Introduction and summary
The second Trump administration is trying to take credit for the historic drop in violent crime across America despite the fact that this trend began before it took office, and it is using these declines to justify expanding policies that are unpopular, ineffective, and costly.1 Even though the data clearly show that violent crime and homicides were already dropping in American cities in 2023 and 2024, the Trump administration has continued to threaten city and state leaders and argue that its extreme actions, such as deploying the National Guard to support law enforcement operations, are improving public safety.2
However, new Center for American Progress analysis finds no evidence that National Guard deployments have reduced violent crime. Moreover, if these deployments are extended and continue through the end of 2026, they could cost American taxpayers more than $1.7 billion.
Background
In June 2025, in response to resistance against the federal immigration enforcement surge, President Donald Trump federalized California’s National Guard and deployed 2,000 guardsmen, along with 700 U.S. Marines, to Los Angeles to “protect ICE and other United States Government personnel who are performing Federal functions.”3 Although this initial deployment was limited in scope, it served as a road map for subsequent authoritarian power grabs in American cities that Trump had been threatening for years.4 Just two months later, on August 11, 2025, President Trump federalized Washington, D.C.’s, police force and sent an initial 800 National Guard troops to the city, declaring a “crime emergency,” despite violent crime in the city being at a 30-year low in 2024.5 Since then, Trump has authorized the deployment of National Guard troops and other federal agents to four more cities—Portland, Oregon; Chicago; Memphis, Tennessee; and New Orleans. He has also threatened to send the National Guard to five additional cities—Baltimore; New York City; St. Louis; San Francisco; and Oakland, California—under the similar pretense of addressing crime and violence.6
Because the president’s legal authority to justify these deployments was questionable, nearly all met immediate legal challenges.7 For example, although the Trump administration sent the National Guard to the Chicago and Portland areas, courts blocked those troops from having an active presence in those cities.8 Facing public backlash and judicial opposition, in a December 2025 social media post, President Trump announced the withdrawal of National Guard troops from all cities except Memphis; New Orleans; and Washington, D.C.,9 by February 2026. Nevertheless, the president has continued to assert that these deployments, along with other federal agent surges focused on immigration enforcement, have reduced crime and made Americans safer.10 At a public roundtable in Memphis in March, Trump claimed that the Memphis Safe Task Force, which sent federal agents and the Tennessee National Guard to Memphis, “stopped crime.”11
In an effort to falsely claim his policies have reduced crime, the president is exploiting the fact that violent crime and murder were already declining in the cities his administration targeted with these extreme interventions. On average, the 11 cities where the National Guard was deployed or threatened to be deployed saw a 14 percent decrease in their 12-month rolling violent crime rate and a 22 percent decrease in their 12-month rolling murder rate from June 2024 to June 2025, before the National Guard was first deployed to Los Angeles.
These recent violent crime trends, predating even Trump’s second inauguration, suggest that, more than likely, violent crime would have continued trending downward in these cities by the end of 2025, regardless of any additional interventions. However, the Trump administration has ignored this fact when reporting year-over-year crime statistics to claim these extreme tactics have made Americans safer.12 Instead of acknowledging that violent crime and homicides have been declining at historic rates since 2023 in most American cities, the Trump administration is disregarding the longer-term trend that began before it took office.13 Any real attempt to understand if these tactics have meaningfully affected crime rates or trends in American cities requires deeper analysis, taking into account the preexisting trend before the National Guard were deployed to the targeted cities.
Recent analyses of National Guard deployment in Washington, D.C., indicate that this strategy has not affected violent crime or shootings.14 But those studies did not use a comprehensive set of data points covering the full deployment periods across all affected cities to determine whether the deployments had a measurable effect on crime trends. New CAP analysis does, however, and provides the strongest evidence to date that the Trump administration’s unpopular and costly deployment of National Guard personnel to American cities has had no measurable effect on the rates of violent crime, homicide, or gun violence.
