Returns claiming the earned income credit and Public-good pages per 1,000 users
Where Returns claiming the earned income credit is higher, Public-good pages per 1,000 users tends to be lower. The link is weak: -0.32 across 3,076 counties, on a scale from −1 (opposite) to +1 (in step). That figure compares places with the same population density, income and education, so those differences can't explain it. It holds inside states too. Measured separately within each of 40 states, the middle value of the link is -0.24.
Holds after adjusting
Every place, plotted
One dot for each place reporting both measures. Hover over or tap a dot to see its name and values. The thin line is the straight line that best fits the dots. The bold line, shown for counties, is the link after adjusting for density, income and education. A dot far from the lines is a place that does not follow the pattern.
Loading every place…
What each measure is
Returns claiming the earned income credit
The percentage of tax returns claiming the earned income tax credit, which goes to working households with low incomes. It is a count from tax records of the working poor, not a survey estimate.
IRS Statistics of Income · IRS SOI N59660 / N1
Public-good pages per 1,000 users
Facebook pages for community groups and public-good efforts, per thousand Facebook users. It is per user, not per resident.
Social Capital Atlas · Opportunity Insights civic_organizations_county
Highest and lowest
Where each measure is highest and lowest among the places reporting both (2023), with the other measure beside it. This is shown for every size of place where both are published. Metro-area values are added up from their counties. A measure that cannot be added up across counties, such as a middle value (median), has no metro figure. Counts are shown per 1,000 residents, since raw totals would only list the largest places. Cities and towns and ZIP codes are shown where the Census survey or the CDC publishes them; most other sources stop at the county. They are shown for places of 5,000 residents or more, and only where the survey’s margin of error is small enough for the figure to be usable. The CDC publishes its health estimates for cities in a version adjusted for age, so places with older or younger people can be compared fairly, but not for ZIP codes.
Counties
By returns claiming the earned income credit
| Returns claiming the earned income credit | Public-good pages per 1,000 users | |
|---|---|---|
| Highest | ||
| Kusilvak Census Area, AK | 48.7% | 0.0082 |
| Humphreys County, MS | 47.6% | 0.00349 |
| Tunica County, MS | 46.3% | 0.0129 |
| Holmes County, MS | 46% | 0.0126 |
| Oglala Lakota County, SD | 45.5% | 0.0165 |
| Lowest | ||
| Skagway Municipality, AK | 0% | 0.0368 |
| Daggett County, UT | 0% | 0.0649 |
| McMullen County, TX | 0% | 0.0214 |
| Glasscock County, TX | 0% | 0.0265 |
| Sully County, SD | 0% | 0.0193 |
By public-good pages per 1,000 users
| Returns claiming the earned income credit | Public-good pages per 1,000 users | |
|---|---|---|
| Highest | ||
| Douglas County, MO | 22.5% | 0.195 |
| Bristol Bay Borough, AK | 4.44% | 0.12 |
| San Juan County, CO | 6.38% | 0.112 |
| Lewis County, ID | 16.7% | 0.106 |
| Alpine County, CA | 8.33% | 0.0828 |
| Lowest | ||
| Humphreys County, MS | 47.6% | 0.00349 |
| Winkler County, TX | 21.6% | 0.00361 |
| Jefferson County, MS | 42.3% | 0.00375 |
| Maverick County, TX | 35.9% | 0.00381 |
| Starr County, TX | 44.4% | 0.00414 |
States — Public-good pages per 1,000 users is not published for states
By returns claiming the earned income credit
| Returns claiming the earned income credit | Public-good pages per 1,000 users | |
|---|---|---|
| Highest | ||
| Mississippi | 26.7% | — |
| Louisiana | 24% | — |
| Georgia | 20.6% | — |
| Arkansas | 20.5% | — |
| Alabama | 20.3% | — |
| Lowest | ||
| New Hampshire | 7.91% | — |
| Massachusetts | 9.56% | — |
| Washington | 9.82% | — |
| Minnesota | 9.98% | — |
| Vermont | 10.1% | — |
Metro areas
By returns claiming the earned income credit
| Returns claiming the earned income credit | Public-good pages per 1,000 users | |
|---|---|---|
| Highest | ||
| Zapata, TX | 45.4% | 0.00436 |
| Rio Grande City-Roma, TX | 44.4% | 0.00414 |
| Clarksdale, MS | 42% | 0.011 |
| Greenville, MS | 41.2% | 0.0106 |
| Greenwood, MS | 38.7% | 0.0109 |
| Lowest | ||
| Los Alamos, NM | 3.7% | 0.0314 |
| Jackson, WY-ID | 6.19% | 0.0262 |
| Heber, UT | 6.24% | 0.0176 |
| Edwards, CO | 6.39% | 0.02 |
| Nantucket, MA | 6.55% | 0.0312 |
By public-good pages per 1,000 users
| Returns claiming the earned income credit | Public-good pages per 1,000 users | |
|---|---|---|
| Highest | ||
| Vermillion, SD | 12.9% | 0.0451 |
| Port Townsend, WA | 8.29% | 0.0442 |
| Vineyard Haven, MA | 7.55% | 0.0418 |
| Barre, VT | 9.11% | 0.0411 |
| Ithaca, NY | 10.1% | 0.0406 |
| Lowest | ||
| Eagle Pass, TX | 35.9% | 0.00381 |
| Rio Grande City-Roma, TX | 44.4% | 0.00414 |
| Laredo, TX | 37% | 0.00422 |
| Provo-Orem-Lehi, UT | 10.4% | 0.00431 |
| Zapata, TX | 45.4% | 0.00436 |
Counties furthest from the pattern
Compared with the straight line that best fits all 3,091 counties: the five with far more public-good pages per 1,000 users than their returns claiming the earned income credit would suggest, and the five with far less.
