A First Look at the Latest ICE Arrest Data Published by the Deportation Data Project
ICE arrest data just arrived with incredibly useful new fields, updated trends in who ICE is targeting and where, deeper understanding of the enforcement system, and some quality issues.
The Deportation Data Project released new ICE data today as a result of their FOIA lawsuit, providing a more up-to-date analysis of the Trump administration’s expansion of immigration enforcement efforts. I spent the last few hours validating and analyzing the data so that you can see what’s new, what’s missing, and what the data says about ICE enforcement across the country. Let’s dive in.
Data Validation
I have gone into much further detail in the past about the essential validation steps to ensure that the data you are working with is reliable and consistent with past data. I won’t go into great depth here, but just note that you should always do your own validation to make sure you know what you’re working with. The current data passes a few validation steps, which also results in some observations and additional documentation.
Sum check. The total sum check of overlapping months between this data and the DDP’s previous data release is well within a margin of error, as are several basic fields, which means that we are working with apples-to-apples data.
Coverage. Normally, when you have overlapping datasets, you have two options: append the current data to the previous data release or backfill the current data as far back as you can go. In this case, the new data goes back further than the beginning of the previous data. The previous data started at September 2023 (a month before the start of FY 2024), but the new data goes back to the start of FY 2023 (in October 2022). I would recommend treating this as a 100% replacement for the old data so that you’re working with the most current numbers and then some.
Final date of data. Based on the partial data for March 10 and 11, March 9 is probably the final date of complete data. So if you’re doing day-by-day analysis, cut off those final two dates. If you’re looking at the total data set, there’s no harm in leaving those in since they are unlikely to represent false arrests—it’s just that the total arrests for those dates weren’t included. Since I like to include all of the latest data, Just use the new data in its entirety.
Fields. There are some changes in the fields that are available. Mostly we get more data, but not always. Here’s my breakdown of the fields.
Minor renamings: Two fields were renamed between releases but appear to represent the same underlying data, with no analytical impact expected.
The field previously called Apprehension State is now simply called State, which is likely a cosmetic change only.
The field previously called Apprehension County is now called County, also likely cosmetic — though it is worth noting that the County field is 100% null across all four fiscal years in the current dataset regardless of what it is labeled.
The field formerly called Unique Identifier has been renamed Anonymized Identifier, with the underlying IDs apparently re-coded as part of the anonymization process. I have not checked to confirm if these can be integrated across datasets.
Dropped fields. Four fields from the previous release are absent from the current dataset.
The Final Program Group, which was a higher-level grouping of the Final Program categories, has been dropped entirely. The impact is likely low since the more granular Apprehension Final Program field remains available, and if the grouping is needed it can be reconstructed by recoding those values.
The three individual case identifiers (Alien File Number, EID Case ID, and EID Subject ID) are also absent, though this has no analytical impact since all three were redacted in the previous release and were therefore non-usable regardless.
New fields. The following new fields have been added to this data release and represent the opportunity for new knowledge and analysis.
Arresting Agency is 100% ICE across all records and has no analytical value in its current form.
Apprehension Type distinguishes Targeted from Collateral arrests but is effectively 100% null through July 2025, so analysis using it should be restricted to August 2025 onward. I’m not entirely sure about the reliability of this field, to be honest. I would prefer to have much more historical data to make sure that documentation is consistent across administrations rather than subject to political influence.
The Operation field identifies named enforcement operations but is 57–83% null depending on the year; the non-null values may be important for understanding discrete enforcement surges and warrant further investigation.
The dataset now includes both Apprehension Criminality and Case Criminality using the same three-value scheme. I address this below.
TOA Current Duty Site is fully populated and may offer sub-AOR geographic detail worth exploring.
Case Threat Level is coded 1/2/3 but is 48–73% null depending on the fiscal year, and its meaning and incomplete coverage require further investigation.
Renaming & addition. The dataset includes two geographic fields that together form a hierarchy: TOA Current Duty AOR (Area of Responsibility) and TOA Current Duty Site.
