Abstract
Large low-skilled immigration flows influence both the distribution of local school resources and also local relative wages, which exert counterbalancing pressures on the local return to schooling. I use the National Education Longitudinal Study (NELS:88) and U.S. Census data to show that low-skilled immigration to an area induces local natives to improve their performance in school, attain more years of schooling, and take jobs that involve communication-intensive tasks for which they (native English speakers) have a comparative advantage. These results point out mechanisms that mitigate the potentially negative effect of immigration on natives’ wages.
I. Introduction
Immigration is a very important feature of many local labor markets in the United States. In 12 of the largest 25 cities in 2009, the foreign-born share in total populations was greater than 20 percent (U.S. Census Bureau 2012). Immigrants potentially influence the lives of the native-born population in many ways, including the likelihood of getting a job, wage offers, local prices, migration incentives, and schooling environments. Such relationships are important for public policy because they are potentially large and also because government policies like visa granting directly influence the number of immigrants in the country.
This paper investigates the impact of immigration on the human capital investment decisions of native-born youth. I focus on the effect of immigrants with relatively low education, a particularly important group in the United States.1 Betts (1998) provides a useful framework for thinking about the effect of immigration on natives’ education through two channels. The first is through the quality of schooling. A large local inflow of low-skilled immigrants tends to reduce the schooling resources available for natives—for example, by shifting teachers toward English-proficiency classes. Diminished school resources reduce the value of education to natives and induce them to get less of it.2
On the other hand, recent immigration has increased the market supply of low-skilled workers and should in theory put downward pressure on wages and employment probabilities for low-skilled residents in many areas. To the extent that low-skilled workers complement the productivity of high-skilled workers, wages in jobs requiring more education may rise. Both mechanisms increase the return to education, and native-born youth in the area with more immigration may have a strong incentive to acquire more schooling.
This paper demonstrates empirically that low-skilled immigration induces native-born youth to increase their investments in human capital. I study behavior of nativeborn children in the National Education Longitudinal Study (NELS:88) responding to immigration flows that I measure in the U.S. Census. I find that low-skilled immigration flows induce local native youth to increase their high school attendance, grades, test scores, and the academic rigor of their curricula. The results use plausibly exogenous variation in local immigration (based on pre-existing ethnic enclaves) and are robust to controls including mother’s education and also characteristics of the student’s school. In addition, I find that low-skilled immigration induces native-born students to attain more secondary and postsecondary schooling.
Finally, I study NELS:88 respondents’ early-career jobs to test a recent hypothesis (Peri and Sparber 2009): low-skilled immigrants, who have relatively low English-language skills, induce low-skilled natives to invest in communication-oriented job skills rather than manual skills. Peri and Sparber (2009) describes the relationship in an equilibrium model of behavior and finds support for it in U.S. Census data. I find that native-born respondents to the NELS:88 with more early immigration exposure choose jobs where they use more word processing and email and perform fewer manual tasks. From changes in natives’ job tasks, I infer that natives invest in communication skills as a way to differentiate themselves from low-skilled immigrants in the local labor market. This finding complements Peri and Sparber (2009) and points to another way that native workers change their behavior to mitigate wage losses due to immigration.
The empirical results in this paper emphasize a potential benefit of immigration that is largely overlooked in the research literature about costs and benefits of immigration. In particular, public policies that let in more immigrants will yield increased natives’ skills. Such increased human capital investment is socially desirable if native-born youth underinvest in their own schooling or if education generates positive externalities.3 In addition, by augmenting their human capital in response to immigration, native-born workers mitigate the effect of immigration on their wages.
My empirical findings also relate to the estimation of immigration’s effect on natives’ wages, a literature with mixed results. One major strand of the literature (for example, Card 2005) demonstrates that native-born workers in high-immigration areas do not earn substantially lower wages than similar workers in low-immigration areas.4 A competing strand of the literature (for example, Borjas 2003) estimates large elasticities of substitution between immigrant and native-born workers, which imply large wage effects. After showing my empirical results, I note that native children’s increased investments in human capital in response to immigration imply that previous studies have mismeasured labor inputs when estimating substitution elasticities. The result is an upward bias: Researchers infer too much substitutability and thereby overly large effects of immigration on natives’ wages. So my findings about natives’ human capital investment responses to immigration also highlight a downward bias in estimates of the effect of immigration on natives’ wages (among studies that estimate substitution elasticities). If this bias were corrected, then the literature’s most-negative estimates of immigration on natives’ wages would move toward zero.
II. Prior Literature on Immigration and Natives’ Schooling
Most of the previous literature on immigration and natives’ education in the United States uses Census data to measure effects on natives.5 Betts (1998) documents a negative relationship between state-level immigration and the probability that native-born black and Hispanic students complete high school in the 1980 and 1990 Censuses.6 Betts (1998) compares schooling of native-born adults (aged 19–25 and 24–30, respectively) with immigration in the state where those natives live at the time of the Census. But schooling decisions should be more influenced by immigration when natives are children, and selective migration may confound the relationship between immigration and education levels of adults.
Hunt (2012) uses 1940 to 2000 decennial Census samples to expand upon the work of Betts (1998) and Betts and Lofstrom (2000): It assigns immigration flows to natives at younger ages based on birth states, distinguishes among more-and less-educated immigrants, measures natives’ education consistently over time, and instruments for state-level immigration flows (with lagged immigrant origins). It finds that the presence of immigrants in a state raises the probability that natives attain 12 years of schooling, with particularly large effects in the black native-born population. Also with U.S. decennial Census data, Jackson (2009) shows that college enrollment rates among the native-born increase with the entry of more low-skilled immigrants to the state’s labor market.7
Such analysis with Census data clearly benefits from very large samples and the ability to measure changes over time (for example, comparing differential changes in California and Oregon across the 1980, 1990, and 2000 Censuses). Large samples allow separate estimation for different racial and ethnic groups (for example, to discern whether immigration differentially affects black and white natives). On the other hand, Census data do not provide ideal outcome measures, and the public-use data do not allow very precise location of individuals.
In this paper, I exploit features of the NELS:88 data set that improve upon analyses of Census data in several ways. While prior studies focus on schooling attainment, the NELS:88 allows me to use information about students’ attitudes, expectations, and specific experiences in school to study the mechanisms behind immigration’s effects on schooling attainment (for example, curriculum choices, study effort). In addition, the NELS:88 collects information from respondents and their parents so I can control for parents’ education when investigating other determinants of schooling choices. The NELS:88 surveys principals and teachers in respondents’ schools so I can control for specific school resources and the student body. The NELS:88’s focus on education translates into more accurate measures of diploma receipt and specific certification in comparison to less-informative Census information about years of schooling. In particular, the NELS:88 distinguishes between a high school diploma and the General Educational Development (GED) credential and also collects information about students’ curricula (for example, Advanced Placement and vocational classes).
