migration effect of education

A

Does Immigration Affect Whether US Natives Major in Science

and Engineering?

Pia M. Orrenius, Federal Reserve Bank of Dallas

Madeline Zavodny, Agnes Scott College

Immigration may affect the likelihood that US natives major in science or engineering. Foreign-born students may crowd US na- tives out of science or engineering, or they may have positive spill- overs on US natives that attract or retain them in those fields. This study uses data on college majors from the 2009–11 American Com- munity Surveys to examine the effect of the immigrant share in US natives’ age cohort while they are in high school or in college. We find some evidence that immigration adversely affects whether US- born women who graduated from college majored in a science or engineering field.

I. Introduction

The number of workers in science, technology, engineering, and math- ematics ðSTEMÞ fields is commonly viewed as critical to the future of the US economy. Concern that not enough US natives are studying sci- ence and engineering to maintain the nation’s pace of innovation and its

We thank Bill Kerr and seminar participants at Georgia Tech and the IZA Migration Meeting for helpful comments and Melissa LoPalo for excellent re- search assistance. The views expressed here are solely those of the authors and do not reflect those of the Federal Reserve Bank of Dallas or the Federal Reserve System. Contact the corresponding author, Madeline Zavodny, at mzavodny@ agnesscott.edu. Information concerning access to the data used in this article is avail- able as supplementary material online.

[Journal of Labor Economics, 2015, vol. 33, no. 3, pt. 2] © 2015 by The University of Chicago. All rights reserved. 0734-306X/2015/33S1-0012$10.00 Submitted January 31, 2013; Accepted November 22, 2014; Electronically published June 29, 2015

S79

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

S80 Orrenius/Zavodny

A

long-term economic competitiveness underlies a number of recent policy proposals. Policy makers have proposed granting permanent resident sta- tus to foreign students who earn a US graduate degree in STEM and revamping undergraduate education in those areas, for example, while cor- porations have advocated for increased resources in STEM education at the K–12 and college levels as well as for more temporary and permanent visas for high-skilled foreign workers.1

Proposals to admit more immigrants trained in science and engineer- ing and to improve STEM education are motivated by the view that not enough US natives earn degrees in those fields. During the period 1977– 2009, the proportion of US citizens and permanent residents earning bachelor’s degrees in science and engineering ðS&EÞ fields—which we de- fine here as including life and physical sciences, engineering, mathematics, and computer science but not the social sciences—fluctuated between 15% and 20% with little discernible trend.2 During that period, the number of bachelor’s degrees in S&E fields awarded to temporary residents—for- eigners on student visas—increased faster than the number awarded to US citizens and permanent residents. The foreign-born proportion of the US population that holds a bachelor’s degree in S&E rose as well, in part due to increases in the proportion of US S&E degrees granted to immi- grants but also as a result of inflows of S&E graduates of foreign univer- sities. The foreign-born share of college graduates living in the United States who majored in S&E increased from 11% in 1990 to 17% in 2000 ðFree- man 2009Þ; it hit 29% during 2009–11.3 This increase stemmed in large part from the creation of the H-1B visa program and an increase in the num- ber of employment-based permanent resident visas implemented in the early 1990s.

1 For example, S.B. 744, the 2013 Border Security, Economic Opportunity, and Immigration Modernization Act, proposed no limit on the number of green cards available to employment-based immigrants holding a PhD; allocating up to 40% of numerically limited employment-based green cards to holders of a master’s degree or higher in STEM earned in the last 5 years from a US university who have a US job offer in a related field; and allocating 25,000 H-1B visas to advanced degree graduates in STEM from US universities. The President’s Council of Advisors on Science and Technology advocated changes to STEM undergraduate teaching ðsee http://www.whitehouse.gov/sites/default/files/microsites/ostp/pcast-engage-to -excel-final_feb.pdfÞ. For corporations, see, e.g., Microsoft’s proposal urging an in- crease in educational resources and immigration in STEM ðavailable at http://www .microsoft.com/en-us/news/download/presskits/citizenship/MSNTS.pdfÞ.

2 Authors’ calculations based on data from National Science Foundation ð2002, 2012Þ. Including psychology and other social sciences as S&E fields—as the NSF does—raises the proportion to about one-third.

