these challenges. With our final research question, we explore relationships between national factors and teacher distribution. We begin with a descriptive examination of factors that may help to explain cross-national inequitable (or equitable) patterns of teacher distribution (Table 6). In examining eight policy-related variables from the 2012 PISA, we find a great deal of cross-country variability in most of the policy-related variables. For example, cross-school ability grouping, which measures the percentage of students whose principals report that student records of academic performance and recommendations of feeder schools are always considered for admission, ranges from 4% in Finland to 97% in the Netherlands, with a cross-country average of 46%. Within-school ability grouping (reported as of the percentage of students whose principal reported no ability grouping for any class in the school) ranges from 1% in England to 59% in the Czech Republic (country-level measures of each of these variables are reported in supplementary Table A2 in the Appendix). We also find large cross-country variability in teacher incentives for career development and financial bonuses, school accountability in terms of publicly posting and tracking achievement data, and school autonomy in curriculum and resource allocation.
Education Policy and Country Characteristics Across 32 Countries and Education Systems
Percentage of students in schools whose principals reported whether student records of academic performance and recommendations of feeder schools are always considered for admittance.
Percentage of students whose principal reported no ability grouping for any class.
Percentage of students in schools whose principal reported that appraisals of teachers lead directly to a change in the likelihood of career advancement.
Percentage of students in schools whose principal reported that appraisals of teachers lead directly to a financial bonus or another kind of monetary reward.
Percentage of students in schools that use achievement data for public posting.
Percentage of students in schools that use achievement data tracked over time by an administrative authority.
The indices of school autonomy were retrieved from the PISA 2012 database.
In Table 6, we also report sample-wide averages of per capita GNI, income inequality as measured by the Gini index, and educational expenditures per pupil in 2012 dollars. These measures illustrate a fair degree of social and economic diversity across the country sample. Per capita GNI ranges from $11,421 in Serbia to $57,799 in Singapore, while the Gini ranges from 0.249 (lowest inequality) in Denmark to 0.55 (highest inequality) in Brazil. Finally, per pupil expenditures range from $1,682 in Malaysia to $12,359 in Norway (Table A2, Appendix).
To examine how policy variables relate to cross-school distribution of teacher qualification variables, we focus on economically disadvantaged students, estimating ordered logit regressions of the percentage of low-SES students in school on each of the teacher qualification variables, adding interactions with the eight policy-related variables (Table 7). We also include country-level controls for GNI per capita, education expenditure per pupil, and Gini index of income inequality, as well as country-specific fixed effects. As with the cross-school analysis reported in Table 3, we find few statistically significant results and only one statistically significant interaction term: Teacher sorting by education level is exacerbated in schools that use cross-school ability grouping, meaning that as this type of ability grouping increases, teacher education level is significantly lower in schools with higher percentages of low-SES students.
Heterogeneity in the Teacher Sorting Across Schools by Education Policy and System
Note. Ordered logit regressions of student characteristics in schools were run on the school-averaged values of teacher quality attributes; country characteristics including GDP per capita, education expenditure per pupil, and Gini index of income inequality; and other country-specific effects. The dependent variables are classified into four categories: 0%~10%, 11%~30%, 31%~60%, and more than 60%. Teachers are weighted with the sampling weight. Robust z statistics in parentheses were calculated based on the standard errors adjusted for clustering of countries.
p < .05. **p < .01.
In Table 8, we extend our analysis of key policy factors to cross-classroom differences. Specifically, we estimate ordered logit regressions of the proportion of low-SES students on teacher characteristics (similar to the model reported in Table 4), interacting each of the eight policy-related variables with the four teacher qualification variables. We include subject matter dummy variables to control for subject-specific differences in teacher sorting. As in the cross-school model reported in Table 7, we also include country-level controls for GNI per capita, education expenditure per pupil, and Gini index of income inequality, as well as country-specific fixed effects.
