10

Research Directions and Data Recommendations

A detailed review of the research literature upon which this report is built reveals that much is known about the economic and fiscal impacts of immigration. A rich portrayal of the roles that immigrants have played in recent U.S. economic history can be drawn, and short-run labor market and public finance outcomes can even be forecast reasonably well (Kerr and Kerr, 2013). But even with the theoretical and empirical advances of recent decades, some questions remain difficult to answer comprehensively and accurately. In some cases, research is constrained by a still emerging conceptual clarity; more often, however, it is hindered by data limitations. Data on immigrants and their descendants—on nativity, education, age and date of arrival, time spent in the United States, and legal status at present and upon entry—are central to analyses of the economic and fiscal impacts of immigration.

In this chapter, the panel recommends next steps for improving the data infrastructure necessary to support continued advances on the research topics detailed in this report. The data needed to study fiscal and economic impacts of immigrants are similar to the data needed to study their integration into society. Therefore, many of the recommendations presented here previously appeared in the report by our sister panel, The Integration of Immigrants into American Society (National Academies of Sciences, Engineering, and Medicine, 2015, hereafter “the Integration report”). In addition to presenting formal recommendations, we identify several opportunities to enhance available data but do not formally recommend them. While these data would be valuable to researchers, they do not rise to the

same level of importance or their collection may be less feasible than the data enhancements we recommend.

10.1 COUNTING AND CHARACTERIZING IMMIGRANTS AND THEIR DESCENDANTS

To understand the effects of immigration on society and the economy, it is necessary to know how many immigrants have arrived in the country, when they arrived, and from where. As discussed in Chapter 2, answers to these seemingly basic questions can be surprisingly difficult to obtain and will continue to be so without further improvement of data sources. Every Decennial Census from 1850 to 2000 included a question on birthplace (foreign-born respondents were also asked about country of birth), which allowed the size of the foreign-born population to be measured. Data on the foreign-born are also collected by the American Community Survey (ACS), a large household survey that replaced the long-form Decennial Census after 2000, and, since 1994, the Current Population Survey (CPS), which is designed for the primary purpose of monitoring labor market trends. These data sources provide information about basic demographic and socioeconomic characteristics—age, sex, marital status, employment status, occupation, income, earnings, and educational attainment—of the foreign-born. The foreign-born population includes permanent residents, persons on temporary work and student visas, and undocumented residents who entered the country either without inspection or have overstayed visas; however, neither the CPS nor the ACS identify the legal status of respondents. The Census Bureau also produces population projections that estimate the future size of the foreign-born population based on a set of demographic assumptions.¹

Although it is important to build into the nation’s statistical infrastructure the capacity to monitor progress of the foreign-born population, it is equally critical to do so for their U.S.-born children who, as native-born citizens, reveal a great deal about how new Americans are integrating into society and helping to shape the nation’s economic and demographic landscape. The ability to identify second generation respondents is extremely desirable for empirical analyses of both labor market and fiscal impacts of immigration. As with the foreign-born themselves, their children may on average attain different education and skill levels (often higher—see

___________________

¹ Because of the inconsistences in the Decennial Census series and the lack of counts of the second generation population, the Pew Research Center also produces projections, including separate projections for the second and third-plus generations, which are used for some of the fiscal impact estimates in Chapter 8 of this report. The Pew population series differs slightly from official census data because of methods of adjustment, estimation, and projection, but the differences are generally less than 1 percentage point, well within the margin of error.

Chapter 8), achieve different occupational outcomes, and generate at least slightly different fiscal impacts compared with the general population. In turn, their presence may affect employment rates and composition (either positively or negatively), as well as per capita earnings, taxes paid, and social program utilization—all integral to fiscal and labor market outcomes.

