alzhub.org
Get help now — free helplines

You can call any of these for free. You do not need a diagnosis, an appointment, or a crisis to call — questions and plain exhaustion are reasons enough.

  • Alzheimer's Association 24/7 Helpline 800-272-3900 24 hours a day, 365 days a year Free, confidential support and information for people with dementia, caregivers, families and the public. Master's-level clinicians. Also Spanish.
  • 988 Suicide & Crisis Lifeline 988 24/7 Call or text 988, or chat at 988lifeline.org. For anyone in emotional distress or a mental-health crisis, including caregivers.
  • Alzheimer's Foundation of America (AFA) Helpline 866-232-8484 9:00 am - 9:00 pm ET, seven days a week Licensed social workers providing support, info and referrals. Also text message and web chat. Text: 646-586-5283

In a medical emergency, call 911. More numbers, and what each line does, on the help page.

Home › Research › Correlations

Correlations in alzhub's own data

48 county tests across 3143 US counties, plus one state-by-year panel. Each was checked twice: once raw, and once within states after holding age structure, rurality, income, education, race and ethnicity, and how hard each county looks for dementia fixed. Most of the large raw associations did not survive the second check.

Looking for relationships the news has raised? Links the news suggests that nobody has tested sorts 53 of them by whether the study library supports them, contradicts them, or has nothing on them.

Counties, not people

Every number on this page compares counties (or states). "Counties with more X had higher Y" does not mean people who have X are more likely to get dementia — reading a group-level association as a personal one is the ecological fallacy, and it is the most common way results like these get misreported. None of these associations shows that one thing causes another.

The outcome is diagnosed dementia among Medicare beneficiaries aged 65 and over, from 2020 claims. A county can have a high rate because it has more dementia or because it diagnoses more of the dementia it has.

What held up

run 2026-10-15

Ranked by mechanism first, then by the size of the adjusted coefficient. The adjusted figure is a partial rank correlation with state fixed effects; its p-value uses standard errors clustered by state, and q is the Benjamini-Hochberg false-discovery-adjusted p across all 48 tests.

Held up, with a coherent mechanism

Counties with more Adults 65+ living alone (%) had higher Diagnosed dementia prevalence, Medicare 65+ (%)

Raw 0.184 n=2572 · 95% CI [0.146, 0.222] · q=6.6e-21
Adjusted 0.116 n=2392 · 95% CI [0.075, 0.155] · p=4.3e-06 · q=5.2e-05 · Bonferroni pass

Spearman rank correlation. Most entangled with: Median household income (USD) (ρ=-0.373), Population aged 85+ (%) (ρ=0.226), Metro county (1) vs nonmetro (0) (ρ=-0.192).

Counties with more Adults with a bachelor's degree or more (%) had lower Diagnosed dementia incidence, Medicare 65+ (%)

Raw -0.323 n=1928 · 95% CI [-0.363, -0.281] · q=1.4e-47
Adjusted -0.110 n=1787 · 95% CI [-0.156, -0.063] · p=3.5e-04 · q=0.002 · Bonferroni pass

Spearman rank correlation. Most entangled with: Median household income (USD) (ρ=0.673), Metro county (1) vs nonmetro (0) (ρ=0.400), Population density (log scale) (ρ=0.384).

Counties with more Adults 65+ living alone (%) had higher Diagnosed dementia incidence, Medicare 65+ (%)

Raw 0.174 n=1928 · 95% CI [0.131, 0.218] · q=1.6e-14
Adjusted 0.091 n=1787 · 95% CI [0.044, 0.137] · p=0.002 · q=0.011

Spearman rank correlation. Most entangled with: Median household income (USD) (ρ=-0.373), Population aged 85+ (%) (ρ=0.226), Metro county (1) vs nonmetro (0) (ρ=-0.192).

Counties with more Adults with a bachelor's degree or more (%) had lower Diagnosed dementia prevalence, Medicare 65+ (%)

Raw -0.276 n=2572 · 95% CI [-0.312, -0.239] · q=7.8e-46
Adjusted -0.085 n=2392 · 95% CI [-0.125, -0.045] · p=0.004 · q=0.015

Spearman rank correlation. Most entangled with: Median household income (USD) (ρ=0.673), Metro county (1) vs nonmetro (0) (ρ=0.400), Population density (log scale) (ρ=0.384).