Heterogeneity in Teacher Sorting Across Classes Within Subjects by Education Policy and System
Note. Ordered logit regressions of student characteristics in class were run on individual teacher quality attributes; country characteristics including GDP per capita, education expenditure per pupil, and Gini index of income inequality; and other country-specific effects. The dependent variables are classified into four categories: 0%~10%, 11%~30%, 31%~60%, and more than 60%. Teachers are weighted with the sampling weight. Robust z statistics in parentheses were calculated from the standard errors adjusted for clustering of schools.
p < .05. **p < .01.
We find many more significant results (both positive and negative) in the cross-classroom analysis, relative to the cross-school analysis. The first panel of Table 8 reports the results of ability grouping policy variables interacted with the four teacher qualification variables. We find that a few of the policy variables may be able to offset consistently inequitable teacher sorting arrangements. First, we find that teachers with higher values of the overall teacher quality factor (Column A) and teachers with more experience (Column B) are significantly more likely to work in classrooms with low-SES students in countries that more frequently practice cross-school ability grouping. We find the same in countries that are less likely to practice within-school ability grouping. In other words, the degree of cross-classroom teacher sorting falls as countries use more cross-school ability grouping and less within-school ability grouping. However, we find no statistically significant ability grouping coefficients in terms of teacher education level or teachers’ self-efficacy in instruction.
In terms of the teacher incentive policy variables (Panel 2 of Table 8), teacher sorting according to both the overall teacher quality factor and experience increases in countries where teacher incentives for career development are implemented. However, in countries implementing bonuses, teachers with higher overall factor scores and more experience are more likely to teach in classes with low-SES students, meaning that such bonuses are negatively associated with cross-classroom sorting by overall teacher quality and seniority.
In the third panel of Table 8, we find that teacher sorting by overall teacher quality factor and experience increases in countries where student achievement data are publicly posted for school accountability, but keeping track of data is positively associated with the likelihood of more experienced teachers working with low-SES students. Finally, teacher sorting by overall teacher quality and seniority increases as school autonomy in curriculum and instruction increases. In contrast, teachers with greater self-efficacy are more likely to work with low-SES students as autonomy in resource allocation increases.
As with the results reported in Table 4, we interpret a select coefficient from the results reported in Table 8. In the second column of Table 8, an increase of 1 SD (16.19%) in the country-level percentage of students whose principals reported no ability grouping for any class is associated with a decrease of 6.7% in the odds of a teacher with 10 more years of experience working with lower SES students. In contrast, in the next row, the odds of a teacher with 10 more years of experience are 6.2% greater with an increase of a country-level standard deviation (21.44%) in the percentage of students in schools whose principals reported that appraisals of teachers lead directly to a change in the likelihood of career development. Additionally, the odds become 10.0% greater for a standard deviation (23.86%) increase in the percentage of students in schools that use achievement data for public posting. 13
In Table 9, we check the robustness of our results by regressing the percentage of low-SES students on the overall teacher quality factor interacted with policy variables across subsamples of teachers divided according to subject: reading/social science and math/science. Among reading/social science teachers, we find many of the same results of the full sample, including equity-enhancing interactions for no ability grouping and financial bonuses for teachers and equity-reducing interactions for publicly posting achievement data and school autonomy in curriculum. In this case, keeping track of data has a positive interaction with the overall teacher quality factor, while autonomy in resource allocation has a negative interaction with overall teacher quality. For math/science teachers, however, we find no significant interactions between overall teacher quality and the policy variables, suggesting that the sorting results may be driven primarily by reading/social science teachers.
Robustness of the Results: Heterogeneity in the Teacher Sorting Across Classes by Education Policy and System by Subject
Note. For each subsample specific to the subject, we repeated the ordered logit regression analysis reported in the first column (A) of Table 7. Teachers are weighted with the sampling weight. Robust z statistics in parentheses were calculated from the standard errors adjusted for clustering of schools.
p < .05. **p < .01.
