Violent Crime Has Run 9-21% Higher in Republican-Leaning States Than Democratic-Leaning States, 2015-2024 — But the Gap Is a Cross-State Pattern, Not a Within-State Effect
FBI Crime Data Explorer state violent-crime rates matched to the AskAmerica State Political Index (SPI), 50 states + DC
Summary
Yes — over the last ten years (2015-2024), FBI-reported violent crime rates have been consistently higher, on average, in Republican-leaning states than in Democratic-leaning states. Using the FBI Crime Data Explorer's state violent-crime rate matched to AskAmerica's State Political Index (SPI), Republican-leaning states averaged 406.8 violent crimes per 100,000 residents versus 336.2 for Democratic-leaning states — 21% higher (an unweighted, one-state-one-vote comparison). Weighting by state population narrows but does not erase the gap: 408.0 vs 374.1 per 100k, 9% higher. The pattern held in every single year from 2015 through 2024 and is not driven by any single state.
Two important qualifications change how strong a claim this supports. First, the gap survives but shrinks by roughly a third after controlling for state poverty rate and median household income, consistent with published research showing socioeconomic confounders explain part — not all — of the raw gap. Second, and more importantly: when the same data are analyzed as a state-and-year fixed-effects panel — asking whether a given state's crime rate moves when its own political lean shifts, rather than comparing different states to each other — the relationship is statistically indistinguishable from zero (coefficient +0.08, p=0.73). This is a persistent difference BETWEEN which states currently lean red or blue, not evidence that a state's crime rate rises or falls as its politics change. The published literature is itself split for exactly this reason: state-level comparisons (Third Way, CDC-based) find red states higher; county-level comparisons (Manhattan Institute, Heritage Foundation, RealClearInvestigations/Lott) find the opposite once local resolution and demographic controls are added.
1. What the published research already says
This question has an active, adversarial literature. Reading it first surfaced three distinct camps, which is itself the main methodological finding:
- Third Way (Feb 2024, using CDC/NCHS mortality data, not FBI data), The 21st Century Red State Murder Crisis: red states (2020 Trump-won) had a murder rate 33% higher than blue states in both 2021 and 2022, and 23-24% higher on average every year from 2000-2022 — a result that held even after removing each red state's largest county.
- Center for American Progress Action Fund (2023): comparing gun-homicide rates in cities, after controlling for population, cities in blue states are consistently safer from guns than cities in red states, Jan 2015-Aug 2023.
- Manhattan Institute (Borjas & VerBruggen, Feb 2024) and the Heritage Foundation: at the STATE level, Democratic areas look safer; at the COUNTY level, the relationship reverses — Republican-voting counties have lower homicide rates than Democratic-voting counties, largely because the highest-crime blue counties are concentrated in large cities that also happen to sit inside otherwise-red states. The Manhattan Institute brief further shows that once state or county models add controls for age, income, poverty, and racial composition, the red/blue homicide gap "can easily" be made to disappear — their central point being that the conclusion is highly sensitive to the researcher's choice of geographic grain and control set, not settled fact in either direction.
This split is the reason the analysis below reports both the raw cross-state gap AND a within-state, over-time test, rather than stopping at the first comparison.
2. AskAmerica data and method
Crime measure: crime.cde_offenses (FBI Crime Data Explorer), offense_code = 'violent-crime', summed across each year's 12 monthly rates to get an annual rate per 100,000 residents, for all 50 states + DC, 2015-2024 (confirmed via data_coverage: no interior gaps in this table for that window).
Political-lean measure: officials.state_political_index (SPI, a 0.50 x presidential + 0.30 x Senate + 0.20 x House composite on a -100 to +100 scale), read PER CONGRESS and matched to the Congress covering each crime year (114th Congress = 2015-16 ... 118th = 2023-24) — deliberately avoiding the trap of applying a single, current-day 'red state/blue state' label backward across a decade in which some states' partisan composition actually changed.
Data-quality control: cde_offenses.population_coverage_pct flags the share of each state's population covered by agencies that actually reported that month. National average coverage dropped from ~96% to ~87% in 2021 (the FBI's SRS-to-NIBRS reporting transition), and individual states/years — e.g. Alabama's 2019 rate implausibly halves alongside a coverage drop to 65% — show clear reporting artifacts, not real crime declines. State-years with under 80% average coverage (31 of 510, or 6%) were excluded from every figure below.