Another weakness of Census-based analyses is that they match respondents with local immigrant flows that may not be relevant to them. Studies tend to associate with respondents the state-level immigration flows where they were born or where they live in their 20s. There are two problems with this. First, young people are geographically mobile, and their birthplace or residence in their 20s might be quite different from their residence when making educational choices. Second, some states are very large and contain locations with differing immigration histories. The more precise timing and location information in the NELS:88 allow me to make a more informative match between immigration waves and natives. Specifically, I measure immigration flows facing natives in local labor markets rather than states, and I match these immigration flows to the place where NELS:88 respondents attend the eighth grade.8
III. Empirical Strategy
The empirical goal in this paper is to estimate the effect of low-skilled immigration on human capital investments among young native-born residents nearby. In practice, I regress measures of human capital investment or attainment on local immigration flows and control variables that might influence schooling decisions and could be incidentally correlated with immigration. The basic regression equation explaining individual human capital (H) investment is:
(1)
I investigate human capital investment decisions (Hi,s,c) in three categories: in-school investments (attendance, curriculum choices, grades), educational attainment (for example, graduating from high school), and job tasks. Native-born youth have a comparative advantage in English-based communication tasks, and immigrants have a comparative advantage in manual tasks (Peri and Sparber 2009). I infer investment in job skills from the task-intensity (communications and manual) of native-born workers.
Human capital investment (H) of individual i is influenced by the immigration flow to i’s origin c (say, city), which is measured by ∆Ic. Individual characteristics like sex and race influence school decisions and might vary across locations, so I control for them in Xi (which also includes a constant term). In some specifications, Xi also includes mother’s education, which is a strong predictor of schooling and might also be related to local immigration (say, if highly educated mothers leave locations with high low-skilled immigrant flows). The vector Wc includes characteristics of the individual’s origin: region, population size, and metropolitan status. To control for local features that influence schooling decisions (other than recent immigration), I also control for the educational distribution of adults living in the origin.
Betts (1998) notes that immigration may decrease educational attainment of natives, as immigrant children use up resources at the school or school district level. Such an effect would reduce the quality of school and thereby its return to natives. Some specifications in Betts (1998) control for state-level school resources (pupil-teacher ratio), but it suggests that school-level controls would be preferable in testing the effect of immigration on natives’ educational attainment. I control in some specifications for school-and district-level measures of educational resources (Zs). One control is the percent of classmates in the respondent’s school who have limited proficiency in English, so the regression compares students in schools facing similar resource needs from immigrants. This should help account for the potential that parents in high-immigration areas choose their children’s schools to reduce exposure to immigrants.9 Additional controls in Zs are indicators for the school being Catholic or private and non-Catholic, school enrollment, school student-teacher ratio, percent of the school’s teachers with post-bachelor’s degrees, average salary at the school for a starting teacher with a bachelor’s degree (adjusted for local cost of living), school-year term length (in hours), and school district expenditures per student.
Studies using Census data tend to take decade-long differences to wipe out all long-term characteristics of states or MSAs. This strategy is not available to me because the outcome variables in my data pertain to a single cohort. However, the control variables in Xi, Wc, and Zs should capture many of the potential schooling shifters that might also be correlated with local immigration flows. Indeed, some of the variables included are not available with Census data (for example, school characteristics, mother’s education for adult respondents). Still, there may be unobserved location-specific features that shift both immigration and natives’ schooling decisions.
In area-based studies of the effects of immigration on wages, there are always concerns about omitted variables bias. In particular, local labor demand shifters likely increase immigration and wages and may be unobserved in a regression. Similar bias may be present when associating educational attainment and local immigration, although the endogeneity story is less compelling than with wages. Nevertheless, there could be unobserved local traits that affect both immigration and human capital investment of local natives. For example, current wage growth may be unmeasured (or mismeasured), but it could yield both higher immigration and less educational attainment among natives by raising the opportunity cost of time in school. With such endogeneity in mind, I estimate specifications that instrument for recent immigration flows with origins of earlier local immigrants and nationwide immigration by origin. Bartel (1989) demonstrates that the strongest predictor of where U.S. immigrants choose to live is the prior presence of members of the same ethnic group. This is most true of less-educated immigrants, the focus of my study. The idea of using such behavior in an identification strategy comes from Altonji and Card (1991) and is employed frequently in the economics literature.
The specific instrument I use for immigration flows follows Smith (2012). Let c index locations of residence and o denote an immigrant’s region of origin. Io,c,t is the number of immigrants from origin region o living in location c in Census year t. Ic,t is the total number of immigrants living in location c (all origins) at time t, and Io,–c,t is the total number of immigrants from region o in locations other than c at time t. The instrument is:
(2)
The instrument identifies variation in immigration flows across locations using nationwide trends in immigration by origin (Io,–c,1980 and Io,–c,1990) and the origins of local immigrants in the previous period (Io,c,1980/Ic,1980 and Io,c,1970/Ic,1970).10 A location would have a high predicted immigration flow if it has a relatively large pre-existing share of immigrants from recent sending countries. Such variation is plausibly unrelated to contemporary (1990) economic conditions that motivate immigrants to settle locally and also motivate young native residents to invest in education.
Note that location differences in the instrument do not arise from pre-existing differences in local immigration levels or growth. Rather, they arise from differences in the origins of prior immigration flows. The instrument predicts higher immigration flows among locations with relatively large shares of their immigrant populations from regions that subsequently sent many immigrants. Using region shares in the previous immigrant population normalizes by prior immigration levels and growth. For example, a location with very low immigration a decade ago may have a large predicted immigrant flow if a large share of its (small) earlier-period immigrant population was from a region that sent many immigrants later.
IV. Data
A. NELS:88
I use the National Education Longitudinal Study (NELS:88) to measure human capital investments of U.S. native youth. The NELS:88 was administered by the U.S. Department of Education. It began with a representative sample of eighth graders in U.S. schools in 1988. Followup surveys were fielded in 1990, 1992, 1994, and 2000, and include responses from students, parents, teachers, and school administrators. Using school and residence zip codes in the restricted-access version of the NELS:88, I assign to each sample member the immigration flow in the local labor market where she attended the eighth grade (when respondents were about 14 years old).
The NELS:88 provides quite a large sample for a longitudinal survey. The base-year survey reached almost 25,000 students. The next two waves of the survey are similarly large. I use these large samples when measuring respondents’ behaviors and attitudes during secondary school. The final followup survey (in 2000) includes a subsample of previous respondents (about 12,140 of them), and I use this sample when measuring final educational attainment and early-career job characteristics.11
My focus is on the reaction of local native youth to immigration flows. Some of the NELS:88 sample members are immigrants themselves. I select only those NELS:88 respondents who were born in the United States (or Puerto Rico) and whose parents were born in the United States (or Puerto Rico). Hence, the sample excludes first and second generation immigrants. I do not restrict the sample by race or ethnicity.12 I also keep respondents who move across local labor markets at any time during the sample, so educational investments and work experiences may not occur where the initial immigration flow was experienced.
B. Measures of School Efforts and Success
The NELS:88 includes a variety of questions about specific attitudes and behaviors while students are in secondary school. Local immigration might change attitudes toward school and work, especially if local native youth believe they will need to compete with immigrants for future jobs. To measure such attitudes, I use responses to three questions in the tenth grade survey. The first is an indicator that the student strongly agrees that education is important to get a job later, which might be more likely among natives who anticipate future competition in labor markets for high school dropouts. The other two expectations variables are indicators for students being sure they will graduate from high school and sure they will continue education after high school.
Native students may attempt to differentiate themselves from abundant immigrant labor by trying harder in school. To measure such behavior, I collect information about school attendance and hours spent on homework outside of school. The eighth grade survey asks respondents how many days they were absent from school and how frequently they skipped classes. Using the final followup wave of NELS:88 (age 26), I regress an indicator for graduating from high school on indicator variables for absence and skipping frequencies. For each respondent in the full sample, I predict a graduation probability using their responses about absences and skipping class. I use this as a school attendance composite variable in the analysis. For an additional measure of school effort, I also use students’ self-reported hours of out-of-school homework per week in the tenth grade.
