Unauthorized immigrants in the U.S.: Estimation methods, microdata & selected results

Unauthorized immigrants in the U.S.: Estimation methods, microdata & selected results Jeffrey S. Passel Senior Demographer Measuring irregular migration: Innovative data practices Expert workshop, Global Migration Data Analysis Centre International Organization for Migration Berlin, Germany 18 19 May 2017

Pew Research Center Pew Research Center is a nonpartisan fact tank that informs the public about the issues, attitudes and trends shaping America and the world. We conduct public opinion polling, demographic research, content analysis and other data-driven social science research. We do not take policy positions. We are a subsidiary of The Pew Charitable Trusts, our primary funder.

Today s presentation [Note: This is an expanded version of the presentation given on 18 May 2017. Additional text has been added to explain and enhance the description of methods and sources.] Measurement methods Description of Residual Method Who is Authorized vs. Unauthorized Microdata Status Assignments Data sources Survey data & microdata ACS & CPS Role of weighting in measurement Cross-year & cross-system comparisons Example results Trends, family & labor force characteristics

US unauthorized immigrant population up +500k/year for 90 07; Then declines -500K/year for 07 09; Stable for 09 14; First real decline below 09 level occurred in 15 13 Unauthorized immigrant population (millions) 12.2 11.3 11.1 11 11.0 9 8.6 7 5.7 Shaded area represents 90% confidence interval 5 3.5 3 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 Source: Pew Research Center estimates based on residual methodology.

Preliminary estimate for 16 is inconclusive about further changes because of wide error margin** 13 Unauthorized immigrant population (millions) 12.2 11 11.3 11.1 11.3 11.0 9 8.6 7 5.7 Shaded area represents 90% confidence interval 5 3.5 3 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 ** 2016 estimate based on March Current Population Survey (CPS); preliminary CPS-based estimates since 2012 have been consistently higher than final ACS-based estimates. Source: Pew Research Center estimates based on residual methodology.

Estimation methods

Outline of methodology Estimate unauthorized in survey Residual after lawful temporary migrants are taken out of survey Compared with demographic estimate of lawful permanent residents (LPRs) Adjust for omissions Uses external sources Assign legal status in microdata Create status variable for individuals, families & households Probabilistic multiple imputations Tabulate microdata

Residual estimates of unauthorized immigrants Unauthorized immigrants = Total immigrants (Survey) minus Lawful immigrants (Estimate) Widely used: DHS-OIS 05 11; Warren 80 00 Passel (et al.) 80 present Binational Study 96 Warren & Warren 13 (Variant)

Equations for Pew estimate of unauthorized immigrants Unauthorized = Total minus Lawful immigrants immigrants immigrants Less Estimated Undercount Unauthorized = Counted minus Counted lawful counted immigrants immigrants Counted = Survey minus Counted lawful immigrants foreign-born non-immigrants Note: All populations are for post-1980/1982 entrants. Immigrants exclude temporary migrants or non-immigrants.

Residual estimate using 2015 American Community Survey (ACS) excludes pre-1980 entrants, who are assumed to be legal, and lawful temporary migrants identified in the survey data 43.239 In millions 34.4 7.171 +1.7 8.881 7.2 7.2 7.2 Pre-'80 lawful immigrants (ACS) Add lawful nonimmigrants Leaves post-'80 immigrants (ACS) Lawful immigrants (post-'80) Residual counted Source: Pew Research Center, consistent with Passel & Cohn 2017.

Residual estimate of unauthorized immigrants includes adjustment for undercount Total = Counted plus Missed unauthorized unauthorized unauthorized immigrants immigrants immigrants Some assumptions: a. Estimated undercount % Undercount for lawful immigrants based on Census coverage studies age-sex-race * 1.75 for recent arrivals % Undercount for unauthorized immigrants 1.83 * specific rates for lawful immigrants b. Estimated undercount (2) New work (Van Hook et al. 2014) shows improvements in ACS/CPS coverage for Mexicans from 90s to late 00s c. Internal migration ACS rates for F-B applied to lawful pop

Residual estimate using 2015 ACS 1. Reduce estimate of lawful immigrants by undercount in ACS to compare with counted post-1980 immigrants 34.358 In millions 24.460-1.5% 24.1 Post-'80 immigrants (ACS) Lawful immigrants (est. for '80+) Counted lawful (est. for '80+) Residual (+uncounted) Unauthorized immigrants, total Source: Pew Research Center, consistent with Passel & Cohn 2017.