3 The 2009–11 share is based on authors’ calculations from 2009–11 American Community Survey ðACSÞ data on college graduates.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Immigration and Majoring in Science and Engineering S81

A

The increase in the foreign-born share of S&E college graduates mir- rors a general rise in the immigrant population share in the United States. The foreign-born proportion of the US population rose from less than 5% in 1970 to more than 12% in 2010. Although most migrants arrive as adults, an increasing number are children or college students. The foreign- born proportions of high-school-age youth and of college students more than tripled from 1970 to 2000. Some of these foreign-born students ac- companied their immigrant parents, while others came to the United States to study with the hope that doing so would increase the likelihood that they would be able to remain in the country to work or live after com- pleting their studies ðe.g., Kato and Sparber 2013Þ; indeed, many of them ultimately did so via temporary or permanent visas. The proportion of US-born college graduates who majored in S&E rose from 15% in 1970 to 17% in 2000.4 This modest rise occurred in the face of robust—and at times booming—demand for STEM workers and steady or increasing rel- ative earnings in STEM occupations ðHanson and Slaughter 2013Þ. This study examines whether these patterns are related. Specifically,

we examine the relationship between whether US-born college graduates majored in an S&E field and the foreign-born share of their age cohort while in high school and in college. We focus on whether US natives chose S&E majors at the undergraduate level because this is a critical step toward an S&E graduate degree and a career in an S&E field. We are able to take advantage of a large-scale survey, the American Community Survey ðACSÞ, that asks about college major. Although there is a large literature on the effects of immigration on natives’ labor market outcomes and a smaller lit- erature on effects on natives’ educational outcomes, no previous research has directly examined whether immigration appears to affect natives’ choice of college major. To address this question, this study uses a novel combi- nation of new and old data to implement a common approach in the im- migration literature, namely, exploiting differences in immigrant shares across areas and over time to examine immigration’s effects on US natives. Specifically, we combine new ACS data on natives’ choice of college ma- jors with older data from decennial censuses on immigrant shares. The census immigrant shares coincide with the years the cohorts of US natives attended high school and college. The paper goes a step further by looking separately at US-born men

and women. Women may be more responsive to the influx of immigrants since they are already considerably less likely than men to major in S&E. In 1970, only 7% of bachelor’s degrees awarded to women were in S&E fields, and 11% in 2010. The corresponding shares for men were 25% in

4 Authors’ calculations based on data from the 2009–11 ACS on US-born col- lege graduates who were age 22 in 1970 or in 2000.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

S82 Orrenius/Zavodny

A

1970 and 23% in 2010. Meanwhile, the share of all bachelor’s degrees awarded to women increased from 43% to 57% during that period ðNa- tional Science Foundation 2013Þ. Female students who choose S&E ma- jors are also less likely to persist in those majors than men ðGriffith 2010; Ost 2010Þ. The National Academy of Sciences ð2007aÞ says that increas- ing the number of women ðand minoritiesÞ who study and work in STEM fields is vital to maintaining the country’s economic competitiveness. The endogeneity of immigrant shares to increased demand for STEM

workers is an important consideration in this type of study. We control for potential endogeneity in immigrant shares by measuring them in US natives’ state of birth and by using instruments based on historical im- migrant settlement patterns. We also control for the attractiveness of STEM jobs using measures of the proportion of college graduates in a state work- ing in STEM occupations, average annual earnings in those jobs relative to non-STEM jobs, and the 10-year changes in those measures. The instru- mental variables results do not indicate an effect of immigration on whether men major in an S&E field, whereas a higher immigrant share in the college cohort has a negative effect among women in our preferred specification. The next section explains why immigration might affect natives’ edu-

cational outcomes and provides an overview of previous research in this area, focusing on the United States. Section III discusses the data and em- pirical methodology. Section IV presents the results. Section V concludes with a discussion of limitations of this study and directions for future re- search.

II. Literature Review and Theoretical Framework

Immigration can have a positive or negative effect on whether natives major in S&E. This study focuses on the effect of immigrants in natives’ own age cohort. This effect can occur through several channels. First, im- migrants may compete directly with natives for educational resources. Foreign-born students may crowd out natives from S&E courses in the short run if the supply of education is inelastic. This is more likely to occur at the college level than at the high school level.5 In the medium to long run, however, larger inflows of immigrant students may lead to more S&E course offerings and to new or larger programs, particularly since for-

5 Bound and Turner ð2007Þ report that the supply of higher education is in- elastic. If natives may have more difficulty getting into required introductory math and science classes their first year of college as the number of foreign-born college students increases, they may become more likely to major in other disciplines. Bet- tinger ð2010Þ reports that students who take more STEM classes their first semester of college are more likely to persist in STEM majors, although the direction of cau- sality is unclear.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicag

o.edu/t-and-c).

