April 23, 2012

Net Migration from Mexico Falls to Zero—and Perhaps Less

Appendix C. Methodology

The measures of migration developed for this report draw on numerous data sources from both Mexico and the U.S. The principal Mexican data sources are:

The principal U.S. data sources are:

This appendix describes the specific items used from these various sources and the development of some of the migration measures used in the report. Data from the Mexican censuses, the Mexican Population Count, the American Community Survey, the U.S. censuses of 1990 and 2000, and the Current Population Survey are based on Pew Hispanic Center tabulations from public-use microdata sets, some of which have been modified or adjusted. As such, the figures reported may differ from published data from the same sources.

Mexican Data Sources

Mexican Censuses: 1990, 2000 and 2010

This report employs Mexican Census (Censo de Población y Vivienda) data from the decennial censuses of 1990, 2000 and 2010. In addition to the basic census form, each census also included a detailed questionnaire administered to 10% of households that collected a variety of demographic, social and economic characteristics. In addition, the 2000 and 2010 censuses included a questionnaire focused on migration. It was administered to households within the 10% sample with recent migrants (anyone who had migrated in the five years prior to the census).

Immigrants to Mexico and Mexicans returning from abroad are identified through a number of different questions. The full census asks respondents their state or country of birth, which is used to measure lifetime migration (identifying people born outside of Mexico). The full census also includes two questions on residence five years before the census. For example, in the case of the 2010 census, respondents are asked about their place of residence in June 2005—the state or country of residence and the municipio or delegación of residence if respondents were in Mexico. This is asked of people ages 5 or older. These two questions can be used to measure return migration of Mexicans during the five-year period before the census or immigration to Mexico during the period by persons not born in Mexico. They also provide data on migration patterns within Mexico. Since the question on residence five years ago pertains only to persons ages 5 and older, the number of immigrants into Mexico by persons younger than 5 years old is determined from the place of birth question, i.e., the number under age 5 born outside of Mexico.

The supplemental sample questionnaire, asked of a 10% sample of households in 2000 and 2010, is focused on international migration. The first question asks whether anyone who “lives or lived with you (in this house) went to live in another country” in the previous five years. Persons identified as leaving the country with this question are designated as “recent migrants” in this report. The census asks how many recent migrants left from the household and then follows with a battery of questions about each migrant: gender, age at most recent migration, month and year of most recent departure from Mexico, state of residence at departure, destination country and place of current residence (for example: U.S., other country, Mexico). For those migrants who are identified as currently living in Mexico, there are further questions: month and year of return and, in 2010, whether the returned recent migrant is in the respondent’s household.

Total migration into Mexico from the United States during the five years before the 2010 and 2000 censuses is estimated from the full census migration questions and the migrant sample of returned recent migrants. The components of migration into Mexico are:

With these questions, it is possible to identify all persons who migrated into Mexico during the previous five years who were still alive at the date of the census (and were still in Mexico). For out-migrants from Mexico who left in the last five years, only those who departed from existing households that still have members in Mexico can be identified; the Mexican census does not provide direct information on outmigration of entire households.

The migrant sample includes only a limited amount of socio-demographic data on the migrants. However, most of the recent migrants who have returned to Mexico (i.e., those who returned to the same household) can be linked to their own record in the full census sample. For 2010, we were able to match about two-thirds of the returned recent migrants (26,700 unweighted cases out of 38,750 returnees in the full migrant sample); for 2000, we were able to match about three-quarters of returned recent migrants (24,915 out of 32,672 returnees). Because some migrants make multiple trips to the U.S., some of the returned recent migrants (i.e., those who made a trip out of Mexico after June 2005 in 2010 or after January 1995 in 2000) were living in the U.S. five years before the census. In measuring total migration into Mexico, it is necessary to remove this group from the estimate to avoid double counting. Using the matched samples, we found that 33% of the returned recent migrants in the 2010 census had been in the U.S. five years earlier; among returned recent migrants in 2000, 14% had been in the U.S. in 1995. Applying these shares to the full count of returned migrants reduced the estimate of recent returns by 102,000 in 2010 and by 37,000 in 2000. Very few of the returned recent migrants were children under age 5 born in the U.S.; reductions from this overlapping group amounted to 728 persons in 2010 and 773 in 2000.