To summarize the results for Research Question 3, we find many more cross-classroom than cross-school interactions. Across schools, only one policy variable has a negative interaction with teacher qualification variables, meaning that it is associated with greater teacher sorting across schools: cross-school grouping, with teacher education level. In contrast, we find many significant interactions across classrooms, both negative and positive. Although cross-classroom teacher sorting appears to be global in nature, at least among education systems participating in TALIS, several national policy variables—including cross-school grouping, lack of within-school ability grouping, financial bonuses for teachers, and keeping track of data—are negatively associated with teacher sorting by overall teacher quality factor or experience. In contrast, teacher incentives for career advancement, public posting of achievement data, and autonomy in curriculum and instruction appear to exacerbate sorting of teachers by overall teacher quality factor or experience. We found similar results in a subsample of reading and social sciences teachers but not math and science teachers, suggesting that the nature of teacher sorting may differ across subjects.
Discussion and Conclusion
The primary question we examined in this study is whether teacher sorting is a global phenomenon. We begin our discussion by noting that most of the countries in our sample are relatively wealthy European, East Asian, and North American members of the OECD and by no means represent the entire globe. Most of the developing world is missing from this analysis. However, our examination of teacher distribution across a diverse set of 32 countries and education systems serves as a warning for all countries, regardless of national income. Although we found evidence of teacher sorting in all countries, the influence of cross-school inequality appears to be higher in lower income countries, especially in the three Latin American countries in our sample, which are all plagued by economic, social, and educational inequality. However, we find little evidence of teacher sorting according to school-level student composition, with the exception of disproportionately lower educated teachers working in schools with lower SES students.
Turning our focus to cross-classroom sorting, we find many reasons for concern; in all 32 countries in our sample, within-school differences explain more of the variation in overall teacher quality than cross-school differences. However, the importance of cross-classroom inequality increases with national income, suggesting that cross-classroom teacher sorting is an even greater concern for higher income countries. Further, while we find ample evidence of cross-classroom teacher sorting, the degree of sorting differs across teacher qualification variables. While teachers with higher overall qualification measures and teachers with more experience are less likely to work in the classrooms of low-SES students and students with special needs, teachers with more years of education are less likely to work with language-minority students. In other words, students from each of the three categories of disadvantage we study face some type of inequity in terms of cross-classroom teacher sorting.
Although our results suggest that teacher sorting may be a global phenomenon, we also find several national exceptions and surprises that merit further investigation. Portugal has lower between- and within-school inequality in the overall teacher quality factor than Finland, which has gained considerable attention due to its strong performance on PISA and relatively high degree of cross-school equity in student performance (Sahlberg, 2007). In fact, Portugal has the lowest within-school inequality in the sample, while England has the lowest cross-school inequality, as measured by mean squares between schools of the overall teacher quality factor. Much less surprising, we find that Mexico has the highest level of cross-school inequality in the overall teacher quality factor; additionally, Brazil, Chile, and Mexico have the highest ICCs of the overall teacher quality, meaning that cross-school differences explain more variation in the overall teacher quality factor in these three Latin American countries than in the other countries in the sample.
Although South Korea has been identified as a country with a high degree of equity in teacher distribution (e.g., Akiba et al., 2007; Luschei et al., 2013), cross-school inequalities in the overall teacher quality factor in Korea are higher than in many other countries, but within-school inequality in Korea is relatively low. This result may reflect rising inequality in Korea and/or increased use of accountability measures and ability grouping. Evidence from PISA (OECD, 2013) shows that between 2003 and 2012, the slope relating student SES to math achievement in Korea steepened, indicating an increase in educational inequality. During this period, the percentage of Korean students in schools with no ability grouping dropped from 26.2% to 10.0%. Accountability practices also increased during this period, as the percentage of students in schools where school performance is compared against regional or national benchmarks increased from 62.0% to 70.2%, and the percentage of students in schools where student assessment is used to monitor teachers’ practice increased from 70.6% to 84.1% (OECD, 2013).