Immigration and Majoring in Science and Engineering S83

A

eign students tend to boost college revenues and to disproportionately study S&E.6

Similarly, immigration may affect the quality of education and natives’ academic preparation. Academic preparation, particularly in math, is crit- ical to students’ interest and persistence in S&E majors ðe.g., Levine and Zimmerman 1995; Smyth and McArdle 2004; Arcidiacono, Aucejo, and Hotz 2013Þ. Foreign-born students may require intensive English-language instruction, reducing the resources that high schools devote to math and science and to academic preparation for college.7 Students’ academic prepa- ration may also worsen if school resources are inelastic and immigration increases the number of students. Alternatively, attending high school or college with high-achieving im-

migrants may increase the quality of natives’ education and their interest in studying S&E through increased rigor, positive peer effects, and other spillovers. Highly skilled immigrants may put pressure on high schools to increase educational resources, particularly in math and science. Foreign students attending US universities tend to be among the better students in their home country, and immigrants often outperform their US-born peers at the K–12 level, particularly within racial/ethnic groups.8 How- ever, attending high school or college with high-achieving immigrants may

6 In addition, foreign graduate students may be complementary with under- graduate courses in S&E if increased inflows of graduate students, who serve as teaching assistants and instructors, result in expansion of undergraduate S&E course offerings and programs. In OLS and IV regressions, Bound and Turner ð2010Þ find that the number of foreign PhD students in the sciences at a given university has a small positive effect on the number of undergraduate science majors; they find little evidence of a significant effect on the fraction of students at a given univer- sity majoring in science, however. Alternatively, foreign graduate students may adversely affect natives’ performance in classes. Borjas ð2000Þ finds that student performance in undergraduate economics principles classes is negatively related to having a foreign-born teaching assistant.

7 For example, Chin, Daysal, and Imberman ð2012Þ find that bilingual education programs in the United States, which typically place limited-English-proficient ðLEPÞ students in separate classrooms, appear to have positive spillovers onto non- LEP students’ standardized test scores, suggesting that attending school with LEP immigrants may adversely affect US natives’ academic preparation. Several non-US studies also find negative effects of immigration on native-born students’ educa- tional outcomes ðe.g., Gould, Lavy, and Paserman 2009; Brunello and Rocco 2011; Jensen and Rasmussen 2011; Hardoy and Schøne 2013Þ. In the United States, Schwartz and Stiefel ð2011Þ find a negative ðpositiveÞ relationship between natives’ test scores and the immigrant share in the school ðclassÞ for children in third through eighth grade in New York City public schools. Neymotin ð2009Þ finds that SAT scores of US natives living in California and Texas are not significantly negatively related to the immigrant share of SAT takers from their high school.

8 For studies comparing educational outcomes among foreign- and US-born K–12 students, see Schwartz and Stiefel ð2006Þ and references therein.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

S84 Orrenius/Zavodny

A

increase the competition for good grades in math and science classes, and natives who move down in the grade distribution may be less likely to major in S&E. Barnett, Sonnert, and Sadler ð2012Þ report that immigrants earn higher grades, on average, than US natives in college calculus classes, which are crucial gateway courses for S&E majors. Luppino and Sander ð2012Þ show that lower-performing white, non-Hispanic students typi- cally respond to tougher competition—higher math SAT scores among students majoring in STEM—by moving away from STEM majors; minor- ities persist in STEM majors but become less likely to graduate. Ost’s ð2010Þ study of a large research university finds that females’ persistence in science majors is more sensitive than males’ to their grades, while Rask’s ð2010Þ study of a liberal arts college finds the opposite. Immigration may also affect natives’ educational outcomes through more

indirect channels. For example, immigration may change the returns to education in general or the returns to S&E majors, leading to changes in natives’ choices regarding educational attainment or field of study ðe.g., Arcidiacono, Hotz, and Kang 2012Þ. Although the immigrant share in the labor market and in STEM occupations may be more important in driving such changes, the immigrant share in natives’ own cohort may matter as well. Students may perceive higher immigrant shares in their own cohort as increasing the competition for STEM jobs or decreasing wages in STEM jobs in the future, reducing their willingness to major in S&E. Borjas ð2009Þ finds that an influx of foreign graduate students into a par- ticular field within S&E reduces the earnings of competing native-born doctorate recipients. Empirical findings on the effect of immigration on US natives’ educa-

tional attainment are mixed. In a two-stage least squares ð2SLSÞ analysis, Hunt ð2012Þ finds that higher immigrant population shares, particularly of immigrants who have not completed high school, when US natives are aged 11–17 increase natives’ high school completion; the effect is largest among blacks and is not significant among US-born Hispanics. Other stud- ies, however, find evidence of a negative relationship between US natives’ high school completion and the immigrant share ðBetts 1998; Betts and Lofstrom 2000Þ. Evidence on the relationship between college attendance or completion and the immigrant share is also mixed ðBetts and Lofstrom 2000; Jackson 2011Þ. At the graduate level, Borjas ð2007Þ concludes from an ordinary least squares ðOLSÞ analysis that foreign students may crowd out white male US natives but not other groups of US natives.9

Immigration also may affect natives’ choice of what college or univer- sity to attend. In OLS and instrumental variables ðIVÞ results, Hoxby

9 Regets ð2001Þ, Zhang ð2004Þ, and Bound and Turner ð2010Þ report results that do not suggest crowd out in graduate programs.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicag

o.edu/t-and-c).