Analysis of National Guard deployments’ impact on crime trends
To evaluate the administration’s claim that the deployment of the National Guard has reduced violent crime in U.S. cities, CAP conducted a three-stage interrupted time-series analysis using citywide homicide, violent crime, and gun victimization data. This approach was employed to identify any statistical changes in crime trends associated with the deployment of the National Guard in specific cities. Interrupted time-series analysis estimates the impact of a particular intervention on a treated group by comparing trends before and after the intervention takes place. This method enables analysts to determine whether a change in the post-intervention trend is attributable to the intervention itself or is more likely a continuation of the pre-intervention trend. For the purpose of this study, the intervention of interest is the deployment of the National Guard, the cities where the National Guard was deployed constitute the “treated” group, and the three outcomes of interest—violent crime rates, homicide rates, and gun victimization rates, are each analyzed in a separate stage.
To ensure that the conclusions of this study are as robust as possible, the researchers conducted the analysis in three stages, using multiple datasets, varied observational windows, and multiple crime measures. The national infrastructure for crime data, particularly regarding gun violence, remains insufficient for analyzing near-real-time trends.15 To address those limitations, this analysis incorporated all available measures across multiple sources to minimize bias from any single reporting mechanism. Furthermore, the researchers controlled for citywide and seasonal trends before and after National Guard deployments, using both monthly and weekly aggregate data to test for result sensitivity. In the third and final stage of this analysis, the researchers used an approximate control group of cities to facilitate comparative analysis against a plausible counterfactual group.
CAP’s analysis indicates that the deployment of the National Guard had no measurable effect on violent crime trends. By restricting the variables in each model and focusing only on deployment status and violent crime trend data, the researchers structured the analysis to be biased toward finding evidence of a statistically significant result, if one existed. This approach also accounts for potential influences from other federal law enforcement operations in these cities. Although this analysis primarily examines National Guard deployments, the coinciding operations of federal immigration enforcement strengthen the conclusion that there is no evidence that any of these extreme tactics have meaningfully improved public safety by reducing violent crime in the United States.
Model 1: Independent discontinuity trend analysis for each city
Tested variable: Seasonally adjusted monthly violent crime rate
Modeled cities: Los Angeles; Washington, D.C.; and Memphis
Data source: Real-Time Crime Index
In the first stage of this analysis, monthly totals of violent crime were obtained from the Real-Time Crime Index (RTCI)16 for each city with available data: Los Angeles; Washington, D.C.; and Memphis. Because the National Guard did not have an active presence in Portland or Chicago, even though they were federalized, they are not included in the Model 1 analysis. New Orleans was also excluded, as the National Guard was not deployed there until December 30, 2025, and RTCI data were only available through February at the time of analysis. The RTCI data were converted to seasonally adjusted monthly violent crime rates using the locally estimated scatterplot smoothing (LOESS) method for each month from January 2021 to January 2026.17 This adjustment was implemented to ensure that the model results reflect changes in crime rate trends attributable to the deployments, rather than to seasonal crime trend fluctuations in each city.
Importantly, the researchers employed the LOESS method to analyze seasonally adjusted rates rather than rolling monthly sums to increase the model’s sensitivity in detecting trend changes immediately after the intervention began. In contrast, using 12-month rolling averages would have meant that the months following the initial intervention would have been calculated using pre-intervention data, reducing the potential for an observable effect. Each city was modeled independently, testing if there was a statistically significant difference in the violent crime rate level or trend before and after the intervention in a given city.
Running this model, CAP found Los Angeles and Washington, D.C., returned statistically insignificant results, meaning that there was no detectable change in the level or trend in violent crime rates since the month of deployment. Only Memphis returned a statistically significant result for both the violent crime rate level and the violent crime rate trend after the intervention. Yet while the model estimates that there was a drop in the violent crime rate level in the month following the intervention, it also estimates that the violent crime rate began trending upward post-intervention as well. This means that even if these results are statistically significant, any immediate effect of the intervention will not be sustained and, in fact, the National Guard deployment to Memphis may have a crime-exacerbating effect over a longer period of time—though, importantly, this is not the conclusion of the researchers.
Upon conducting additional robustness checks by expanding the observational window to include 2023 and 2024 data as part of the pre-intervention trend, the analysis found that only Memphis demonstrated a statistically significant result across all tests, consistently indicating an associated decrease in the violent crime rate level and trend post-intervention. However, it is important to note that this initial model was designed specifically to bias the results toward finding a statistically significant estimate and this evidence is not conclusive. Subsequent stages of this analysis include controls for confounding variables, therefore more accurately estimating the actual effect of National Guard deployments on crime trends and levels in cities. Furthermore, the finding that only one city exhibited statistically significant results among the three tested suggests that the result is more likely attributable to natural variance rather than to a strong association between the intervention and changes in violent crime.