Residual estimate using 2015 ACS 2. Subtract counted lawful immigrant population to get estimate of unauthorized immigrants counted in the survey 34.358 34.358 In millions 24.460-1.5% 10.3 24.1 Post-'80 immigrants (ACS) Lawful immigrants (est. for '80+) Counted lawful (est. for '80+) Residual (+uncounted) Unauthorized immigrants, total Source: Pew Research Center, consistent with Passel & Cohn 2017.

Residual estimate using 2015 ACS 3. Adjust for unauthorized immigrants missed by ACS to get estimate of the total unauthorized immigrant population 34.358 34.358 In millions 24.460-1.5% 10.3 11.0 +6.3% 24.1 Post-'80 immigrants (ACS) Lawful immigrants (est. for '80+) Counted lawful (est. for '80+) Residual (+uncounted) Unauthorized immigrants, total Source: Pew Research Center, consistent with Passel & Cohn 2017.

Who are the lawful immigrants? Components of lawful immigrants: a. Refugee arrivals (ORR**, DHS/INS, State) b. Asylum approvals (DHS/INS) c. Cuban-Haitian entrants, Amerasians, some parolees (ORR) d. Other entrants, other parolees (DHS/INS) e. IRCA legalizations approved SAWs & LAWs (INS) f. DHS/INS new arrival green cards (except Amerasians in d.) g. Adjustments to LPR status, except from a.-e. above (DHS/INS) h. Pre-1980 (or 1982) arrivals (ACS/CPS counts) a.-g. are combined year-by-year using allowances for mortality, emigration, & state-to-state movement to estimate the legal population ** ORR is Office of Refugee Resettlement; DHS is Department of Homeland Security; INS is its predecessor, the Immigration and Naturalization Service; SAWs are Special Agricultural Workers; LAWs are other persons legalized under the 1986 immigration law (IRCA).

Who are the unauthorized? Not lawful or nonimmigrants (temporary) Overstays (~40-45% or more) & EWIs Quasi -lawful, grouped with unauthorized: a. DACA** Beneficiaries (700,000+) b. TPS** & DED**, especially Central Americans (250,000+) c. Asylum applicants not yet adjudicated d. Adjustment applicants (esp. K, V visas) Overlapping categories (& little data) Possibly 10% of unauthorized have DACA and TPS (and work authorization) ** DACA is Deferred Action for Childhood Arrivals, a temporary legal status, including work authorization for some individuals brought to the US as children. TPS is Temporary Protected Status. DED is Deferred Deferred Enforced Departure (much rarer than TPS). EWI is Entry Without Inspection (i.e., clandestine entry.)

Illustration of measuring the statistical significance of change in unauthorized immigrant populations. The standard error of the residual estimate is the standard error of the estimated post- 82 arrivals because the lawful population is a demographic estimate. The change from the 2014 ACS to 2015 ACS is not significant at 90% confidence (Z = -1.183) 33.8 ACS 32.9 ACS In millions ±0.113 s.e. 11.0 11.1 ±0.187 90% interval (1.645 s.e.) 22.8 21.8 ±0.106 s.e. ±0.175 90% interval (1.645 s.e.) Diff = -0.184 ±0.155 s.e. ±0.256 90% interval (1.645 s.e.) Z = -1.183 2015 ACS Residual* Standard error (for post-'82 F-B) 2014 ACS Residual* Standard error (for post-'82 F-B) Change 2014-2015 & std. error * Adjusted post-1982 immigrants. Source: Pew Research Center, consistent with Passel & Cohn 2017.