Immigration and Majoring in Science and Engineering S85

A

ð1998Þ finds that foreign students appear to displace black and Hispanic natives from selective colleges but not from other colleges. Neymotin ð2009Þ finds that US natives’ likelihood of applying to selective colleges is not significantly negatively related to the immigrant share of SAT test takers from their high school after controlling for school fixed effects. The potential effect on selectivity matters in our study’s context because stu- dent persistence in S&E majors is higher at more selective schools for better-prepared students but lower for less-prepared students ðArcidiacono et al. 2013Þ. The returns to majoring in the sciences are higher at more selective schools ðArcidiacono 2004Þ, which may prompt more students to major in the sciences at such schools. Griffith ð2010Þ concludes that the sorting of women and minorities into different types of undergraduate institutions may play a significant role in gender and racial/ethnic differ- ences in persistence in STEM majors. On balance, previous research thus suggests that immigration does not

reduce US natives’ educational attainment, including college attendance and completion, and may even increase it. Immigration may affect how selective of a college or university natives attend, however. Student per- sistence in S&E majors depends on school selectivity and students’ own academic preparedness, both absolutely and relative to their competition. Immigrants account for a growing share of the competition in S&E ma- jors, making it important to understand their effect on whether US natives major in S&E.

III. Data and Methodology

A. Data

We use data from the 2009–11 ACS together with immigrant shares derived from decennial censuses. The ACS is a large-scale survey that re- placed the long-form decennial census but is conducted every year. The ACS asks about demographic characteristics and labor market outcomes. Since 2009, it has asked respondents who have at least a bachelor’s degree to report their college major. We examine only US-born respondents in the ACS who hold at least a bachelor’s degree.10 We classify those who report majoring in biology and life sciences, physical sciences, engineer- ing, computer and information sciences, or mathematics and statistics as S&E majors. The main advantage of the ACS is that it offers very large, nationally

representative samples of multiple cohorts. However, it lacks data on im- portant factors that are related to the choice of college major, such as fam-

10 We term people who have at least a bachelor’s degree “college graduates” throughout; all measures of college graduates shown here include people who hold an advanced degree as well as people who hold only a bachelor’s degree.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

S86 Orrenius/Zavodny

A

ily background and academic ability.11 Other data sets that include such measures, such as the Baccalaureate and Beyond Studies and the National Longitudinal Surveys of Youth, have much smaller samples and include fewer age cohorts. We examine four 5-year age cohorts in the 2009–11 ACS. The cohorts

are all US-born college graduates aged 28–32, 38–42, 48–52, and 58–62 in the 2010 ACS ð1 year younger for the 2009 ACS and 1 year older for the 2011 ACSÞ. These cohorts were chosen because they were aged 18–22, or traditional college age, when a decennial census was conducted. For ex- ample, people aged 28–32 in the 2010 ACS were aged 18–22 at the time of the 2000 census, and those aged 58–62 in the 2010 ACS were aged 18–22 at the time of the 1970 census. Figure 1 shows the proportion of US-born college graduates who ma-

jored in S&E by the year in which they were age 22, the modal age at college graduation. This proportion in the ACS data corresponds well to similar measures from the 1993 and 2003 National Surveys of College Graduates ðNSCGÞ and from the National Science Foundation ðNSFÞ.12 The proportion of US natives majoring in S&E rose from 1974 until 1986, declined precipitously from 1986 to 1991, and was fairly flat from the mid- 1990s through 2008. The increase in the 1970s and early 1980s likely re- flects the emphasis on science and math while those age cohorts were in school ðthe “Sputnik generation”Þ. The Internet boom of the late 1990s did not have a major effect on the proportion of S&E majors; if it led students to switch from science to computer science majors, that would be not be discernible here. Figure 1 also shows an estimate of the foreign-born proportion of US

college graduates who majored in S&E. This measure includes only peo- ple living in the United States at the time of the ACS and so misses for-

11 Altonji, Blom, and Meghir ð2012Þ provide an overview of the relatively sparse literature on the determinants of college majors.