Controls and robustness: partial correlation controlling for state poverty rate (census.poverty_rate) and median household income (census.income_summary); cluster-robust regression (clustering on state, since each state contributes up to 10 correlated year-observations); leave-one-state-out sensitivity analysis; and a two-way (state + year) fixed-effects panel regression, which is the correct tool for 'does a state's own crime rate move when its own politics move,' something diff_in_diff or a plain cross-sectional regression cannot answer.
3. Findings
3.1 The raw comparison. Over 2015-2024, Republican-leaning state-years (SPI < 0) averaged 406.8 violent crimes per 100k vs. 336.2 for Democratic-leaning state-years (SPI > 0) — Republican-leaning 21.0% higher, unweighted (one state-year, one vote). Weighting by each state-year's population — which matters because Republican-leaning states, on average, are somewhat less populous — narrows the gap to 408.0 vs 374.1, or 9.0% higher. Breaking out SPI's five-category classification, 'Solid Republican' states carry the single highest population-weighted rate of any category (417.7 per 100k); 'Lean Republican' states, interestingly, carry the LOWEST (355.6) — the pattern is concentrated in the most reliably red states, not spread evenly across all right-leaning ones.
3.2 Every year, but narrowing. Republican-leaning states had a higher population-weighted rate in every single year from 2015 through 2024. The gap was largest in 2021 (443.1 vs 353.4, +25%) and has narrowed sharply since — by 2023-2024 the two groups are within 2-5% of each other (390.3 vs 384.8 in 2023; 371.5 vs 360.7 in 2024). Whether this convergence continues or is a temporary post-2021 reversion is not something 10 years of data can resolve.
3.3 Confounders explain some, not all, of the gap. The plain (zero-order) correlation between SPI and the annual violent-crime rate is r = -0.243 (more Republican-leaning associates with higher crime; p < 0.001, n=420). Holding state poverty rate and median household income fixed, the partial correlation shrinks to r = -0.165 but remains statistically significant (p < 0.001). This is consistent with the Manhattan Institute's finding that socioeconomic controls attenuate the gap — it is NOT consistent with their finding that controls can make it disappear entirely, at least with these two controls at the state grain.
3.4 Statistical significance is sensitive to how correlated state-year observations are treated. A pooled cross-sectional regression of the annual rate on SPI, with standard errors clustered by state (49 clusters, reflecting each state contributing up to 10 correlated year-observations rather than 10 independent ones), gives a coefficient of -0.415 (crime rate falls by about 0.42 per 100k per point of SPI toward Democratic) but is only borderline significant, p = 0.10 — the raw n=469 overstates the effective number of independent observations, which is closer to 49 states.
3.5 The gap does not survive as a within-state, over-time effect. This is the most important qualifying result. A two-way fixed-effects panel (absorbing everything constant within a state and everything common to all states in a given year, then asking whether a state's OWN crime rate tracks its OWN SPI as that SPI shifted across five Congresses) finds a coefficient of +0.08 with p = 0.73 — no detectable within-state relationship at all. In plain terms: the cross-state gap in section 3.1 reflects who currently leans red or blue and how much crime those particular states happen to have, not a demonstrated tendency for a state's crime rate to rise when it becomes more Republican or fall when it becomes more Democratic (or vice versa). States' SPI scores did not move enough within this 10-year window, and crime moved for reasons unrelated to that modest partisan drift.
3.6 Robust to any single state. A leave-one-state-out sensitivity analysis on the pooled cross-sectional coefficient found no state whose removal flips the sign or crosses the p=0.05 significance threshold; the coefficient ranges only from -0.32 (dropping Alaska) to -0.55 (dropping New Mexico) across all 49 leave-one-out refits. The cross-sectional pattern is broad-based, not an artifact of Mississippi, Louisiana, or any other single high-crime red state.
4. Caveats
- Grain matters enormously. This analysis is at the STATE level, matching most of the political literature (Third Way, CAP). The Manhattan Institute and Heritage Foundation both show the relationship reverses at the COUNTY level — Republican-voting counties have LOWER homicide rates than Democratic-voting counties, because high-crime cities cluster inside otherwise-red states. This report cannot and does not adjudicate which grain is 'correct' — they answer different questions (which STATE governments preside over more violence vs. which local political culture is associated with more violence).