Change in unauthorized immigrant population from 2009 ACS to 2015 ACS is statistically significant at 90% confidence (Z = -2.499) 33.8 ACS 11.0 22.8 ±0.113 s.e. ±0.187 90% interval (1.645 s.e.) 30.9 ACS 11.3 19.6 ±0.093 s.e. ±0.153 90% interval (1.645 s.e.) In millions Diff = -0.367** ±0.147 s.e. ±0.252 90% interval (1.645 s.e.) Z = -2.499 2015 ACS Residual* Standard error (for post-'82 F-B) 2009 ACS Residual* Standard error (for post-'82 F-B) Change 2009-2015 & std. error * Adjusted post-1982 immigrants. Source: Pew Research Center, consistent with Passel & Cohn 2017.

Change in unauthorized immigrant population from 2015 ACS to 2016 CPS is not statistically significant at 90% confidence (Z=+0.871), in part, because of much larger sampling error in the 2016 CPS 33.8 ACS 34.6 CPS In millions 11.0 ±0.113 s.e. 11.3 ±0.187 90% interval (1.645 s.e.) 22.8 23.3 ±0.400 s.e. ±0.659 90% interval (1.645 s.e.) Diff = +0.363 ±0.416 s.e. ±0.685 90% interval (1.645 s.e.) Z = +0.871 2015 ACS Residual* Standard error (for post-'82 F-B) 2016 CPS Residual* Standard error (for post-'82 F-B) Change 2015-2016 & std. error * Adjusted post-1982 immigrants. Source: Pew Research Center, consistent with Passel & Cohn 2017.

Status assignments, geographic data & characteristics

Types of data & estimates We distinguish the residual estimates of lawful residents and unauthorized immigrants as analytic estimates from the survey-based estimates that result from the microdata status assignments Analytic estimates of unauthorized immigrants a. Counted in survey and corrected for undercount b. Totals by country/region of birth (i.e., Mexico, Latin America, Asia, Rest of World) c. 6 individual states (CA, FL, IL, NJ, NY, TX) and balance of the US d. Some limited demographic information: * Age and Sex * Period of Entry Some of the analytic estimates are used as control totals in making the microdata status assignments into the survey The resulting survey-based estimates may differ somewhat from the analytic estimates for some countries, age groups, regions, etc. because of the level of control and the random nature of status assignments The following slides describe the status assignment process and the resulting data

Status assignments (I) Lawful temporary immigrants a. Students (F, M visas) & scholars (J visa) b. High-tech guest workers (H-1B), intracompany transfers (L visas), nurses (H-1A) and physicians (J visas) c. International organizations (G visas), diplomats and embassy employees (A visas) d. Religious workers (R visas), athletes, artists, entertainers and outstanding talent (O,P visas) e. Au pairs and exchange visitors (J visas) -- Uses occupation, period of entry, relationships in household, other characteristics -- No demographic targets; results tend to understate DHS estimates, but adjustments for undercount compensate -- Many more lawful temporary immigrants appear in ACS than CPS

Status assignments (II) Refugees/asylees (at entry) a. Country of birth identifies refugee-sending areas b. Period/year of entry isolates specific in-flows c. Demographic targets used to control assignments Naturalized citizens a. In US < 6 years Edit to alien if spouse is not a US citizen b. In US >= 6 years Mexico Treat as potential unauthorized immigrant (as below); may end up as unauthorized or naturalized Central America Same as Mexico All other countries Accept as naturalized c. No demographic targets

Status assignments (IIa lawful) Definite lawful immigrants a. Naturalized citizens, refugees & legal temps (as above) b. Entered US before 1980 or 1982 (depending on survey year) c. Occupations/employers Government workers, veterans & military Law enforcement and related Court and legal Licensed occupations (e.g., medical) Security occupations (esp. with licensing; e.g. police, air traffic) Assorted other minor occupations d. Program participation** (SSI, TANF, Medicaid, Medicare) e. Family members (most) of lawful residents Others: Potential unauthorized ** SSI is Supplemental Security Income; TANF is Temporary Aid for Needy Families. In most states, Medicaid eligibility is restricted to lawful immigrants and US citizens.