12 One concern with retrospective data is recall bias. Comparisons across data sets and between retrospective and contemporaneous data suggest that survey re- spondents can reasonably accurately recall their major. The correlation between the proportion of US-born college graduates in S&E by likely year of bachelor’s degree for the period 1966–89 ð1999Þ in the 2009–11 ACS data and the 1993 ð2003Þ NSCG data is 0.83 ð0.75Þ. The correlation between the proportion of likely US college graduates in S&E by likely year of bachelor’s degree for the period 1966–98 in the ACS data and the NSF data is 0.55. The correlation between the proportion of college graduates in S&E by year of US bachelor’s degree for the period 1966–89 ð1998Þ in the NSF data and the 1993 ð2003Þ NSCG data is 0.86 ð0.80Þ. Further, the proportion of college graduates in a given cohort reporting that they majored in S&E is rea- sonably well correlated over time. The correlation between the proportion of college graduates in S&E by year of bachelor’s degree for respondents who earned their bachelor’s degree between 1966 and 1989 in the 1993 NSCG and the 2003 NSCG is 0.79.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

FIG. 1.—The proportion of college graduates majoring in S&E by nativity. Calculations based on 2009–11 American Community Survey data for college graduates ages 25–65 at the time of the survey. Proportion of S&E US college grad- uates who are foreign born is based on US natives and immigrants who arrived in the United States by age 20 and only includes people living in the United States at the time of the survey. S&E does not include psychology or the social sciences.

Immigration and Majoring in Science and Engineering S87

A

eign students who have returned home. We excluded immigrants who ap- peared to have earned their bachelor’s degree abroad, based on their age and year of arrival in the United States.13

The foreign-born proportion of US college graduates who majored in S&E rose fairly steadily from the mid-1970s until 1995 and shows a slight downward trend since then. The downward trend may be due to the United States attracting fewer foreign students who want to study S&E or to fewer foreign students remaining in the United States after receiv- ing a bachelor’s degree in S&E. The difficulty foreign students experi- enced getting visas for a few years after the 9/11 terrorist attacks and ex- pansions in higher education in other countries in recent years may have contributed to fewer foreign students earning bachelor’s degrees in STEM in the United States; the number of international undergraduates study- ing in the United States fell from 2001–2 through 2005–6 and then began rebounding. The Great Recession of 2007–9 and relatively strong eco- nomic growth in China and India may have led to a decline in the number of foreign-born S&E graduates staying in the United States.

13 We assume that immigrants who have a bachelor’s degree or higher and ar- rived in the United States by age 20 are US college graduates. We use the ACS data here because the National Science Foundation data on degree recipients groups to- gether US citizens and permanent residents, making it impossible to distinguish between immigrants and US natives.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

S88 Orrenius/Zavodny

A

The top row of table 1 reports the share of US-born college graduates who majored in S&E for the four cohorts we examine. This share reflects the pattern shown in the figure—it is higher for the cohort who were college age in 1980 than for the 1970 cohort, and it is lower for the 1990 and 2000 cohorts than for the 1980 cohort. The rest of the analysis fo- cuses on these four college cohorts of US natives and how the proba- bility that they majored in S&E is related to immigrant shares while they were in high school and college. We created two measures of immigrant shares from the 1970–2000 de-

cennial censuses. The first is the share of the age cohort—the population aged 18–22 in the census—that is foreign born. Immigrants who report arriving in the United States in the 5 years prior to the census are not in- cluded in the construction of this variable so that it approximates the immigrant intensity of US natives’ high school ðand earlierÞ experience and does not capture immigrants who arrived as undergraduates or graduate students.14 The second immigrant share variable is the foreign-born share of 18–22-year-olds who are enrolled in college at the time of the decen- nial census.15 This measure captures the immigrant intensity of US natives’ college educational experience. Since 1950, college students have been in- structed to give their current residence, not their parents’ residence, in the census if they live in campus or off-campus housing instead of with their parents. Foreign students studying in the United States are instructed to answer the census with their household location in the United States ðCork and Voss 2006Þ. The rise in immigrant shares across cohorts, reported in rows 4 and 5

of table 1, reflects the general increase in the foreign-born fraction of the US population since the 1960s. The immigrant share while in college is

14 The fraction of US natives who are the children of immigrants—the second generation—also may affect the likelihood of majoring in S&E. We attempted to control for the proportion of the age cohort that is second generation ðthe children of immigrantsÞ by merging children’s and parents’ records in the census 10 years prior, when most children were living with their parents ðthe census stopped ask- ing about parental birthplace in 1970, making it impossible after then to calculate the second-generation share like we do the immigrant shareÞ. However, first-stage regressions using either of the instruments used here did not have enough power to instrument for this variable, which may be endogenous if parents’ locational choices depend on factors that also affect whether US natives major in S&E. We also cannot distinguish second-generation immigrants from the children of US natives in the ACS.

