What methodologies do researchers use to estimate the unauthorized immigrant population, and how do their results differ?
Executive summary
Three families of methods dominate efforts to count the U.S. unauthorized immigrant population: the residual estimation approach that subtracts estimated lawful immigrants from total foreign‑born counts (used by Pew, DHS and others), survey‑imputation methods that assign legal status to individual survey respondents (used by MPI), and administrative‑linkage or “nonmatch” and bottom‑up flow methods that use Social Security, court, and microdata to infer unauthorized status; these approaches often produce similar mid‑range estimates but diverge sharply at the margins because of different assumptions about undercount, exits, and who is counted as “unauthorized” [1] [2] [3] [4].
1. The residual estimation workhorse: subtracting lawful from total foreign‑born
The residual method constructs a demographic tally of lawful immigrants from admission records and applies it against survey or census measures of the total foreign‑born population; the difference—after adjustments for survey undercount and emigration—is treated as the unauthorized population, a method underpinning Pew’s widely cited estimates (including its 2023–24 high of roughly 14 million) and the DHS approach [1] [5] [6].
2. Survey imputation: assigning legal status inside the ACS and SIPP
Migration Policy Institute and related analysts impute legal status for noncitizen respondents in the American Community Survey and SIPP by combining questionnaire responses, administrative patterns, and external weights to produce local and national profiles; MPI’s approach yields rich characteristic data and produces estimates close to—but not identical with—residual totals after researchers apply upward adjustments for ACS undercount [2] [7].
3. Administrative‑linkage and the CPS‑SSA nonmatch alternative
A distinct line of work links household survey reports to Social Security Administration records (the CPS‑SSA nonmatch method) to count people without valid SSNs or using others’ SSNs; proponents argue this avoids some residual method assumptions and produces broadly similar overall totals while offering a continuous time series for monitoring trends [3].
4. Bottom‑up flow models and microdata: entries, exits and monthly dynamics
Researchers such as the Dallas Fed, CBO and other teams reconstruct flows by tracking microdata on admissions, Notices to Appear, detentions, and birth/death records to estimate monthly entries and exits and thus net unauthorized migration; this “bottom‑up” approach can show rapid reversals in net flows that cross‑sectional residual snapshots miss, and it can yield different trends even when aggregate levels align [4].
5. Alternative estimators and why totals can differ by millions
Some analyses that tweak assumptions about survey coverage, mortality, and emigration—using death registration, birth registration, or net‑migration matching—produce much larger estimates (one review noted an alternative over 22 million), exposing how sensitive results are to undercount assumptions and methodological choices; other studies report wide uncertainty bounds around point estimates to reflect this sensitivity [8] [9].
6. Key sources of divergence: undercount, classification, and temporal scope
Differences among methods trace to three practical decisions: how heavily researchers adjust for survey undercount of hard‑to‑reach migrants, how they classify those in limbo (asylum seekers, people with pending cases, long‑term overstays), and whether the method measures a stock at a point in time or reconstructs flows—choices that affect whether estimates cluster in the low‑teens of millions or extend substantially higher [1] [9] [4].
7. What convergence and disagreement mean for policy and reporting
While leading methods often produce similar mid‑range magnitudes—explaining why Pew, MPI, CMS and DHS estimates are frequently comparable—differences at the margins matter politically and operationally: advocates, policymakers, and courts may favor estimates that emphasize growth, decline, or uncertainty; analysts and commentators should therefore cite method, year, and key assumptions when quoting numbers, a persistent theme in expert discussions documented in podcast and technical forums [10] [8].
8. Limits of available reporting
The sources reviewed document methods, recent point estimates and sensitivities, but they do not resolve every dispute: several methods report only national aggregates or series through 2024–25 and differ in how they treat temporary parolees and surges, so some finer disputes—such as precise counts by legal‑status category in 2025 onward—remain contingent on continually updated microdata and administrative releases [5] [4] [3].