What methodologies do researchers use to compare crime rates between undocumented immigrants and U.S.-born residents?
Executive summary
Researchers compare crime rates between undocumented immigrants and U.S.-born residents by combining administrative crime records (arrests, convictions, incarcerations) with population estimates of undocumented people, using multiple cross-checks and robustness tests to address measurement gaps; the consensus from several peer‑reviewed and national studies is that immigrants—including undocumented people—tend to have lower measured rates of arrests and incarceration than the U.S.-born, but substantive data limitations remain [1] [2] [3].
1. "Count the arrests, then estimate who’s in the pool": reliance on administrative arrest and incarceration records
At the state and national level, researchers commonly start with police and justice system administrative data—arrest records, prosecution and conviction files, and incarceration rolls—to measure criminal events, for example the Texas study that used Texas Department of Public Safety arrest logs that record immigration status at booking to compare felony arrest rates across groups [1] [4]; broader national work uses incarceration counts and conviction data to build comparable rates over time [5] [2].
2. "Who’s the denominator?": residual methods to estimate the undocumented population
Because the undocumented population is not directly counted, most studies estimate it with a residual methodology—subtracting the number of authorized immigrants from total foreign‑born counts in Census/ACS data—using sources such as the Center for Migration Studies (CMS) or Pew estimates; these modeled population denominators are central because an inflated or deflated denominator will materially change per‑capita crime rates [6] [7] [5].
3. "Triangulation and robustness checks": testing alternative denominators and classifications
Top papers explicitly test robustness to alternative undocumented‑population estimates (CMS vs. Pew), to varying definitions of “undocumented” at arrest, and to substituting arrests with convictions or misdemeanors, reporting that core findings often hold across these specifications—an approach featured in the Texas study and archived data/code that replicate main results under different assumptions [8] [9] [4].
4. "Beyond arrests — using incarcerations, convictions and historical series to reduce bias"
Researchers also analyze incarceration gaps and long‑run trends: multi‑century incarceration analyses and national studies look at conviction and imprisonment to limit distortions from policing practices, finding immigrants historically have lower incarceration rates than the U.S.-born, a pattern that complements arrest‑based findings [2] [5] [3].
5. "Addressing bias and missing data": limitations, underreporting and alternate triangulation methods
Methodological caveats persist: undocumented people may be less likely to report victimization or interact with police, arrest classification can be inconsistent, and residual population estimates have critics—so researchers supplement with triangulation using death and birth records, linear interpolation for missing years, and sensitivity tests to assess whether estimation error could overturn results [7] [1] [8]; papers acknowledge that data constraints prevent measuring "who actually committed every crime" and therefore stop short of definitive causal claims [5].
6. "Competing analyses and interpretive choices": why different studies reach different headlines
Some non‑peer‑reviewed reports (for example, analyses from Cato or the Crime Prevention Research Center) have yielded conflicting estimates on conviction/arrears depending on data sources and methods, and researchers point to differences in peer review, measurement choices, and robustness testing as reasons for divergent conclusions—highlighting the need to weigh methodology and transparency when evaluating claims [6] [1] [10].
7. "Best practices moving forward": transparency, multiple measures, and cautious interpretation
The methodological consensus is pragmatic: use administrative crime records when possible, estimate undocumented denominators with established residual methods (and alternate sources for sensitivity), report results for arrests and convictions separately, perform triangulation with other vital records, and be explicit about limits in attributing causation—approaches used in the Texas study and by national researchers form a template for responsible inference [4] [7] [5].