Thus, for analyzing the earnings and occupational integration of immigrants and their descendants, and for a range of other research purposes, a question on parental birthplace is needed for a large representative sample of the population. Such a question was first added to the 1890 Decennial Census² but was dropped for the 1980 and subsequent Decennial Censuses. In 1994, the CPS helped to ameliorate the situation by adding two questions about parental birthplace: “In what country was your father born?” and “In what country was your mother born?” The CPS is, however, not exactly comparable to the Decennial Census or to the ACS; it only covers the civilian, noninstitutionalized population, and it includes data on a different set of potential covariates. Massey (2010) provides a definitive discussion of immigration measurement issues and explains why a question about parents’ birthplace is crucially needed on the ACS. In so doing, he also notes the primary constraint inherent in the relatively small sample size of the CPS (compared to the ACS or the long-form Decennial Census): the CPS sample size is often inadequate to address questions about immigrants by nationality groups, and the problem is intensified for smaller geographic areas.³ As Massey (2010, p. 128) notes, the CPS allows one to “study second-generation Mexican immigrants in California, but is of little use if one seeks information about second-generation Koreans in Oregon—the sample will just be too small.” For cases in which the CPS is inadequate for studying subgroups, the ACS would often provide the sample size needed to do so. For this reason, and others cited above, a modification to the ACS is warranted:

___________________

² From 1890 through 1930, parental birthplace questions were asked of all census respondents. With the advent of sampling in the 1940 census, these questions were asked only to a subset of the population: for every 20th person (5%) in the 1940 census, for every 5th person (20%) in the 1950 census, for 25 percent of households in the 1960 census, and for 15 percent of households in the 1970 census.

³ The relevant part of the CPS (the March supplement) has a sample size of around 75,000 households, which yields, on average, information on more than 11,000 foreign-born households and 26,000 foreign-born individuals. The March supplement also significantly oversamples Hispanics and, to a lesser degree, Asians. A list of all the surveys that collect data on immigration can be found at the University of California, Berkeley, Population Center. Available: http://www.popcenter.berkeley.edu/resources/migration_data_sets/data_by_region.php [November 2015].

With such an enhancement to the ACS, fiscal analyses such as those reported on in Chapter 9 of this report would be more robust because more characteristics of the foreign-born and the second generation could be compared against the rest of the population at the state or substate level.

During this panel’s work estimating the fiscal impact of immigration, it also became clear that, in addition to asking about parental birthplace, it would be useful to have data on parents’ educational attainment. The absence of this information even in the CPS, which includes information on parental birthplace, means that both The New Americans (National Research Council, 1997) and the analyses in this report (Chapter 8)—along with many other studies—must rely on average intercohort education levels (a comparison of the mean values for different cohorts), a simplifying assumption that affects a large research literature, not just that on immigration. If microdata existed to compare individuals and their parents, estimates of intergenerational transmission of educational attainment and the determinants thereof would be much more precise.

As discussed below, some research questions about immigrants and their descendants are best addressed by tracking populations over a number of years. Longitudinal studies of the second generation are needed to provide information about their economic and social contributions, about their labor market and fiscal impact, and about how they integrate along various dimensions over time—essential aspects of the country’s overall immigration experience. Past academically sponsored efforts, such as the New Immigrant Survey (Jasso et al., 2006), have attempted to do this, but that particular survey was limited to legal immigrants arriving in certain years. A survey similar to the National Education Longitudinal Studies, but focused on a large second generation sample followed from early adolescence into adulthood, would enhance immigration research.⁵

___________________

⁴ This recommendation is replicated from the Integration report (p. 429, Recommendation 10-1).

⁵ Detailed, individual-level data of this kind, often required for capturing and analyzing processes as they unfold, require safe access that protects privacy and confidentiality.

In addition, the Integration report recommends that a number of currently operating national longitudinal surveys “should oversample the foreign-born, especially the smaller Asian and non-Mexican Hispanic groups that, when combined, make up a significant share of the immigrant population.” Existing models of how to oversample key populations can be found in a range of surveys, such as the National Health Interview Survey, the Panel Study of Income Dynamics, and the National Health and Nutrition Examination Survey (National Academies of Sciences, Engineering, and Medicine, 2015, p. 432).