Model 2: Multigroup interrupted time series (ITS)
Tested variable: Seasonally adjusted weekly homicide rate
Modeled cities: Los Angeles; Washington, D.C.; and Memphis
Data source: Publicly available city data18
The second stage of the analysis uses a multigroup interrupted time-series model to compare pre-intervention and post-intervention trends across all cities where the National Guard was deployed in 2025. This approach accounts for city-specific trends before and after intervention, enabling a more precise estimate of the effect of National Guard deployment on crime, while controlling for random variation that may arise when analyzing each city independently. Furthermore, instead of relying on monthly aggregated violent crime data, which is less sensitive to immediate level and trend changes, this analysis uses seasonally adjusted estimates of weekly homicide rates (per 100,000) from cities with available data: Los Angeles; Memphis; and Washington, D.C. New Orleans is excluded from this model due to data limitations and access.
This analysis found no statistically significant crime-reducing effect of National Guard deployment on the level or trend of seasonally adjusted weekly homicide rates across the sample. In other words, after controlling for both seasonality and city-specific trends in variance, this model estimated that none of the declines in weekly homicides reported in Washington, D.C.; Memphis; and Los Angeles could be attributed to the deployment of the National Guard.
Additional robustness checks, which narrow the window of pre-intervention trend data to after January 1, 2025, yielded similar results. However, while the estimates were mostly insignificant, this additional test suggested that the National Guard deployments may be associated with an increase in the homicide trend after deployment. Specifically, the average weekly homicide rate trend increased in cities post-deployment compared with the estimated trend of these cities had the National Guard not been deployed. For an additional test, the model was run again with Chicago added. While Chicago did not see the National Guard actively deployed, this model was run with and without Chicago included to assess whether including Operation Midway Blitz in the treatment group affected the results.19 Again, the model found no evidence that these deployments or federal law enforcement surges reduced the homicide rate.
Model 3: ITS with difference-in-differences extension
Tested variable: Seasonally adjusted weekly gun victimization rate
Treated cities: Los Angeles; Washington, D.C.; Memphis; and New Orleans
Control cities: New York, Oakland, St. Louis, San Francisco, and Baltimore
Data source: Gun Violence Archive
The final model introduces another layer of complexity by comparing crime trends in cities where the National Guard was deployed with those in a plausible control group. In addition to the cities where the Trump administration sent National Guard troops, the president made threats to deploy the guard to five other cities—New York, Oakland, St. Louis, San Francisco, and Baltimore—under the same premise that crime was out of control.
If these cities were indeed just as “crime ridden”20 as Los Angeles, Memphis, New Orleans, and Washington, D.C., then crime trends in this group of threatened cities serve as a plausible counterfactual for what may have occurred in the treated cities had the National Guard not been deployed. While there are obvious limitations to constructing a control group based solely on these criteria, it is reasonable to expect that the post-intervention trends of the treated cities would differ significantly from the control group cities if in fact the National Guard deployments had a significant impact on violent crime trends. However, if the violent crime trends following National Guard deployments do not differ from the control group, this provides an additional layer of confidence that these deployments had no observable effect on violent crime.
Due to data infrastructure limitations, the only viable data source for this model is the Gun Violence Archive, which provides weekly gun victimization rates for all 11 cities in the sample.21 This final model used seasonally adjusted weekly gun victimization rates to compare trends across cities. As with Model 2, this model was executed under multiple scenarios: one excluding Chicago from the sample and another with Chicago included as a treatment city.
Consistent with the previous models, the analysis identified no statistically significant effect of National Guard deployments on either the level or trend of gun victimization rates post-deployment.
Taken together, these three models provide the most compelling and complete evidence yet that the National Guard deployments to Los Angeles; Washington, D.C.; Memphis; and New Orleans have had no measurable effect on violent crime, homicides, or gun victimizations.
National Guard and federal enforcement surges are costly
In addition to being ineffective at reducing violent crime, the National Guard deployments have been widely unpopular22 and could cost American taxpayers more than $1.7 billion by the end of 2026. Each day National Guard personnel are on active duty in American cities, taxpayers cover the cost, which includes:23
- Additional pay and benefits granted to active-duty National Guard personnel.
- Daily lodging and food costs for all active-duty personnel.
