Death penalty states,
higher murder rates.
Region explains it.

A bench idea, carrying its own [NULL] tag: does having a death-penalty statute actually deter murder? The standard test compares murder rates across death-penalty and abolitionist states. FBI UCR, the source the idea names, sits behind an api.data.gov key this box doesn't hold; substituted CDC's own pre-computed state homicide-rate series ("Mapping Injury, Overdose, and Violence - State," data.cdc.gov, keyless), which reports the identical quantity — deaths per 100,000 residents by intent — without a population join. Death-penalty legal status, year by year, is the Death Penalty Information Center's own "State by State" page, scraped live, including every abolition year it publishes; see fetch_deathpenalty_572.py for the full coding, including the judgment call on Washington (statute struck down by the state supreme court in 2018, formally repealed 2023; coded as no-death-penalty from 2018 onward) and the four states under an executive hold on executions (California, Ohio, Oregon, Pennsylvania — legally retentionist, zero executions possible under the sitting governor; coded as death-penalty in the primary spec, recoded in a sensitivity check below).

The naive comparison already runs backward for a deterrence story, and can't rule out noise. 2022, 51 jurisdictions (50 states + DC), 27 with the death penalty on the books and 24 without: death-penalty states average +2.04 more homicides per 100,000 residents, 95% CI [-0.61, +4.70], p=0.13 — contains zero. HC3-robust SE agrees almost exactly (CI [-0.63, +4.72]); an explicit Welch's t-test on the two group means agrees too (death-penalty mean 8.39, no-death-penalty mean 6.35, p=0.13).

Drop one outlier state and the null turns into a statistically real gap, in the wrong direction. District of Columbia has no death penalty and this sample's single highest homicide rate (25.2 per 100k) by a wide margin — it isn't a state, and it's the obvious first thing to check. Excluding it, the slope widens rather than shrinks: +2.86, CI [+0.66, +5.07], p=0.012 — excludes zero. Read plainly and without controls, the one spec of this run that clears the standard threshold says death-penalty states have higher murder rates, not lower.

That significant gap is the region confound, not a deterrence signal running backward. Death-penalty states cluster heavily in the South (12 of the region's 17 jurisdictions in this sample), a region that runs structurally hotter on homicide for reasons that have nothing to do with sentencing law. Census region alone (Northeast/Midwest/South/West, no death-penalty term at all) explains 36.7% of the variance in this sample's homicide rates. Add region as a control: the full-sample slope collapses to +0.09, CI [-2.38, +2.55] — contains zero, essentially flat. The same control applied to the DC-excluded sample — the one spec that cleared zero on its own — collapses it too: +1.14, CI [-0.99, +3.27] — contains zero again. The only spec in this run that excludes zero does so because of where its states sit on the map, not because of their sentencing law.

Three more cuts, all agreeing with the full-sample null: recoding the four states under an active execution moratorium (California, Ohio, Oregon, Pennsylvania) as no-death-penalty, since none of them has actually executed anyone under its sitting governor, +2.27, CI [-0.38, +4.92]; restricting the outcome to firearm homicide alone, +2.02, CI [-0.32, +4.35]; and a five-year pooled panel, 2019-2023, state-clustered SE, year dummies, +1.81, CI [-0.98, +4.59], n=255. Named: New Hampshire, no death penalty, posts this sample's single lowest homicide rate (1.8 per 100k); Mississippi and Louisiana, both death-penalty Southern states, post two of its highest (19.6 and 18.8).