Implications for Policy
Our results suggest that education policymakers across the globe must work harder to ensure access of disadvantaged children to qualified teachers. Even countries with relatively high cross-school equity in teacher distribution nonetheless face daunting challenges in terms of cross-classroom inequities. The contrast between cross-school and cross-classroom distribution of teachers raises several important considerations for policymakers. First, prior evidence suggests that nations can, with some concerted effort, keep cross-school teacher sorting in check through policies like centralized assignment of teachers to schools, mandatory rotation of teachers across schools, and incentives for teachers to work in difficult-to-staff schools, as is the case in South Korea (Kang & Hong, 2008). However, efforts to enact such approaches in decentralized countries like the United States must contend with the fact that related policies are often implemented at the district level, not regionally as in Korea. One possible solution to this challenge is the centralization of teacher assignment processes to county offices of education or other supradistrict entities that have a broader view of teacher vacancies and needs and may be better positioned to deploy teachers in a way that enhances equity (Akiba & LeTendre, 2009). However, such policies are likely to have limited influence on the sorting of teachers across classrooms within schools, which has emerged from our results as a pressing concern, especially in more developed countries.
Because school conditions are controlled, cross-classroom sorting is likely to result from teachers with more experience (or other qualifications) exercising influence to choose classrooms based on their preferences for more advantaged students, who may also be easier to teach. By the same token, teachers with advanced skills or knowledge (such as those who teach Advancement Placement Calculus in the United States) may, by virtue of these skills, teach more advantaged students who are more likely to enroll in such courses. Although little research has examined cross-classroom teacher distribution patterns, a few studies from the United States have found that teachers with stronger qualifications are more likely to work in the classrooms of more advantaged students (Grissom et al., 2015; Kalogrides et al., 2013). Researchers have attributed these differences to social or micropolitical processes, such as teachers’ use of school-based seniority or social capital to select classes. However, this research has not given as much consideration to the role of within-school tracking or segregation of students, which creates differences in student composition and the desirability of teaching different classrooms within schools.
The challenge of cross-classroom teacher sorting suggests two types of policy interventions: (1) incentives or support for teachers who work in classes with stronger concentrations of disadvantaged students (Feng, 2010), or (2) reducing within-school ability grouping or segregation of students so that classrooms are more homogeneous and teachers have fewer differences to sort across. Although the first option may be more feasible in the short term, the second is likely to have a more positive impact on educational equity due to its potential to not only equalize teacher resources across classrooms but also provide disadvantaged students with positive peer influences of higher achieving or more advantaged classmates (Harris, 2010). In terms of equity, either of these two approaches will be superior to current practices in the United States that allow more experienced teachers to select their classrooms (Grissom et al., 2015; Kalogrides et al., 2013).
Our analysis provides some evidence regarding national policies that may mediate or exacerbate cross-classroom teacher sorting. First, we find that in countries that are more likely to practice ability grouping across schools, teachers with higher teacher quality factor scores and levels of experience are significantly more likely to work in classrooms with more low-SES students. We find parallel increases in countries that are less likely to use ability grouping within schools. In other words, ability grouping across schools is negatively associated with cross-classroom teacher sorting, while ability grouping within schools is positively related to sorting. Both results confirm our hypotheses discussed above. In the first case, cross-school ability grouping may reduce cross-classroom sorting by making schools more homogeneous and creating fewer student differences to sort across. Second, grouping of students within schools is likely to make schools more heterogeneous across classrooms. In countries that avoid this practice, there may again be fewer differences to sort across.
In contrast to our hypothesis that teacher performance incentives and accountability mechanisms will consistently create incentives for teachers to sort across schools and classrooms, we find that career advancement incentives are negatively related to teacher sorting, while financial bonuses are positively related to sorting. Similarly, whereas publicly posting achievement data is negatively related to teacher sorting by overall teacher quality factor and experience, keeping track of data is positively associated with teacher sorting by experience. These results suggest that there are multiple pathways among teacher performance incentives, accountability mechanisms, and teacher sorting, none of which is completely predictable.
