What independent datasets (Deportation Data Project, TRAC) provide yearly ICE arrest and removal time series and how have researchers reconciled them?
Executive summary
Two independent, public projects provide the most widely cited yearly time series for ICE arrests and removals: the Deportation Data Project, which republishes and processes individual-level ICE, CBP and EOIR releases (including arrests, detainers, detentions and—when reliable—removals) [1] [2] [3], and TRAC at Syracuse University, which has compiled FOIA-obtained ICE removals records into a continuous removals time series dating back to FY2003 and maintains monthly-updated tools and QuickFacts for arrests and detention counts [4] [5] [6].
1. The two independent record-keepers: what they publish and why it matters
The Deportation Data Project republishes original ICE datasets and provides processed, linkable tables that allow researchers to trace enforcement paths from arrest through detention to deportation using anonymous A-number identifiers when available, and it hosts documentation and code for reproducible analysis [1] [2] [3]. TRAC’s long-running project assembled ICE’s removals records through FOIA litigation and now supplies an indexed, case-level removals series from FY2003 forward plus compiled monthly and semi-monthly arrest and detention aggregates used routinely in media and scholarship [4] [5] [6].
2. What the series actually measure and where they diverge
The Deportation Data Project focuses on raw, individual-level ICE and CBP tables (arrests/book-ins, detainers, detentions, encounters and removals) and flags dataset limitations such as missing or unreliable removals tables in some ICE releases [2] [1]. TRAC produces curated time series and summary statistics—arrest monthly totals and cumulative fiscal-year removals—that are suited to year‑over‑year comparisons, but those aggregates depend on ICE’s semi‑monthly reporting cadence and on TRAC’s FOIA-updated snapshots [6] [7] [4].
3. How researchers reconcile the datasets in practice
Researchers reconcile the sources by treating them as complementary: using Deportation Data Project files for event-level tracing (linking arrest → detention → flight/removal by anonymous A-number where present) and using TRAC’s curated removals series for annualized counts and historical continuity [3] [4]. Where ICE releases contain both individual records and agency summaries, analysts merge tables via the unique identifiers and cross-check totals against ICE’s semi‑monthly statistics hosted on ICE’s site, thereby triangulating counts [2] [8].
4. Sources of disagreement and the fixes researchers apply
Discrepancies arise from duplicated rows, missing removals tables, cumulative-versus-period reporting, and ICE’s intermittent corrections; the Deportation Data Project documents and removes apparent duplicate arrests, warns when removals tables are unreliable, and provides processed datasets and code so others can reproduce cleaning steps [9] [1] [2]. TRAC has documented and adjusted for ICE reporting oddities—pointing out cumulative fiscal‑year reporting and correcting for ICE’s late revisions—and has highlighted GAO criticisms of ICE data practices that affect comparability [7] [6].
5. What reconciliation produces: consensus findings and contested interpretations
When reconciled carefully, the combined sources allow robust year-over-year arrest trends, geographic concentration analyses (for example, state-level arrest rates), and long-term removals counts, and have undercut simple narratives of “mass deportation peaks” without nuance; TRAC’s analyses show removals and arrests can fall or rise depending on measurement window and ICE reporting cadence, while Deportation Data Project users have shown high daily arrest levels concentrated in cooperating jurisdictions after cleaning duplicates [6] [9] [10]. Alternative readings persist because ICE’s semi‑monthly summaries, the Deportation Data Project’s raw tables, and TRAC’s aggregates each encode different assumptions about deduplication, time‑windowing, and which events count as “removals” [7] [1].
6. Limitations, open questions and best practices for future work
All reconciliation work depends on ICE’s raw releases and on researchers’ transparency about cleaning: analysts should publish code and decision rules, prefer person‑level A-number merges when present, flag removed or missing removals tables, and compare processed counts against ICE’s own semi‑monthly metrics and GAO critiques to assess stability; where source material is silent about a specific claim, reporting must acknowledge that limitation rather than assert falsity [2] [8] [6].