The AOR field identifies one of 26 regional jurisdictions while the Site field identifies the specific office or facility within that region. There are 632 across the dataset. These are nested: nearly 100% of the Site fields fall cleanly within the corresponding AOR.
No change. The following fields are common across both the old and new datasets with no change: Apprehension Date, Apprehension Method, Apprehension Criminality, Case Status, Case Category, Departed Date, Departure Country, Final Order Yes No, Final Order Date, Birth Date, Birth Year, Citizenship Country, Gender, and Apprehension Site Landmark.
Significant errors. Deportation Data Project notes significant data issues with the data on encounters and deportations. The deportations data has been particularly rife with errors throughout the entirety of DDP’s lawsuit, so that’s disappointing but not a complete surprise. What is frustratingly common is that even in FOIA lawsuits under the order of a federal judge, ICE will still turn over bad data without doing a quality check first. No matter how many times the public sues and no matter how many times ICE is ordered to do this, the agency still trips over itself, acting like it’s learning to walk for the first time every single time. These issues do not seem to affect the arrest data directly, but if you’re trying to stitch together multiple datasets, be sure to read DDP’s notes and proceed with caution.
Immigrants with No Criminal History Continue to Make up Largest Group of ICE Arrests
ICE data continues to show spikes in arrests driven by immigrants with no criminal history. Since the last update to detailed data in the middle of October, ICE’s total arrests of what it classifies as “other immigration violators” (OIV, people with suspected immigration violations but no other criminal history) shot up sharply while the corresponding number of people with charges and convictions actually slightly declined. By February 2026, the last full month of data, arrests of OIV declined sharply to pre October levels, as did the total number of people with criminal convictions. Only arrests of people with pending charges increased and only slightly.
The key takeaway here is pretty straightforward: despite the public safety justifications of mass detention and deportation, including the unprecedented planned use of warehouses to increase detention to over 100,000, the number of people ICE is arresting on a monthly basis has not increased substantially since late Spring 2025 and has, in fact, objectively declined since October.
What’s interesting when you dig into the arrest data is just how obvious it becomes that ICE is strategically and intentionally targeting certain regions over time. We know this from reporting and from public statements from the administration about targeting urban areas run by Democrats, but the data provides an additional view on this that you don’t get just from the rhetoric.
I’m using the state field for this analysis and not trying to geolocate the arrests any more precisely than that for this initial run. Across the entire new dataset, 15% of the values in the state column are null, so we have to ignore those. Look at Minnesota and Illinois below. We see massive spikes in the new data that correspond to known ICE surges in urban areas in those states. Much of the increase in arrests in those states were for people with no criminal history. The spike in Minnesota in January corresponds to Operation Metro Surge, and the spike in October in Illinois corresponds to known increase in enforcement in Chicago. We saw this in the previous data, but since that data only went through half of October, we didn’t get a full picture.
What’s possibly more interesting is West Virginia. I haven’t seen a lot of reporting about West Virginia (might be my own media bias), but the state has seen a massive spike in ICE arrests that really deserves more attention. ICE increased arrests in that state dramatically between October and February. If you have more information about this or are from an area hit by ICE activity, please share more information in the comments.
A First Look at ICE’s Field Office Geographies
As I noted above, the new data includes a field that allows us to associate arrest with a specific ICE field office rather than just the general AOR. ICE’s Areas of Responsibility has been a challenge for researchers because, although we know what that geography is, it's so general and often crosses state boundaries that, except for specific cases where it falls entirely within a state, the geography is not always particularly useful. But with the new field office data, we can drill in from the state level in places like Minnesota to the ICE field office listed in St. Paul to capture the more precise number of arrests associated with this field office. This represents a huge leap forward in reliably mapping the geography of ICE arrests and, given the nearly complete coverage, should be much better than attempts to impute location from Apprehension Site Landmark.
