August 11, 2025 to September 30, 2026 · Metropolitan Police incidents and gunshot detections · Martin Ngoh
Violence in DC fell by close to half after the National Guard arrived on August 11, 2025. It fell just as much where no troops were posted, so nothing in the data shows the troops themselves adding to the decline.
It fell. In the first Guard year, August 11, 2025 to August 10, 2026, reported crime was −34% against the average for the same dates over the previous 10 years. The year before, it was already −11%.
Homicides and gunfire fell too, and they do not depend on anyone calling the police. Homicides fell from 167 to 94 (−44%). Gunshots picked up by sensors fell −44% from the year before (August 11 to June 30). From September to June, DC's homicide drop was the second largest of 36 large police departments.
It did not fall more where troops were posted. In the first months, gunshots fell −57% in the two wards holding nearly all the Guard posts and −60% in the wards without them. When the Guard began patrolling Anacostia in 2026, gunshots there fell −35% against −40% in neighboring Ward 7, which had no documented patrols.
Tests written down before they ran agree. Across about 800 half-kilometer squares, crime near troop posts fell no more than in comparable areas on any of six measures, and the areas around the Metro stations where troops stood did worse than most random sets of other stations. Against other large cities, DC's property crime fell about 22% more after August 2025, but citywide, not near the posts.
Federal agents arrived the same day, and violent crime was already falling, so this data cannot say what drove the decline. It can say the decline does not follow the troops.
The first Guard year (August 11, 2025 to August 10, 2026) against the average of the same dates in the 10 years before. The second line is the year before the deployment, against the same average.
| Offense | First Guard year | 10-year average | Guard year vs average | Year before vs average | Change from year before |
|---|---|---|---|---|---|
| All reported crime | 21,027 | 31,606 | −34% | −11% | −25% |
| Violent crime | 2,556 | 4,330 | −41% | −32% | −13% |
| Violent crime with a gun | 941 | 2,058 | −54% | −19% | −44% |
| Homicide | 94 | 176 | −46% | −5% | −44% |
| Robbery | 1,189 | 2,339 | −49% | −24% | −33% |
| Assault with a dangerous weapon | 1,165 | 1,597 | −27% | −43% | +29% |
| Sex abuse | 108 | 218 | −51% | −59% | +20% |
| Property crime | 18,471 | 27,275 | −32% | −8% | −27% |
| Car theft | 2,398 | 3,613 | −34% | +42% | −53% |
| Theft from cars | 4,023 | 9,218 | −56% | −29% | −38% |
| Other theft | 11,321 | 13,070 | −13% | −4% | −10% |
| Burglary | 728 | 1,368 | −47% | −34% | −19% |
A 10-year average removes the seasons, not the trend. These two checks ask how much of the drop was new.
| Offense | Against level before | If the trend continued |
|---|---|---|
| All reported crime | −24%−30% to −17% | −25%−35% to −13% |
| Property crime | −25%−31% to −19% | −27%−36% to −16% |
| Violent crime | −12%−26% to +5% | −5%−30% to +29% |
| Violent crime with a gun | −43%−54% to −29% | −28%−48% to +1% |
| Homicide | −48%−60% to −32% | −49%−61% to −32% |
DC murders fell −51% from September to June against −15% nationally. The year before, DC fell −22% and the nation −17%. Against the 12 months before, the step is −48% (−60% to −32%).
Against the 12 months before, violent crime shows a step of −12% (−26% to +5%), which cannot be told apart from no change. Robbery fell −34%, but it had fallen −30% the year before.
MPD's crime data is under investigation for downgraded reports, and fewer calls to police can look like less crime. Homicides and gunshots picked up by sensors avoid most of that. Sensors cover Wards 1, 2 and 4 through 8; Ward 3 has none.
Since 2024, only 3 single days had no detections. There were no longer gaps.
Of 61 one-kilometer squares with 20 or more detections the year before, 0 went silent. The typical square fell −46%.
The "probable gunfire" category left out here fell too, so detections did not move between labels.
Shootings counted by the Gun Violence Archive were down about two thirds from a year earlier (The Trace, October 2025).
The Guard's posts moved, so the test runs in two phases. Phase 1, August 11 to November 26, 2025: posts and patrols were downtown, on the Mall, at Metro stations, Union Station and Navy Yard, and along H Street, 14th Street, Dupont Circle, Georgetown and Capitol Hill, almost all in Wards 2 and 6, compared with Wards 3, 4, 5, 7 and 8 (Ward 1 is left out because patrols reached 14th Street by mid-September). Phase 2, January to June 2026: Guard patrols in Anacostia, MPD's 7th District, compared with the neighboring 6th District (Ward 7), where no patrols are documented. Each is the change from the same dates a year earlier.