- Transportation costs, both to deployment sites and for daily operations.
A letter from the nonpartisan Congressional Budget Office (CBO) reports that sending National Guard and U.S. personnel to Los Angeles; Washington, D.C.; Memphis; Portland; and Chicago cost $496 million in 2025.24 This figure does not include the cost of deployments to New Orleans, which started in December 2025, or the ongoing deployments in Washington, D.C.; New Orleans; and Memphis in 2026. As of the publication of this report, the Louisiana National Guard deployment in New Orleans is set to last only through August25 but could be extended through the end of the year or beyond if the White House continues to push for it. In May 2026, U.S. Marshals Service Director Gadyaces Serralta requested an additional 1,500 National Guard troops to be deployed to Washington, D.C., for the Fourth of July, which would bring the total deployment level to 5,000 troops.26 If the D.C. deployment stays at 5,000 troops27 and deployments to D.C., New Orleans, and Memphis are extended through the end of the year, the total estimated cost of the National Guard deployments from June 2025 through the end of 2026 will exceed $1.7 billion.
In addition to requesting a more than $1 billion increase in Department of Defense funding to support National Guard Army personnel,28 the president’s fiscal year 2027 budget asks for $3 billion for “Combat[ing] Violent Crime in the Nation’s Cities … to surge law enforcement resources” in American cities.29 Taken together and given the proposed budget’s specific reference to Operation Make D.C. Safe and Beautiful and the Memphis Safe Task Force as a “model” for combating violent crime in American cities, this unprecedented request should shock Americans as it signals the administration’s plan to not only continue but expand the use of this harsh and ineffective tactic in cities across America, including potentially additional National Guard deployments.
In contrast, the Trump administration terminated an estimated $500 million in DOJ Office of Justice Programs grants that went to assistance for survivors of crime, juvenile justice and child protection programs, and other justice system improvements in 2025. The budget cuts included many programs that went directly to violence intervention programs employing evidence-based models, such as community violence intervention programs and hospital-based violence intervention programs, that have been shown to reduce violence and save taxpayers money.30
Conclusion
The findings of this analysis confirm what many already suspected: At the time of the administration’s federal law enforcement and National Guard deployments, the primary goal was never to stop crime. Despite multiple tests using multiple data sources, measures, and windows of time, there is no compelling evidence that these deployments reduced homicides, violent crime, or gun violence. Instead, these actions effectuated a dangerous power grab by the Trump administration; increased unrest in the cities that felt their independence being threatened; and cost taxpayers millions of dollars, with millions more likely to be spent because of this political charade. This analysis should give policymakers confidence to reject out of hand the Trump administration’s claims that its actions have materially improved public safety in these cities, call out its falsehoods, and demand better solutions for the American people.
Appendix: Methodology and results
Previous studies and analysis of the effect of the National Guard deployments to American cities on public safety outcomes have been limited by data constraints and single-city study design.31 To more accurately estimate the impact of the National Guard deployments on violent crime outcomes, CAP conducted a three-stage analysis to control for time- and city-variant effects. Using monthly and weekly longitudinal data from the Real-Time Crime Index, Gun Violence Archive, and officially reported city data from January 2023 to February 2026, we investigated the relationship between National Guard deployments and crime outcomes across three models:
- Independent single-group interrupted time-series analysis of monthly violent crime rates.
- Multigroup interrupted time-series analysis of weekly homicide rates.
- Multigroup interrupted time-series analysis of gun victimization rate data with a difference-in-differences extension.
This study is designed to better isolate the causal relationship between the National Guard deployments on the three tested outcomes of interest: violent crime, homicides, and gun victimization. The proposed multistage impact analysis design allows us to overcome data limitations unique to each source used while maximizing the robustness of results and validity by testing across multiple observational windows and outcome variables. Each citywide crime measure analyzed is adjusted by city population and seasonality to account for autocorrelation and improve estimation sensitivity. This methodological appendix provides additional insight into the city and crime data utilized, the model design, and additional results for each model (including formulas, coefficient result tables, and key statistical measures).
CAP’s analysis finds strong evidence that the deployment of the National Guard did not result in a measurable decrease in violent crime rates.