The current challenge is that the new ICE Field Office data did not come with addresses. However, we are in luck! Just days ago on March 12, my colleague Jacqueline Stevens at Northwestern University released a detailed spreadsheet of ICE offices across the country complete with address information. You can find that information here. Using that spreadsheet with the new DDP data and a little help from Claude, we can use fuzzy matching to combine the two to create a more complete list of field offices and address locations that are easier to map. I am sharing the preliminary results here, but this should be rigorously validated before using for peer-reviewed publication. I will try to do that over the next week and report back any major concerns.
The map below shows the total number of ICE arrest in 2026 so far (January 1—March 9) by ICE field office, with the top 20 locations in total arrests shown in orange and the rest in blue. Note that 75,333 of 78,500 (more or less) total 2026 arrests are mapped for 96% coverage. A total of 268 of the 408 sites relevant for ICE arrests in 2026 were mapped, which seems low, but the ICE sites that were unable to be associated with an address only account for a very small number of overall arrests. I excluded a small number of arrests in outlying islands and territories (224 in 2026 so far) to make the visualization easier.
For full transparency, here’s the crosswalk table that I used for this map. It’s not perfect, but based on a spot check of the field offices with the largest numbers of arrests and field offices that I am personally aware of, it seems solid enough to count as preliminary research. Feel free to improve upon this in any way or to critique it if you find that it’s too inaccurate even for preliminary mapping. I’m open to feedback.
A First Look at How Criminal History Changes Over Time
Because the latest ICE data includes two fields for criminal history—one at time of arrest and one that is “current” knowledge for ICE at the time of data export—we can finally evaluate the quality of the information that ICE has at arrest. One argument about the arrest criminality data is that ICE might not know something when they arrest a person that they find out later. Valid research question. According to this hypothesis, the longer the duration between the time of arrest and now, the greater the likelihood that something in the criminal history has changed. For example, a person with a charge at time of arrest might pick up a conviction, or a pending charge against a person at time of arrest will get dropped—and so on.
The data shows that this hypothesis is true, but it’s not a huge difference. In general, when changes occur, they tend to get worse for non-citizens. People with charges tend to get convictions, and people with no criminal history at time of arrest can pick up a charge or conviction. But note that it can go the other way: some people who have a charge do have their charges dropped, possibly because (as many cops have told me) why prosecute a charge when the person will get deported, or because the underlying charge isn’t legitimate. Overall, this fact could be statistically significant for certain academic studies, but I don’t think this is substantial for most reporting. For example, I wouldn’t say that the result of this finding is that we can’t trust the “criminal history at time of arrest” data, but I would just note that the final data by criminal history definitely does shift between arrest and some later point.
The Work of Transparency Continues
As usual, I’m grateful to the Deportation Data Project for their ongoing work. I hope this initial analysis is useful to reporters, researchers, analysts, and the broader public. The tireless work of transparency is crucial under any administration, Republican or Democrat, but it’s especially under this one since the administration continues to make unverified claims, withhold records, hide essential reports, and reduce transparency in an effort to control the public narrative. We are not powerless. We can use data to increase awareness and understanding, and lay the foundations for a future where Congress requires these agencies to produce this data on a monthly basis. The public shouldn't have to litigate at great expense for records the agency ends up releasing anyway.
If you who haven’t taken the leap to become a paid subscriber to consider supporting this work with a few of your hard-earned dollars. I’m honored that this Substack continues to drive media reporting, research, and public debate—but it does require effort and expense. If you can pitch in a few bucks a month or year, it would help enormously. Find an option that fits your budget at austinkocher.substack.com/subscribe.
A note about Final Program: in the detentions table, this release has unfortunately dropped the Final Program field, replacing it with Apprehension Final Program. There was sometimes a difference between Final Program in the arrests and detentions tables in previous releases, particularly in certain jurisdictions (e.g., San Francisco, Spokane). The now-gone detentions Final Program sometimes helped distinguish between different types of arrests (immigration court, check-in, community):
https://missionlocal.org/2026/03/sf-ice-data-arrests-breakdown-check-ins/
FYI, the Syracuse NY sub-office is at 100 Northern Concourse, Syracuse, NY 13212
Thank you for all you do!!