For gunshots, homicide, reported gun crime and violent crime, in both phases, by ward and within 500 meters of a post, no comparison shows the troop areas falling more by a margin outside its 95% range. Property crime fell less with troops in both phases. The ranges come from resampling two- and four-week blocks, so ordinary swings in crime are not mistaken for an effect.
Other theft in the downtown business district fell −29% in the first Guard year against −7% elsewhere, but the year before it had risen about 26% faster than elsewhere, so the drop mostly undoes that rise. Theft from cars downtown fell −56% against −37%, and part of that gap was already there before August.
| Ward | Guard posts | Homicides | Gunshots detected |
|---|---|---|---|
| Ward 1 | none | 13 to 7 | 140 to 77 (−45%) |
| Ward 2 | 7: downtown Metro stations, the Mall | 5 to 5 | 32 to 17 (−47%) |
| Ward 3 | none | 2 to 4 | no sensors |
| Ward 4 | none | 7 to 1 | 151 to 105 (−30%) |
| Ward 5 | none | 15 to 8 | 308 to 205 (−33%) |
| Ward 6 | 5: Union Station, Eastern Market, Waterfront, NoMa, L'Enfant Plaza | 12 to 4 | 140 to 89 (−36%) |
| Ward 7 | 1, on its western edge (Stadium-Armory) | 38 to 20 | 1,731 to 906 (−48%) |
| Ward 8 | 1, on its western edge (Navy Yard); Anacostia patrols from January 2026 or earlier | 58 to 24 | 1,880 to 1,029 (−45%) |
August 11 to June 30, year before to Guard year. Posts as reported by WJLA, Guard releases, a Guard daily update filed in federal court (September 12, 2025) and a November 2025 court opinion; there is no official map.
On October 8, 2026, before running them, I published a test plan fixing the areas, dates, measures and what would count as an effect: a reduction of 15% or more near posts. Every test it lists is reported here. Afterward, a Guard daily update filed in court (September 12, 2025) showed more patrol sites than the plan had, so the tests were rerun with them. Both versions are shown.
| Measure | Near posts, planned sites | With the September list | Random station sets that did better | Reading |
|---|---|---|---|---|
| Gunshots detected | 0.920.73 to 1.14 | 0.940.76 to 1.17 | 93% | Inconclusive |
| Homicide, robbery, sex abuse | 1.431.16 to 1.76 | 1.301.05 to 1.60 | 100% | Planned sites: increase near posts. September list: increase near posts, not significant across six measures. |
| Reported gun crime | 0.950.75 to 1.22 | 0.850.66 to 1.09 | 90% | Inconclusive |
| Property crime | 1.010.91 to 1.12 | 1.020.92 to 1.14 | 92% | Rules out a 15% reduction |
| Other theft | 0.970.84 to 1.13 | 0.980.85 to 1.14 | 96% | Planned sites: inconclusive. September list: rules out a 15% reduction. |
| Theft from cars | 0.980.81 to 1.18 | 0.960.80 to 1.16 | 28% | Inconclusive |
None of the six measures shows fewer crimes near posts than in comparable areas. For property crime, the range rules out a reduction of 15% or more with either list of sites; for other theft, only with the September list. For gunshots and gun crime, the range is too wide to rule out a modest reduction, and it also allows an increase.
I drew 12 stations at random from the 24 DC stations away from any known Guard site, 1,000 times, and ran the same test on each set. For gunshots, 93% of the random sets did better than the real Guard stations; for homicide, robbery and sex abuse, 100%. The Guard probably chose stations with growing problems, which would explain part of this.
Homicide, robbery and sex abuse near posts ran +43% against comparable areas with the planned sites, which holds after correcting for testing six measures, and +30% with the September list, which does not hold after that correction. About half of the gap was opening before the troops came: measured against only the 24 weeks before, it is +19% (0.93 to 1.53).
Metro Transit Police blotters (January 2022 to August 2026, crimes with victims, placed at stations by address) show the stations that later got troops had been getting worse than other stations for two years, peaking in the three months before August 2025. Afterward they fell back 18 to 46% from that peak, but changed like other stations against the year before (+7% and +7%). A Guard effect and an ordinary fall from a peak look the same here.
Against a synthetic DC built from other large agencies, September 2025 to June 2026: property crime −22% (p = 0.04), murder −27% (p = 0.10), robbery +4% (p = 0.70). With the tests near posts, the property drop came citywide, not where troops stood.
Troops fell from 5,148 in July 2026 to 3,124 by September 7 and 2,864 by October 7. Against normal, citywide crime did not move (−26% in the eight weeks before the first state left, −26% after). Near posts, violent crime rose against an unusually quiet June, back to its February to May level. Gunshots for this period come out in November.
Gunshots near the two Anacostia posts fell 25% more than in comparable areas, and reported gun crime along the patrolled corridors fell about 48% more, although gunshots there rose. Among 36 wave-level estimates, one or two significant results in each direction is what chance alone produces; the combined tests the plan relied on show no reduction.
