Liberty Evidence Engine

Evidence · 2026-09-11

Most defensive gun uses don't involve shots fired

William English's 2026 National Firearms Survey (Wave 2) asked 51,399 U.S. adults about defensive gun use. The public microdata show 6,673 lifetime DGUs among respondents who answered that question. English's published finding that most incidents did not involve firing the gun is the lead — reported here as his result, not a LEE recompute. Large sample; survey, not NCVS; not a Kleck war.

Download data Methods

Data

label count share_of_answered_pct n_answered n_survey source
Once 2893 43.4 6673 51399 file_unweighted
Twice 2345 35.1 6673 51399 file_unweighted
Three times 920 13.8 6673 51399 file_unweighted
Four times 159 2.4 6673 51399 file_unweighted
Five or more times 356 5.3 6673 51399 file_unweighted

Methods & limits

Question. When a 2026 survey of 51k U.S. adults asks gun owners about defensive firearm use, what does the public microdata show for lifetime frequency — and what stays English's published result about shots fired?

  • This page reports William English, 2026 National Firearms Survey: Results from the Second Wave (SSRN working paper, posted August 24, 2026). Fielded July–August 2026. The public microdata file on Harvard Dataverse has 51,399 rows (paper: 51,398 adults; 16,688 internally validated gun owners).
  • Chart and CSV are unweighted counts from question Q15 in the Dataverse Qualtrics dump: how many times the respondent defended themselves, property, family, or a household member with a firearm. 6,673 respondents answered with at least one lifetime incident (once 2,893; twice 2,345; three 920; four 159; five or more 356). 44,725 rows are blank on Q15 (skip / not asked). This file has no survey weights, so LEE does not nationalize these counts into 88 million owners or 2.2 million incidents per year.
  • Lead finding, as cite: English reports that in most defensive incidents the gun was not fired (published share about 76%). That is his weighted incident result. v1 does not recompute fire / show / neither from the skip-path follow-ups. Same rule as the annualized 2.2 million figure: labeled as English's published number, not a LEE engine.
  • Instrument: online survey (Centiment), not NCVS, not CDC, not a crime-count. Lifetime recall, not a one-year victimization series. Compare instruments on Why defensive-gun-use estimates disagree.
  • Downloadable CSV is the unweighted Q15 distribution above. Re-run from the Dataverse .tab if you want to check the counts. Urban homicide concentration remains the Lee demo page; this is a survey report sitting beside it.
Limitations. No weights in the public tab, so this page cannot honestly produce national stock or annual DGU totals. Q15 is lifetime frequency, not incidents per year. Fire / show / neither follow-ups are skip-pathed across multiple Qualtrics blocks; v1 leaves English's ~76% not-fired share as a cite. Survey self-report: telescoping, social desirability, and rare-event false positives are real and not adjudicated here. Not a carry-law causal model. Not NCVS microdata.

Sources.