Students may also choose their school curricula conditional on their expected work environment. The NELS:88 identifies students who took Advanced Placement (AP) classes, which are academically rigorous and useful in transitioning to college. The surveys also ask students whether they have taken a vocational class. I use indicators for taking any AP classes and any vocational classes to identify students’ curriculum choices. I view taking an AP class as a proxy for effort in academics and the intention of continuing education past high school. I infer the opposite intention for students taking vocational classes.
To measure success in school—a partial indicator of human capital investment—I collect information about grades and test scores. NELS:88 students reported their grades in math, English, history, and science classes during eighth grade. Their responses for each subject were “mostly As,” “half A and half B,” “mostly Bs,” etc. and also “not taking subject.” With the final followup sample of 26-year-olds, I regress an indicator for graduating from high school on indicators for each response about grades and use the resulting coefficient estimates to predict graduation conditional on grades for each student. This is a grades composite to measure academic success in high school. Finally, I use measures of test scores in eighth and 12th grades to infer schooling investments. In each grade, the NELS:88 reported standardized test scores in reading and math. I calculate the percentile of each student’s scores in the sample’s test score distribution (separately for reading and math tests in eighth and 12th grades). The test score measures are the averages of a student’s reading and math percentile scores in eighth and 12th grades.
C. Measures of Educational Attainment
The first measure of education attainment that I use is receipt of a high school diploma. The NELS:88 separately identifies students who graduated from high school and those who obtained a GED credential, and I treat GED holders as high school dropouts.13 Respondents could have received their high school diploma at any time before the final NELS:88 interview in 2000 (when respondents were mostly 26 years old). I also investigate a second measure of educational attainment: school attendance after high school. The NELS:88 asked respondents whether they had attended any “college, university, or vocational, technical or trade school for academic credit.” Many of the postsecondary attenders did not finish their program. For a third education measure, I create an indicator for the receipt of any postsecondary certificate, license, or degree (associate’s, bachelor’s, or graduate degrees).
D. Measures of Job Tasks
Peri and Sparber (2009) studies “communication” and “manual” tasks of less-educated workers. The NELS:88 allows me to provide complementary evidence. Instead of inferring tasks from workers’ occupations as in Peri and Sparber (2009), I observe direct responses about tasks that workers perform on the job. The NELS:88 asked respondents about job conditions where they work at the time of the 2000 survey or their most recent job if not currently working. Workers responded “never,” “occasionally,” or “a lot” to multiple prompts about tasks they did at work. I generate indicators for workers doing “a lot” of “read letters, memos, or reports,” “write letters, memos, or reports,” “use computer,” “word processing,” “send and receive email,” “search the Internet,” and “measure or estimate the size or weight of objects.” I interpret reading and writing and email, word processing, and Internet computer use as communication tasks. I interpret the estimation of sizes and weights of objects as manual tasks (like spreading mulch on a landscaping job).
To get a sense for what responses about job tasks imply, Table A2 categorizes specific occupations by task responses. The first column shows the percent of NELS:88 respondents in each occupation that read letters and memos “a lot” at work, and the second column shows the percent in each occupation that estimate the size and weight of objects “a lot.” The third column takes the ratio of the first two columns, a measure of relative communication-oriented tasks in each occupation. Legal professionals are clearly intensive employers of communication tasks. The occupations near the bottom of the table include manual jobs (for example, cooks, laborers, farm laborers), consistent with my interpretation of the “estimate size and weight” responses as implying manual tasks. These occupations are also common among immigrant workers. A native-born worker wanting to avoid labor market competition with low-skilled immigrants would prepare for other jobs, where he would probably “estimate size and weight” less and “read letters and memos” more.
E. Control Variables
The individual controls (Xi) include an indicator for female. Race/ethnicity categories are black, Hispanic, Asian, and American Indian (non-Hispanic white is the omitted category). I measure parental education as years of schooling attained by the respondent’s mother. I focus on mothers because I am more likely to have a NELS:88 respondent’s mother’s education than father’s education. Mother’s and father’s education are highly positively correlated.
The vector Wc includes characteristics of the individual’s eighth grade local area: region, population size, and metropolitan status. I control for region with indicators of the four Census regions: Northeast, Midwest, West, and South. To control flexibly for the eighth grade location’s size, I use indicators for six categories: small town rural, small nonmetro, larger nonmetro, small metro, medium metro, and major metro.14 To control for local features that influence schooling decisions (other than recent immigration), I also control for the percent of adults in the 1990 Census with less than high school education and the percent of adults with a bachelor’s degree or more education.
In some specifications, I control for characteristics of the NELS:88 respondent’s eighth grade school (Zs). School administrators of NELS:88 schools were asked how many students in their school’s eighth grade cohort had limited English proficiency, which I interpret as a proxy for immigrants’ needs at the school level. In practice, I use indicator variables for 11–20 percent, 21–30 percent, and 31 and higher percent eighth graders with limited English proficiency. The omitted category is 10 percent or less. Also from the school administrator survey, I create indicators for the school being Catholic or private and non-Catholic, school enrollment (number of students), school student-teacher ratio, and percent of the school’s teachers with post-bachelor’s degrees. The school administrator reported the average salary at the school for a starting teacher with a bachelor’s degree, and I divide this value by an index I created to measure housing costs in the area relative to the rest of the country.15 I measure the school year’s term length in hours by multiplying days and hours per day. I collect the school district’s total expenditures per student from the Common Core of Data (National Center for Education Statistics 1990).16
F. Location Definition and Local Immigration Measure
In this paper, local areas are commuting zones (CZs), which are collections of adjacent U.S. counties.17 Studies of immigration in the United States commonly analyze states or metropolitan statistical areas (MSAs). In contrast to states, commuting zones are good approximations of self-contained local labor markets. Their boundaries are very similar to metropolitan statistical areas (MSAs) in cities but CZs describe local markets in rural areas as well. Similarly skilled people living in the same CZ (for example, immigrants and natives) apply for and work in roughly the same jobs. Outside the CZ, jobs are mostly out of commuting distance and would require a relatively long-distance move to accept.
For the purpose of assessing immigration’s effect on native human capital investment, the immigration measure could be either a stock or a flow. The presence of many immigrants in the local labor market may induce native youth to invest more in education. On the other hand, recent growth in immigration may be more salient than current stocks in influencing natives’ decisions. I focus here on immigration flows.18
That is, I associate local natives’ skill investment decisions with recent increases in the local foreign-born population. One reason is that I expect large increases to be more salient (noticed by local natives) than large stocks. In addition, large increases probably imply more about future conditions than large current stocks imply. Suppose there are two cities: A and B. City A’s population is 5 percent foreign-born in 1980 and 10 percent foreign-born in 1990. City B’s population is 10 percent foreign-born in both 1980 and 1990. It seems reasonable to expect, based on trends, that in later years, City A will have the larger share of immigrants. Hence, students attending high school in City A in 1990 will have a greater incentive to invest in schooling because they reasonably expect greater competition with immigrant laborers than do students in City B. The alternative stock-based immigration measure would treat City A and City B identically and miss the important dynamic incentives that students face when investing in skills with future payoffs.