Finally, we find that autonomy in curriculum and instruction is negatively associated with teacher sorting by overall teacher quality factor and experience, while autonomy in resource allocation is positively associated with cross-classroom sorting by teacher self-efficacy. In the first case, curricular/instructional autonomy may provide teachers with less concern about the nature of their students, or it may provide greater say to all teachers (not just teachers with greater seniority) to choose classroom assignments.
Together, these results suggest that policies designed to promote educational efficiency—such as student tracking or grouping, teacher performance incentives, school accountability, and curricular/instructional autonomy—have unintended consequences in terms of teacher sorting. Evidence of policy’s influence on teacher sorting also shifts some of the blame for sorting from teachers and their preferences to policies and their consequences.
Implications for the United States
Surprisingly, given previous research on teacher sorting in the United States, our ANOVA of the overall teacher quality factor finds that the United States is in the middle of the pack in terms of teacher sorting. Although the mean squares of the overall teacher quality factor between schools in the United States (0.916) is higher than peer countries like England and Canada (Alberta), this measure is lower than most countries in our sample, as is the mean squares within schools. In fact, the United States actually falls below the OECD average in several PISA-reported measures of horizontal differentiation of students, including the number of educational tracks in school, the percentage of students in general (rather than vocational or modular) tracks, and the percentage of principals who report that students’ academic performance or recommendations from previous schools are always considered for admission (OECD, 2013, Figure IV.2.4). This suggests that U.S. challenges in terms of cross-school sorting of experienced teachers may be more related to teacher labor market features and related policies like seniority-based hiring, single salary schedules within districts, and reliance on local property taxes (Boyd, Lankford, Loeb, & Wyckoff, 2003) than to cross-school tracking of students. Further, this finding demonstrates the importance of further investigating teacher sorting in other countries, which may lead to the identification of even greater inequality in teacher distribution than in the United States.
According to previous literature, the United States also faces equity challenges in terms of within-school teacher sorting (Grissom et al., 2015; Kalogrides et al., 2013). Yet according to the relatively low mean squares within schools of the overall teacher quality factor, the United States is not alone in this regard. The U.S. measure of within-school inequality is lower than measures in every other country in the sample except Portugal, Malaysia, Alberta, Abu Dhabi, and Poland. However, PISA-reported within-school measures of student tracking suggest that the United States practices relatively high degrees of grouping of students by ability within schools. Only 6% of U.S. principals report no ability grouping for any classes within their schools, compared to the OECD average of 24%. Further, 63% of U.S. principals report one type of grouping for some classes, compared to 41% across the OECD (OECD, 2013, Figure IV.2.4). These results demonstrate the persistent challenge of within-school tracking of students in the United States (Oakes, 2005) and suggest that reduction of within-school ability grouping may represent one promising avenue for reducing sorting of teachers across classrooms. However, reducing grouping or tracking practices in U.S. schools will face considerable resistance, especially considering a general trend toward student resegregation in the United States (Orfield, 2014). As Orfield (2014) argues, desegregation of our schools (and by extension our classrooms) will require a “new civil rights agenda for American education.”
Limitations and Areas for Future Research
Several data limitations preclude our investigation of critical questions related to teacher sorting across schools, classrooms, and nations. First, we cannot claim the existence of causal relationships between the national policy variables we examine and teacher distribution patterns, although we did attempt to control for unobserved, country-specific differences. This research is broad, exploratory, and primarily descriptive, indicating the importance of future studies that investigate both causal relationships across countries and local policies and practices. Further, while the ANOVAs give us some perspective on between- and within-school differences in teacher sorting, the TALIS data allow exploration of sorting of teachers according to various teacher qualification measures across classrooms but not within schools. 14 Finally, while the TALIS sample of countries is fairly diverse, they are all mid- to high-income countries, with no truly low-income countries in the sample. Although we believe that the study has global implications, we cannot claim that the sample is globally representative.
