School Results vs. Demographics — Every Connecticut District

How far each district's reading and math results sit above or below what its students' family circumstances predict — the ranking that holds income and need constant.

← All statewide rankings · School results at a glance

Picking a town highlights it in every chart and table on this page, and the highlight follows you to the other ranking pages. Clear the box to remove it.

Read this before ranking districts: this page ranks operating districts, not towns, because results are produced by districts and a town in a regional district shares its figure with the other member towns. Pick a town and the district (or districts) that teach its students light up. Each bar is a residual: the state's Performance Index the district earned, minus the index a regression on six family-circumstance variables predicts for a district with its demographics. Zero means results match the prediction. The residuals average to zero across districts, because the fit includes an intercept, so a large share of districts sit below zero (the boxes under the charts print the count); being below zero is a position, not a finding, and a district inside its 90% interval cannot be told apart from zero. The residual is relative to other Connecticut districts and correlational: it does not say why a district sits where it does, and it is not a measure of teaching. Rank 1 is the most negative residual: a sort order, not a verdict. How the residual is built.
—
Furthest below prediction, reading
—
Median residual, reading
—
Furthest above prediction, reading

Reading results minus what demographics predict — every modeled district

ELA Performance Index (0–100, three-year mean ending —) earned, minus the index predicted from the district's share of students eligible for free or reduced-price meals, English learners and students with disabilities, and its median household income, adult college attainment and share of children living with one parent. Points on the index scale. Sorted lowest to highest.

first quarter of the ranking second quarter third quarter fourth quarter your picked town
Loading…

Source: CSDE EdSight PerformanceIndexExport (ELA, all students), three-year mean; demographic shares from EdSight October 1 enrollment; income, attainment and family structure from the US Census ACS 5-year estimates. Regression and residuals built by Data/build_school_peer_model.py into school_performance_model.json, read here from school_performance_page.json (districts[].resid.index_ela_all). Ordinary least squares, one observation per district, unweighted; interval = 1.645 × residual SD × √(1 + leverage).

—
Furthest below prediction, math
—
Median residual, math
—
Furthest above prediction, math

Math results minus what demographics predict — every modeled district

The same construction for the Math Performance Index.

first quarter of the ranking second quarter third quarter fourth quarter your picked town
Loading…

Source: as above, districts[].resid.index_math_all.

The full table

Every modeled district, ranked on the reading residual, with the index earned, the index predicted, the 90% interval, whether the district sits within it, and the math residual. Type a town name to find the district that teaches its students.

Loading…
How this ranking is made

Residual = index earned − index predicted. The prediction is an ordinary-least-squares regression of the three-year mean Performance Index on six variables, fitted across every modeled Connecticut district with complete data; the coefficients, R² and residual standard deviation are in the model file and printed on each town's own School Results page. The 90% interval is the range within which a district's residual is not distinguishable from zero given how much the model leaves unexplained; a district whose residual is smaller than its interval is “within the band”.

  • Districts, not towns. A regional district's residual belongs to every member town alike; the table lists the towns each district serves, and the picker highlights every district that teaches a town's students.
  • The residuals average to zero. The regression includes an intercept, so the residuals sum to zero across districts. How many fall below zero is observed data, printed under each chart, not a fixed share; being below the line is a position, not a finding.
  • Correlational. Six variables cannot capture everything about a community; a residual can reflect what the model omits as easily as what a district does.
  • Small districts. Districts with fewer than 500 tested students are marked in the table because their results can move more from year to year, so a single year’s position is less certain. The mark is a sample-size flag: it does not change the printed interval, which comes from the model’s residual spread and the district’s leverage.
  • Not shown. Districts with fewer than 20 tested students, and districts that are not a town's operator (charters, regional education service centers, the technical high school system).

The full method, every check and the research memo behind it: METHODOLOGY.md and RESEARCH-school-performance-vs-spending.md in the repository.

More school results: test scores, grades 3–8 · academic growth · SAT scores · attendance · spending per student