I use U.S. Census data in the Integrated Public Use Microdata Series (IPUMS) (Ruggles et al. 2010) to count immigrants by commuting zone (CZ). Let c index CZs of residence. Ic,t is the number of (low-skilled) immigrants in CZ c in Census year t. The specific measure of a local immigration flow is:
(3)
I collect data from the 1970, 1980, and 1990 Censuses.19 I identify the commuting zone (CZ) where each respondent lives using the county group of residence variables in IPUMS.20 Immigrants are respondents who were born outside of U.S. territories and either a naturalized U.S. citizen or not a citizen.21 To focus on relatively low-skilled immigration, I select only immigrants with a high school education or less. Evidence that high school dropouts and high school graduates are close-to-perfect substitutes with each other but imperfect substitutes with college-educated workers motivates this working definition of “low-skilled” (Card 2009).
Table 1 shows summary statistics describing immigration in the 741 CZs. The first row displays the distribution of 1990 CZ immigrant shares. Not surprisingly, there is a large variance: Some CZs have almost no immigrants while in some CZs immigrants account for more than 20 percent of the population. The second row describes the percent less-educated immigrants (with a high school degree or less) in total CZ population. Again, the variance across CZs is large. The third row shows that less-educated immigrants make up a large share of immigrants in all CZs and the majority in most CZs. The fourth row of Table 1 documents flows of low-skilled immigrants between 1980 and 1990, which is the main independent variable in the analysis below. Most CZs in the 1980s actually experienced reductions in the number of low-skilled immigrants, but there were some CZs with very large increases. The very large percent increases included CZs with both large and small populations so the variety across CZs is not just a consequence of tiny immigrant populations doubling, for example.
The rest of the table describes measures of the sample sizes used to measure local immigration. I cannot always report sample sizes for CZ means because they are weighted averages of PUMA means. Instead, I report the distribution across CZs of a sample size approximation that I call , where c indexes CZs, wpc is the share of CZ c population that is in PUMA p, and np is the PUMA p sample size. Appendix 3 derives and provides justifications for using , including the fact that is the same as the CZ c sample size when it can be observed. The bottom three rows of Table 1 describe the distribution of across CZs in 1970, 1980, and 1990. The sample sizes used to calculate CZ averages are in the thousands, which I interpret as large enough for acceptable precision. Sample sizes are somewhat large even for very small CZs because, in such cases, I assign the average of a relatively large PUMA to several small CZs that it contains. As long as such very small adjacent CZs are similar to one another, I expect their immigration measures to be reliable.
G. First Stage Results: Explaining Local Immigration Flows with Prior Immigration
Table 2 shows that the prior immigration instrument is a strong predictor of immigration flows. The observations are NELS:88 respondents. The dependent variable is the actual immigration flow (∆Ic,1990) they experienced. In addition to the instrument , all specifications include sex and race/ethnicity variables and characteristics of the respondent’s CZ: indicators for urbanicity and region, percent of the 1990 adult population with a bachelor’s degree, and percent of the 1990 adult population with less than high school education. The local education distribution is meant to capture potential local traits other than immigration flows that shift human capital investment of locals. Some (second-stage) specifications below include mother’s education or school quality measures, so Columns 2 and 3 include these variables. In all three specifications of Table 2, predicted immigration flows are strongly associated with actual flows between 1980 and 1990. The F-statistics for the instrument’s coefficients equaling zero are above 100: This is a strong instrument.22
H. Summary Statistics from the NELS:88
Table 3 displays average characteristics of NELS:88 sample members. Column 1 shows each variable’s sample size (rounded to the nearest ten for confidentiality). Sample sizes differ because of missing data for some variables and because some variables (for example, educational attainment) are measured in the smaller (subsampled) final followup survey. Column 2 in the first panel of the table shows that half of the sample is women and most respondents are white and not Hispanic (the omitted race/ethnicity category). The average respondent’s mother had a little more than 13 years of school. Columns 4 and 6 break down NELS:88 respondents’ characteristics by the 1980–90 low-skilled immigrant growth rates in their eighth grade CZs. The differences (in Column 7) are somewhat small, except that the share of Hispanic respondents in high-immigration CZs is much higher than in low-immigration CZs. Because much of the contemporary immigration was from Central America, and ethnic groups tend to cluster near each other, this is not surprising. Mother’s education is also higher in places with more immigration. I control for all of these demographic and background variables when assessing the relationship between local immigration and natives’ schooling levels.
The second and third panels of Table 3 describe NELS:88 respondents’ attitudes and behaviors in school. Students in high-immigration CZs think they are more likely to go to college, do more homework, take more AP classes, take fewer vocational classes, and get higher grades. This is consistent with the hypothesis that local natives distinguish themselves from low-skilled immigrants by attaining more education. Of course, Table 3 also shows native-born youth in higher-immigration areas attending less school. In addition, the simple differences in means mask potential confounding factors and other explanations. The mean differences do not control for student demographics and backgrounds that surely influence schooling expectations and efforts. In addition, they do not account for potential local factors that both induce low-skilled immigration and raise the return to schooling of local natives, like a local positive shock to labor demand. Empirical specifications in the next section address both of those issues.
The fourth panel of Table 3 describes highest schooling attainment of NELS:88 respondents. The difference in Column 7 shows that native-born eighth graders in high-immigration CZs stay in school longer than those in lower-immigration CZs. The difference is statistically indistinguishable from 0 for high school graduation, but the likelihood of getting postsecondary schooling increases as local low-skilled immigration increases. From the lower panels of Table 3, native-born workers from high-immigration origins tend to read somewhat more on the job, use computers more frequently, and use fewer manual tasks. These mean differences are consistent with natives differentiating their skills from local immigrants, but they could also reflect other features of CZs that are incidentally correlated with low-skilled immigration. I control for such potential confounding factors in specifications below. The final row shows that the majority of respondents live in the same CZ in eighth grade and when they are 26 years old and that those in higher-immigration areas are less likely to move away.
V. Results About Immigration and Natives’ Efforts and Success at School
This section describes the relationship between local immigration flows and students’ efforts in secondary school. I exploit the rich information about students’ experiences in the NELS:88 to illuminate mechanisms behind immigration’s role in natives’ education attainment. Overall, native-born students in relatively high-immigration CZs appear to invest more in academics.
Table 4 displays results from ordinary least squares (OLS) and two-stage least squares (2SLS) regressions with various specifications and dependent variables (Equation 1). Each cell in the table reports the coefficient (and standard error) on the eighth grade CZ low-skilled immigration flow. Each row describes a different dependent variable, and columns contain different specifications. The first four columns reflect the sample of all respondents with nonmissing data for these variables measured in secondary school. The sample sizes are quite large (above 10,000) because these variables are measured prior to subsampling for the fourth followup sample. Column 1 of Table 3 illustrates the range of sample sizes for these regressions (for example, larger when measuring attendance and smaller for the 12th grade test score). Columns 5 through 7 of Table 4 select only NELS:88 respondents whose mothers had no more than 12 years of schooling. Children of less-educated parents are likely to compete in labor markets with less-educated immigrants, so their behaviors are of particular interest.
The first panel of Table 4 reports the effects of local immigration flows on attitudes and expectations of native-born tenth graders. The first row includes little evidence that immigration increases the extent to which tenth grade natives think education is important for their careers. The first column’s coefficient (0.0025) reflects a baseline OLS specification that controls for local immigration, individual sex and race/ethnicity, and eighth grade CZ controls, while the second column’s coefficient (0.0281) is from the analogous 2SLS specification that instruments for immigration. These coefficients are not statistically significantly different from zero.
