Student Achievement by School, Poverty, and Race

Atlanta Public Schools has one of the largest racial achievement gaps in the country, which reflects one of the largest racial income gaps in the country. The Education Opportunity Project shows this relationship at the district level for both the black-white gap and the Hispanic-white gap. However, the district-level relationship can hide large variation at the school-level. The graph below shows the relationship between achievement, poverty, and race at the school level for elementry math results.

The graph shows there is little variation in poverty levels for white student populations in APS. The highest challenge index for white students is 11.8 at Bolton. Black and Hispanic populations have much more variation in the challenge index. Black student populations, for example, range from 13.6 to 90.4 in the challenge index. These populations also have a much wider range of content mastery scores than white students as well.

Content mastery is an achievement metric used by the state’s CCRPI system, and is a weighted average of student achievement, where beginner is worth 0, developing is worth 50, proficient is worth 100, and distinguished is worth 150.1

The challenge index measures the percentage of students who are either directly certified2 for free lunch or English Language Learners (ELL). Note that for white and black APS populations, the challenge index is essentially a poverty measure, while the inclusion of English Language Learners has more of an effect on the challenge index for Hispanic populations. This makes comparisons between the Hispanic population and the black and white populations more difficult to interpret.

Notice that there is little overlap between the black and white distributions. There is no overlap in the challenge index and only slight overlap for content mastery (elementary math). The two highest performing schools for black students, Drew and Morningside, have higher content mastery scores than the lowest performing school for white students, Classical. Black students at Drew also out perform white students at Toomer.

Bubble size in this graph is proportional to the number of students represented. Populations of less than 15 are not included. Note that the smallest populations have more variable test results, because a different score for a couple students could change the achievement level significantly.

The next graph includes a trend line for the relationship between content mastery and the challenge index for black students. Notice that the trend line crosses the y-axis at a content mastery score of about 90, near the white student results for Classical and Toomer. This suggests that if the average APS school had a population of black students who were at a similar challenge index to the white student populations, performance for those black students would be similar to white students at Classical or Toomer, but well below most schools. This suggests that some of the district’s black-white gap can be explained by income differences, but not all.3

However, some high performing schools such as Drew or Hope-Hill might be able to close the black-white gap in the absence of income gaps. Note that Hope-Hill’s black student population scored almost 20 points above the trend line for black students. If the poverty level here were on par with the district’s white student population, then this pattern would suggest that Hope-Hill’s black population would be near the district average for white students.4

The next graph uses a different performance metric, the percentage of students who scored distinguished. The view more clearly shows differences in performance for higher-income populations on the left-hand side. Use the filters on the right to change the performance metric, subject, school level, or race/ethnicity categories. When changing the level to high school, note that high school has different test names, so subject also needs to be changed.

The last graph shows the percentage of students scoring proficient at the high school level for all subjects combined except algebra and physical science5.

Milestones subgroup data is from GOSA. Subgroup challenge index is calculated internally.

To get email alerts for similar posts, please subscribe to the blog.


  1. See here for a description of the achievement levels. CCRPI defines content mastery on a scale from 0 to 1.5. We’ve multiplied each weight by 100 for easier interpretation.
  2. A student is directly certified for free lunch if their household receives state anti-poverty aid (SNAP or TANF) or if the student is homeless, an unaccompanied youth, or a migrant. This measure is used instead of the more general “free and reduced lunch” because many APS schools use the CEP option for whole-school free lunch. Poverty rates are lower when using direct certified instead of free lunch, but the rank-ordering of schools is mostly the same.
  3. This is consistent with findings from districts that do not have income gaps. For example, see the third graph in this New York Times article. Student-level analysis with more detailed socio-economic status data also finds that test-score gaps remain after controlling for income. See this article by Fryer and Levitt.
  4. This assumes two ideas: that the trend line for black student test scores and the challenge index would continue linearly for low-poverty black student populations, and that schools like Drew or Hope-Hill that out-perform the trend line would continue to out-perform at a similar level for a very low-poverty population of black students.
  5. The default view leaves out algebra because some clusters have large middle school advanced math programs, so their high school algebra I population is not representative. Physical science is left out because it is not a compulsory course and also not a representative sample.