Held up, mechanism indirect

Counties with more Median household income (USD) had lower Diagnosed dementia prevalence, Medicare 65+ (%)

Raw -0.378 n=2580 · 95% CI [-0.412, -0.343] · q=2.2e-87
Adjusted -0.183 n=2392 · 95% CI [-0.222, -0.144] · p=1.7e-12 · q=8.1e-11 · Bonferroni pass

Spearman rank correlation. Most entangled with: Adults with a bachelor's degree or more (%) (ρ=0.673), Metro county (1) vs nonmetro (0) (ρ=0.445), Population density (log scale) (ρ=0.331).

Counties with more Median household income (USD) had lower Diagnosed dementia incidence, Medicare 65+ (%)

Raw -0.393 n=1936 · 95% CI [-0.431, -0.353] · q=1.0e-71
Adjusted -0.157 n=1787 · 95% CI [-0.203, -0.111] · p=1.4e-05 · q=1.1e-04 · Bonferroni pass

Spearman rank correlation. Most entangled with: Adults with a bachelor's degree or more (%) (ρ=0.673), Metro county (1) vs nonmetro (0) (ρ=0.445), Population density (log scale) (ρ=0.331).

Held up, but probably measures diagnosis, not disease

Counties with more Certified nursing-home beds per 1,000 pop had higher Diagnosed dementia prevalence, Medicare 65+ (%)

Raw 0.260 n=2495 · 95% CI [0.222, 0.297] · q=1.6e-39
Adjusted 0.158 n=2316 · 95% CI [0.117, 0.198] · p=1.1e-08 · q=2.6e-07 · Bonferroni pass

Spearman rank correlation. Most entangled with: Population density (log scale) (ρ=-0.476), Population aged 85+ (%) (ρ=0.427), Metro county (1) vs nonmetro (0) (ρ=-0.421).

Counties with more Nurse practitioners per 1,000 pop had higher Diagnosed dementia prevalence, Medicare 65+ (%)

Raw 0.205 n=2572 · 95% CI [0.167, 0.242] · q=1.4e-25
Adjusted 0.104 n=2392 · 95% CI [0.063, 0.144] · p=6.0e-07 · q=9.5e-06 · Bonferroni pass

Spearman rank correlation. Most entangled with: Population density (log scale) (ρ=0.332), Black population (%) (ρ=0.216), Adults with a bachelor's degree or more (%) (ρ=0.172).

Counties with more Certified nursing-home beds per 1,000 pop had higher Diagnosed dementia incidence, Medicare 65+ (%)

Raw 0.285 n=1905 · 95% CI [0.242, 0.327] · q=1.1e-36
Adjusted 0.082 n=1765 · 95% CI [0.035, 0.129] · p=5.6e-04 · q=0.003 · Bonferroni pass

Spearman rank correlation. Most entangled with: Population density (log scale) (ρ=-0.476), Population aged 85+ (%) (ρ=0.427), Metro county (1) vs nonmetro (0) (ρ=-0.421).

Counties with more Nurse practitioners per 1,000 pop had higher Diagnosed dementia incidence, Medicare 65+ (%)

Raw 0.119 n=1928 · 95% CI [0.075, 0.163] · q=1.9e-07
Adjusted 0.056 n=1787 · 95% CI [0.009, 0.103] · p=0.001 · q=0.008

Spearman rank correlation. Most entangled with: Population density (log scale) (ρ=0.332), Black population (%) (ρ=0.216), Adults with a bachelor's degree or more (%) (ρ=0.172).

Held up for one outcome only - fragile

Counties with more PM2.5 modelled pop-weighted mean 2016-2019 (ug/m3) had higher Diagnosed dementia incidence, Medicare 65+ (%)

Raw 0.354 n=1928 · 95% CI [0.313, 0.394] · q=1.2e-57
Adjusted 0.091 n=1779 · 95% CI [0.044, 0.138] · p=0.014 · q=0.049

Spearman rank correlation. Most entangled with: Black population (%) (ρ=0.514), Population density (log scale) (ρ=0.486), Medicare Advantage penetration (%), June 2026 (ρ=0.379).