Across all stages of the analysis, only one city in one model produced a statistically significant result associated with a reduction in violent crime (p < 0.05). Model 1 estimates that the National Guard deployment to Memphis, Tennessee, is associated with an initial decrease of 35.4 fewer violent crimes per month per 100,000 residents (p < 0.05), while also estimating that deployment was associated with an ongoing monthly violent crime rate increase of 19.2 incidents for every month following the initial deployment (p < 0.01). Taken on its face, this implies the deployments could have had a crime-exacerbating effect over the entire deployment period. Across all other cities and across all other models, when accounting for city-variant effects, no statistically significant reductions in violent crime, homicides, or gun victimizations were found.
The results for Memphis in Model 1 do not support the Trump administration’s claims for two reasons. First, because no other city experienced a similar decrease, we know that the case of Memphis lacks necessary external validity to claim that the decrease was attributable to the National Guard deployments. Secondly, if Memphis, as the outlier result, is indicative of the impact of the National Guard deployments, the model estimates that the initial level drop in violent crime (35.4 fewer violent crimes per month at p < 0.05) was completely erased within two months of deployment (+19.2 violent crimes per month at p < 0.01), indicating, at best, that this intervention scheme is not a sustainable solution to reduce crime.
Model 1: Independent single-group interrupted time-series regression analysis
Tested variable: Seasonally adjusted monthly violent crime rate
Modeled cities: Los Angeles; Washington, D.C.; and Memphis
Data source: Real-Time Crime Index
In the first model, we use monthly violent crime totals from the RTCI for Los Angeles; Washington, D.C.; and Memphis. New Orleans is not included in the sample for Model 1 due to data availability and the lack of post-intervention periods. New Orleans did not see the National Guard deployed until December 30, 2025, and only had one month of available data for the treated period at the time of this study.
Seasonal crime patterns are a widely understood and generally accepted feature of crime trend analysis.32 To estimate discrete changes in crime trends that are seasonally agnostic therefore requires some data smoothing. However, this introduces a different bias in interrupted time-series analysis, as over-smoothing can interfere with and mute immediate changes around the discontinuity of interest because post-intervention periods will be weighed by pre-intervention data. For this reason, instead of using standard rolling averages, we use the Locally Estimated Scatterplot Smoothing method (R stats package version 3.6.2) to estimate the seasonally adjusted rate of violent crime in a given month while maintaining maximal period-to-period measure sensitivity.33 The entire data sample from the RTCI (January 2018–January 2026) was used to distinguish a stable seasonal pattern from which the seasonally adjusted violent crime rates were estimated. The seasonally adjusted monthly violent crime estimates are also adjusted to account for city population (per 100,000). The base model is then run on all city data between January 2025 to January 2026, with additional sensitivity analysis done by extending the pre-intervention window from January 2024 to January 2026 and January 2023 to January 2026.
We then estimate the effect of the National Guard deployments using the following formula for each city:
\[\begin{aligned} vc\_pop\_adj_{t} = {} & \beta_0 + \beta_1 runner_t + \beta_2 deployed_t \\ & + \beta_3(\textit{time_since_deployment}_{t}) + \alpha_i + \epsilon_t \end{aligned}\]
Such that \(vc\_pop\_adj_{t}\) is the estimated seasonally adjusted violent crime rate in month \(t\); \(runner_t\) is the continuous secular trend variable representing time elapsed in the pre-intervention and post-intervention observational window in month \(t\) (January 2025–January 2026); \(deployed_t\) is the binary variable indicating whether period t occurs in the pre-intervention period (\(deployed_t\) = 0) or post-intervention period (\(deployed_t\) = 1); and \(\textit{time_since_deployment}_{t}\) is the interaction term representing the post-intervention trend. As the deployments do not align exactly with the start or end of months, the intervention binary variable is assigned to the first month in which more days have the National Guard deployed in the city than not. For example, the National Guard was first deployed to Los Angeles on June 7, 2025.34 Therefore, where \(t\) = June 2025, for Los Angeles, \(\textit{time_since_deployment}\) = 1, where \(t\) = July 2025, \(\textit{time_since_deployment}\) = 2, and so forth. Finally, \(\epsilon_t\) is the error term not explained by the model in period \(t\).