One standard deviation of the low-skilled immigration flow distribution across CZs is about 0.43 (53.7 percent). It is informative to multiply the immigration regression coefficient by 0.43, which yields the predicted human capital investment change given a one standard deviation increase in low-skilled immigration flow. In this case, the result from Column 2 (2SLS) is a 1.2 percentage point increase in the likelihood of claiming that education is important for a career. The third column adds a control for mother’s education, and the fourth column instead adds controls for resources at the respondent’s eighth grade school. Columns 5 through 7 refer to the subsample of respondents whose mothers have 12 or fewer years of schooling, where the effect of immigration on natives’ attitudes toward education and careers is very small.
The second results row in Table 4 shows in similar specifications that increased local immigration does not induce tenth graders to increase their expectations of completing high school. However, the third row implies that students experiencing high local immigration are more likely to expect to continue their education after high school. This is true with different specifications that control for family background and early school environment, and in both the full sample and those with less-educated parents. This expectations result is consistent with actual postsecondary attainment increases reported in the next section.
The fourth results row of Table 4 shows that low-skilled immigration tends to increase school attendance among native-born students. If this is the sum of both the (negative) effect of fellow immigrant students—working through school resources—and the (positive) effect of immigrants in the labor market, then it is particularly strong evidence that natives increase their human capital investments in the face of competition with immigrants. Indeed, this interpretation is consistent with the stronger result in Column 4 that controls for school resources. The finding in Columns 5 through 7 that the effect is somewhat stronger among natives with less-educated parents lends further weight to this interpretation. The next row shows results about homework hours, providing some weak evidence of positive immigration effects.
I hypothesize that relatively large local flows of immigrants with little formal schooling would raise the labor market return to AP classes and lower the labor market return to vocational classes. The sixth and seventh rows in Table 4 confirm this hypothesis. Native-born students in higher-immigration CZs are more likely to take AP classes and less likely to take vocational classes. The effects are stronger among students with less-educated parents (Columns 5–7).
The eighth row in Table 4 shows a positive effect of immigration on native-born students’ grades, and the effect is very consistent across 2SLS specifications. The final two rows of results in Table 4 display the effect of low-skilled immigration to the CZ on the test scores of native-born students. The effects are uniformly positive and statistically significant, using alternative controls for the subsample of students with less-educated parents (Columns 5–7). A coefficient of 5 (as in Column 2 for eighth grade tests) implies that a one standard deviation (across CZs) increase in low-skilled immigration flow (53.7 percent) causes natives’ test scores to increase by 2.15 percentiles. The test score increases are somewhat larger in 12th grade, which is consistent with cumulative effects.
The results about test scores are related to a previous literature that mostly emphasizes how native-born students are affected by immigrants in their own school. The focus is on the school quality effect of immigration rather than the effect of labor market expectations. For example, Diette and Oyelere (2012) shows that immigration flows to North Carolina affected test scores of natives. Interestingly, low-ability natives increased their scores while high-ability natives decreased theirs, which is consistent with the dual mechanisms affecting natives differently by pre-existing ability. Perhaps native students who are likely to drop out of high school (“low-ability” in Diette and Oyelere 2012) are more likely to increase their motivation and performance in the midst of expected labor market competition. On the other hand, higher-ability natives appear more affected by a reallocation of schooling resources and see their test scores fall.
VI. Results About Immigration and Natives’ School Attainment
Table 5 illustrates the relationship between local low-skilled immigration and educational attainment of local native-born youth. The three rows of results describe different levels of education: receiving a high school diploma, attending postsecondary school, and earning a postsecondary credential. The regressions reflect smaller—though still sizable—samples than those used in the previous section because this section’s samples include only respondents kept after subsampling for the final followup survey. Table 5’s specifications are analogous to those in Table 4. Overall, low-skilled immigration increases the educational attainment of natives.
The dependent variable in the first row is an indicator for obtaining a high school diploma (not a GED credential). Column 2’s 2SLS coefficient of 0.0501 implies that if the local immigration flow increases by one standard deviation of its distribution across CZs (0.43 or 53.7 percent), the high school graduation rate among natives will increase by about two percentage points. This is a large effect. The high school graduation rate is an important social metric that has been stubbornly low in the United States.
Column 3 includes a control for mother’s years of schooling. Although the specifications already control for the educational distribution of adults in the CZ, it is still possible that something other than immigration about high-immigration CZs induces children to get more education. For example, growing labor markets may quickly attract highly educated workers23 whose children tend to get plenty of education as well. The control for mother’s education should capture such an effect directly. The control predicts higher schooling among respondents and reduces the coefficient on immigration though it remains positive. The specification in Column 4 of Table 5 controls for school resources and generates a larger, positive effect. Columns 5 through 7 show the same results for the subsample of the NELS:88 student population whose mothers did not pursue education past high school. Because mother’s education predicts own education, children of less-educated mothers are probably more likely to compete with low-skilled immigrants in the labor market. However, the first row of results in Table 5 implies that children of more-and less-educated mothers react about the same to low-skilled immigration. The standard errors in Columns 5 through 7 are higher because of the smaller sample size, but the point estimates are similar.
The second row of results in Table 5 repeats specifications from the first row using a new dependent variable: an indicator for the respondent attending any postsecondary schooling. All specifications reveal strong positive effects. Column 2 implies that a one standard deviation (across CZs) increase in local immigration flow raises the college-going rate among natives by 6.5 percentage points (100×0.43×0.152). The control in Column 3 for mother’s education lowers the effect somewhat but it remains statistically and economically significant. Inclusion of school quality controls in Column 4 strengthens the result. The effect on postsecondary attendance is larger than the effect on high school completion, implying that native-born high school graduates (not just dropouts) are particularly affected by local immigration. Columns 5 through 7 show that immigration’s effect on postsecondary attendance is even larger in the subsample of youth with less-educated mothers. There exists evidence of a positive return to postsecondary credits, even for students who do not earn a degree (Kane and Rouse 1995). So, immigration’s positive effect on natives’ postsecondary attendance probably increases their future earnings, holding other things fixed.
The third results row in Table 5 documents the relationship between local immigration in eighth grade and the likelihood that a native-born student acquires a postsecondary credential. The dependent variable is an indicator for earning (by age 26) any postsecondary degree, including certificates, licenses, and associate’s and bachelor’s degrees. The results here imply that immigration induces local natives not just to start postsecondary school but also to finish. The effects are somewhat smaller than the effects on attendance only, which is not surprising if immigration induces marginal college-goers to stay in school after getting the high school diploma. It is significant that immigration induces local natives to complete postsecondary educational programs, which yield high returns in the labor market.
The larger effect on postsecondary attendance than on high school completion implies that the effect of immigration is strongest on high school graduates on the margin of attending postsecondary education. These are not all four-year college degree programs. Immigration appears to induce native high school graduates to start (and—to a lesser extent—complete) license and certificate programs, perhaps to differentiate their labor skills from those of immigrants.
VII. Results About Immigration and Natives’ Early-Career Jobs
This section describes another way that native-born residents change their human capital investments in response to low-skilled immigration: augmenting their comparative advantage in communication skills. My findings agree with Peri and Sparber’s (2009): Native-born workers in the midst of less-skilled immigration tend to supply relatively more communication tasks. Table 6 assesses the relationship between early immigration exposure (in eighth grade) and job characteristics in the early careers of native workers (at age 26). The dependent variables in the table are indicators for “a lot” being the response to questions about how much the respondent performs each set of tasks at work. Table 6 includes OLS and 2SLS specifications that control for respondents’ sex and race/ethnicity and characteristics of the CZ where they live (and perform their work tasks) in 2000. The coefficients on immigration exposure in regressions predicting reading and writing tasks (the first two rows of results in Table 6) are uniformly positive. There is not much statistical precision, but the results are consistent with low-skilled immigration inducing natives to invest more in communication skills and use them at work.