Status assignments (III-unauthorized) Targets (analytic) a. 6 States (CA, FL, IL, NJ, NY, TX) and balance of US b. Total, under 18 c. ACS: Mexico, Latin America, Asia, all other (collapsed in some states due to small sample sizes) d. CPS: Mexico, all other Random assignments of Potentials a. Replicate households for multiple imputation (1) Weights are fractionated (2) Separate random assignments for individuals in each replicate (3) Like assignments collapsed at end of process (4) ACS: 5-10 replicates; CPS: 100 replicates b. Initial p s = Target estimate / Potential unauthorized (1) Where Target defined above and Potential from previous slide (2) Parents & non-parents done separately (3) Assignments for parents use targets for children

Status assignments (IIIa-unauthorized) Final status assignments of Potentials a. Household edits for consistency of parents, children & relatives (based in part on date of arrival) b. Adjustment of p s from (Target/Potential unauthorized) (1) Initial modification based on IRCA LPS** by occupation group (2) Possibility of using other data here, e.g. SIPP (3) Adjust each state-region based on convergence to targets (4) Separate adjustments for parents & non-parents by state-region c. Iterate assignments and edits (a. & b. as above) until targets are hit d. Relatively insensitive to LPS p s since targets set initial values at 80 95% of potential unauthorized population Adjust weights for undercount ** LPS is Legalized Population Survey

Legal status microdata Dataset with Legal status variable a. Categories unauthorized, LPR, naturalized citizen, refugee arrival, lawful temporary immigrant Other variables created a. Nuclear families ( MHUs or Minimal Household Units) including parent-child linkages b. Family legal status hierarchical based on head & spouse c. Household legal status hierarchical based on head & spouse Uses of data a. Geography, including all states; most metropolitan areas b. Detailed data by country of birth may differ from analytic estimates (except for Mexico) c. Family data and other characteristics of legal and unauthorized immigrant populations (as illustrated below)

Immigration measurement issues (I) Problematic demographic components a. Lawful nonimmigrants, including K & V visas b. Some categories of refugees/parolees difficult to identify c. Backlogs in categories defined as US residents (not in a.) d. Emigration, especially of lawful immigrants e. Deportations (& emigration) of unauthorized immigrants f. Counts of other quasi-legal categories Categorical issues a. Definition of and identification of true US residents (i.e. those living in US) as opposed to more transient groups; this especially affects Mexicans and lawful temporaries b. Gross flows (into and out of US); also flows between categories of immigrants.** ** Because the estimates are cross-sections in time, flows from year-to-year must be inferred from annual differences. The residual nature of the estimates means that the annual differences have very large error bands, so flow estimates can be problematic.

Immigration measurement issues (II) Data issues a. Reliance on Census/CPS/ACS population controls to set magnitude of foreign-born population b. Lack of direct measures of census/survey undercounts for immigrants (either in total of by legal status) c. Emigration measurement d. Interpretation of date of entry in CPS/ACS versus data on admissions from DHS/INS Government/survey issues a. Better models for measuring immigration and emigration as components of population change (Census Bureau) -- Consistent population controls and weighting over time b. More microdata, especially from DHS (OIS) c. Direct collection of legal status in more surveys (perhaps as input to models differentiating lawful, temporary and unauthorized immigrants)

Data sources and weighting

American Community Survey Universe: Total population** 2005 Sample size (random, representative with minor stratification) a. Full sample: 250,000 households per month ( rolling sample or 3 million per year) Subsampling for non-response within small geographic areas Expansion in 2012 to 3.2 million households Weighting issues a. 12 merged monthly samples (3 or 5 merged years for published data) b. Sub-state geography comparable across groups of years (2005 2011, based on 2000 census geography; 2012 based on 2010 census geography) c. New population controls every year (!) can hamper comparability over time 1. For 2007 2009 (i.e., Vintage 07-09 population estimates) change in methods created significant discontinuities with previous (and subsequent) years 2. Change in 2010 to 2010 Census-based population estimates also created significant discontinuities with 2009 and earlier years d. Intercensal reweighting has LARGE impact on 2007-2009 estimates; the Pew Research Center has created new population weights with methods approximating those used by the Census Bureau for the public-use microdata Key Variables a. Country of birth, US citizenship, year of entry (I.e. year came to live in US) b. No data on parental place of birth (to identify 2 nd generation) c. Measurement issues on some rolling sample questions (e.g., current year entrants) d. Public-use sample 1% of US population (i.e., ~3 million individuals) ** 2005 ACS covered household population only, not group quarters (GQ). The Pew Research Center has developed representative GQ data from the 2006 ACS and official population estimates so that the dataset used for 2005 covers the entire US population.