The estimated coefficients are thus:
- \(\beta_0\) = model intercept
- \(\beta_1\) = pre-intervention trend
- \(\beta_2\) = estimated level change associated with the intervention
- \(\beta_3\) = estimated trend (or slope) change associated with the intervention
Running this model, CAP finds that Los Angeles and Washington, D.C., return statistically insignificant results (p > 0.05), meaning that we cannot reject the null hypothesis that the National Guard deployments to these cities had no effect on violent crime rate levels or trend. Only Memphis returned a statistically significant result for both the violent crime rate level and the violent crime rate trend after the intervention. Model 1 estimates that the National Guard deployment to Memphis is associated with a level drop of 35.4 fewer violent crimes per month per 100,000 residents (p < 0.01) and an ongoing monthly population-adjusted violent crime rate increase of 19.2 incidents for every month following the initial deployment (p < 0.01).
Upon conducting additional robustness checks by expanding the observational window to include 2023 and 2024 data to the pre-intervention trend, only Memphis returned a statistically significant result across all tests (see Table A1; Model 1.A, 1.B, and 1.C), consistently estimating an associated level drop in the violent crime rate post-intervention. Again, however, it should be emphasized that this initial model is designed specifically to bias the results to finding a statistically significant result and additional stages to this analysis include controls for confounding variables and more accurately estimate the discrete effect of the National Guard deployments on violent crime trends and levels.
Model 2: Multigroup interrupted time series
Tested variable: Seasonally adjusted weekly homicide rate
Modeled cities: Los Angeles; Washington, D.C.; and Memphis
Data source: Publicly available city data
To better account for city-variant trends when estimating the causal effect of the National Guard deployments on violent crime, the second model uses a multigroup interrupted time-series approach. To ensure the model is more sensitive to level and trend changes in the immediate window following the intervention start, we use official weekly crime data reported by the cities in our sample. Given the different reporting systems across the sample cities, we use weekly homicide totals to ensure consistency across cities. Using this method, we can construct a dataframe of weekly reported homicides in Washington, D.C.; Los Angeles; and Memphis from January 1, 2024, to January 1, 2026. Again, New Orleans is not included due to a lack of publicly available data. Homicide weekly totals are adjusted for population (per 100,000) and seasonality using the LOESS method of smoothing. Seasonally adjusted estimates were made using the entire data sample. We then estimate the effect of the National Guard deployments using the following formula and cluster standard errors by city:
\[\begin{aligned} hom\_adj_{it} = {} & \beta_0 + \beta_1 runner_t + \beta_2 deployed_t \\ & + \beta_3(\textit{time_since_deployment}_{it}) + \alpha_i + \epsilon_t \end{aligned}\]
Such that \(hom\_adj_{it}\) is the estimated seasonally adjusted homicide rate in week \(t\) in city \(i\).
The estimated coefficients thus can be interpreted as:
- \(\beta_0\) = model intercept
- \(\beta_1\) = pre-intervention trend
- \(\beta_2\) = estimated level change associated with the intervention across cities
- \(\beta_3\) = estimated trend (or slope) change associated with the intervention across cities
- \(\alpha_i\) = the city-specific level of weekly homicides compared to the model’s intercept
Unlike the previous model, this model tests across the trends of multiple cities pre- and post- deployment, while controlling for city-specific trends and seasonality—making a more robust test and controlling for omitted variables that bias the estimated effect of the National Guard deployments. The model detects no statistically significant crime-reducing effect of the deployments on the level or trend of seasonally adjusted homicides, represented by the coefficients of variables \(deployed\) and \(\textit{time_since_deployment}\). To treat Memphis, we ran the model using both the reported start date of the Memphis Safe Task Force, September 29, 2026, and the date that National Guard began patrolling, October 10, 2026, according to the city’s official website.35 This is replicated in Model 3. Both start dates returned similar results and did not change the direction of the coefficients. Additional robustness checks narrowing the window of pre-intervention trend data to after January 1, 2025, return similar results. (see Table A2; Model 2.A, 2.B, 2.C, and 2.D) Note that while mostly statistically insignificant, this additional test estimates that the average effect of the National Guard deployments may actually be associated with an increase in the homicide trend since deployment, meaning the weekly homicide rate trend increased in cities post-deployment compared with the estimated trend of these cities had the National Guard never been deployed.