The next panel of Table 6 is labeled “Computer and communication tasks.” There is not a consistent effect of immigration on computer use but natives originating in CZs with more immigration tend to use word processing, email, and the Internet more frequently. Effects are somewhat noisy for the subsample of natives with less-educated parents (Columns 5–7) but the coefficients are somewhat large and statistically significant for Internet use. The last row of the table investigates the effect of immigration exposure on manual tasks at work. The effect is negative in all specifications and both samples. Overall, the results are consistent with Peri and Sparber’s (2009) hypothesis that natives respond to low-skilled immigration by augmenting communication skills at the expense of manual skills.24
My findings about immigration and job tasks are related to previous research showing that manufacturing firms in cities experiencing large low-skilled immigration waves are less likely to invest in automation machinery (Lewis 2011), and firms near higher skill supplies are more likely to adopt personal computers (Beaudry, Doms, and Lewis 2010). In light of this previous research about business firms, it would appear that workers overall are less likely to use computers in cities with many low-skilled immigrants, but I find that native-born workers experiencing low-skilled immigration waves early in life use computers more frequently at work later on. The increased early-career computer use I observe is probably due more to individual natives’ human capital investment and occupational choices than to local firms’ production decisions. Both workers and firms are choosing skill and task mixes in production. I have focused on choices of workers to invest in particular tasks, and I believe this is appropriate given that the behavior I observe is in response to immigration waves early in life. However, the effects I estimate are probably also partially due to changes in firms’ productive processes.
VIII. Robustness of the Empirical Results
The results described above are robust to several changes in specification. As already mentioned, local immigration has a positive effect on natives’ human capital investments when controlling for mother’s education or a variety of school quality measures. Estimated effects are also positive when I use the share of less-educated immigrants in the local population to measure immigration instead of growth in the less-educated immigrant population (results available upon request).
Table 7 shows results from several alternative specifications. Each row represents a separate dependent variable used in prior tables. While prior results measure immigration at the commuting zone (CZ) level, Column 1 of Table 7 uses the state level. Results are similar to those above: More low-skilled immigration is associated with more human capital investments, educational attainment, and email use on the job. Column 2 shows results using an alternative instrumental variable. I categorized immigrants in the Census samples based on their countries of origin and placed them into 16 groups. For each group, I calculated the share of immigrants with a high school degree or less education. I selected four country groups with consistently high shares of less-educated immigrants: Central America, Southern Europe, Caribbean, and Oceania. The alternative immigration measure is based on the number of all immigrants from those four country groups coming to the local area. Results in Column 2 of Table 7 show that immigration measured in this way is associated with increases in human capital investment decisions.
Columns 3 and 4 of Table 7 show results when the local immigration is matched to NELS:88 respondents where they are last observed in secondary school (12th grade for most of them) and at the final followup survey (when most were 26 years old). The estimated effects are similar across immigration timing measures, which is consistent with the strong links between early and later locations (the majority of respondents stay where they are). The estimated effects of immigration measured at later times are relatively weakest in explaining respondents’ secondary schooling efforts. Estimated effects of immigration measured at later times are relatively strong in explaining behavior that takes place later in life, such as schooling attainment and job tasks. This is consistent with contemporary immigration being more salient in making decisions than earlier experiences with immigration.
A potential concern with my interpretation of results is that immigrants tend to settle in places with relatively high education, and the instrumental variables strategy might fail to solve the problem. In such a case, it might be true that any immigration (high-skilled or low-skilled) is associated with higher human capital investment of natives. However, Columns 5 and 6 of Table 7 show that the influence of high-skilled immigration on human capital investment of natives is very different from the estimated effect of low-skilled immigration. Column 5 defines high-skilled as having more than a high school degree, and Column 6 defines high-skilled as having a bachelor’s degree. Point estimates for high-skilled immigration tend to be negative, although confidence intervals are very large in these specifications. The instruments for high-skilled immigration are quite weak, which means that this robustness check is not a high-power test. Nevertheless, as far as I can tell, the observed positive effect of immigration on natives’ human capital is specific to low-skilled immigrants, which is consistent with my interpretation that natives increase their skills to avoid competing with an abundant local supply of low-skilled labor.
IX. Bias in Substitution Elasticity Estimates
The previous sections argued that low-skilled immigration induces local natives to increase their investments in human capital. In this section, I describe how reactions of native-born workers imply that previous estimates of the substitution elasticity between immigrant and native-born workers are upward-biased. I begin by describing how the estimation procedure typically works. Adopting the notation in Borjas, Grogger, and Hanson (2011), is the labor input from people in education group j, experience group k, year t, and nationality group n. Nationality F means a foreign-born worker while D (domestic) means a native-born worker. The total labor input in an education-experience-year cell is . The elasticity of substitution is σN = 1/(1-λ). If wages equal marginal products, then
(4)
where ϕ ≡ ln[ψ/(1-ψ)]. See Appendix 2 for a derivation of Equation 4.
Borjas, Grogger, and Hanson (2011) and Ottaviano and Peri (2011) are two of the studies employing this framework. Their method is to regress log relative wages of foreign-and native-born workers on log relative numbers of foreign-and native-born workers, where observations describe cells defined by education levels, experience levels, and years. Fixed effects help make the case that variation in relative employment is related to exogenous shifts in immigrant supply rather than demand-side factors. Call the regression coefficient on log relative employment , so that the estimate of the elasticity of substitution is (motivated by Equation 4). If is effectively zero (Borjas, Grogger, and Hanson’s finding), then conclude that the elasticity of substitution between foreign-born and native workers is infinite (perfect substitutes). This implies that immigration’s effect on natives’ wages should be large. The intuition is that when foreign-born labor becomes suddenly more abundant, its compensation does not change relative to native-born workers’ wages: wages of both walk down the demand curve together (at the same rate) following the immigration supply shock. The idea follows Card and Lemieux (2001) and Borjas (2003).25
But the framework remains very simple on the labor supply side. What if immigration into an education-experience-year cell induces native workers in that cell to attain more human capital? Competition for jobs provides a strong incentive. Such behavior can be incorporated into the framework above. To simplify notation, I will suppress the jkt subscripts from now on. Suppose the number of foreign-born workers (NF) is the same as effective foreign-born labor: LF = NF. However, native-born labor within the same (observational) education-experience-year cell augments its effectiveness in response to immigration. Let the fraction of foreign-born workers in an education-experience-year cell be s ≡ NF/(NF+ ND). Native-born workers respond to the share of foreign-born workers by augmenting their human capital (labor productivity) in the following way:
δ(s) is a continuous function such that δ′(s) > 0, to capture increased effectiveness of native-born labor as more foreign-born workers enter the market. A result of this setup is that counts of workers (or hours) mismeasure the effectiveness of native-born labor in the production function. This will induce bias in estimates of the elasticity of substitution between foreign-born and native-born labor (σN).26
To see this, plug LF = NF and LD = δ(s)ND into Equation 4 (still suppressing jkt subscripts):
(5)
The empirical exercise in Borjas, Grogger, and Hanson (2011) and Ottaviano and Peri (2011) is to regress log relative wages of foreign-and native-born labor on log relative numbers of foreign-born and native workers. The coefficient on the log ratio of foreign-born and native workers is (differentiating Equation 5)
(6)
The third and fourth equalities use the derivations in the footnote below.27 The elasticity of substitution estimate from the empirical exercise is . Using Equation 6, this estimate is related to the true elasticity of substitution (σN) as follows:
(7)
It is clear that s(1 – s)δ′(s) > 0, since s is the proportion foreign-born in the population, and δ′(s) > 0 by assumption (native workers increase their skills in the presence of more immigrants). In addition, δ(s) > 0, since otherwise effective labor units would be negative. Therefore, the term in brackets is greater than one so long as s(1 – s) δ′(s) < δ(s), which is likely.28
If natives’ human capital stock is fixed, so δ′(s) = 0, then Equation 7 implies that : there is no bias. But if δ′(s) > 0 and Equation 7’s term in brackets is greater than 1 (which is likely), then . This means that prior estimates of the elasticity of substitution between foreign-born and native-born labor are upward-biased. In addition, the bias is greater ( is further from σN) the greater is natives’ human capital investment response to immigration (δ′(s)).