The data from Mexican censuses shown in the report were developed from tabulations of microdata samples. The microdata come from a 10% sample of the full census; all cases in the 10% recent migrant sample are included in the microdata. For 2010 and 2000, microdata samples were downloaded from the INEGI website (entered at http://www.inegi.org.mx/est/contenidos/espanol/soc/sis/microdatos/default.aspx). The sample sizes for the microdata are:

For 1990, the microdata sample was downloaded from the Integrated Public-Use Microdata Series International (IPUMS-I) website (https://international.ipums.org/international/). The IPUMS-I sample was generated from the original INEGI microdata sample, but also includes some additional IPUMS-generated variables, notably variables for linking respondents with their mothers and fathers (if present in the household) developed from the relationship variable. The sample size for the 1990 IPUMS-I person sample is 8,118,242 cases.23

Our estimates attempt to differentiate U.S.-born immigrants to Mexico who are children of Mexican migrants who were in the U.S. from other U.S.-born immigrants to Mexico. This requires that we ascertain the country of birth of a respondent’s parents in the microdata. Unfortunately, none of the Mexican censuses include a direct question on country of birth of the respondents’ mother or father. However, for 2010, the INEGI microdata sample provides a pointer to each respondent’s mother and father if they were present in the household. For 1990 and 2000, the IPUMS-I microdata samples include similar pointers developed from the information collected on relationship. With these pointers, we were able to classify most U.S.-born individuals as the child of a Mexican-born mother or father (75% in 2010, 73% in 2000, and 66% in 1990). Only a small percentage of U.S.-born individuals could be linked to a parent or parents not born in Mexico (less than 2% of the U.S.-born in each census). The remaining 25-34% were not in households with their parents so they could not be identified definitively as having Mexican parents.

Mexican Conteo of 2005

The 2005 II Conteo de Población y Viviendais a mid-decade census conducted by INEGI to identify changes in the Mexican population and update various population projections for the country and constituent geographic areas. The Conteo collected less information than the full censuses of 2000 and 2010. The Conteo included information on residence five years earlier but did not ask for place of birth, nor did it include the special migrant questionnaire that was used in the 2000 and 2010 censuses. The Conteo provides information on movement into Mexico by persons ages 5 and older but not the more detailed information on other components of movement between Mexico and the U.S.

For 2005, we used 10% microdata samples from both INEGI and from IPUMS-I. Each sample included 10,282,760 people.

Survey of Migration in the Northern Border of Mexico or Encuesta sobre Migración en la Frontera Norte de México (EMIF-Norte)

The EMIF-Norte was first designed and conducted in 1993 by the Mexican Population Council (CONAPO), the Ministry of Labor and Social Welfare (STPS), and the Colegio de la Frontera Norte (COLEF). In later years, the Mexican Migration Institute (INM), the Ministry of Foreign Relations (SRE) and the Ministry of Health (SS) joined the survey’s technical committee. The estimates shown in this report were obtained from the EMIF-Norte surveys of 1995, and 2000-2010 calendar-year datasets published by COLEF.24

Survey Design

EMIF-Norte applies probabilistic sampling techniques for populations moving to and across the U.S.-Mexico border. The survey uses a multi-stage stratified design to collect information from migrants and potential migrants of different types across the length of the border and throughout the year.

The sampling framework of the EMIF-Norte characterizes four distinct flows:

  1. Flow from the South into the U.S.;
  2. Flow from the northern Mexican border areas into the U.S.
  3. Individuals handed over to Mexican authorities by U.S. authorities (including so-called “voluntary returns” and removals);
  4. Other flows from the U.S.

In this report, we analyze data from the third flow exclusively—a randomized sample of Mexican nationals ages 15 years and older, who had been either returned or removed by the U.S. Department of Homeland Security and handed over to Mexican authorities at specified points along the Mexico-U.S. border.

Sample Selection for EMIF-Norte

The EMIF sample is selected at various points along the U.S.-Mexico border (which are estimated to include about 94% of the migration flow). Within geographic units or sampling points, data are collected within sampled time intervals (“sample hour” or “jornada”). Migrants are identified within each time-space sampling unit with a short set of four to five screening questions that distinguish them from other subjects within the flow, such as tourists, people born in the U.S. or residents of the border cities who are cross the border regularly. Once a migrant is identified, a longer questionnaire is administered to collect a range of information specific to each type of migration.