| Returns claiming the earned income credit | Public-good pages per 1,000 users | |
|---|---|---|
| Above the line | ||
| Douglas County, MO | 22.5% | 0.195 |
| Bristol Bay Borough, AK | 4.44% | 0.12 |
| San Juan County, CO | 6.38% | 0.112 |
| Lewis County, ID | 16.7% | 0.106 |
| Cheyenne County, CO | 15.5% | 0.0822 |
| Below the line | ||
| Utah County, UT | 10.4% | 0.0042 |
| Divide County, ND | 4.76% | 0.00812 |
| Harding County, SD | 3.45% | 0.00878 |
| Dallas County, IA | 7.55% | 0.00825 |
| Lincoln County, SD | 7.21% | 0.00842 |
Two series, one county
Both measures over time for a county you pick. The two charts are stacked, so they share only the years along the bottom. You can pick from the thirty largest counties reporting both. This panel covers 2010–2024, and a measure published only once shows as a single dot.
Loading…
Every version of the link
Each row measures the same link, from −1 to +1, in a different way. If the rows mostly agree, the link does not depend on one choice of method.
| Before adjusting for anything | -0.340 | 3,076 counties reporting both |
| After adjusting for density, income and education | -0.319 | the part of the link that is not just how urban, rich and educated a place is |
| Also adjusting for population size | -0.301 | size taken out as well as density |
| Also comparing only within each state | -0.240 | each county is compared only with others in its own state, so this is no longer a comparison between states |
| Giving bigger counties more weight | -0.272 | the adjusted link for the average resident rather than the average county. With this weighting, the data counts for about 286 equally weighted counties |
| Middle value inside single states | -0.236 | the link worked out inside each of 40 states with at least 20 counties reporting |
| Using ranks, after adjusting | -0.381 | uses each county’s rank, not its value, so a few extreme counties cannot create the link |
How much to believe it
Could it be chance?
The p-value is the chance of seeing a link this strong if there were really none. A value of 0.05 or less means the link is unlikely to be chance. Very small values are written in short form: 3.1e-16 means 0.00000000000000031, and the bigger the number after “e-”, the smaller the value. Treating every county as a separate case gives p = 9.0e-74. But counties in the same state tend to resemble each other, so the better figure allows for that: p = 1.3e-14. The site tested 20,910 pairs, and testing so many turns up some flukes. The q-value allows for that too. For this pair it is 6.1e-73.
Do nearby places look alike?
Moran’s I measures how much nearby places look alike: 0 means no map pattern, and higher means neighbours are more similar. For the part of each county that the adjustment leaves unexplained, it is +0.12 (p = 0.001). That unexplained part still clusters on the map. So some regional factor shared by both measures is not accounted for.
Did they change together?
Not measurable: at least one of these is published for too few of the years a change needs (2014–2019 or 2019–2023), or for too few counties in both.
Related pairs
- Food insecurity and Returns claiming the earned income credit — +0.84
- Lack of reliable transportation and Returns claiming the earned income credit — +0.84
- Housing insecurity and Returns claiming the earned income credit — +0.83
- Self-care disability and Returns claiming the earned income credit — +0.82
- Vision disability and Returns claiming the earned income credit — +0.81
- Threat of utility shut-off and Returns claiming the earned income credit — +0.80
- Stroke and Returns claiming the earned income credit — +0.78
- Diagnosed diabetes and Returns claiming the earned income credit — +0.78
Association only. Nothing here shows that either measure changes the other. This describes counties, not the people in them. Two measures going together across places does not mean the same people have both. How the US county data works · Explore Returns claiming the earned income credit and Public-good pages per 1,000 users