Key dates for population controls Vintage 2007 (ACS 2007, CPS 2008) a. ½ of post- 00 immigration by old method (FB difference) b. ½ of post- 00 immigration by new method (residence one year ago ROYA) Vintage 2008 (ACS 2008, CPS 2009) a. All post- 00 immigration by new method; results in less immigration and smaller Hispanic & Asian populations and, hence fewer foreign-born Vintage 2009 (ACS 2009, CPS 2010) a. Changes in age structure of migration, but not level Vintage 2010 (CPS 2011) a. More changes in age structure of migration, but not level Census 2010-based controls (ACS 2010, CPS 2012) a. Large differences in Hispanics and Asians because 2010 Census counts for these groups are much higher than previous estimates (i.e., Vintage 2009) b. New base for population estimates used for 2010 and later ACS and for 2012 and later CPS c. Intercensal population estimates (consistent with 2000 and 2010 Censuses) result in revisions for 2001 2009 estimates Revisions to weights a. Census Bureau does NOT generally revise survey weights to reflect new population estimates; exceptions are 2000 CPS and 2010 2011 CPS but not the ACS b. Pew has revised weights for ACS 2005 2009 & CPS 1995 2009 c. Pew created new data for group quarters in 2005 so 2005 ACS covers full population

ACS foreign-born trend using original weights is somewhat erratic. Large increase for 05 06 is followed by smaller increases and a decline for 07 08. Then, 09 10 shows the largest increase of the decade when other data show a large drop in in-flows. Foreign-born population (millions) 39.9 40.4 37.5 38.1 38.0 38.5 Original weights ('05 HH only) 35.8 Reweighted to intercensal estimates Original weights ('06-'11) 2004 2005 2006 2007 2008 2009 2010 2011 2012 Source: Pew Research Center tabulations of IPUMS data.

Revised ACS weights smooth overall trend and lead to sensible estimates of year-to-year change. Smaller annual changes for 07 10 than earlier. Revisions directly affect residual estimates of unauthorized immigrants. Foreign-born population (millions) 39.9 40.4 39.3 38.4 38.8 37.5 38.1 38.0 38.5 36.4 35.8 Original weights ('05 HH only) Reweighted to intercensal estimates Original weights ('06-'11) 2004 2005 2006 2007 2008 2009 2010 2011 2012 Source: Pew Research Center tabulations of IPUMS data.

Current Population Survey Universe: Civilian, noninstitutional population (omits ~4 million in institutions) Sample size (state-based design, not directly proportional to population) a. Monthly: 45,000 Households b. March: 1994-2001, 50,000+ HHs; 2001-present, ~75,000 HHs Weighting issues (March supplement) a. 1994 1995: Many problems affect weights and country-of-birth data; do not recommend using 1994; use 1995 with caution; special revised weights are available (developed at Urban Institute) b. 1994 2001: 90-based weights; use with great caution; 1. Revised intercensal weights for 1994 1999 by Pew Research Center 2. Census Bureau has released weights for March 2000 based on 2000 Census 3. For March 2001, use alternative data set c. 2001 (SCHIP): Expanded sample based on 2002 redesign; weights consistent with 2000 Census d. New population controls every year can hamper comparability over time 1. For 2008 2010 (i.e., Vintage 07 09 population estimates) change in methods created significant discontinuities 2. Change to 2010 Census-based population estimates (for 2012 CPS and later) also created discontinuities 3. Census Bureau has released alternative 2010-based weights for March 2010 and 2011 CPS e. Intercensal reweighting has LARGE impact for some years; the Pew Research Center has created new population weights for 1994 2009 with the same methods used by the Census Bureau. New weights have very LARGE impact on 1995 1999 and significant effects on 2008 Key Variables a. Place of birth, US citizenship, period of entry (2-year period from 1980; 2+ or 3+ for most recent years) b. Parental place of birth (allows identification of 2 nd and 3 rd + generations for comparison and analysis) c. Public-use sample Full CPS sample

Original CPS series has major breaks after 00 and 10 Censuses, with an especially large discontinuity following the 00 Census. Year-to-year changes are erratic particularly when population estimation methods changed in 07 09. Foreign-born population (millions) 36.8 38.2 30.1 29.9 23.0 26.5 28.4 Original weights ('95-'01) Reweighted to intercensal estimates Original weights ('00-'12) 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 Source: Pew Research Center tabulations of March CPS supplements.

