What methodologies do academic studies (e.g., UCLA) use to standardize and compare ICE arrest rates across states?
Executive summary
Academic teams studying ICE arrests standardize comparisons across states by using shared administrative arrest records combined with population denominators (noncitizen or undocumented estimates), applying normalization and smoothing techniques, and running multivariate models that control for confounders; UCLA’s work exemplifies this approach while also flagging data linkage and geographic-assignment limitations in ICE releases [1] [2] [3]. These methodological choices are designed to avoid misleading, scale-driven comparisons and to isolate policy- and implementation-driven variation in enforcement [1] [2].
1. Data sources: administrative ICE records plus public demographic estimates
Researchers begin with ICE’s Enforcement and Removal Operations administrative arrest files as the core numerator for comparisons, often using FOIA-archived releases reposted by projects like the Deportation Data Project to ensure provenance [4] [5], and then augment those counts with public demographic estimates—most commonly the American Community Survey’s multi-year noncitizen counts and alternative undocumented-population estimates from DHS or Pew—to build meaningful denominators [1] [2].
2. Choosing denominators: noncitizen vs. undocumented benchmarks to normalize rates
To avoid scale-driven distortions where populous states simply have more arrests, UCLA and related studies normalize arrest counts by the “population at risk,” comparing multiple benchmarks: noncitizen totals from the ACS and competing undocumented-population estimates from DHS and Pew to test robustness; presenting rates per 100,000 noncitizens (or per estimated undocumented population) is a standard practice in their California–Texas comparison and national visuals [2] [1].
3. Temporal standardization and smoothing: rolling averages and period breaks
Studies smooth volatile daily arrest counts with rolling averages (Prison Policy and other trackers use 14‑day rolling averages) and sometimes split observation windows into discrete policy periods to compare apples to apples over time—e.g., breaking a year into two policy phases—so trends reflect operational changes rather than day‑to‑day noise [3] [6].
4. Geographic assignment and area-of-responsibility complications
Assigning arrests to states isn’t straightforward: researchers rely on ICE office-area information and reported apprehension locations, but gaps force imputations or AOR aggregation (e.g., Washington, D.C. sometimes combined with Virginia); UCLA visualizations explicitly include D.C. and Puerto Rico and note AOR-driven distortions in local comparisons such as San Diego vs. Bay Area [1] [7] [3].
5. Multivariate controls: separating enforcement from crime and demographics
To assess whether variation reflects crime or policy, studies run multivariate regressions controlling for variables such as overall crime rates, estimated immigrant crime rates, and demographic factors; UCLA reports that arrest rates showed no statistically significant association with overall crime or estimated immigrant crime in their models, suggesting implementation differences drive much of the cross‑state variation [1].
6. Data linkage, cleaning, and algorithmic reconstruction
Because ICE datasets can omit identifiers, duplicate records, or split detention stints, researchers and third‑party groups (e.g., Vera Institute) build algorithms to link records, de‑duplicate transfers, and reconstruct individual pathways; these technical fixes are necessary to compute accurate counts and longitudinal measures but introduce analytic decisions that affect results [8] [4].
7. Limitations, transparency gaps, and competing interpretations
Methodological transparency matters because ICE’s own public reporting is sometimes vague about what’s counted, prompting warnings that odd arithmetic or missing metadata can produce contestable narratives; critics and advocates therefore reach different conclusions about the meaning of arrest-rate spikes—academic approaches mitigate this by documenting assumptions and testing alternative denominators, but residual uncertainty about linkage and counting remains [6] [8] [9].
8. Why methodology matters: from numbers to policy stories
Choosing denominators, smoothing windows, geographic assignment rules, linkage algorithms, and control variables aren’t neutral technicalities; they determine whether headlines emphasize a national surge, regional variation, or policy effects—UCLA’s multi‑benchmark normalization and multivariate testing aim to move beyond raw counts so that comparisons across states attribute variation to enforcement choices rather than population size or data artifacts [1] [2].