Child Poverty on Native Land

Purpose

This page documents exactly how we built the Native-land child poverty comparison, so anyone can check our work, challenge our choices, or rebuild it from scratch. The opinions in the main post are ours. The numbers are not. We want the foundation to be undeniable, and that only works if we show all of it.

What this page is

It is the full recipe behind one chart: a side-by-side comparison of child poverty in school districts that sit on Native land versus every other school district in the country. We define what counts as Native land, how we matched it to school districts, and how we turned that into the rates you see. Every number traces back to two public federal files.

The data source

Two sources, both public, both reproducible.

For child poverty, we use the U.S. Census Bureau's Small Area Income and Poverty Estimates (SAIPE) for 2024, school district file. SAIPE is the Census Bureau's official annual model-based estimate of poverty for every school district in the country. The specific measure is related children ages 5 to 17 in families in poverty. We use the count of those children in poverty as the numerator and the total count of related children ages 5 to 17 as the denominator. That ratio is the child poverty rate for the district.

For the land, we use the National Center for Education Statistics (NCES) EDGE School District Geographic Relationship File for 2025, specifically the crosswalk between school districts (LEAs) and American Indian, Alaska Native, and Native Hawaiian (AIANNH) areas. For every place a school district overlaps one of these areas, this file gives the land area of the overlap. That is what lets us measure how much of a district sits on Native land.

We joined the two files on the NCES district identifier (LEAID). Of 13,132 districts in the SAIPE file, 13,128 matched the geographic file. The four that did not are newly created or recoded districts whose identifiers differ between the two vintages. They are immaterial to the result.

Step 1: Decide what "Native land" means

This is the choice that matters most, so we made it in the open.

The AIANNH file contains several kinds of areas, and they are not the same thing. Federal reservations and off-reservation trust lands are governed land bases. Oklahoma Tribal Statistical Areas (OTSAs) are statistical overlays that cover nearly all of Oklahoma, including the Tulsa and Oklahoma City metros, on former reservation land that is no longer governed as a reservation. There are also Alaska Native village areas, Hawaiian Home Lands, state-recognized reservations, and various designated statistical areas.

We define Native land as federal reservations, off-reservation trust lands, Alaska Native village areas, state-recognized reservations, and Hawaiian Home Lands. In plain terms: the places that function as Native land bases.

We deliberately exclude Oklahoma Tribal Statistical Areas. Including them sweeps in whole metro areas, diluting the comparison until it stops measuring what we are asking about. We tested this directly. Adding OTSAs more than triples the district count and drags the median down from 22.4% to 19.1%, almost entirely because of Oklahoma metros, not because the underlying reality changed. We chose the definition that isolates the question, not the one that flatters or muddies it.

Step 2: Decide how much overlap counts as "on" Native land

A district can clip the edge of a reservation or sit entirely inside one. We needed a threshold.

We require that at least 75% of a district's land area fall within Native land for it to count. We chose 75% rather than a looser bar because we tested the full range. At "any overlap," the comparison nearly vanishes, because most of those districts are ordinary towns that merely border Native land. As the threshold tightens, the gap widens, because the higher cuts isolate districts that genuinely are Native-land districts and nothing else. We picked a level that is demanding enough to mean something and not so extreme that it captures only a handful of places.

To compute each district's share, we sum the land area of all its overlaps with qualifying Native land, then divide by the district's total land area. The total is the sum of every row for that district in the geographic file, including the portion that is in no AIANNH area at all. That keeps the denominator honest.

Step 3: Compute the two metrics

For each group, Native-land districts and everyone else, we report two numbers.

The weighted-average rate is the total count of children in poverty across the group divided by the total count of children ages 5 to 17 across the group. This is the rate a child in the group actually experiences, on average, and it is robust to noise in small districts because it pools the counts.

The median rate is the middle district's own poverty rate, with every district counting equally regardless of size. This describes the typical district rather than the typical child.

We report both on purpose, because they answer different questions. In both groups the weighted average sits above the median, which tells you that poverty is concentrated in the larger districts: the biggest places are poorer than the typical place, so pooling the counts pulls the rate up. The gap between the two measures is similar in each group (about 3 points on Native land, about 2 points elsewhere), so the contrast between the groups holds up whichever measure you use.

Why Hawaii does not appear

Hawaii has no qualifying district, and the reason is geometry, not definition.

Hawaii is a single statewide school district covering all 216,723 of the state's children ages 5 to 17. Hawaiian Home Lands are real, but they are scattered parcels that together make up only about 5% of that one district's land area. No threshold we would reasonably set, and certainly not 75%, can capture a 5% overlap. So Hawaii falls out for the same structural reason it would fall out under almost any version of this analysis: there is no school district that is predominantly Hawaiian Home Land, because there is only one district and it is the entire state.

How we built and checked it

The analysis is a short, reproducible pipeline in Python using pandas. We read the two federal files, classified each AIANNH area by type from its official name, summed overlaps per district, computed the share, applied the 75% threshold, and calculated the two rates for each group.

We checked it three ways. We confirmed the district-to-geography join rate (13,128 of 13,132). We swept the overlap threshold from any-overlap through 100% and confirmed the gap behaves as expected, widening as the definition tightens, which is the signature of a real effect rather than an artifact. And we spot-checked named districts against the source files by hand, confirming that the high-poverty places the chart surfaces (Dupree on the Cheyenne River Reservation at 55%, McLaughlin on Standing Rock at 54%, districts across the Navajo Nation past 50%) match the underlying SAIPE values.

How to read the result

The headline: 144 school districts sit at least 75% within Native land. Their median child poverty rate is 22.4% and their weighted-average rate is 25.6%. The other 12,983 districts sit at a 12.4% median and a 14.6% weighted average. Native-land child poverty runs close to double the rest of the country, and the median Native-land district is poorer than roughly three out of four districts everywhere else.

The chart shows the full spread, not just these summary numbers, so you can see that this is a shift in the center of the distribution and not a few extreme places dragging up an average.

Honest notes and limitations

The two files are one year apart. SAIPE is 2024; the geographic crosswalk is 2025. School district boundaries move slowly, so a one-year offset introduces only small slippage, but it is real and we note it.

SAIPE estimates carry margins of error, and they are larger for small districts. We lean on the weighted average, which pools counts and is robust to this, and we treat the medians at the very strictest thresholds, where the sample is thin, as directional rather than precise.

The median in the chart is unweighted, meaning each district counts equally regardless of size. That is the right choice for describing how districts differ from one another, but it is a different question than how the average child fares, which the weighted average answers.

This is a measure of where children live, not of their race or tribal enrollment. A school district that is 75% within reservation land is not necessarily 75% Native by population. We are measuring child poverty in the places that sit on Native land, which is a geographic statement, not a demographic one.

And as noted above, this analysis cannot speak to Hawaii, because of how that state's single district is drawn.