Reversed sign after adjustment

Counties with more Primary-care physicians per 100k had higher Diagnosed dementia prevalence, Medicare 65+ (%)

Raw -0.108 n=2529 · 95% CI [-0.146, -0.069] · q=7.3e-08
Adjusted 0.107 n=2348 · 95% CI [0.067, 0.147] · p=1.3e-05 · q=1.1e-04 · Bonferroni pass

Spearman rank correlation. Most entangled with: Adults with a bachelor's degree or more (%) (ρ=0.535), Population density (log scale) (ρ=0.289), Median household income (USD) (ρ=0.274).

Appeared only after adjustment - not trusted

Counties with more Social associations per 10,000 pop had higher Diagnosed dementia prevalence, Medicare 65+ (%)

Raw -0.021 n=2580 · 95% CI [-0.060, 0.017] · q=0.309
Adjusted 0.076 n=2392 · 95% CI [0.036, 0.116] · p=0.004 · q=0.015

Spearman rank correlation. Most entangled with: Population aged 85+ (%) (ρ=0.382), Population aged 65+ (%) (ρ=0.267), Metro county (1) vs nonmetro (0) (ρ=-0.245).

Counties with more Adult depression (%), PLACES modelled had higher Diagnosed dementia prevalence, Medicare 65+ (%)

Raw 0.004 n=2392 · 95% CI [-0.036, 0.044] · q=0.879
Adjusted 0.048 n=2392 · 95% CI [0.008, 0.089] · p=0.009 · q=0.032

Spearman rank correlation. Most entangled with: Medicare Advantage penetration (%), June 2026 (ρ=0.257), Hispanic population (%) (ρ=-0.237), Median household income (USD) (ρ=-0.193).

What did not hold up

27 tests

These have sizeable raw associations that fell to near zero once the confounders and state were held fixed. Quoting any raw number here as a finding would be wrong. The chronic-disease columns are CDC PLACES small-area model estimates, partly predicted from the same demographics used as controls.

This table is wider than a phone screen — swipe it sideways to see the rest.

Exposure → outcomeRawn Adjustednp (clustered)q
Adult physical inactivity (%), PLACES modelled → Diagnosed dementia incidence, Medicare 65+ (%) 0.4521787 0.0381787 0.1960.348
Adult physical inactivity (%), PLACES modelled → Diagnosed dementia prevalence, Medicare 65+ (%) 0.4442392 0.0352392 0.1420.323
Population below poverty (%) → Diagnosed dementia prevalence, Medicare 65+ (%) 0.4212572 0.0382392 0.0860.228
Adult diabetes (%), PLACES modelled → Diagnosed dementia incidence, Medicare 65+ (%) 0.4181787 -2.2e-051787 0.9990.999
PM2.5 modelled pop-weighted mean 2016-2019 (ug/m3) → Diagnosed dementia prevalence, Medicare 65+ (%) 0.4082571 0.0582383 0.0470.141
Population below poverty (%) → Diagnosed dementia incidence, Medicare 65+ (%) 0.3941928 0.0061787 0.7870.911
Adult diabetes (%), PLACES modelled → Diagnosed dementia prevalence, Medicare 65+ (%) 0.3892392 -0.0202392 0.3550.509
Adult hypertension (%), PLACES modelled → Diagnosed dementia incidence, Medicare 65+ (%) 0.3831787 -0.0191787 0.4970.635
Uninsured (%) → Diagnosed dementia incidence, Medicare 65+ (%) 0.3691936 0.0241787 0.2680.414
PM2.5 annual mean (ug/m3), CHR ~2021 → Diagnosed dementia prevalence, Medicare 65+ (%) 0.3642579 0.0052391 0.8350.911
Adult hypertension (%), PLACES modelled → Diagnosed dementia prevalence, Medicare 65+ (%) 0.3642392 -0.0062392 0.8190.911
Adult current smoking (%), PLACES modelled → Diagnosed dementia incidence, Medicare 65+ (%) 0.3631787 0.0221787 0.4810.635
Adult obesity (%), PLACES modelled → Diagnosed dementia incidence, Medicare 65+ (%) 0.3481787 0.0311787 0.2280.391
Adult current smoking (%), PLACES modelled → Diagnosed dementia prevalence, Medicare 65+ (%) 0.3452392 0.0352392 0.1660.323
Adult obesity (%), PLACES modelled → Diagnosed dementia prevalence, Medicare 65+ (%) 0.3432392 0.0442392 0.0600.170
Uninsured (%) → Diagnosed dementia prevalence, Medicare 65+ (%) 0.3322580 0.0042392 0.8710.929
Households with broadband (%) → Diagnosed dementia incidence, Medicare 65+ (%) -0.3091936 -0.0491787 0.0450.141
PM2.5 annual mean (ug/m3), CHR ~2021 → Diagnosed dementia incidence, Medicare 65+ (%) 0.2891936 0.0011787 0.9660.987
Medicare Advantage penetration (%), June 2026 → Diagnosed dementia prevalence, Medicare 65+ (%) 0.2872572 0.0552392 0.1160.293
Adult binge drinking (%), PLACES modelled → Diagnosed dementia incidence, Medicare 65+ (%) -0.2641787 -0.0221787 0.4290.588
Adult binge drinking (%), PLACES modelled → Diagnosed dementia prevalence, Medicare 65+ (%) -0.2612392 0.0032392 0.9150.955
Households with broadband (%) → Diagnosed dementia prevalence, Medicare 65+ (%) -0.2552580 -0.0402392 0.1680.323
Medicare Advantage penetration (%), June 2026 → Diagnosed dementia incidence, Medicare 65+ (%) 0.2291928 0.0651787 0.1810.334
Adults with a routine check-up in past year (%) → Diagnosed dementia prevalence, Medicare 65+ (%) 0.2272392 0.0192392 0.5030.635
Primary-care physicians per 100k → Diagnosed dementia incidence, Medicare 65+ (%) -0.1911928 0.0381779 0.1630.323
Adults with a routine check-up in past year (%) → Diagnosed dementia incidence, Medicare 65+ (%) 0.1861787 0.0271787 0.3610.509
Metro county (1) vs nonmetro (0) → Diagnosed dementia incidence, Medicare 65+ (%) -0.0721928 0.0401787 0.1630.323

