How do researchers estimate crime rates for undocumented immigrants when administrative data are incomplete?
Executive summary
Researchers combine imperfect arrest or conviction records with independent population estimates and statistical adjustments to estimate undocumented immigrant crime rates; where immigration status is missing from standard crime datasets they rely on special administrative sources (like Texas DPS), demographic modeling from groups such as the Center for Migration Studies, and sensitivity checks to bound uncertainty [1] [2] [3]. Those methods produce consistent findings in multiple studies that undocumented immigrants are not more criminal than the U.S.-born, but every estimate carries caveats because mainstream crime data systems usually do not record immigration status [4] [1].
1. The core problem: numerator and denominator both under-specified
Estimating a crime rate requires a numerator (offenses, arrests, or convictions) and a denominator (the number of people in the group); for undocumented immigrants both parts are imperfect because most widely used crime systems — the FBI’s UCR, the NCVS, and NIBRS — do not record immigration status, and population tallies of the undocumented are produced by specialist research centers rather than by routine census counts [1] [2] [3].
2. When administrative arrest data include status: the Texas model
The clearest approach uses jurisdictions that capture immigration status at arrest; Texas Department of Public Safety records enabled researchers to separate arrests by undocumented, legal immigrant, and native-born groups and to compute arrest-rate proxies across 2012–2018, a dataset cited in multiple reports and used for peer-reviewed publication and NIJ summaries [1] [4] [5]. Such “natural experiments” are powerful because they avoid imputing status, but they are limited to the places that collect the information and still rely on arrests as a proxy for offending [1] [4].
3. Constructing denominators: CMS, Pew and undercount adjustments
Because routine population counts do not enumerate undocumented residents directly, researchers use independent, peer-reviewed estimates from organizations like the Center for Migration Studies or Pew, which generate state and national undocumented-population estimates by origin country and adjust for underenumeration and length of residence; those controls matter because undercount rates are higher for recent entrants and vary by national origin [2] [3].
4. Imputation, survey linkage, and residual methods when status is missing
Where administrative systems lack status fields, analysts combine multiple data sources: they link nativity and legal-status indicators from surveys, apply statistical imputation to assign undocumented status probabilistically, or use residual estimation techniques that subtract estimated lawful noncitizen counts from total foreign-born totals to infer undocumented numbers; these approaches require assumptions about underreporting and nonresponse that researchers test through robustness checks [3] [2].
5. How researchers handle measurement bias and sensitivity
Because arrests can reflect policing practices rather than underlying offending, studies test alternative measures (arrests vs convictions vs misdemeanors), substitute different undocumented-population estimates, and run models across jurisdictions and years to check stability; the Texas-focused work, for example, reports findings robust to alternate population controls and classification rules, and notes that homicide rates and rare-event statistics are especially noisy [4] [1] [6].
6. What the evidence says — and what cannot be concluded from it
Multiple analyses using the methods above consistently find undocumented immigrants have equal or lower arrest and offending-rate proxies than U.S.-born residents, and macro studies find no evidence that undocumented immigration increases violent crime — conclusions advanced by academic papers, policy centers, and advocacy groups such as Migration Policy Institute, Brennan Center, and the American Immigration Council [1] [3] [7] [8]. However, these conclusions rest on the assumption that population estimates and arrest-to-offense relationships are accurate; mainstream data gaps mean researchers must be candid about possible nonresponse bias, geographic heterogeneity, and limits to generalizing single-state findings nationally [3] [1] [9].
7. Reading the methods as evidence: strengths, limits, and policy implications
The methodological takeaway is not certainty but convergent evidence: jurisdictions that record status provide direct numerators, independent demographic groups provide defensible denominators, and careful sensitivity analyses narrow plausible ranges — together these practices form the best current approach for an otherwise data-poor question, but they also reveal why sweeping national claims should be treated with caution unless backed by consistent multi-jurisdictional data [4] [2] [3].