15 The census educational categories and questions have changed over time. The 1990 census does not ask what grade the student is attending, so we classify peo- ple ages 18–22 who are enrolled in school and are at least high school graduates as enrolled in college; we classify people in the 2000 census in the same manner for comparability. For the 1980 and earlier censuses, which did not ask specifi- cally about high school graduation, we include people 18–22 who report being college undergraduates.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Table 1 Descriptive Statistics

Cohort

All 2000 1990 1980 1970

% S&E major 16.4 16.8 15.8 18.2 14.7 % of men majoring in S&E 24.3 24.1 23.5 27.3 22.4 % of women majoring in S&E 9.1 11.1 9.1 9.9 6.3 % foreign born while in high school 3.8 6.3 4.5 2.9 1.7

ð4.1Þ ð5.4Þ ð4.6Þ ð2.3Þ ð1.2Þ % foreign born while in college 5.9 9.2 7.2 5.1 2.3

ð5.4Þ ð6.7Þ ð6.0Þ ð3.4Þ ð1.6Þ Number of observations 641,971 149,731 159,212 165,099 167,929

NOTE.—Shown are sample means ðstandard deviationsÞ. The sample is US-born college graduates who were aged 18–22 in the year indicated as the cohort. Data are from the 2009–11 American Community Surveys. Percentages foreign born are from decennial census data based on state of birth; see text for details. Observations are weighted using the person weights rescaled to sum to the same total for each cohort.

Immigration and Majoring in Science and Engineering S89

A

consistently higher than the immigrant share while in high school. This pattern primarily reflects foreign students coming to the United States for college, although it also captures a slightly higher likelihood of college en- rollment among immigrants than among US natives in the 1980 and 1990 cohorts. The immigrant share measures are created from the decennial censuses

at the state level. We merge the immigrant share measures with the ACS data by individuals’ cohort and their state of birth, which is the only place of residence available in the ACS data besides current place and place 1 year ago. Other studies similarly use state of birth to examine state-level variables related to education.16 Using state of birth mitigates endogeneity bias that would arise if immigration affects US natives’ place of high school or college attendance, although we also instrument for the immigrant share below. Because only about 80% of people go to high school in their state of birth ðand about 65% go to college in their state of birthÞ, the estimates below are likely to underestimate any effect of immigration on US natives’ college major. The estimates we present are robust to controlling for the share of US natives in an individual’s state of birth who are no longer liv- ing in that state during high school or during college; this variable should control for migration in response to immigration.

16 Studies using state of birth to proxy for state of high school attendance in- clude Card and Krueger ð1992Þ, Dynarski ð2008Þ, and Bound, Hershbein, and Long ð2009Þ. Dynarski ð2008Þ reports that about 20% of high school–age youth live outside their state of birth. The College Board reports that about 13% of full- time undergraduate students at public 4-year colleges and universities are study- ing out of state ðhttp://trends.collegeboard.org/college-pricing, accessed April 15, 2013Þ.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

S90 Orrenius/Zavodny

A

As noted earlier, women are less than half as likely as men to major in S&E. Rows 2 and 3 of table 1 report the percentage of college graduates majoring in S&E by cohort and sex. The share of women majoring in S&E increased over time, although it dipped between 1980 and 1990. The share of men majoring in S&E fields is more similar to the pattern shown in figure 1, with the 1980 cohort having the highest share of S&E majors. The share of S&E majors who are female increased across cohorts, as

the top row of table 2 reports. This reflects both the increase in the share of women majoring in S&E and also the increase in the share of bache- lor’s degrees awarded to women. The share of bachelor’s degrees in non- S&E fields awarded to women also increased across cohorts, although not as much as the share of S&E degrees. Table 2 also reports the share of S&E and non-S&E degrees awarded

to non-Hispanic whites, blacks, Asians, and Hispanics among our sam- ple of US natives. We control for race/ethnicity in these categories in the regressions discussed below.17 The share of bachelor’s degrees earned by each minority group increased across cohorts, with the exception of blacks between 1990 and 2000. The last two rows of the table report the im- migrant share measures for S&E majors and non-S&E majors. They are similar, indicating that S&E and non-S&E majors are not systematically from different states.

B. Empirical Methodology

We use linear probability regressions to examine the relationship be- tween whether US-born college graduates majored in S&E and the various measures of the immigrant share. The basic regression model is

S&Eist 5 a 1 bln Immigrant Shareð Þst 1 dCharacteristicsist 1vLabor Marketst 1 jStates 1 tCohortt 1 εist:

ð1Þ

The dependent variable equals one if individual i who was born in state s and cohort t majored in S&E; the ACS sample is conditioned on hav- ing graduated from college. Immigrant Share is one of the two immi- grant share variables described above, with one focused on the high school

17 We attempted to estimate regressions stratified by race/ethnicity. We do not present those estimates because the first-stage regressions for the samples of blacks and Asians using either of the instruments discussed below did not have enough power to instrument for the immigrant share variable. This is because of differ- ences in the distribution of different races/ethnicities across states. Blacks are con- centrated in the South while Asians are highly concentrated in California ðand HawaiiÞ. Since the South and California experienced substantial changes in their relative shares of immigrants during 1970–2000, the instrumental variable strategy used here does not work well for those groups.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicag

o.edu/t-and-c).