- Violent crime (FBI UCR), not murder (CDC). The most widely cited study (Third Way) uses CDC mortality data restricted to homicide; this analysis uses the FBI's broader violent-crime rate (murder, rape, robbery, aggravated assault combined). The two series can diverge and are not directly comparable in magnitude, though both show the same directional gap here.
- FBI reporting-coverage gaps. The FBI's 2018-2021 shift from the legacy Summary Reporting System to NIBRS-only reporting caused real, non-crime-related swings in several states' reported rates (see 2021's national coverage drop to 87%, and Alabama's 2019 coverage drop to 65% alongside an implausible rate halving). 6% of state-years were excluded on this basis; a different coverage threshold would move the exact percentages slightly but not the direction of the finding.
- SPI construction. SPI blends presidential electoral-vote share, Senate composition, and House composition per Congress — it is a composite political-lean score, not a simple 'who won the state in the last election' label, and can occasionally classify a state counter to casual expectation in an earlier period (e.g., Florida scored 'Lean Democratic' in the 114th Congress, 2015-16, reflecting Obama-era electoral and congressional composition, not 2015 partisan perception).
- Causality. Nothing here establishes that political control causes crime rates, in either direction. Section 3.5's null within-state panel result is itself evidence against a simple causal story running from a state's partisan lean to its crime rate over this window.
Sources
- Third Way — The 21st Century Red State Murder Crisis (Feb 2024) — CDC/NCHS mortality-based murder rate comparison, 2000-2022
- Center for American Progress Action Fund — Cities in Blue States Experiencing Larger Declines in Gun Violence in 2023
- Manhattan Institute — The 'Red' vs. 'Blue' Crime Debate and the Limits of Empirical Social Science (Borjas & VerBruggen, Feb 2024) — State vs. county grain reversal; effect of demographic/economic controls
- Planetizen — Red Cities, Blue Cities, and Crime
- Independent Institute — Are Blue or Red States Worse on Crime?
- AskAmerica: population-weighted and unweighted violent crime rate by political lean, 2015-2024
Show SQL
WITH crime_annual AS (SELECT state_abbr, CAST("year" AS INTEGER) AS yr, SUM(offense_rate) AS annual_rate, AVG(population_coverage_pct) AS avg_cov, MAX(population) AS population FROM crime.cde_offenses WHERE offense_code = 'violent-crime' AND state_abbr <> 'national' AND "year" BETWEEN '2015' AND '2024' GROUP BY state_abbr, CAST("year" AS INTEGER)), pol_base AS (SELECT state_name, congress_start_year, cpi, political_lean FROM officials.state_political_index WHERE congress IN (114,115,116,117,118)), pol AS (SELECT sr.state_abbr, pb.congress_start_year + t.gs AS yr, pb.cpi, pb.political_lean FROM pol_base pb CROSS JOIN (VALUES (0),(1)) AS t(gs) JOIN geo.state_ref sr ON sr.state_name = pb.state_name), joined AS (SELECT ca.state_abbr, ca.yr, ca.annual_rate, ca.avg_cov, ca.population, p.cpi, CASE WHEN p.cpi < 0 THEN 'Republican-leaning' ELSE 'Democratic-leaning' END AS lean_bin FROM crime_annual ca JOIN pol p ON p.state_abbr = ca.state_abbr AND p.yr = ca.yr) SELECT lean_bin, COUNT(*) AS n_state_years, AVG(annual_rate) AS unweighted_avg_rate, SUM(annual_rate*population)/SUM(population) AS pop_weighted_avg_rate FROM joined WHERE avg_cov >= 80 GROUP BY lean_bin - AskAmerica: partial correlation of SPI vs. violent crime rate controlling for poverty and income
Show tool call
partial_correlation(x="cpi", y="annual_rate", controls=["poverty_rate_pct","median_household_income"]) - AskAmerica: cluster-robust regression of violent crime rate on SPI, clustered by state
Show tool call
robust_regression(outcome="annual_rate", predictors=["cpi"], cluster_col="state_abbr") - AskAmerica: two-way (state + year) fixed-effects panel regression
Show tool call
panel_fixed_effects(outcome="annual_rate", predictors=["cpi"], entity_col="state_abbr", time_col="yr", cluster_col="state_abbr") - AskAmerica: leave-one-state-out sensitivity analysis on the cpi coefficient
Show tool call
sensitivity_analysis(outcome="annual_rate", predictors=["cpi"], group_col="state_abbr")