Model 3: ITS with difference-in-differences extension
Tested variable: Seasonally adjusted weekly gun victimization rate
Treated cities: Los Angeles; Washington, D.C.; Memphis; and New Orleans
Control cities: New York, Oakland, St. Louis, San Francisco, and Baltimore
Data source: Gun Violence Archive
The final model adds one more layer of complexity, using an approximate control group to compare violent crime trends of the cities that saw guardsmen deployed against those that did not. In addition to the cities in which the Trump administration deployed the National Guard, Trump made threats to send National Guard troops to five other cities: New York, Oakland, St. Louis, San Francisco, and Baltimore. This group of five cities can approximate a control group to compare their trends against those to which the National Guard was deployed. While there are obvious limitations to using these five cities as a pure counterfactual, adding in these cities as a control to the model still improves the validity of our findings that National Guard deployments did not improve violent crime trends.
Due to data access limitations, in this final model, we use seasonally adjusted weekly gun victimization rates from the Gun Violence Archive to create the full sample of cities: Washington, D.C.; Los Angeles; Memphis; and New Orleans as the treated groups and New York, Oakland, St. Louis, San Francisco, and Baltimore as the control groups. To approximate a counterfactual, “deployment” start dates were assigned to the control cities based on Trump’s initial threat to send the National Guard to that city. In using the date of threat as a proxy for the discontinuity, we estimate the “post-intervention” trend period of the control group cities as if the National Guard were deployed there like in the treated cities. If deployments had a significant effect on gun victimization compared with these placebos, we would expect statistically different estimated effects associated with the “treated * deployed” interaction term and the “treated * time_since_deployment” interaction term. Weekly gun victimization totals are adjusted for population and seasonality using the LOESS method of smoothing. Seasonally adjusted estimates were made using a sample range from January 1, 2022, to January 1, 2026. We then reduced the sample to January 1, 2025, to January 1, 2026, to estimate the effect of the National Guard deployments using the following formula and clustered standard errors by city:
\[\begin{aligned} gun\_vics\_adj_{it} = {} & \beta_0 + \beta_1 runner_t + \beta_2 deployed_{it} \\ & + \beta_3(\textit{time_since_deployment}_{it}) \\ & + \beta_4 treated_i + \beta_5 (treated_i \textit{* }runner_{it}) \\ & + \beta_6 (treated \textit{* }deployed_{it}) \\ & + \beta_7 (treated \textit{* time_since_deployment}_{it} \\ & + \alpha_i + \epsilon_t \end{aligned}\]
Such that \(treated_i\) defines for city \(i\) if it is assigned to the treated or control group, and the introduced coefficients can be interpreted as:
- \(\beta_4\) = the secular difference in level between the treated and control groups
- \(\beta_5\) = the estimated difference in the gun victimization trend pre-intervention
- \(\beta_6\) = the estimated difference in level change associated post-intervention between the control and treatment groups
- \(\beta_7\) = the estimated difference in trend change between the control and treatment group post-intervention
Again, after controlling for city-specific variation and comparing the treated cities against a control, the model detects no statistically significant crime-reducing effect of National Guard deployments on the level or trend of gun victimization rates. Additional sensitivity analysis is done by expanding the pre-intervention observation window to include data from January 1, 2024, to January 1, 2026, and running the model again with Chicago’s Operation Midway Blitz included as a treatment city and using both activation dates for Memphis. Similarly, no statistically significant crime-reducing effects are found in the additional testing.
Limitations
The National Guard deployments to the treated cities were not the only federal interventions occurring in the post-intervention period. Most of the National Guard deployments occurred in conjunction with other surges in federal law enforcement and even additional U.S. military personnel.36 While it confounds the results of these models to not just be estimating the causal impact of the National Guard deployments, importantly, if anything, the model would assign the value of the impact of other interventions to National Guard deployments. If the federal law enforcement surges had a measurable effect on crime, that effect should have been detected by these models. This is at worst an upper-bound estimate or overestimate of the effect of the National Guard on the measured crime outcomes. Seeing no effect gives even more confidence that the guard’s deployment had no measurable effect on crime and some degree of confidence that none of the federal law enforcement surges influenced violent crime levels or rate.
As discussed above, this study faces various data limitations, and the full sample of treated cities was only able to be tested using data from the Gun Violence Archive. Lastly, this study does not account for nor test if there was a measurable magnitudinal effect from the size of deployments. The size of deployments has fluctuated within the initial deployment and the sampled post-intervention window, in addition to facing legal challenges, causing disruptions to intervention fidelity.