10.2 INFORMATION ON LEGAL STATUS

A second major limitation of Decennial Census, ACS, and CPS data for studying immigration is that neither current visa status nor visa status at time of arrival are recorded, making it impossible to distinguish between lawful permanent residents (“green card” holders), persons on temporary nonimmigrant visas for work or study, persons with other types of visas, and persons who lack an official visa. As a result, it is common statistical practice to refer to the foreign-born population in a census or survey as “immigrants” even though such a categorization will typically include foreign students, various workers on temporary employment visas, those on temporary residence visas, and migrants who are not authorized to be in the country.⁶ For this reason, better data are needed on visa status, initially and currently, as well as on time and age at arrival (which is already collected).

There is considerable mobility across visa categories as well, and current visa status does not always predict who stays permanently. Legal status has been shown in a number of surveys to be a dynamic variable that changes over time, as immigrants’ circumstances change. As highlighted in the Integration report (National Academies of Sciences, Engineering, and Medicine, 2015, p. 430), “The attainment of legal status and eventual citizenship are likely to be crucial steps in the process of economic and social integration, yet researchers presently lack the means to model them.”

___________________

Such protections are a feature of the U.S. Census Bureau’s Federal Statistical Research Data Centers, which enable researchers—albeit with some level of burden relative to public access sources—to access and analyze microdata and small-area data (for details, see http://www.census.gov/fsrdc).

⁶ There are many types of temporary visas that permit people to reside (and sometimes work) legally in the United States—usually for 1 year or less, although some temporary visas can be renewed for several years. Temporary resident visas are issued for visitors; fiancés and spouses of U.S. citizens; entertainers, athletes, and religious workers; Canadian and Mexican professionals; business trainees; and others that are allowed to reside in the United States for short periods of time. See Chapter 3 of the Integration report for details on the various visa and other statuses for temporary and long-term entry to the United States.

Because there is no official count of persons who are in the United States without a valid visa—the unauthorized population⁷—an additional question should be considered for the CPS:

Following this guidance provides a good starting point but undocumented persons are likely to be under-enumerated in surveys and censuses. The purpose of the recommended pretest is to determine whether the inclusion of such questions might have a deleterious effect on survey participation. For these reasons, in addition to the “process of elimination method” suggested in the above recommendation, creative use of administrative and other kinds of data is desirable to identify immigrant populations of interest, such as the authorized and nonauthorized.⁹

It is also possible to tap into legalization programs to learn more about the subset of immigrants applying for citizenship. As an example of how this opportunity has been exploited in the past, the Integration report cites the 1986 Immigration Reform and Control Act (IRCA), which mandated a survey of immigrants who legalized. The survey, which collected data on “how they entered the United States, where they fit into the labor market, demographic characteristics, family composition, use of social services, migration behavior and origins . . . illuminated the behavior of a population for which there previously was little systematic information” (Integration report, p. 430). The potential of this kind of instrument points to a clear strategy for additional systematic data collection:

___________________

⁷ The expert consensus is that the unauthorized population peaked at approximately 12 million in 2007, then fell to about 11 million in the wake of the Great Recession (Baker and Rytina, 2013; Passel et al., 2013).

⁸ This recommendation is adapted from the Integration report (p. 430, Recommendation 10-2).

⁹Van Hook et al. (2014) presented evidence about coverage of the Mexican-born population in the 2000 U.S. Decennial Census and in the ACS using death and birth registrations and a net migration method. “For the late 1990s and first half of the 2000–2010 decade, results indicate that coverage error was somewhat higher than currently assumed but had substantially declined by the latter half of the 2000–2010 decade . . . [and] that U.S. census and ACS data miss substantial numbers of children of Mexican immigrants, as well as people who are most likely to be unauthorized: namely, working-aged Mexican immigrants (ages 15–64), especially males” (Van Hook et al., 2014, p. 699).

Legalization programs certainly create targeted opportunities to learn more about individuals who were previously living without legal status in a way that provides a window on the broader group; however, it is important for data users to recognize that those who legalize are a selected group that is not fully representative of their counterparts who have not legalized.