Sample Size

The module of the survey we analyzed—persons handed over to Mexican authorities by DHS— has a sample size of about 6,000 to 10,000 Mexican migrants ages 15 years and older for each calendar year.

Possible Sampling Issues

The sample for this flow is representative of about 95% of all Mexicans ages 15 years and older who were received by Mexican migration authorities at eight specific points along the U.S.-Mexico border and a few international airports. The remaining 5%, not captured in the survey, may have been handed over at less transited points along the border where the survey is not conducted.

However, this module of the EMIF appears to have some issues related to either coverage or weighting. The weighted number of Mexican nationals received and documented by the Mexican authorities and reported by the Mexican Migration Institute (INM)—which form the basis for the EMIF weights—is considerably smaller than that reported by the U.S. Department of Homeland Security (DHS). In some years, the estimated number of Mexican nationals received and registered by Mexican authorities is 58% below the number reported by DHS, though in recent years the gap has narrowed to less than 30%. (See appendix figure C1.)

National Survey of Demographic Dynamics (Encuesta Nacional de la Dinámica Demográfica—ENADID): 2006 and 2009

The National Survey of Demographic Dynamics or ENADID is a national household survey conducted by the Mexican government to collect a wide range of information about population change in Mexico. In addition to a module of questions related to international migration, the survey covers fertility and pregnancy history of women in detail, births and deaths, contraceptive usage and preferences, and marriage. The survey was conducted in 1992, 1997, and 2009 by the Mexico’s National Institute of Statistics and Geography (INEGI) and by the National Population Council (CONAPO) in 2006.

The international migration questions are similar in concept and design to those included in the census and its migration module. The ENADID provides detailed migration information but on a much smaller sample of the population. The survey collects information from current household residents as well as data on persons who used to live in the household but had migrated to the United States in the previous five years. Items collected from current household members include: residence one year ago (asked in 2009), residence five years ago, and place of birth. For the “recent migrants” (i.e., persons who had left the household in the last five years to make a trip to the U.S.), ENADID collects the same information as the 2010 census: age at migration, state of origin for the migrant, month and year of departure, current residence, and, if the migrant is back in Mexico, month and year of return to Mexico. In addition, ENADID provides information not collected in the census: destination within the U.S., documentation on migration and reason for migrating (asked in 2006). The data from the 2006 and 2009 ENADID were used in this report mainly to provide indications of trends in migration based on the question regarding residence five years earlier.

In terms of coverage of migration, ENADID has the same limitations as the census. All migrants into Mexico in the period before the survey (who are still alive and still in Mexico) can be identified. For recent out-migrants, only those migrants from households where some members remained in Mexico can be identified. ENADID is not able to measure outmigration of whole households.

The 2006 ENADID is a sample of 41,926 households. The resulting sample and microdata contain information on 142,951 residents and 5,632 recent migrants. For 2009, the basic ENADID sample is much larger—89,266 households. The microdata sample we analyzed included 343,887 residents. The migrant sample had 4,872 recent migrants.

National Survey of Occupation and Employment (Encuesta Nacional de Ocupación y Empleo or ENOE): 2005-2011

The National Survey of Occupation and Employment is conducted by INEGI to measure Mexico’s labor force and its employment characteristics. The survey has a complex design with a national sample of approximately 120,000 households. Each sample household is interviewed five times at three-month intervals. The sample is divided into five roughly equal rotation groups, and each quarter a new rotation group enters the sample and the group that has completed five interviews rotates out of the sample.

With the reinterview and overlapping rotation group structure, changes in household composition can be monitored after the initial interview. ENOE can thus provide information on births (added to the household), deaths (subtracted from the household), and migration into and out of the household. When a migrant is identified in interviews two through five, either because the person is no longer in the household or a new household member arrives, there are a number of questions concerning destination of the out-migrant or origin of the in-migrant (i.e., somewhere else in Mexico or outside the country) and reasons for migration.

