ContentsCorrelation by level
Correlation by level
In New York City, median household income and the share of adults with a bachelor's degree correlate at r = 0.86 across 55 PUMAs, 0.85 across 181 ZCTAs, 0.77 across 2,192 tracts and 0.71 across 5,455 block groups. The estimates are ACS five-year figures for 2019 to 2023, each place counted once whatever its population.
Change the city or level
City
Level
Income against Education, by census tract, New York
2,192 of 2,324 tracts have both measures and appear in the plot.
Map colour: Adults 25+ with a bachelor's degree or higher
No estimate for education
Income horizontal · Education verticalDrag to brush · Shift-drag to add · click empty space to clear
r = 0.77 across 2,192 areas, unweighted, so every area counts once whatever its population. 132 of 2,324 areas are excluded for missing a value (51 missing Income, 0 missing Education, 81 missing both), and which ones drop out is not random, because an estimate is usually suppressed for a reason that correlates with the thing being measured. This is a correlation between area averages, which is not evidence about individual people and not evidence of cause: two measures can move together across tracts while moving apart inside every one of them.
Correlation
Select areas first: brush the plot, click the map, or activate a row in the table.r = 0.77 across 2,192 areas, unweighted, so every area counts once whatever its population. 132 of 2,324 areas are excluded for missing a value (51 missing Income, 0 missing Education, 81 missing both), and which ones drop out is not random, because an estimate is usually suppressed for a reason that correlates with the thing being measured. This is a correlation between area averages, which is not evidence about individual people and not evidence of cause: two measures can move together across tracts while moving apart inside every one of them.
Click a polygon or a point to pin one census tract here. Shift-click or Shift-drag adds to the selection, and Escape clears it. Changing city or level clears the selection, because these files do not record which tracts contain which block groups.
| Id | ||||
|---|---|---|---|---|
| 36005000100 | not reported | 1.4% | 3,538 | |
| 36047000100 | $168k | 85.7% | 4,839 | |
| 36061000100 | not reported | not reported | 0 | |
| 36081000101 | $129k | 82.4% | 4,600 | |
| 36081000102 | $213k | 87.0% | 6,324 | |
| 36081000103 | not reported | 100.0% | 31 | |
| 36081000104 | not reported | 0.0% | 303 | |
| 36005000200 | $121k | 34.0% | 5,177 | |
| 36047000200 | $81k | 23.6% | 1,203 | |
| 36081000200 | $75k | 26.0% | 2,344 | |
| 36061000201 | $39k | 15.7% | 2,568 | |
| 36061000202 | $40k | 36.6% | 6,934 | |
| 36085000300 | $77k | 55.7% | 2,072 | |
| 36047000301 | $179k | 82.6% | 4,031 | |
| 36005000400 | $98k | 37.1% | 6,481 | |
| 36081000400 | $81k | 15.3% | 4,684 | |
| 36061000500 | not reported | not reported | 0 | |
| 36047000501 | $144k | 87.1% | 4,504 | |
| 36047000502 | $179k | 86.8% | 2,787 | |
| 36061000600 | $34k | 25.3% | 10,179 | |
| 36081000600 | $105k | 24.8% | 3,888 | |
| 36085000600 | $34k | 36.9% | 2,696 | |
| 36047000700 | $189k | 88.1% | 3,983 | |
| 36061000700 | $197k | 84.7% | 9,008 | |
| 36085000700 | $62k | 40.4% | 5,195 | |
| 36081000701 | $137k | 78.8% | 3,559 | |
| 36081000702 | $122k | 66.5% | 4,131 | |
| 36061000800 | $34k | 27.6% | 8,889 | |
| 36081000800 | $113k | 29.7% | 3,617 | |
| 36085000800 | $103k | 27.0% | 4,908 | |
| 36047000900 | $236k | 92.5% | 4,792 | |
| 36061000900 | $222k | 89.0% | 1,860 | |
| 36085000900 | $64k | 43.3% | 2,127 | |
| 36081001000 | $95k | 32.4% | 4,899 | |
| 36061001001 | $130k | 73.2% | 1,720 | |
| 36061001002 | $24k | 12.0% | 6,098 | |
| 36047001100 | $164k | 89.1% | 1,585 | |
| 36085001100 | $59k | 22.3% | 4,203 | |
| 36061001200 | $106k | 58.5% | 4,364 | |
| 36081001200 | $85k | 27.5% | 4,596 | |
| 36047001300 | $173k | 86.4% | 2,408 | |
| 36061001300 | $194k | 87.0% | 5,371 | |
| 36081001400 | $84k | 20.8% | 3,966 | |
| 36061001401 | $94k | 61.9% | 3,395 | |
| 36061001402 | $50k | 47.1% | 3,449 | |
| 36047001501 | $115k | 53.8% | 4,925 | |
| 36061001501 | $124k | 60.6% | 7,777 | |
| 36047001502 | $135k | 82.2% | 5,418 | |
| 36061001502 | $182k | 88.9% | 8,485 | |
| 36005001600 | $43k | 18.3% | 6,011 | |
| 36061001600 | not reported | 42.9% | 7,040 | |
| 36081001600 | $75k | 20.7% | 2,501 | |
| 36085001700 | $53k | 43.2% | 1,518 | |
| 36061001800 | $80k | 51.3% | 6,702 | |
| 36081001800 | $88k | 21.2% | 2,865 | |
| 36085001800 | $90k | 49.3% | 465 | |
| 36047001801 | not reported | 0.8% | 1,564 | |
| 36047001802 | not reported | not reported | 0 | |
| 36047001803 | not reported | 33.3% | 6 | |
| 36047001804 | not reported | not reported | 0 |
Showing 60 of 2,324 matching areas. Sort or search to bring others into view. The map and the plot show all 2,324. Suppressed and unreported values sort last in both directions rather than as zero.
ACS 2023 5-year · boundary vintage 2020 · areas missing either measure are absent from the plot and from the coefficient, and are still drawn on the map in the no-data colour
Change the axes or the weighting
Horizontal
Vertical
- Income unavailable: Already on the horizontal axis.
Weighting
A choropleth is good at where and structurally bad at how do these two things relate. A scatterplot is the reverse: it shows the relationship and throws the geography away. One selection shared between the two shows which places sit off the trend.
Change the level and the coefficient moves, although nothing about the city has changed. A correlation across tracts and the same correlation across block groups are different numbers computed from the same households, because aggregation averages away the variation inside each unit, and how much it averages away depends on how big the units are. Coarser geographies usually produce stronger correlations, which is a property of the arithmetic and not of the places.
That is why the coefficient here never appears without its complete-pair count, its exclusions broken out by axis, and the sentence about area averages. Areas drop out of a correlation for reasons that correlate with the thing being measured (median home value is suppressed exactly where there are few owner-occupied units), so the number describes the areas that survived, and which ones those are is part of the finding.
- Shows
- Which specific places sit off a relationship, with the marginal values and the spatial pattern in view at once.
- Hides
- Anything about individual people. Both axes are area averages, so two measures can move together across tracts while moving apart inside every one of them, and the coefficient still depends on the geography chosen and on which areas had both numbers to begin with.