T ab

le 2

C h ar ac te ri st ic s o f S & E an

d N o n -S & E M aj o rs , b y C o h o rt

S & E M aj o rs

N o n -S & E M aj o rs

A ll

2 0 0 0

1 9 9 0

1 9 8 0

1 9 7 0

A ll

2 0 0 0

1 9 9 0

1 9 8 0

1 9 7 0

% fe m al e

2 9 .3

3 7 .0

3 1 .0

2 8 .3

2 0 .5

5 6 .9

6 0 .0

5 8 .0

5 7 .4

5 2 .6

% n o n -H

is p an ic

w h it e

8 5 .5

7 8 .4

8 3 .0

8 8 .7

9 1 .7

8 5 .0

8 0 .3

8 3 .1

8 6 .6

8 9 .4

% n o n -H

is p an ic

b la ck

6 .2

7 .3

7 .5

5 .7

4 .2

7 .9

8 .2

8 .9

7 .8

6 .7

% A si an

3 .0

6 .2

3 .3

1 .5

1 .3

1 .5

2 .8

1 .4

0 .9

0 .8

% H is p an ic

3 .6

5 .6

4 .5

2 .8

1 .6

4 .1

6 .6

5 .0

3 .2

2 .0

% fo re ig n b o rn

w h il e in

h ig h sc h o o l

3 .8

6 .2

4 .4

2 .9

1 .8

3 .8

6 .3

4 .5

2 .9

1 .7

ð4 .1 Þ

ð5 .4 Þ

ð4 .6 Þ

ð2 .3 Þ

ð1 .3 Þ

ð4 .1 Þ

ð5 .4 Þ

ð4 .6 Þ

ð2 .3 Þ

ð1 .2 Þ

% fo re ig n b o rn

w h il e in

co ll eg e

5 .9

9 .0

7 .2

5 .1

2 .4

5 .9

9 .3

7 .2

5 .1

2 .3

ð5 .4 Þ

ð6 .6 Þ

ð6 .0 Þ

ð3 .4 Þ

ð1 .6 Þ

ð5 .5 Þ

ð6 .7 Þ

ð6 .0 Þ

ð3 .4 Þ

ð1 .6 Þ

N u m b er

o f o b se rv at io n s

1 0 4 ,7 3 4

2 5 ,3 2 8

2 5 ,1 1 8

3 0 ,0 0 9

2 4 ,2 7 9

5 3 7 ,2 3 7

1 2 4 ,4 0 3

1 3 4 ,0 9 4

1 3 5 ,0 9 0

1 4 3 ,6 5 0

N O T E .—

S h o w n ar e sa m p le

m ea n s ðst

an d ar d d ev ia ti o n sÞ.

T h e to ta ls fo r ra ce /e th n ic it y d o n o t su m

to 1 0 0 b ec au se

n o n -H

is p an ic

o th er

ra ce

is n o t sh o w n .

A

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM

ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

S92 Orrenius/Zavodny

A

experience and the other on college. Only one of the measures is included at a time because they are highly collinear.18 Again, these variables are mea- sured when individuals were aged 18–22 and are for the state of birth. They are not sex specific. Characteristics includes age and its square and, as noted above, dummy variables for race/ethnicity ðblack, Hispanic, Asian, and other are the mutually exclusive categories, with non-Hispanic whites as the omitted categoryÞ. These variables control for systematic differ- ences in the probability of majoring in S&E across races/ethnicities. The relative attractiveness of STEM jobs may influence a student’s de-

cision whether to major in S&E. Some specifications control for condi- tions in the STEM labor market by including variables that measure the fraction of college graduates who are employed in STEM occupations ðvs. non-STEM occupationsÞ in the state, the change in that fraction during the past 10 years, the ratio of log average annual earnings of college grad- uates working in STEM occupations to log average annual earnings of col- lege graduates working in non-STEM occupations in the state, and the change in that ratio during the past 10 years. These measures of the rela- tive labor market attractiveness of STEM jobs are calculated from the de- cennial censuses. The regressions also include state and cohort fixed effects. The state-of-

birth fixed effects control for unobservable factors that are specific to a state but constant over time, such as proximity to international borders, size, and climate. The cohort fixed effects control for unobservable fac- tors that are specific to a college cohort, such as changes in the national labor market and the economy and changes in the national emphasis on math and science education. Our preferred specification adds state-specific linear time trends, which control for any unobservable, smooth changes within states that affect the likelihood that US-born students major in S&E. The standard errors are robust and clustered on the state and cohort.19

C. Instrumental Variables

An important concern about the regression model is whether immi- grant shares are exogenous. Factors that affect the propensity of immigrants ðand US nativesÞ to live in a certain state may also affect whether natives

18 If both immigrant share measures are included, the standard errors on the immigrant share variables increase dramatically and their coefficients occasionally change signs.