Nonetheless, our results do suggest important contextual differences between lower and higher income countries. In particular, our results leave us with a hypothesis regarding teacher distribution in developing countries relative to developed countries: Given the difficulties of living in remote rural areas in developing countries, we expect that cross-school sorting is likely to be much greater in low-income environments due to teachers’ reluctance to live in such areas. In industrialized countries where rural areas are more developed, sorting is more likely to occur according to classroom-level teaching conditions, especially student composition. If future research is able to confirm this hypothesis, policy implications must be adjusted accordingly, with a greater emphasis on rural development and teacher incentives in developing countries, in contrast to concerted efforts to integrate students and even out cross-classroom differences in developed countries.
Finally, whereas much of the U.S.-based literature on teacher sorting has identified racially or ethnically based gaps in teacher qualifications, the TALIS data contain very limited information on students’ ethnicity and race. While we do find some differences in teacher characteristics between language-minority and non-language-minority students, future cross-national research with richer measures of student race or ethnicity should examine whether students of all races, ethnicities, and cultural backgrounds have similar access to qualified teachers.
Footnotes
Appendix
Education Policy Characteristics Across 32 Countries and Education Systems
| Ability Grouping
a
|
Teacher Incentives
b
|
School Accountability
c
|
School Autonomy
d
|
Other Country Characteristics |
|||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Country | Cross-School | Within-School, No Grouping | Salary | Promotion | Data in Public | Track Data | Curriculum | Resources | GNI/Capita (2012 US$, PPP) | Per Pupil Expenditure (2012 US$, PPP) | Income Gini Index |
| Singapore | 82 | 3 | 61 | 96 | 50.8 | 98.8 | −0.25 | −0.36 | 57,799 | 8,528 | 0.463 |
| UAE | 70 | 14 | 58 | 89 | 46.7 | 91.5 | −0.44 | 0.39 | 46,916 | N/A | N/A |
| United States | 36 | 6 | 11 | 57 | 92 | 98.4 | 0.93 | 1.1 | 46,548 | 11,596 | 0.351 |
| Norway | 7 | 54 | 9 | 51 | 53.6 | 84.2 | −0.55 | −0.18 | 44,825 | 12,359 | 0.253 |
| The Netherlands | 97 | 6 | 22 | 70 | 90.5 | 82.1 | 0.96 | 1.26 | 41,682 | 9,507 | 0.281 |
| Australia | 44 | 2 | 13 | 68 | 69 | 91.7 | 0.13 | 0.06 | 40,801 | 9,803 | 0.326 |
| Denmark | 15 | 24 | 4 | 15 | 39.7 | 69.9 | −0.05 | 0.18 | 40,600 | 10,975 | 0.249 |
| Canada (Alberta | 39 | 7 | 3 | 44 | 61 | 92.7 | −0.49 | −0.35 | 40,136 | 8,040 | 0.315 |
| Sweden | 10 | 16 | 87 | 61 | 80.4 | N/A | −0.25 | 0.63 | 39,251 | 9,583 | 0.274 |
| Belgium (Flanders) | 27 | 21 | 0 | 23 | 3.1 | 51.3 | −0.11 | −0.29 | 37,878 | 9,713 | 0.268 |
| Finland | 4 | 36 | 19 | 27 | 1.6 | 47.6 | −0.05 | −0.28 | 36,030 | 8,623 | 0.26 |
| United Kingdom (England) | 28 | 1 | 66 | 87 | 87.1 | 89.9 | −0.39 | 0.08 | 35,299 | 9,802 | 0.39 |
