Liberty Evidence Engine

Evidence · 2026-08-27

Why defensive-gun-use estimates disagree

CDC republished a 60,000–2.5 million yearly range. That is a literature envelope, not a CDC count. Compare the instruments — who was asked, whether the survey was anonymous, and whether anyone had to already say they were a victim — and keep the CDC/NAS band as an honest overlay.

Download instruments CSV Compare instruments Methods

Interactive explorer · instruments, not a count

Pick the survey. Keep CDC as an overlay, not a third number.

Loading instruments…

Selected estimate

—

Instrument class

—

Survey n

—

DGU cases in sample

—

Log scale so ~80k and ~2.5 million can share a chart without a fake linear winner. Recomputed in your browser from the downloadable CSV — no black box.

Instrument reference table

Static reference table. Use the explorer above to try other settings.

source_label year estimate_low estimate_high gun_type instrument_class anonymous asks_dgu_directly recall n_sample n_dgu_cases
NCVS assault and robbery (as cited by Kleck 1995) 1993 68000 68000 any government_victimization no no bounded_6mo
NCVS plus household burglary (as cited by Kleck 1995) 1993 80000 82000 any government_victimization no no bounded_6mo
Hart survey 1981 1981 640000 640000 handgun gun_survey unknown yes five_year
Mauser survey 1990 1990 700000 700000 any gun_survey unknown yes five_year
Kleck and Gertz NSDS 1993 (1-year person-based) 1993 2200000 2500000 any dedicated_dgu_survey yes yes one_year 4977 222
Kleck and Gertz NSDS 1993 handguns 1993 1500000 1900000 handgun dedicated_dgu_survey yes yes one_year 4977 222
CDC about page citing NAS Priorities range 2013 60000 2500000 any agency_cite_range n/a n/a n/a

Methods & limits

Question. When people quote ~80,000 or ~2.5 million defensive gun uses a year, what was actually asked, of whom, and why does CDC overlay a range instead of picking a number?

  • Beacon rule: LEE does not run a 2.5 million DGU engine, and it does not hide Kleck. This page compares instruments. The downloadable CSV is literature metadata — not a time series LEE can recompute, and not NCVS microdata.
  • CDC overlay: the CDC firearm-violence about page (restored 2026) cites the National Academies report Priorities for Research to Reduce the Threat of Firearm-Related Violence for a range of 60,000 to 2.5 million defensive gun uses each year. CDC did not measure that range. Toggle it as a band, not as a third ‘CDC number.’
  • Low pole (as Kleck cites NCVS): about 68,000 DGUs/year for assaults and robberies, or about 80,000–82,000 with household burglaries. NCVS is a non-anonymous Census/DOJ victimization survey. It does not directly ask ‘did you use a gun for self-defense.’ The self-protection question is asked only after the respondent already reported a crime, and gun use must be volunteered.
  • Kleck’s NCVS critique (methods, not a dunk): reporting a public-place DGU to a federal interviewer can look like confessing unlawful carry. Location is asked before self-protection. Sensitive crimes (including many assaults inside families) are already under-reported, so DGUs attached to those crimes stay hidden.
  • High pole: Kleck and Gertz, Armed Resistance to Crime, JCLC 86(1) 1995. National Self-Defense Survey — anonymous random-digit-dial, February–April 1993, adults 18+ in the lower 48, 48 state strata, weighted n = 4,977, 222 DGU cases, 61% completes where a person answered. Asked everyone; own and household; 1-year and 5-year recall; any gun type; excluded animals and occupational uses; follow-ups on what was done with the gun. Best cell: 1-year firsthand person-based estimates of about 2.2–2.5 million civilian DGUs against humans (about 1.5–1.9 million handguns). A estimates vs more conservative B estimates that drop dirty-looking cases.
  • Earlier gun surveys Kleck treats as better than NCVS but still thin: Hart 1981 about 640,000 handgun DGUs/year; Mauser 1990 about 700,000 any-gun DGUs/year. National; excluded animals and occupational uses; typically one or two questions; five-year household recall. Kleck says eleven usable surveys implied over 700,000/year; none were NCVS-small.
  • Explorer recomputes in the browser from the CSV: pick an instrument, keep or drop the CDC/NAS overlay, copy a shareable URL (?inst=&cdc=). Log-scale bars so 80k and 2.5 million can sit on the same chart without a fake linear ‘winner.’
  • What truth requires in prose: (1) ‘How many DGUs’ is an instrument question before it is a policy number. (2) CDC’s public overlay is an envelope, not a measurement. (3) Kleck is 222 phone reports from 1993 — not a map, not a county panel, not a yearly series. (4) NCVS microdata is the planned lower-bound path if LEE later pulls it; this page does not pretend we already have it.
  • Out of scope here: carry-law causal models, personal-odds calculators, state DGU maps, and treating NSDS as downloadable microdata.
Limitations. We do not have Kleck’s 222 incident files. Table 1 in the 1995 paper lists additional surveys; this v1 CSV only includes poles and the two gun surveys Kleck singles out as relatively good (Hart, Mauser), plus CDC/NAS as overlay. NCVS figures here are Kleck’s 1995 citation of NCVS, not a fresh BJS extract. CDC overlay uses the restored about-page range, not WONDER mortality (deaths are a different domain). False-positive / telescoping / external-validity critiques of rare-event DGU surveys are real and not adjudicated as a winner on this page. Next iron if ordered: NCVS public-use self-protection items as a true lower-bound series; optional extra literature rows after a second pass on Table 1.

Sources.