The bias implied by Equation 7 is potentially large. If δ(s) = ln(1+s), then the true elasticity σN is about one quarter the size of the estimated elasticity when the immigrant share s is 0.2 and about half the size of when s = 0.4. Such large discrepancies imply significantly different degrees of substitutability between foreign-born and native workers. The empirical results in this paper imply that native workers respond to immigration by investing more in their human capital: δ′(s) > 0. Therefore, prior estimates of the elasticity of substitution between native-born and immigrant workers are probably too large.
X. Conclusion
I show that U.S. native-born respondents to the National Education Longitudinal Study of 1988 increased their human capital when their local labor markets experienced inflows of low-skilled immigrants. In particular, they increased high school attendance, got higher grades, achieved higher test scores, and took more academically rigorous curricula. They completed more years and credentials at school. In addition, young native-born workers differentiated themselves from low-skilled immigrant workers by taking jobs that involve more communication tasks, rather than manual tasks. These results use instrumental variables to account for endogeneity of local immigration flows and control for potentially confounding environmental factors like family background, peers at school, and local labor market characteristics.
The immigration effects I find are large, but I believe they are plausible. Hunt (2012) offers somewhat comparable results. It finds that a one percentage point increase in the immigrant share is associated with a 0.3 increase in the percent of natives completing high school. An immigrant share increase of one standard deviation in its sample (9.8 percentage points in 2000 from Appendix Table A) is then associated with a 2.94 percentage point increase in high school completion. My analogous findings imply a similar effect of about two percentage points.29 My results for postsecondary attendance and attainment are larger but within the same order of magnitude.
While much of the debate about immigration policy tends to focus on potential costs to pre-existing residents, I emphasize a potential benefit. The increased human capital investments I observe make up a positive effect of immigration if native-born youth tend to underinvest in education. Some probably do. Human capital externalities are likely positive. In addition, high school dropout rates are very high, even though the returns to schooling are large in terms of labor market success and also nonpecuniary benefits like health (see Oreopoulos 2007). Nevertheless, to the extent that native-born students invest optimally in their schooling, my findings reveal a cost of immigration in terms of extra schooling costs.
My findings also add to the debate about what effect immigration has on natives’ wages. Earlier research produced large estimates of the elasticity of substitution between foreign-born and native-born workers (for example, Borjas, Grogger, and Hanson 2011). I demonstrate that those estimates are upward-biased when native-born workers augment their human capital in the face of labor market competition from foreign-born workers. My empirical work implies that they do. So the degree of substitutability between immigrants and natives in the labor market is probably smaller than previously estimated. This implies lower effects of immigration on natives’ wages, which is consistent with more direct evidence of immigration effects from area-based studies (for example, Card 2005).
Appendix 1 Supplemental Tables
Appendix 2 Derivation of Marginal Products
Here I derive the ratio of marginal products of foreign-born and native-born workers in the CES production function of Section IX. To simplify notation, I suppress jkt subscripts. Let total output produced from immigrant (F) and native-born (D) labor inputs be Y. Then,
So,
and since ∂ln Y / ∂LF = Y−1 ∂Y / ∂LF, the marginal product of foreign-born workers is
Similarly, the marginal product of native-born workers’ labor is
These imply that the ratio of marginal products of foreign-born and native labor is:
Equilibrium implies that wages equal marginal products of labor:
Taking logs yields Equation 4.
Appendix 3 Derivation of a Sample Size Measure for Estimates of Commuting Zone Characteristics
This appendix describes asymptotic results that justify a measure of the sample size used to calculate commuting zone (CZ) average characteristics from U.S. Census data (for example, average education level). The calculation of CZ averages is complicated because the Census location definition is not the CZ but rather the county group or public-use microdata area (PUMA). Both CZs and PUMAs are collections of counties but some PUMAs overlap multiple CZs. Consequently, it is not straightforward to assign some respondents to CZs and thereby assess CZ-specific sample sizes. This appendix compares the sample size in a simple calculation of a mean (where we get our intuition about what is a “large” or “small” sample) to the corresponding expression in the more complicated calculation of a CZ mean from PUMA data.
Let CZs be indexed by c, PUMAs be indexed by p, and people be indexed by i. x is the variable of interest (for calculating averages). n denotes a sample size. wpc is the share of CZ c population that is in PUMA p. The CZ c average of x is calculated as:
Assume that Var(xip) = σ2 for each i,p. The asymptotic variance of the CZ mean estimate is:
So the standard error of the estimate is .
Suppose I knew each person i’s CZ. Then, the calculation would be much simpler. The asymptotic variance of the CZ mean would be and the standard error of the estimate would be . This hypothetical simple standard error is the true σ multiplied by . What I actually see in the data with PUMAs—but not CZs—identified is the true σ multiplied by . So the intuition we have for sample size nc in the simple case should apply correspondingly to the following expression:
This is the sample size measure that I calculate for each CZ to describe the “sample size” used in calculating CZ means and to gauge how much confidence to have in them.
This sample size measure is the same as the CZ c sample size when it can be observed. If there is a single PUMA p = 1 with the same boundaries as CZ c, then w1c = 1 and . More generally, is the same as the obvious sample size measure when a CZ consists of multiple PUMAs that are each entirely contained in the CZ. Let the total sample size across those PUMAs be N ≡ ∑p np. Assume that wpc = np / N for each p, which is a sufficient condition for the PUMAs under consideration to be wholly contained in the CZ (the share of each PUMA’s population in the CZ’s population is the same as the share of that PUMA’s population in the set of PUMAs). With this assumption, the sample size measure becomes
For example, suppose CZ c consists of three PUMAs with n1 = 200, n2 = 400, and n3 = 400. Also, none of the PUMAs includes any population outside of CZ c. If the individual PUMA sample sizes are proportional to their shares in the CZ population, then the assumption that wpc = np / N for each p is met. In this example, , which is just the sum of sample sizes across contributing PUMAs. In the simple case where PUMAs do not cross CZ boundaries, this is the obvious way to assess the CZ sample size. The added value of is that it provides a sample size measure for other cases, where PUMAs cross CZ boundaries.
Footnotes
Peter McHenry is an assistant professor of economics and public policy at the College of William and Mary. He thanks Daifeng He, Melanie Khamis, Fabian Lange, Melissa McInerney, John Parman, Kaj Thomsson, seminar participants at the College of William and Mary and the Society of Labor Economists 2013 meeting, and three anonymous reviewers for helpful comments. He is grateful to the Thomas Jefferson Program in Public Policy at the College of William and Mary for research support. Some of the analysis uses restricted-access data (NELS:88) that are available from the United States Department of Education to researchers with institutional affiliations. The author would be happy to help interested researchers access the data. For assistance, please contact Peter McHenry, College of William and Mary, P.O. Box 8795, Williamsburg, VA 23187, pmchenry{at}wm.edu.
