How do we know this? Tracing four numbers to the source

Every number on this site is one of two things: computed from public data with a query you can re-run, or cited to a named source you can check. No number is hand-typed or estimated. To prove that isn't just a slogan, here are four numbers walked end to end — from the claim, to the source, to the exact file, to how you'd pull the raw data yourself.

If these four hold up to your inspection, the other two hundred work the same way.

1 · The why — "about 1 in 3"

About 1 in 3 registered young Black men (18–32) don't cast a ballot in a federal election: 31.7%

• Source. The U.S. Census Bureau's Current Population Survey (CPS) — the November Voting Supplement, 2000–2024 — via the IPUMS (Integrated Public Use Microdata Series) extract cps_00002.
• Definition. "Registered" = everyone who voted (registered by definition under Census rules) plus everyone the survey records as registered-but-didn't-vote. The rate is registered non-voters ÷ all registered, weighted by the survey's population weight.
• Sample + margin. 6,960 unweighted respondents; generalized-variance standard error ±0.7 points.
• Where it's computed. From src/data/cps_voting_clean.parquet. The exact query:

  SELECT 100.0 * sum(case when voted_clean = 0 then voter_weight end) / sum(voter_weight)
  FROM cps_voting_clean
  WHERE race_group = 'Black' AND sex_label = 'Male' AND age BETWEEN 18 AND 32
    AND (voted_clean = 1 OR voreg_code = 2);
  

• Re-pull it yourself. Register free at cps.ipums.org, pull VOTED, VOREG, AGE, SEX, RACE, HISPAN, VOSUPPWT for November 2000–2024, and run the query above. You will land on 31.7%.

2 · The proof — "50% logistical, not apathy"

The leading reason registered young people don't vote is logistics, not apathy: 50.0% cite a logistical reason; 23% engagement; 9% access.

• Source. Same CPS supplement, the VOWHYNOT variable — the Census directly asks registered non-voters why they didn't vote.
• Definition. We harmonize the individual reason codes into five policy-relevant buckets (logistical / engagement / access / personal / other). "Logistical" = too busy, out of town, forgot, transportation, weather.
• Sample + margin. 34,031 unweighted registered youth non-voters; ±0.4 points.
• Where it's computed. Precomputed to src/data/findings/finding-04/reason_national.json by the manifest at scripts/precompute/manifest.py. The number on the page reads from that file — it is not typed in.

3 · The verification — "50.7% youth turnout, 2024"

Youth (18–29) turnout in 2024 was 50.7% by our own CPS computation — versus 74.7% for seniors, a 23.9-point gap.

• Source + method. CPS, Census P20 method: only an explicit "voted: yes" (VOTED = 2) counts as a voter; everyone else is a non-voter; the denominator is all citizens 18–29.
• A cross-check, named. CIRCLE at Tufts reports a 2024 youth estimate of ~47% using a different (voter-file-modeled) method. We show our own CPS figure (50.7%) for internal consistency, and name CIRCLE's so you can see the gap is a method difference, not an error. CPS self-report runs a few points high — disclosed on the methodology page.
• Where it's computed. src/data/findings/finding-01/gap_by_year.json, regenerable from the parquet.

4 · The support — "$0–5 per additional voter"

Same-day registration moves turnout at roughly $0–5 per additional voter — a cited number, not one we computed.

• Source. This one is not ours to compute — it comes from the peer-reviewed literature: Gerber & Green, Get Out the Vote (4th ed.) and Brennan Center cost-benefit analyses.
• How we carry it. Every intervention on the Recommendations page is one row in data/external/evidence_roi/intervention_effect_size_catalog.csv, and every row names its source and its effect range — we present ranges, never point estimates, and we include a "do-not-fund" section for interventions the evidence doesn't support.
• Why this station matters. It shows the other kind of traceability: where a number isn't ours to derive, we tell you exactly whose it is and where to read it.

That's the whole site

Those four are not special. Every cell-level number carries a hover-or-tap ⓘ provenance chip with its source, sample size, margin, and year range. Every source on the Sources & References page links to the public original — the Census CPS, McDonald's Voting-Eligible Population (VEP) rates, the National Conference of State Legislatures (NCSL), and the Election Assistance Commission (EAC) — all free, all re-pullable. The methodology page documents every coding decision, including the ones that didn't work.

You don't have to trust us. You can check us.