Our national discourse on school segregation, whether income- or race-/ethnicity-based, tends to focus on the separation of students between schools within districts. There are good reasons for this, including the fact that the majority of desegregation efforts have been within-district efforts. Sometimes lost in this focus, however, is the importance of segregation between districts.

This distinction can be confusing, so consider a large metro area with a central city district surrounded by a group of suburban districts. There may be extensive racial/ethnic segregation of students between schools within those districts, with students of color concentrated in some schools and their White peers concentrated in others. But total segregation across the entire metro area is also a function of segregation between districts - i.e., the degree to which students of certain races or ethnicities are concentrated in some districts and not others (e.g., students of color in the city, white students in the suburbs). In a sense, if we view diversity as a resource, there are multiple "chokepoints" at which that resource is distributed down to the next level—from states to metro areas to districts to schools—and this can exacerbate segregation.

A recent working paper provides one of relatively few pieces of recent evidence suggesting that, in addition to racial and ethnic segregation between districts and between schools within districts, there may be an additional important "layer": segregation within schools.

The paper, written by Charles Clotfelter, Helen Ladd, Calen Clifton, and Mavzuna Turaeva, and published by CALDER, uses detailed administrative data from (no surprise) North Carolina. These data allow the researchers to determine the racial and ethnic composition not only of schools and counties (which in North Carolina mostly define districts), but of individual classrooms as well, including elementary school (4th grade) classrooms, and all math and reading courses and sections in middle schools (7th graders) and high schools (10th graders).

In the paper, segregation is measured with the dissimilarity index, a conventional indicator that essentially compares the composition of smaller units (e.g., schools) to that of the larger units (e.g., counties/districts) in which they are nested. To the degree the composition (in our case, race and ethnicity percentages) of the smaller units varies from that of the larger unit(s), there is segregation. If, in contrast, all of the smaller units are basically the same as the larger unit, this represents integration.

In this analysis, there are three units, or levels (from most to least diaggregate): classrooms, schools, and counties. Total segregation is measured by comparing the "top" and "bottom" levels. That is, how much, on average, do the racial makeups of classrooms differ from those of the counties in which they are located. From the perspective of a "standard" segregation analysis, this is kind of like treating each individual classroom as a separate school.

But there's a chokepoint level between classrooms and counties: schools. That means total segregation can be broken down into two components: 1) compositional differences between classrooms and schools (i.e., within school segregation) and; 2) compositional differences between schools and counties (i.e., between school segregation). Both components determine total segregation, and both might be important in terms of how much they "contribute" to it. Clotfelter et al. calculate the relative importance of both, on average, across the state.

There are a number of interesting results presented in the paper, and it's worth reading in full, but the authors' primary finding is that within school segregation—the sorting of students by race and ethnicity into classes within the same school—is a pretty important factor in North Carolina, particularly in middle and high schools.

For example, looking at 10th grade math classes in 2017, on average, approximately 40 percent of total county-level White/Black and White/Hispanic segregation was within schools (and about 33-35 percent in reading classes). In other words, on average, as much as two-fifths of total (countywide) segregation exists in the form of racial/ethnic differences between classrooms and schools, while the remainder is attributable to differences between schools and counties (between school segregation). In 7th grade math and reading classes, the "contribution" of within school segregation was roughly 25 percent for both race/ethnicity comparisons. 

(As a rough illustration, these results imply that we could theoretically reduce total segregation—or at least total segregation of math and reading classes in later grades—by up to 40 percent without a single student switching schools.)

We would certainly expect to find some degree of segregation within schools (due to tracking), but these figures are surprisingly large, at least to me (though they are not inconsistent with previous analyses of different locations, such as: Kalogrides and Loeb 2013; Conger 2005).

Also noteworthy is the authors' finding that where between school segregation is high, within school segregation tends to be low, and vice versa. You can see this when comparing results between grade levels. For example, high schools in North Carolina tend to be relatively racially/ethnically heterogeneous (lower between school segregation), but there is more sorting of students into classes within those schools (greater within school segregation). Elementary schools (4th grade), in contrast, are comparatively homogeneous in terms of race and ethnicity (more segregation between schools), but segregation within those schools is pretty weak (about 12 percent of total segregation in 2017). Similarly, looking at trends over time (the authors have data for 1998, 2006, and 2017), within school segregation tends to increase as between school segregation decreases, and vice versa. 

The most general and probably most important implication of this paper (and its similar predecessors) is that decades of segregation analyses may be missing a big part of the picture. Without detailed classroom-level data, within school segregation is essentially invisible. Even students in a seemingly integrated district may in a sense still be segregated, and suffer some of the negative effects of segregation, if those students are separated within schools. You can even say that school segregation might be considerably worse than we think it is.

This is also an unsettling idea from a policymaking perspective. For example, if, as the authors show, within school segregation tends to increase when between school segregation decreases. then desegregation efforts, which are almost always focused on reducing between school segregation, might be partially offset by concurrent increases in within school segregation. 

To be clear, this offsetting is not really a zero sum affair. Within school segregation is qualitatively different from that between schools. For instance, two students in a given school, particularly a middle or high school, would still have far greater opportunity for interaction and more equal access to school resources than would students attending different schools, even if the former pair take different courses. But within school segregation nonetheless limits these opportunities, and there is evidence, cited in the paper, that students in advanced courses have greater access to resources, such as more experienced teachers and better curricula (Kalogrides and Loeb 2013; Mickelson 2015). Access to resources varies within as well as between schools.

Within school segregation is also distinct for a different reason: it is basically independent of where people live. It's much easier to move students between classes than it is to move them between schools or districts. That means addressing within school segregation, while far from easy, is a real possibliity (though between-school/district desegregation should of course remain the most urgent goal).

This brings us to the question of why students might be systematically sorted into classes by race and ethnicity, as you can't really address within school segregation without understanding how it happens. This question is complicated and deserves a lot more attention than I will give it here, but suffice it to say that academic tracking is the primary and most obvious culprit (tracking is essentially a form of within school segregation). Clotfelter et al. provide some relevant evidence in this paper. As just one example, they find that 7th and 10th grade Black and Hispanic students were less likely than their white peers in the same school to be enrolled in advanced courses. This would of course generate within school segregation. It also means that some part of within school segregation may represent unequal educational opportunity.

In any case, this is an important paper in that it adds to a relatively thin body of recent work on within school segregation. Substantial sorting of students by race and ethnicity within schools carries pretty serious implications for segregation research and policy alike. It's also a good example of how better data can lead to greater understanding of even the most heavily-researched phenomena.