ENOE is more limited in measuring in- and out-migration than the census or ENADID. ENOE does not collect information on previous residence, so it measures only movement into and out of existing households. Movement of entire households out of Mexico is not measured (as is the case with the census and ENADID). Unlike these other data sources, ENOE does not measure migration of whole households into Mexico. Although whole households who return to Mexico can fall into the ENOE, there is no mechanism for identifying these migrants because ENOE does not include questions on previous residence.

Microdata samples from ENOE are available from INEGI (http://www.inegi.org.mx/est/contenidos/Proyectos/encuestas/hogares/regulares/enoe/Default.aspx). Information in this report was obtained from publications of migration rates and numbers rather than tabulations of microdata.

United States Data Sources

Current Population Survey (CPS): 2000-2012

Monthly CPS

The Current Population Survey (CPS) is a monthly survey conducted by the Census Bureau for the Bureau of Labor Statistics. The CPS is a stratified probability sample of about 60,000 households designed to give state-specific information on employment and unemployment (http://www.census.gov/cps/methodology/). Generally, about 50,000-55,000 households are interviewed. The sample has overlapping rotation groups in which each household is interviewed in four consecutive months, is out of the sample for eight months, and then returns to the sample for four more consecutive months.

The monthly CPS has a range of questions focused on labor force participation, but also collects information on demographic characteristics, education and immigration through questions on country of birth, parents’ country of birth and citizenship (since 1994). The citizenship information identifies respondents as U.S. natives, U.S. citizens through naturalization and non-citizens (but with no further information on legal status). The latter two groups comprise the foreign-born population. For persons born outside the United States, the CPS asks when the individuals “came to live in the United States.” All CPS cases are included in public-use microdata files, available from a variety of sources.

March Annual Social and Economic Supplement (ASEC)

Each March, the basic CPS sample and questionnaire are expanded for the Annual Social and Economic Supplement (ASEC). The sample is augmented to about 80,000 households with a double sample of Hispanic households and oversampling of households with children and households headed by persons who are not white. The questionnaire is expanded to include questions about health insurance, detailed sources of income, program participation and residence the previous March. This makes the March ASEC supplement the main source of information on poverty and lack of health insurance. The question on residence one year prior to the survey date provides information on current migration into the United States. In this report, the March CPS is the principal data source on the size and characteristics of the unauthorized population (see below for estimation methodology).

The published information from the CPS and the CPS microdata use survey weights based on the most current information available to the Census Bureau at the time the survey is conducted. Because additional data on population change can become available and because of changes in the methods used to measure population change, the weights for the monthly CPS and the March supplements are not necessarily consistent across time. Consequently, comparisons of population numbers across different releases of the CPS can conflate actual population change with methodological changes. To minimize the impact of methodological change on comparisons across time, the Pew Hispanic Center has developed alternative weights for the March CPS supplements of 2000-2007 and monthly CPS for 2000-2008 that use a consistent set of population estimates and permit more accurate comparisons over time. The methodology for developing the alternative weights is described in Appendix C of Passel and Cohn (2010).

American Community Survey (ACS): 2005-2010

The American Community Survey (ACS) is a continuous survey that collects detailed information from a sample of the U.S. population on a wide range of social and demographic topics. Each month the ACS samples about 250,000 households. Interviews are conducted by mail and in person; follow-up is conducted on a sample of initially non-responding households. The nominal sample size is about 3.1 million households per year; about 2.1 million households are included in the final sample.25 The monthly samples do not overlap within five-year periods so that detailed information can be obtained for various geographic levels by combining samples across months.

Data from the ACS are released on an annual basis covering interviews conducted during calendar years. Information from a single year of ACS interviews is published for the nation, states, and “recognized legal, administrative, or statistical areas” with populations of 65,000 or more. Data for three consecutive calendar years are combined to provide tabulations for areas with populations of less than 20,000; data for five consecutive years provide information for all areas down to census tracts and block groups. The ACS began in 2005 with a sample of the household population and was expanded to full operational status in 2006 when the household and group quarters population were included.

The ACS includes questions on place of birth (state or country), citizenship and residence one year before the interview. For people born outside the U.S., the ACS asks when the person came to live in the United States. These data items provide information on the foreign-born population and movement to the United States. To the extent that ACS data are used in this report, the information comes from tabulations of microdata obtained from the Integrated Public-Use Microdata Series (IPUMS). For each year, the microdata set represents a 1% sample of the U.S. population or about 3 million individual cases per year.