Currently, data on legal immigrants entering the United States and those applying for benefits such as naturalization collected by the Department of Homeland Security (including USCIS), the State Department, and the Office of Refugee Resettlement are generally limited to data items needed for processing cases. The collection of additional information would make it possible to maximize the research value of these administrative data and to allow specific questions of interest to be addressed.

Additional data, such as on occupation and education, could be collected from all applicants for lawful permanent resident status. Information on family members admitted at the same time could be linked and information on sponsors added. These additional data items could be collected from a sample of the people processed every year. A 10 percent sample of the admissions/naturalizations each year, for example, would generate a dataset with about 100,000 awards of lawful permanent residence and 75,000 naturalizations every year. Of course, as pointed out in the Integration report (National Academies of Sciences, Engineering, and Medicine, 2015, p. 431), such an expansion in administrative data collection only creates value if the information can be made available to researchers and the public in secure data centers.

___________________

¹⁰ This recommendation is adapted from the Integration report (p. 431, Recommendation 10-4).

Understanding of the unauthorized and other immigrant populations could be further enhanced by exploiting longitudinal data sources. This panel supports the idea behind the recommendation advanced in the Integration report (National Academies of Sciences, Engineering, and Medicine, 2015, p. 430) to add questions about legal status to a select set of longitudinal surveys that contain significant numbers of foreign-born respondents. The New Immigrant Survey, the Survey of Income and Program Participation, and the Los Angeles Family and Neighborhood Study are examples of surveys that include direct questions on legal status. This modification could be considered for the Panel Study of Income Dynamics, the National Health Interview Survey, the National Education Longitudinal Survey, and the National Health and Nutrition Examination Survey. However, careful pretesting would be needed to assess the potential impact on response rates overall, and of undocumented immigrants in particular, of asking respondents about legal status. The integrity of these very important surveys should not be risked unless it can be convincingly established that eliciting truthful answers about legal status from respondents will not create undue risks to the entire enterprise.

10.3 MEASUREMENT OF IMMIGRATION AND EMIGRATION PATTERNS

Longitudinal data are also essential for uncovering the correlates of a range of social and economic outcomes of immigration (National Research Council, 1996). Likewise, the ability to follow individuals and cohorts over time is crucial to understanding factors behind geographic movements—for example, those affecting emigration, circular migration, and interstate migration—and analyzing selection effects associated with these behaviors (in this case, the factors or characteristics that are causally linked with immigration and emigration). Given that the earnings, tax payments, or program use of those who stay are systematically different from those who leave, measures of return and circular migration are especially important for estimating long-term economic impacts.¹¹

For the same reasons, longitudinal data that are valuable for tracking changing legal status of individuals, return or circular migration, or changes in patterns of program use are also essential for projecting fiscal impacts with precision. In their discussion of return migration, Kerr and Kerr (2011) pointed out that analyses of fiscal impacts often assume that immi-

___________________

¹¹ The U.S. Immigration and Naturalization Service ceased publishing emigration data in 1958 because the data available were thought to be incomplete, but alternative estimates based on recent research suggest that current emigration levels are not insignificant (Van Hook et al., 2006).

grants remain permanently in the host country after arrival; public service use and taxes paid are then estimated on the basis of cross-sectional patterns. The authors conclude that, in order to “provide a better estimate of the mean effect and also characterize the heterogeneity in immigrant types,” calculations of both labor market and fiscal impacts need to consider rates of return migration and identify selective outflow (Kerr and Kerr, 2011, p. 69). This advice is followed in the forward-looking fiscal projections presented in Chapter 8, which incorporate population projections by the Pew Research Center that include adjustments to account for out-migration.¹²

Better data on remittances would also enhance immigration research. Remittances dampen the contribution of immigrants to aggregate demand in the host country while stimulating aggregate demand in the origin country into which the funds flow; by extension, some fiscal benefits in the host country attributable to immigrants may likewise be weakened (Kerr and Kerr, 2011). If questions on respondents’ own and parental nativity were added to an existing survey, such as the Survey of Consumer Finances and the Consumer Expenditure Survey, the resultant data could prove useful for refining understanding of spending and remittance behavior among immigrants. The fiscal accounting exercises in Chapters 8 and 9 build in adjustments to account for the impact of remittances on consumption and sales taxes paid; however, these adjustments were based on data for Germany because adequate U.S. data were unavailable.