The Niskanen Center (May 2026) found a property crime drop of about 24% and no clear effect on violent crime. A Senate Homeland Security Committee minority staff report (2026) found "no directly attributable impact on crime."
Planned sites. Each column changes one choice from the plan: the radius around a post, the start dates for waves 2 and 3, leaving out periods with July 4 or December 31, leaving out the National Mall, and using only control squares as busy as the posts. No version shows fewer crimes near posts by a margin outside its 95% range. The rise in homicide, robbery and sex abuse appears in every version. A negative binomial model, run for gunshots in the first wave, gives 1.09 (0.78 to 1.53). Leaving out the Mall changes almost nothing because MPD records very little crime there: the Park Police patrol it, and its squares have no gunshot detections, so they are not in the gunshot measure at all.
| Measure | Main | 250 m | 1,000 m | Other start dates | No holiday periods | No Mall | Busier controls |
|---|---|---|---|---|---|---|---|
| Gunshots detected | 0.920.73 to 1.14 | 0.830.55 to 1.27 | 1.020.85 to 1.24 | 0.890.73 to 1.09 | 0.910.70 to 1.19 | 0.920.73 to 1.14 | 0.950.74 to 1.21 |
| Homicide, robbery, sex abuse | 1.431.16 to 1.76 | 1.461.09 to 1.96 | 1.191.00 to 1.42 | 1.351.09 to 1.66 | 1.411.13 to 1.76 | 1.431.16 to 1.76 | 1.561.26 to 1.93 |
| Reported gun crime | 0.950.75 to 1.22 | 1.330.98 to 1.81 | 0.920.75 to 1.13 | 0.880.69 to 1.14 | 0.900.71 to 1.16 | 0.950.75 to 1.22 | 1.060.82 to 1.37 |
| Property crime | 1.010.91 to 1.12 | 1.050.88 to 1.26 | 1.010.92 to 1.11 | 1.020.92 to 1.14 | 1.010.91 to 1.13 | 1.010.91 to 1.12 | 1.030.91 to 1.16 |
| Other theft | 0.970.84 to 1.13 | 1.000.78 to 1.28 | 0.990.85 to 1.16 | 0.990.85 to 1.15 | 0.970.84 to 1.13 | 0.970.84 to 1.13 | 0.980.83 to 1.15 |
| Theft from cars | 0.980.81 to 1.18 | 1.250.99 to 1.56 | 0.970.85 to 1.11 | 0.950.79 to 1.15 | 0.980.81 to 1.19 | 0.980.81 to 1.18 | 1.130.92 to 1.39 |
The data shows violence fell and that the fall does not follow the troops. It cannot name the cause. Possibilities include the federal agents, MPD's own work, a trend already under way, or some mix.
On August 11, 2025 the federal government also took control of MPD (to September 10), sent in federal agents and stepped up immigration enforcement. The White House said about half of arrests by federal authorities were in Wards 7 and 8, where there were almost no troops. Federal agents kept working alongside MPD after September 10.
A House Oversight report (December 2025), an MPD internal affairs report (May 2026) and the DC Inspector General (July 2026) found misclassified and downgraded reports. Assault with a dangerous weapon rose +29% as knife and other-weapon cases roughly doubled while gun cases fell, which looks like a change in how assaults are recorded. That is why this page leans on homicides and gunshots.
People avoiding police during immigration enforcement, or federal agencies taking reports that never reach MPD's data, would both lower counts. Neither can be measured here. Crimes on the National Mall are mostly handled by the Park Police and are not in this data.
The two wards with troops had 17 homicides the year before, too few to detect much. Gunshots, with thousands of detections, carry the comparison.
There is no official map of Guard posts. Locations come from news reports, Guard releases, a Guard daily update filed in federal court and a court opinion. The update counts 18 Metro stations but names none; 12 are known from news reports, so some comparison stations probably had troops. Records requests for the post list are drafted. The Guard also did occasional patrols and cleanup elsewhere, so "without troops" means light presence, not none.
Reported crime: MPD's public incident data (homicide, sex abuse, assault with a dangerous weapon, robbery, burglary, car theft, theft from auto, other theft, arson), January 2015 to September 30, 2026, by report date. Data for October 2026 is left out as incomplete. The 10-year average for any window is the mean count on the same calendar dates from August 11, 2015 to August 10, 2025, so it contains no days after the deployment. Wards are the 2022 boundaries, assigned by each incident's location.
Gunshots: MPD's ShotSpotter data through June 30, 2026, single and multiple gunshots only. Comparisons that use gunshots run August 11 to June 30 in every year. National comparison: the Real-Time Crime Index (AH Datalytics), monthly through June 2026, a fixed sample of 590 agencies with DC removed. The step model is a Poisson regression of monthly counts with the 10-year norm as offset and Newey-West errors. Troop comparisons are ratios of changes with block-bootstrap ranges and a placebo year.