What methods do US authorities use to estimate the number of undocumented immigrants?
Executive summary
United States estimates of the undocumented (or unauthorized) immigrant population rest on a mix of survey “residual” calculations, administrative flow accounting, microdata bottom‑up reconstructions and alternative mathematical models — each with explicit adjustments for census undercounting and departures and each subject to methodological debate and political pressure [1] [2] [3] [4].
1. The residual method: survey totals minus legally present immigrants
The dominant, long‑standing approach — used by DHS, Pew and many academic groups — starts with Census Bureau or American Community Survey (ACS) counts of the foreign‑born and subtracts estimates of legally resident immigrants (green card holders, refugees, temporary visa holders), leaving a “residual” interpreted as unauthorized residents; Pew and others rely on that basic relationship while adding refinements and updated Census population controls [5] [1] [2].
2. Imputation and profile methods: who in the surveys is likely unauthorized
Institutes such as the Migration Policy Institute (MPI) and the Center for Migration Studies (CMS) take Census/ACS microdata and impute legal status for noncitizens using characteristics (year of arrival, occupation, household structure) and weight those imputations to national benchmarks, producing state and county profiles and adjusted totals [6] [7] [8].
3. Administrative‑flows and bottom‑up counting: entries and exits
An alternative is flow accounting that pieces together entries (border apprehensions, paroles, overstay models) and exits (removals, voluntary departures, mortality) using agency microdata; researchers and the Dallas Fed use notice‑to‑appear records, CBP parole data and other administrative sources to produce monthly net‑immigration reconstructions rather than a single residual snapshot [3] [9].
4. Survey alternatives and CPS efforts
Some analysts use the Current Population Survey (CPS) instead of the ACS to replicate DHS methods or to explore sensitivity; organizations such as CIS have applied DHS’s framework to CPS microdata as a test of robustness, acknowledging CPS’s own coverage and sample‑size limits [10].
5. Mathematical models and ‘hidden‑population’ approaches
A smaller literature applies mathematical modeling used for hidden populations (capture–recapture, parameterized demographic models) that can yield substantially different totals — for example, a Yale study using diverse demographic and operations data produced estimates far above the consensus, illustrating how model choice and parameter assumptions drive wide variance [4].
6. Adjusting for undercount, benchmarking and contentious assumptions
All methods must account for undercount in household surveys; a commonly debated correction (often traced to small, localized studies) assumes a nonresponse adjustment for undocumented people — an inherently uncertain parameter that materially alters totals — and analysts explicitly benchmark or reweight estimates to other administrative totals when new Census controls arrive [11] [1].
7. Why estimates diverge: data, timing and policy signals
Differences across Pew, MPI, DHS, CBO and advocacy analysts stem from data choice (ACS vs CPS vs DHS microdata), whether estimates are cross‑section snapshots or flow reconstructions, the treatment of paroles and asylum processing, and the timing of new population controls — changes to Census migration measures in 2024–25, for instance, forced many groups to revise upward their 2021–23 immigration figures [1] [12] [3].
8. Political stakes and methodological agendas
Numbers are not neutral: advocacy and policy groups adopt or promote methods that align with priorities — FAIR’s much higher estimates frame an enforcement narrative and criticize Census lag, while academic centers emphasize methodological transparency and sensitivity testing; users should read source methods and explicit assumptions because methodological choices map to policy arguments [13] [8].
9. What cannot be settled from available reporting
The sources document the main techniques and their disagreements but cannot resolve the true count because no universal registry exists and each approach contains unverifiable assumptions (undercount rates, outmigration, parole residence). Researchers reduce uncertainty with microdata, benchmarks and transparent sensitivity checks; no single source in the reporting definitively proves one method as “correct” over the others [2] [3] [1].