Every school in the explorer gets a single number, the Demographic Shift Index. It answers one plain question: between the two chosen years, what share of a school’s students would have to be a different race or ethnicity for the earlier year’s makeup to match the later year’s?
A score of 0 means the makeup didn’t change at all. A score of 100 means it changed completely — not one part of the mix stayed the same. Most New Jersey schools, over a span of a few years, land somewhere in the single digits.
Take a school of 100 students. Here is its racial and ethnic makeup in two years. (With 100 students, each share is also a head count, which keeps the arithmetic simple.)
| Group | 2015 | 2025 | Change |
|---|---|---|---|
| White | 50% | 30% | −20 |
| Black | 30% | 30% | 0 |
| Hispanic | 10% | 25% | +15 |
| Asian | 10% | 15% | +5 |
| Native Am. | 0% | 0% | 0 |
| Sizes of the changes, added up | 40 |
Halve that total: the Shift Index is 20.
Three things moved between 2015 and 2025: the White share fell 20 points, the Hispanic share rose 15, and the Asian share rose 5. Add up the sizes of those moves, ignoring whether each was a gain or a loss, and the total is 40. Then take half of it. The Shift Index is 20: about 20 of every 100 students would have to be a different race for the 2015 mix to look like the 2025 mix.
Why halve it? Because every share that leaves one group has to land in another — the shares always add to 100%. The 20 points White lost are the very same 20 points that Hispanic and Asian gained. Counting both the leaving and the arriving counts that movement twice; halving the total fixes the double-count and leaves the real figure.
The index runs from 0 to 100. Some anchors for what a score means:
That last one is the ceiling: a school that is 100% White in the first year and 100% Hispanic in the second scores exactly 100 — White falls 100, Hispanic rises 100, the sizes add to 200, and half of that is 100. There is nowhere further to move.
The index compares shares, not raw numbers, so a school that simply grows or shrinks is not mistaken for one that changed character. A school that doubles from 300 students to 600 while keeping the exact same racial proportions scores 0 — its makeup is identical, there is just more of it. Enrollment still appears next to every school, though, because a sharp jump or drop in the same window is often the clue to what caused a shift: a rezoning, a closure, a new housing development.
The Shift Index measures how much a makeup moved, not which way. A school going from mostly White to mostly Hispanic and a school going the other direction register the same size of change. It is not a measure of "more diverse" or "less diverse," and it does not favor any group — it is a yardstick for the amount of change. The colored dot and the per-group figures beside each school show the direction; the index gives the magnitude.
There are other ways to put a number on the distance between two compositions. Statisticians reach for chi-square distances, the Hellinger distance, information-theoretic divergences like Kullback–Leibler. Most of them share a property we deliberately avoided: they magnify change in the smallest groups. Under those measures a group going from 1% to 3% of a school can count for more than the largest group sliding ten points, because they weigh each change against how small the group already was.
That is the wrong instinct for this question. The Shift Index treats every student as an equal part of the whole: a point of share is a point of share, whether it moves in the biggest group or the smallest. The goal is to understand how much the overall makeup changed, and a child counted in a two-percent group is exactly one child — the same as a child in a sixty-percent group.
The difference is not academic. Tested across New Jersey, a small-group-sensitive measure routinely flags schools whose makeup is all but unchanged — one group ticking from near zero to a couple of percent, often just the year-to-year noise of a few students — and ranks them above schools that genuinely transformed. Total variation distance does not. It answers the plain question, weights students equally, and holds steady against that noise. That is why it is the index.
The index compares years only within one state reporting scheme. When both years are before 2006–07, it uses the five categories collected then: White, Black, Hispanic, Asian (which then included Pacific Islander), and Native American. When both years are 2006–07 or later, it uses all seven: White, Black, Hispanic, Asian, Pacific Islander, Native American, and two or more races. The explorer does not calculate a shift or rank across that boundary. Dropping the later Two-or-more category or silently folding groups would produce a differently defined measure under the same name. The data explainer walks through that history.
A source blank is not automatically zero. The explorer accepts it as a mathematical zero only when the other exhaustive race counts add exactly to the authoritative enrollment total, and names the component whose zero was inferred in the school detail. Otherwise that school is excluded from the chosen pair. One conflicting 1999–00 ID, 3530-050, describes two different institutions in the source; both rows for that school-year are quarantined instead of guessed, summed, or silently overwritten.
A single number cannot explain itself. The index says a school changed; it does not say whether that was a slow drift, as families move in and out over years, or a one-time event, like a rezoning, a grade reconfiguration, or a school closing and its students landing elsewhere. A sudden enrollment swing in the same window is often the tell. Small schools also move more on a handful of students, so a very small school can post a big score from ordinary year-to-year noise — raising the minimum-enrollment filter steadies the comparison. And the index is whole-school: it does not look inside at particular grades or programs.
In symbols, the index is ½ · Σ |change in each group’s share| — half the sum of the absolute changes. Statisticians know this quantity as the total variation distance between two distributions; here it is simply "how much the mix moved." It is computed from the exact reported enrollment counts — taking shares first, rounding only the final score for display — so schools that sit close together are ordered by their true values, not by a rounding artifact.
