Methodology

The IRS builds this data from year-to-year address changes on individual income tax returns. A file labeled 2022 to 2023 compares the address on the 2022 return with the address on the 2023 return. It reports the number of returns filed (an approximation of households that moved), the number of personal exemptions (an approximation of individuals), and the total adjusted gross income (AGI) associated with those returns.

The data covers people who filed federal income tax returns. Non-filers, which include many lower-income and some elderly people, are not represented. Treat these counts as filer migration, not total population migration.

AGI is reported in thousands of nominal dollars and is not adjusted for inflation. We store it exactly as the IRS reports it and convert to dollars only for display, labeled nominal. Comparisons across years are nominal unless stated otherwise.

A common misreading is that the income of someone who moves is lost to the place they left. It usually is not: the job often stays and is refilled. We present net AGI as movement of filers and their reported income, nothing more. We do not frame it as economic gain or loss, and we make no claim about why anyone moved.

The data lags about two years behind the present. Two methodology breaks affect long comparisons: the series starting 2011 to 2012 used an improved approach, and the series starting 2022 to 2023 used an enhanced matching process. The 2022 to 2023 county file also adopted Connecticut’s nine planning regions in place of its legacy counties.

The migration timelapse reaches back into the 1990s using the IRS’s pre-2011 state-to-state files, which are kept separate from the modern series. Those files did not report income (AGI) until the 1995 to 1996 year, so earlier years show people only. They also captured a slightly smaller share of filings (returns filed after September were excluded). The 2011 to 2012 break is marked on the timeline, and figures before and after it are not presented as strictly comparable.

Genuine county-to-county and state-to-state flows are loaded from the IRS outflow files. Aggregate, total, and non-migrant rows are separated out and used for the headline totals. Every flow is validated against the official FIPS crosswalk, and rows that fail validation are quarantined and logged rather than loaded.

Headline inflow, outflow, and net figures use the IRS’s own “Total Migration, US and Foreign” totals, which include disclosure-suppressed small flows. The top origin and destination lists show only individual flows the IRS published above its disclosure threshold, so they will not sum to the headline total.

The “Who lives here” and “Housing” sections on state and county pages come from the U.S. Census Bureau’s American Community Survey (ACS) 5-year estimates: median age, household income, educational attainment, the poverty rate, average commute, broadband access, race and ethnicity, and home value, rent, and homeownership. The ACS is public domain. We pull the figures directly from the Census Data API and store them in the units the Census reports, converting only for display.

A 5-year estimate is an average over a five-year window, labeled by its ending year (the 2024 figure averages 2020 through 2024). Consecutive windows overlap by four years, so year over year changes are not independent; compare years at least five apart. A few measures start later than others because the survey added the question later, most notably home broadband, which begins with the 2017 window. Reported home value is what owners estimate their home is worth, not a record of recent sale prices.

This is a different universe from the IRS migration figures. The ACS counts all residents; the IRS data counts tax filers and their dependents. We present the two on the same pages because they describe the same place, but we never combine them in a single chart, map, or computed figure, and we make no claim that one explains the other. Race and ethnicity follow the Census definitions: White, Black, and Asian shares count people who are not Hispanic or Latino, and Hispanic or Latino residents may be of any race.

The home-price appreciation figures in the Housing section come from the Federal Housing Finance Agency House Price Index (annual, all-transactions), public domain. It is a repeat-sales index: it measures the price change of the same homes over time from mortgages bought or guaranteed by Fannie Mae and Freddie Mac, so it captures how a market moved rather than the current price level (which the ACS home value shows). The five-year change is computed from the index, this year against five years earlier.

FHFA labels its county-level series developmental, and small counties can be missing or revised because the sample is thin, so not every county has a figure. Because the index is built from conforming (Fannie and Freddie) mortgages, it under-weights homes bought with cash or financed outside those limits. The figures are index changes, not a claim about any individual sale or a reason anyone moved.

The new-construction figure in the Housing section is the number of housing units authorized by residential building permits that year, from the U.S. Census Bureau Building Permits Survey (public domain). We read the annual county files and sum them to the state and national figures. The single-family share counts one-unit structures; everything else (duplexes through large apartment buildings) is grouped as multifamily.

These are permit authorizations, not completions or sales, so they lead actual construction and some permitted units are never built. A small number of places do not issue or report permits and will show no figure. It is a count of new supply, not a claim about prices or about why anyone moved.

Source: IRS SOI Tax Stats, U.S. Population Migration Data. Public domain. Please attribute How America Moves when you cite figures from this site.

The elections section shows county-level presidential results for 2000 through 2024, compiled from certified state returns by the MIT Election Data and Science Lab (Harvard Dataverse, CC0 public domain). We validate each load against FEC national totals and refuse it if any year drifts more than 1 percent. Margins are the Democratic-minus-Republican gap as a share of all votes cast; popular vote only, no electoral votes.

Known reporting limits: Alaska reports results by state legislative district rather than borough, so it has statewide figures but no county map. Kansas City, Missouri reports separately from the counties it spans, so those counties' results exclude the city's votes (the state total includes them). A few statewide units (Connecticut write-ins, Maine overseas ballots, Rhode Island's federal precinct) are likewise counted in state totals only. Two county recodes are merged for continuity: Shannon County, SD (renamed Oglala Lakota) and Bedford City, VA (merged into Bedford County).

Senate results come from MEDSL's U.S. Senate statewide returns and House results from U.S. House district returns (both CC0). Races decided by a runoff show the runoff result, including Georgia's January 2021 runoffs, which are counted under the 2020 election they decided. In fusion-voting states a candidate's lines are summed and the candidate keeps their principal party. Congressional district boundaries come from the U.S. Census Bureau for the specific Congress elected that year; because districts are redrawn, House maps do not offer a swing comparison across cycles. A few of the most recent race totals are marked unofficial by MEDSL while certification paperwork settles.

Election results and migration figures are separate datasets from separate sources. The site presents them on the same geography pages but never combines them in a chart, a map, or a computed figure, and it makes no claims about how one relates to the other.