| Japan | 94 | 37 | 27 | 53 | 5.5 | 7 | 1.15 | −0.27 | 35,238 | 8,972 | 0.34 |
| France | 31 | 44 | 42 | 64 | 45.9 | 75.2 | −0.1 | −0.54 | 34,395 | 8,358 | 0.306 |
| Italy | 66 | 24 | 16 | 34 | 40.4 | 30 | 0.36 | −0.59 | 32,110 | 8,442 | 0.327 |
| Spain | 4 | 8 | 9 | 23 | 12.8 | 81 | −0.47 | −0.42 | 31,574 | 8,218 | 0.335 |
| Korea | 67 | 10 | 47 | 63 | 71 | 89.9 | 0.71 | −0.44 | 28,829 | 6,904 | 0.307 |
| Israel | 56 | 2 | 23 | 79 | 48 | 92.7 | 0 | −0.24 | 26,552 | 5,701 | 0.371 |
| Portugal | 37 | 38 | 21 | 42 | 52.4 | 88.7 | −0.68 | −0.48 | 25,519 | 7,037 | 0.338 |
| Czech Republic | 58 | 59 | 72 | 59 | 44.1 | 57.5 | 0.75 | 1.22 | 25,364 | 5,452 | 0.256 |
| Slovak Republic | 53 | 28 | 49 | 72 | 77.1 | 80.6 | 0.48 | 0.78 | 23,194 | 5,316 | 0.25 |
| Estonia | 38 | 11 | 38 | 58 | 34.8 | 78.2 | 0.49 | 0.14 | 20,093 | 5,552 | 0.338 |
| Poland | 19 | 42 | 34 | 57 | 47.8 | 78.1 | 0.37 | −0.34 | 20,034 | 5,764 | 0.298 |
| Croatia | 96 | 8 | 15 | 91 | 25.3 | 87.5 | −0.86 | −0.34 | 19,026 | 3,899 | 0.32 |
| Chile | 39 | 36 | 38 | 67 | 64.5 | 84.9 | 0.12 | 0.57 | 17,312 | 3,225 | 0.503 |
| Latvia | 29 | 18 | 44 | 64 | 32.5 | 57.7 | −0.19 | 0.6 | 16,902 | 4,534 | 0.347 |
| Mexico | 51 | 26 | 42 | 78 | 43.5 | 92.7 | −0.87 | −0.31 | 15,195 | 2,391 | 0.457 |
| Malaysia | 55 | 4 | 75 | 93 | 35.1 | 96.9 | −0.88 | −0.49 | 15,077 | 1,682 | 0.462 |
| Romania | 35 | 10 | 30 | 72 | 67.9 | 69.8 | −0.52 | −0.57 | 14,531 | N/A | 0.273 |
| Bulgaria | 81 | 7 | 29 | 85 | 55.4 | 89.2 | −0.84 | 0.86 | 14,203 | 3,194 | 0.354 |
| Brazil | 21 | 18 | 36 | 57 | 40.9 | 92.3 | −0.42 | −0.32 | 12,537 | 2,677 | 0.55 |
| Serbia | 87 | 5 | 13 | 45 | 57.1 | 56.9 | −0.86 | −0.39 | 11,421 | N/A | 0.387 |
Note. N/A indicates “not available.”
The cross-school grouping variable indicates the country-level percentage of students in schools where principals reported that student records of academic performance and recommendations of feeder schools are always considered for admission. The within-school ability-grouping variable indicates the country-level percentage of students whose principals reported no ability grouping for any class in the school.
The teacher career incentive variable indicates the country-level percentage of students in schools whose principals reported that appraisals of teachers lead directly to a change in the likelihood of career advancement. The teacher monetary incentive variable indicates the country-level percentage of students in schools whose principals reported that appraisals of teachers lead directly to a financial bonus or another kind of monetary reward.
The school accountability variables indicate the country-level percentage of students in schools that use achievement data in the following way: (1) posted publicly and (2) tracked over time by an administrative authority.
School autonomy variables represent principal-reported, country-specific indices of school autonomy in curriculum and instruction and resource allocation, with greater values representing more autonomy.