What moves people between US states?

A honest first look at economics, migration, health & politics across 50 states, 2011–2017 (350 state-years). Net migration, unemployment, income, foreign-born share, age-adjusted death rates (incl. HIV), and presidential vote share — from Census, BLS, NCHS, CDC & MIT/county election returns.

Bottom line. Interstate migration tracks the economy — unemployment pushes people out a year later, and rising income pulls them in — and shows no robust link to disease burden or partisan lean. People move for jobs and income, not (measurably) away from cancer or heart disease — and the "fleeing blue states" pattern, real in raw cross-state numbers, does not survive a within-state test: it's the cost and jobs beneath the politics, not the politics.
Why this isn't just spurious. Over time almost everything trends together, so raw correlations mislead. Everything here uses only within-state variation (each state vs its own norm and the national year), and the key claims are tested with a state-clustered bootstrap (resampling the 51 states, not the rows). One line makes the point: income ~ cancer looks like a strong -0.62 across states, but collapses to +0.05 within states — the raw number was confounding, not a relationship.

What holds up

r = -0.19
95% CI [-0.29, -0.03] · n=250 · lag 1yr
Unemployment leads domestic OUT-migration by 1 year (economic push)
r = -0.16
95% CI [-0.27, -0.06] · n=300
'Red-state welfare': does govt-aid usage track partisan lean?

What doesn't — including health, politics & companies

These do not survive the within-state, clustered test (confidence interval spans zero): log(median income)→Domestic migration /1k, Unemployment→Domestic migration /1k, Unemployment→Intl migration /1k, All-cause mortality→Unemployment, Cancer deaths→Domestic migration /1k, Heart deaths→Domestic migration /1k, HIV diagnoses /100k→Domestic migration /1k, HIV diagnoses /100k→Foreign-born %, Dem pres vote %→Domestic migration /1k, Dem pres vote %→Unemployment, Corp HQ net moves→Domestic migration /1k, Corp HQ net moves→Unemployment, Public aid %→Unemployment. The health↔migration nulls are the notable ones: at this scale and frequency, disease burden does not visibly drive where Americans move.

Do people follow companies? Corporate HQ relocations (from SEC filing addresses) point the same way as people in the raw numbers — California and New York lose both companies and residents; Texas and Florida gain both — but HQ moves are rare, lumpy events (a dozen a year even for the biggest states), so they show no year-to-year within-state co-movement with people migration. Read that as both being pulled by the same economics, not people literally chasing headquarters.

Levels mislead — the same pairs, two ways

pairnaive level rwithin-state r
log(median income) ~ Cancer deaths-0.62+0.05
Public aid % ~ Dem pres vote %+0.16+0.06
Dem pres vote % ~ Domestic migration /1k-0.12+0.21
Unemployment ~ Domestic migration /1k-0.07-0.43

The gap between the columns is confounding (or, for unemployment→migration, real signal that levels hide).

Within-state correlation matrix

UnemploymentDomestic migration /1kIntl migration /1kForeign-born %log(median income)Public aid %All-cause mortalityCancer deathsHeart deathsHIV diagnoses /100kDem pres vote %Corp HQ net moves
Unemployment+1.00-0.43+0.07+0.07-0.29+0.12+0.01+0.01+0.04+0.02-0.12-0.00
Domestic migration /1k-0.43+1.00+0.01-0.12+0.25-0.31+0.17+0.15+0.29+0.19+0.21-0.01
Intl migration /1k+0.07+0.01+1.00+0.29+0.08-0.11-0.05+0.05-0.04-0.08+0.01-0.01
Foreign-born %+0.07-0.12+0.29+1.00-0.04+0.12-0.05-0.17-0.16+0.01-0.04+0.08
log(median income)-0.29+0.25+0.08-0.04+1.00-0.32+0.09+0.05+0.02-0.23+0.23+0.01
Public aid %+0.12-0.31-0.11+0.12-0.32+1.00-0.22-0.10-0.19-0.07+0.06+0.02
All-cause mortality+0.01+0.17-0.05-0.05+0.09-0.22+1.00+0.36+0.39+0.08-0.18+0.07
Cancer deaths+0.01+0.15+0.05-0.17+0.05-0.10+0.36+1.00+0.21+0.22-0.04-0.05
Heart deaths+0.04+0.29-0.04-0.16+0.02-0.19+0.39+0.21+1.00+0.08+0.07-0.01
HIV diagnoses /100k+0.02+0.19-0.08+0.01-0.23-0.07+0.08+0.22+0.08+1.00-0.21+0.00
Dem pres vote %-0.12+0.21+0.01-0.04+0.23+0.06-0.18-0.04+0.07-0.21+1.00+0.03
Corp HQ net moves-0.00-0.01-0.01+0.08+0.01+0.02+0.07-0.05-0.01+0.00+0.03+1.00

Blue = positive, red = negative co-movement within states (state+year fixed effects).

Every hypothesis tested

relationshipr (Δ)95% CInverdict
Unemployment → Domestic migration /1k (t+1)-0.19[-0.29, -0.03]250holds
log(median income) → Domestic migration /1k+0.11[-0.01, +0.19]300noise (CI spans 0)
Unemployment → Domestic migration /1k-0.11[-0.22, +0.03]300noise (CI spans 0)
Unemployment → Intl migration /1k-0.04[-0.11, +0.02]300noise (CI spans 0)
All-cause mortality → Unemployment-0.04[-0.13, +0.06]300noise (CI spans 0)
Cancer deaths → Domestic migration /1k+0.03[-0.10, +0.13]300noise (CI spans 0)
Heart deaths → Domestic migration /1k+0.07[-0.02, +0.16]300noise (CI spans 0)
HIV diagnoses /100k → Domestic migration /1k+0.11[-0.13, +0.30]300noise (CI spans 0)
HIV diagnoses /100k → Foreign-born %-0.10[-0.20, +0.10]300noise (CI spans 0)
Dem pres vote % → Domestic migration /1k+0.10[-0.12, +0.27]300noise (CI spans 0)
Dem pres vote % → Unemployment-0.05[-0.22, +0.13]300noise (CI spans 0)
Corp HQ net moves → Domestic migration /1k-0.00[-0.10, +0.09]300noise (CI spans 0)
Corp HQ net moves → Unemployment-0.01[-0.08, +0.05]300noise (CI spans 0)
Public aid % → Dem pres vote %-0.16[-0.27, -0.06]300holds
Public aid % → Unemployment-0.02[-0.12, +0.08]300noise (CI spans 0)
Read with care. 7 years is short and 2011–2017 is a single recovery period (limited range). This is state-level (ecological ≠ individual). A lagged correlation makes reverse-causation less likely but is not proof of cause. Several pairs were tested, so a borderline result could be chance. Cancer, heart disease and HIV are all in this analysis (all show no robust migration link); COVID is in the panel but starts in 2020, outside this 2010–2017 window, so it is not in this correlation. Government-aid usage (households on SNAP or public assistance) is a STATE rate, not a person: a state where more households receive aid tells you nothing about whether aid recipients lean one party — the classic ecological fallacy, and a famous trap for the "red-state welfare" claim. Partisan lean is the state's 2-party presidential vote share, forward-filled between elections (so it moves slowly within a state); Alaska is absent from the county-level election source for 2008–2016, so the panel here is 50 units, not 51.