Revised, consistent CPS weights create continuous series with more sensible estimates of year-to-year change. Foreign-born estimates are more larger for the 90s, leading to higher estimates of the unauthorized immigrant population. New weights also give more credible estimates for 08 11. Foreign-born population (millions) 39.5 37.5 38.2 36.8 24.8 23.0 31.5 30.7 29.0 30.1 29.9 28.4 26.5 Original weights ('95-'01) Reweighted to intercensal estimates Original weights ('00-'12) 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 Source: Pew Research Center tabulations of March CPS supplements.

Selected results

About one-in-four US immigrants are unauthorized Unauthorized immigrants 11.0 million (24%) Lawful immigrants (total) 33.8 million (76%) Temporary lawful residents 2.1 million (5%) Lawful permanent residents 11.9 million (27%) Naturalized citizens 19.8 million (44%) Total U.S. foreign-born population: 44.7 million Source: Pew Research Center based on augmented 2015 ACS.

Unauthorized immigrant population dominated by Mexico, Latin America Other regions Asia 1,600 Caribbean 350 South America 650 1,100 Unauthorized immigrant population (thousands) Total 11.0 million 5,650 Mexico Central America 1,950 Source: Pew Research Center based on augmented 2015 ACS.

For the first time in a decade, Mexicans are NOT a clear majority of unauthorized immigrants 7 6 Unauthorized immigrants (millions) Mexico 6.9 6.4 5.9 5.8 5.6 5.7 5.6 5 4.5 5.3 5.0 5.3 5.3 5.4 4 3 2.9 4.1 Other countries 2.0 2.8 Shaded/dotted areas represents 90% confidence interval 2 1.5 1 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 Source: Pew Research Center estimates based on residual methodology; 2016 based on March CPS is preliminary.

Immigrant men work more than natives; Women work less 91% 84% 79% Unauthorized Immigrants Lawful Immigrants U.S. Born 61% 67% Percent in labor force, 2015 72% Adult Men (18-64) Adult Women (18-64) Source: Pew Research Center based on augmented 2015 ACS.

Immigrant families, especially unauthorized, more likely to be couples with children U.S. native households 18% Percent of group s households that are couples with children, 2015 Legal immigrant households 33% Unauthorized immigrant households 44% Source: Pew Research Center based on augmented 2015 ACS.

Increasing % of unauthorized immigrant adults in the US for many years Share of unauthorized immigrant adults 34% 60% 10 or More Years in US 55% 50% 47% 34% 36% 38% 41% 43% 64% 66% 66% 35% 34% 37% 37% 32% Less than 5 Years in US 27% 24%22%20% 17% 15% 14% 15% 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 Source: Pew Research Center based on augmented 1995-2003 March CPS and 2005-2015 ACS.

Unauthorized immigrant families include 6 million legal residents, mostly children Unauthorized men 5.6 million 54% of adults Unauthorized women Unauthorized children (<18) 750,000 Only 7% of All unauthorized U.S. citizen children (<18) 5.0 million 87% of kids 4.7 million 46% of adults Other adults 990,000 17 million in unauthorized immigrant families, 2014 Source: Pew Research Center based on augmented 2014 ACS.

Conclusions Unauthorized immigrants appear in US surveys a. Methods for estimating size of population and determining characteristics are available Direct impacts on estimates from weighting methods and population controls a. Be very careful that estimates are consistent when measuring trends and making cross-survey comparisons b. Consistent weights are available at the Pew Research Center for the American Community Survey and the Current Population Survey Better data on survey coverage is needed a. For foreign-born population especially b. Separate estimates for coverage of unauthorized immigrants would be valuable Much can be learned from ACS & CPS a. Careful analysis is required

Thank you!

Contact Information Jeffrey S. Passel Senior Demographer jpassel@pewresearch.org 202-419-3625 (direct) 202-527-2146 (mobile)