10.4 EXPLOITING MULTIPLE DATA SOURCES

For a wide range of information needs underpinning immigration research, strategic linking of administrative datasets—on visa status for example—and other sources beyond traditional household surveys can greatly enhance the capacity to track variables of interest, particularly at the individual level, over time. USCIS and other federal agencies compile administrative data containing detailed information about immigrants, including flows of new arrivals by visa status and data on newly naturalized U.S. citizens. However, the published data are aggregated in a way that offers only very basic cross-tabulations. It is impossible to use these data for fine-grained analyses, which typically require micro-level data on individuals and the ability to link to additional information sources, such as aggregate data on localities.

___________________

¹² The cumulative return rates used in the analysis are segmented by age and by duration in the United States. The return rate is about .24 for immigrants in their first 10 years in the country and about .31 during the first 50 years after arrival. These estimates are within a percentage point or two of the return rates used in the forward looking fiscal analysis presented in The New Americans (National Research Council, 1997).

Sometimes key pieces of information cannot be gleaned from household surveys. An example, used in the estimation of state and local fiscal impacts, is the cost of bilingual education and of educating students for whom English is a second language (not necessarily in a bilingual education program). The costs of such programs cannot be estimated from a household survey because they are incurred by schools, not parents. The source of data used in this report for modeling the added costs for language-assistance instruction is a now fairly outdated study by the Urban Institute (Clark, 1994). Updated information would be useful for sharpening estimates of education costs associated with immigration.

Beyond the survey data realm, another action that would be useful for generating fiscal projections would be for the Congressional Budget Office to make its budget projection engine public and give users the ability to experiment with different scenarios to see how changes affect estimated fiscal flows, tax rates, the size of the national debt, etc. For federal fiscal estimates, such as those produced in Chapter 8 of this report, this capability would have provided the opportunity to generate additional scenarios and to flesh out more exhaustively how reasonable each one appears. Achieving this is a complex proposition, but the capability would benefit research projects estimating future fiscal impacts of various policies—immigration related or otherwise.

Exploiting multiple data modes also has the potential to advance research on employment dynamics. To quantify the mobility of workers, or the extent to which displacement of pre-existing workers occurs, longitudinal data that “measure layoffs, unemployment spells, changes of residence and occupational and industrial mobility” are critical (Longhi et al., 2008, p. 25). Record linkages between surveys and detailed administrative records are now available to study firm and worker interactions and status changes. For the United States, the pioneering Longitudinal Employer-Household Dynamics Program¹³ has proven highly useful for analyzing how labor markets adapt to changing circumstances and, in so doing, has expanded opportunities for more sophisticated studies of employment effects associated with immigration inflows.¹⁴ In general, research in the United States has more frequently examined wage impacts than employment effects; European scholars have given more attention to analyzing employment impacts.¹⁵Foged and Peri (2015) analyzed labor market outcomes of low-

___________________

¹³ For details, go to http://lehd.ces.census.gov/research.

¹⁴Mouw et al. (2012) and Rho (2013) examined worker displacement in high immigration industries using evidence from the Longitudinal Employer-Household Dynamics Program.

¹⁵ As reviewed in Chapter 5, there has been some work in the United States on employment impacts. Smith (2012) analyzed the impact of immigration on hours worked of low-skilled native-born workers and found that the largest negative effect was on teenagers.

skilled natives in response to an inflow of low-skilled immigrants using longitudinal employer-employee data from Denmark.