No association in either specification: Low access to a grocery store (%), 2019 → diagnosed dementia prevalence, medicare 65+ (%); Metro county (1) vs nonmetro (0) → diagnosed dementia prevalence, medicare 65+ (%); Any drinking-water violation (1/0) → diagnosed dementia prevalence, medicare 65+ (%); Adult depression (%), PLACES modelled → diagnosed dementia incidence, medicare 65+ (%); Social associations per 10,000 pop → diagnosed dementia incidence, medicare 65+ (%); Low access to a grocery store (%), 2019 → diagnosed dementia incidence, medicare 65+ (%); Any drinking-water violation (1/0) → diagnosed dementia incidence, medicare 65+ (%).

The one test with a time dimension

343 state-years

Fine-particle air pollution (PM2.5) and the Alzheimer's death rate, by state and year

Pooled r = 0.266 n=343 · slope 1.773 per µg/m³ · p=0.009
State + year fixed effects slope -0.183 95% CI [-0.579, 0.213] · p=0.359 · 49 states

49 states, 2016–2022. States with more PM2.5 had higher age-adjusted Alzheimer's death rates in the pooled data, but within a state, years with more PM2.5 did not have higher rates. Death rates largely reflect how each state codes dementia deaths, which the fixed effects remove. Seven years of a slow-moving exposure is a weak test: this is no evidence of an effect, not evidence of no effect.

Method, briefly

Spearman rank correlation for continuous exposures, point-biserial for the two yes/no exposures. Adjusted: partial rank correlation after removing state averages and these controls: Population aged 65+ (%); Population aged 85+ (%); Metro county (1) vs nonmetro (0); Population density (log scale); Median household income (USD); Adults with a bachelor's degree or more (%); Black population (%); Hispanic population (%); Adults with a routine check-up in past year (%); Medicare Advantage penetration (%), June 2026. Benjamini-Hochberg FDR over all 48 adjusted tests, on state-clustered p-values (q<0.05); Bonferroni pass (p<0.05/48) also reported. The county outcome exists for 2020 only, so there is no within-county comparison over time. Every number here comes from the county analysis run on 2026-10-15.

Earlier correlation work