Each year’s ACS is weighted to the Census Bureau’s population estimates for that year. The 2010 ACS is the first to be weighted to results from the 2010 census. The use of annual population estimates for weighting can create discontinuities in making comparisons across years when the estimation methods change or when the results of a new census are introduced (as in 2010). To help minimize comparison issues related to changes in population estimates, the Pew Hispanic Center has produced alternative ACS weights for 2005-2009 that are consistent with results from the 2010 census and the 2000 census (Passel and Cohn, 2012). These alternative weights are used in ACS results for 2005-2009.26 Note that estimates of the size of the foreign-born population from the ACS differ from those based on the CPS for a number of reasons. The surveys differ in weighting and coverage; the CPS universe is of the civilian, noninstitutional population while the ACS universe is of the total resident population. Additionally, our estimates from the March CPS are adjusted for survey undercoverage.

Decennial Censuses: through 2000

U.S. decennial censuses from 1850 through 2000 have provided information on the foreign-born population via a question on place of birth. Through 1970, these censuses also asked mother’s country of birth and father’s country of birth, which permit identification of the second generation. Data on the Mexican-born population from 1850 through 1990 are from these census results presented by Gibson and Jung (2006) and in the Historical Statistics of the United States (U.S. Census Bureau, 1975).

For 1980 through 2000, we used a 5% public-use sample of census records from IPUMS to generate information on the foreign-born population. These sources also collect information on citizenship and year of entry to the U.S.

Estimation Methods

Two principal sets of new estimates presented in this report were generated by the Pew Hispanic Center using U.S. data sources described above and demographic estimation methods—estimates of the size and characteristics of the unauthorized immigrant population in the U.S. and estimates of the annual inflow of Mexican immigrants into the U.S. for 1991-2010. Previous versions of these estimates and the methods used to derive them have been published elsewhere; see, for example, Passel and Cohn (2011) and Passel (2011). This section includes a brief description of the estimation methods used in this report.

Residual Method for Estimating Unauthorized Immigrant Population

The data presented in this report on unauthorized and legal immigrants from Mexico were developed with essentially the same methods used in previous Pew Hispanic Center reports (Passel and Cohn 2010; Passel and Cohn, 2009). The national and state estimates use a multistage estimation process, principally using March Supplements to the Current Population Survey (CPS).

The first stage in the estimation process uses CPS data as a basis for estimating the number of legal and unauthorized immigrants included in the survey and the total number in the country using a residual estimation methodology. This method compares an estimate of the number of immigrants residing legally in the country with the total number in the CPS; the difference is assumed to be the number of unauthorized immigrants in the CPS. The legal resident immigrant population is estimated by applying demographic methods to counts of legal admissions covering the period from 1980 to the present obtained from the Department of Homeland Security’s Office of Immigration Statistics and its predecessor at the Immigration and Naturalization Service. The initial estimates here are calculated separately for age-gender groups in six states (California, Texas, Florida, New York, Illinois and New Jersey) and the balance of the country; within these areas the estimates are further subdivided into immigrant populations from 35 countries or groups of countries by period of arrival in the United States. Variants of the residual method have been widely used and are generally accepted as the best current estimates. See also Passel and Cohn (2011, 2010, 2008) and Passel (2007) for more details.

Then, having estimated the number of legal and unauthorized immigrants in the March CPS Supplements, we assign individual foreign-born respondents in the survey a specific status (one option being unauthorized immigrant) based on the individual’s demographic, social, economic, geographic and family characteristics. The data and methods for the overall process were developed initially at the Urban Institute by Passel and Clark (1998) and were extended by work of Passel, Van Hook and Bean (2004) and by subsequent work at the Pew Hispanic Center.

The final step adjusts the estimates of legal and unauthorized immigrants counted in the survey for omissions. The basic information on coverage is drawn principally from comparisons with Mexican data, U.S. mortality data and specialized surveys conducted at the time of the 2000 census (Bean et al. 1998; Capps et al. 2002; Marcelli and Ong 2002). These adjustments increase the estimate of the legal foreign-born population, generally by 1% to 3%, and the unauthorized immigrant population by 10% to 15%. The individual survey weights are adjusted to account for immigrants missing from the survey

The estimates for 2000-2008 use specially developed survey weights for the CPS to ensure consistency across the years in the underlying population figures. (See Passel and Cohn 2010 for a detailed discussion of the need for these weights and their development.)