Another area in which multiple data sources could advance research on the impact of immigration on wages and employment is in measuring capital formation. As discussed in Chapter 5, the demand for labor and the capacity of the economy to absorb new workers, including immigrants, is strongly influenced by the speed at which firms invest and adjust their capital stock and production technologies. Assumptions are often made or implied about this process which, in certain kinds of models, strongly influence wage and employment impact estimates. At this point, there is little empirical basis for these assumptions because the temporal characteristics of how capital formation occurs in response to changing factors of production is an under-researched topic (Longhi et al., 2008, p. 25). Better microdata on investment and capital stock at industry and regional levels are needed and might be supplemented by a variety of non-survey-data sources such as firm administration records or commercial databases.

Long-term multisource data projects are also important for studying economic and social mobility—a topic that has recently gained heightened visibility among researchers, policy makers, and the general public. Concerns about growing income and wealth inequality and about the health of the “American dream” have spurred research into intergenerational issues, which often have even more acute implications for immigrants and their descendants. The Integration report points out that “matched individual-level records from Decennial Censuses (and the ACS) with income data from Internal Revenue Service and the Social Security Administration would allow for longitudinal studies of the socioeconomic progress of immigrants in American society and allow for the measurement of both intracohort change and intercohort change (for cohorts based on time of arrival in the United States) for successive waves of immigrants.” Additionally, “matched Census and USCIS records would allow for in-depth studies of pathways to legalization and also the impact of legal status on socioeconomic outcomes of individuals and their children”¹⁶ (National Academies of Sciences, Engineering, and Medicine, 2015, p. 431). This opportunity should be pursued:

___________________

¹⁶ Similar data are collected in the French Permanent Demographic Sample (EDP), which is a pioneering longitudinal database maintained by the French National Institute of Statistics and Economic Studies, a central government agency located in Paris. The EDP is a panel survey based on immigrant arrival and census data that comprises a 1 percent sample of all immigrants that have entered France since 1967. The panel database includes information on immigrants at the time of arrival, linked to the General Population Census of 1968 and later censuses. It provides a rich database on the social and economic adjustment of immigrants over recent decades. A study by Richard (2013) provides an example of the usefulness of EDP data for studying labor outcomes.

The American Opportunity Study (AOS) is a new project, still under way, that takes as its goal to digitize and link data across Decennial Censuses, the ACS, and other administrative sources (such as Internal Revenue Service datasets) for the purpose of studying social mobility and related topics such as the following (Grusky et al., 2015):¹⁸

• Parent-child social mobility across a variety of dimensions (income, education, occupation) and with repeated measurements
• Social mobility within small geographic units
• Three-generation analyses (and beyond)
• Subgroup analyses (e.g., immigrants from specific countries or regions)
• Study of complex families (distinguishing social, biological, and financial parents)
• Intergenerational inheritance of program participation

A key topic motivating the AOS project is to improve the measurement of intergenerational changes in the immigrant population, ultimately improving the evidence base for policy.

Due to its high profile and its centrality among policy issues, research will continue on immigration regardless of whether the changes recommended in this chapter are implemented, and much of the focus of this research will be on the fiscal and economic consequences topics covered in this volume. However, initiatives such as the AOS and others that create a coordinated data infrastructure will, if successful, greatly enhance these

___________________

¹⁷ This recommendation is replicated from the Integration report (National Academies of Sciences, Engineering, and Medicine, 2015 p. 431, Recommendation 10-5).

¹⁸ The linkages across Census Bureau and administrative data will be designed to promote social, behavioral, and economic research in a way that creates savings on survey costs, improves data accuracy, and increases the ability to understand the long-term consequences of economic and social change. A longitudinal panel of the population, with identifiers for immigrants and later generations, could be constructed, and research using it could be conducted in restricted data environments such as the Census Bureau’s Research Data Centers (National Academies of Sciences, Engineering, and Medicine, 2015).

research efforts. In this chapter, the panel has briefly identified next steps for pushing the knowledge frontier forward so that a report published 20 years from now will be able to present an even more comprehensive assessment of how immigration contributes to the economy and affects those engaged in its activities.

This page intentionally left blank.