↵1. Card (2005) and others have documented that immigrants to the United States since the late 1960s are much less educated than natives on average. Reasons include global population shifts and the 1965 Immigration Act, which widened the national origins of immigrants to the United States.
↵2. The literature on the effects of immigrant students on native students’ school performance yields mixed results. See Diette and Oyelere (2012), Jensen and Rasmussen (2011), and Neymotin (2009).
↵3. Among many points of disagreement in a symposium on human capital policy, Carneiro and Heckman (2003) and Krueger (2003) agree that there are many U.S. residents who would be better off if they invested more in human capital. Lange and Topel (2006) and Moretti (2004) describe some evidence of local positive spillovers from education although they both note that the empirical evidence is mixed.
↵4. Identification often comes from changes over time as well. Typical studies accommodate unobserved area-specific factors that affect local wages and immigration by instrumenting for immigration (typically with pre-existing ethnic enclaves). Critics have suggested that natives migrate out of local areas with large local immigration flows, which would attenuate estimates of wage effects even if competition for jobs and true (nationwide) earnings effects are large (Borjas 2003). However, empirical work on natives’ migration responses is also controversial, including findings of essentially zero and also substantial responses (Card and DiNardo 2000; Peri and Sparber 2011; Borjas 2006).
↵5. An exception is Llull (2010). It estimates a dynamic structural model of human capital attainment, occupational choice (blue-or white-collar), and wages with the National Longitudinal Survey of Youth 1979 and the Current Population Survey (CPS). It finds that immigration reduces wages even though natives increase their education in response. Eberhard (2012) uses the CPS to calibrate a general equilibrium model of immigration, human capital accumulation, and wages. In counterfactual exercises, it finds that natives increase their human capital accumulation in response to nationwide immigration shocks.
↵6. Betts and Lofstrom (2000) is an extension with similar findings.
↵7. Smith (2012) estimates immigration effects on youth outcomes, including whether the teenager is in school. It uses decennial Census and annual American Community Survey data. It finds small positive effects of local immigration flows among white girls and smaller positive effects for white boys. It also obtains somewhat noisy estimates of immigration on a cohort’s future earnings.
↵8. I measure immigration flows at the commuting zone (CZ) level, described in detail below.
↵9. Betts and Fairlie (2003) reports evidence that parents in higher-immigration areas are more likely to send their children to private school.
↵10. There are 16 origin regions. Table A1 lists them. I assign people to origins based on their countries of birth in the Census.
↵11. For details about the sampling procedure, see Curtin et al. (2002).
↵12. Results reported below are similar in NELS:88 samples that include children of immigrants. I experimented with analyses on samples of only black respondents or only Hispanic respondents. The results were mostly consistent with, but somewhat less precisely estimated than, those reported below. Separating by race and ethnic category is a productive exercise, though. See Betts (1998) and Hunt (2012) with larger Census samples.
↵13. Evidence in Heckman, Humphries, and Mader (2010) that GED recipients do not earn a positive return in the labor market from their GED credential motivates me to categorize them as high school dropouts. However, I also ran specifications that treat GED recipients as high school completers and obtained similar results.
↵14. Locations are sets of adjacent counties called commuting zones (CZs), which I describe in the next subsection. Tolbert and Sizer (1996) categorize CZs into six size categories based on the largest population center in each CZ. Small towns have fewer than 5,000 residents, small nonmetro areas have between 5,000 and 20,000, and larger nonmetro areas have at least 20,000 but no MSAs in the CZ. The remaining three categories are CZs with at least one MSA in their territory. They are classified according to the size of the largest MSA, where small metro centers have fewer than 250,000 residents, medium metro centers have between 250,000 and 1 million, and major metro centers have more than 1 million. These population figures refer to 1990.
↵15. I assign households in the 1990 U.S. Census to commuting zones and calculate average monthly rental prices for two-and three-bedroom dwellings that are not group quarters. I then divide each CZ’s rent average by the average across CZs to form the local cost of living index.
↵16. I use expenditures data for the 1989–90 school year as the year prior was not available on the NCES web page. For students enrolled in private schools, I impute the school district expenditures where they live.
↵17. See Tolbert and Sizer (1996) for a description of how CZs were identified. It uses journey-to-work data from the 1990 Census to identify counties with strong labor market links. There are 741 CZs in the United States.
↵18. Alternative analyses that measure immigration as a stock (that is, percent of local residents who are low-skilled immigrants) yield findings similar to those described below.
↵19. The data sets are the 1970 form 1 metro and form 2 metro samples, the 1980 5 percent sample, and the 1990 5 percent sample. The citizenship variable is not available in the 1970 form 2 metro sample. For that sample, I impute citizenship status based on the likelihood of citizenship in the 1970 form 1 metro sample conditional on respondents’ birthplace, age, and education.
↵20. The data identify the county group where each respondent lives [called “county groups” in 1970 and 1980 and called “public-use microdata areas” (PUMAs) in 1990]. Most county groups by these definitions are completely enclosed in a CZ so the identity of the respondent’s CZ is clear. Sometimes county groups intersect with more than one CZ; in these cases, I assign Census respondents to CZs based on the proportion overlap between county group and CZ populations.
↵21. People born in any of the 50 states, Washington D.C., or outlying areas and territories (American Samoa, Guam, Puerto Rico, U.S. Virgin Islands) are “natives” in the analysis.
↵22. Table 2 describes the fourth followup NELS:88 subsample. In the larger sample of high schoolers (prior to subsampling), the first stage is similarly strong.
↵23. Wozniak (2010) shows that highly educated workers are more likely to move in response to local labor market conditions than less-educated workers.
↵24. Results were very similar in specifications that add controls for respondents’ completed education. I also investigated whether immigration affected native-born respondents’ job training activities. The NELS:88 asks whether respondents ever participated in a job training program and also whether they got training in order to improve basic communication skills. I regressed indicators for training and communication-specific training on immigration flows in 2SLS specifications analogous to those in Table 6. The coefficients on the eighth grade immigration flow variable were positive but mostly statistically insignificant.
↵25. Jaeger (2007) employs a similar strategy to estimate large elasticities of substitution, exploiting variation in immigration across cities rather than education-experience cells.
↵26. I thought earlier that natives’ human capital augmentation may be captured more simply as a reduction of ψ, the parameter that measures the relative effectiveness of immigrant labor inputs. As a result, human capital augmentation would show up empirically in estimates of ϕ, leaving estimates of σN unbiased. However, natives’ responses must work through changes in LD as above. When immigrants increase in a particular education-experience cell, the natives who respond are those in that education-experience cell. This increases the effective labor supply of native-born workers in that cell, not overall. But ψ measures the relative effectiveness of all immigrants. Because it does not have a jkt subscript, ψ is too blunt an instrument for modeling the native human capital augmentation idea.
↵27. The third equality:
The fourth equality:
↵28. δ(s) is probably in the neighborhood of one, since foreign-born and native-born workers with similar education and experience are probably similarly productive. Since s is a fraction, s(1 – s) is at most 0.25. So δ′(s), the slope of the human capital augmentation function, would need to be very large (likely greater than 4) to make the term in brackets negative.
↵29. This comes from the 0.0501 coefficient in Column 2 of Table 5 multiplied by one standard deviation in the immigration flow variable (0.43).
- Received June 2013.
- Accepted January 2014.