Annual In-Flows of Immigrants from Mexico

Detailed, accurate estimates of flows back and forth across the U.S.-Mexico border have been difficult to develop, particularly since the flow has been largely unauthorized. Census and survey data from the United States provide measures of the Mexican-born population in the United States and of flows of Mexicans who are living in the U.S. on a more or less permanent basis. The volume of temporary, seasonal or circular migration is harder to assess accurately.

Total immigration during any interval can be estimated from a demographic identity—change in the foreign-born population equals immigration less foreign-born emigration and deaths:

FBt+nFBt = It,t+nDt,t+nEt,t+n

or, immigration equals foreign-born population change plus deaths and emigration:

It,t+n = (FBt+nFBt) + Dt,t+n + Et,t+n

These boundary conditions provide the framework for measuring flows into the U.S. from Mexico. Survey-based estimates of all elements are subject to various measurement issues, including undercount, definitional inconsistencies, sampling and other errors. Developing consistent measures of annual immigration involves coping with these problems.

Estimates were developed for three periods—1990-1999, 2000-2004, and 2005-2009—using a combination of data sources, assumptions and measurement techniques, depending on the nature of the available information.

Estimated Flows, 1990–1999

Based on the boundary conditions specified above, the total implied migration of Mexicans to the U.S. for 1990 through 1999 is about 5 million:27

The initial distribution of annual immigration flows for 1990 through 1999 is estimated as the average annual arrivals from the year of arrival data of the 2000 census and the 2001 and 2002 ACS (unadjusted for undercount). The annual totals are adjusted for undercoverage using information from the March 2000 CPS (Passel and Cohn, 2011) and then inflated further by 0.5% for each year before 2000 to account for mortality and emigration between arrival and 2000. The resulting distribution, which sums to 4.6 million, is then adjusted upward by roughly 9% to agree with the total implied by the boundary conditions.

Estimated Flows, 2000–2004

Total immigration for this five year period is based on arrivals observed in the March 2005 CPS for the 2000 2005 period, or 3.175 million new arrivals. The distribution of these arrivals across the period is based on a complex average involving distributions from several measures: CPS arrivals from the annual March supplements where the relevant years are fully observed (i.e., March 2002–2008), the same data from monthly CPS averages, residence one year ago estimates from the March 2001 2004 CPS, estimates based on annual change in arrival cohorts from monthly CPS data for 2000 2006 (Passel and Cohn, 2009), and averages across three ACS samples where the cohorts are fully observed (2002–2008).

Estimated Flows, 2005–2010

The total number of arrivals for 2005–2009 is estimated from two measures which give almost identical results of 1.65 million arrivals for 2005–2009: (a) arrivals for 2006–2009 from the 2010 March CPS plus 2005 arrivals estimated as the difference of the most recent arrival cohorts observed in the March 2006 and 2005 CPS supplements; (b) a survival calculation for 2005–2010 similar to the 1990 2000 calculation shown above but allowing for the large number of returned migrants observed in the Mexican Census of 2010. The distribution of this total to individual years for 2005–2009 uses an index based on the same measures as for the 2000–2004 period noted above. For 2006–2010, the distributions of apprehensions at the U.S. Mexico border and departures for Mexico as measured by ENOE are incorporated into the index. The estimate for 2010 uses the index value based on the three available measures (apprehension, ENOE and March 2011 CPS residence one year ago) in comparison with the index values for 2005-2009.

  1. The IPUMS-I microdata sample for the 2000 Mexican census was also used for some estimates.
  2. Datasets were downloaded from COLEF’s EMIF website: http://www.colef.net/emif/, accessed in November 2011.
  3. For more information on ACS sample sizes and methods, see http://www.census.gov/acs/www/methodology/sample_size_data/index.php
  4. The alternative dataset for 2005 not only includes the alternative weights, but also incorporates cases to represent the group quarters population for 2005.
  5. Some 3.575 million legal immigrants,1.490 million unauthorized, 0.035 million nonimmigrants with 4.5 million counted in the 1990 census.