19 We cluster on state � cohort because that is the level of variation in the im- migrant shares. The standard errors are slightly larger if we instead cluster on state but patterns of statistical significance tend to be similar. Most notably, the nega- tive relationship between whether women major in S&E and the immigrant share while they are in college remains significant at the 10% level in the IV specifica- tion with the lagged immigrant share.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicag

,

o.edu/t-and-c).

Immigration and Majoring in Science and Engineering S93

A

from that state opt to major in S&E. Labor market shocks, such as the Internet boom in Silicon Valley in the 1990s, may attract immigrants to a state and also make US natives from that state more interested in ma- joring in S&E. If our measures of the relative attractiveness of STEM jobs and the state and cohort fixed effects and state trends do not fully con- trol for this, ordinary least squares ðOLSÞ estimates will have a positive or upward bias. Another source of upward bias is if immigrants are attracted to states with educational systems that put more emphasis on math and science and therefore generate more S&E majors. Alternatively, the OLS estimates may display a negative or downward

bias. Families whose children disproportionately major in S&E in col- lege ði.e., parents who themselves work in STEM occupationsÞ may move away from states that are receiving large numbers of immigrants. In this case, the likelihood that a US native majors in S&E ends up lower ðhigherÞ in states with high ðlowÞ immigrant shares than it would be otherwise, making OLS estimates too negative. To bias our estimates, such endoge- nous mobility would have to take place before children are born since our immigrant share measures are for the state of birth. However, if families move out of high-immigration states after having children and this increases the likelihood that those children ultimately major in S&E in college, then the OLS estimates have a positive or upward bias. We control for the potential endogeneity of the immigrant shares by

using one of two instrumental variables. Both instrumental variables are closely based on Smith ð2012Þ, who also uses only one instrument at a time.20 The first instrument is based on immigrants’ historical settlement patterns. Immigrants tend to settle in the same areas as earlier immigrants from their country of origin ðalthough this tendency lessened in the 1990s as immigrants moved to new destinations in the South and West, as dis- cussed by Massey ½2010�Þ. This instrument is based on reallocating im- migrants across states based on their distribution, for 16 countries or re- gions of origin, across states in 1960.21 Specifically, we calculate the predicted immigrant share in state s in census year t as

20 Smith ð2012Þ also uses distance from Mexico as an instrument. We do not since variation in immigrant shares driven by Mexican immigration is of less interest in our study’s context. Other research using instruments for the immigrant share based on historical settlement patterns includes Card and DiNardo ð2000Þ, Card ð2001Þ, Saiz ð2007Þ, and Hunt ð2012Þ.

21 The 16 countries or regions are the United Kingdom, Canada, Ireland, Italy, Germany, Poland, Russia/USSR, other Europe, Mexico, Central America, South America, Cuba, other Caribbean, Southeast Asia and other Asia, Middle East and South Asia, and the rest of the world. These 16 groups were chosen because they are identified in the 1960–2000 census and the ACS and because they are each a relatively large share of immigrants in 1960. This avoids having large numbers of states with zero immigrants from a given group.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

S94 Orrenius/Zavodny

A

Predicted immigrant sharest 5 o16

j51 Number of immigrantsj

t � % of immigrants living in s j1960

Population st

;

ð2Þ where j represents country or region of origin. The immigrant share in- strument is calculated separately for each of the immigrant share variables based on the relevant immigrants and US natives. However, we calculate the distribution of immigrants in 1960—the second term in the numerator in equation ð2Þ—using all immigrants from a given country or region, not just young adults or college students. To be valid, this instrument requires assuming that the distribution

of immigrants by country or region of origin across states in 1960 is not correlated with shocks occurring 10–40 years later that affect whether US natives born in a given state majored in S&E. In other words, shocks that affect both the distribution of immigrants across states in 1960 and US natives’ college major do not persist over time. This seems reasonable given that 1960 predates much of the “space race” and the computer rev- olution. Further, immigration patterns were considerably different in 1960 than in later years. At the time, the foreign-born population share was below 6%, versus above 12% in 2010. Most immigrants were from Eu- rope and had been in the United States—mainly in the Northeast and Midwest—for many years. The 1965 Immigration and Nationality Act led to dramatic increases in immigrant inflows and a shift in origins to Latin America and Asia. Immigrant destinations also have shifted to the West and South over time. The instrument also requires assuming that immigrant inflows are not

systematically related to factors that affect whether US natives major in S&E in a way that is also related to immigrants’ distribution across states in 1960. If more immigrants come from China, for example, because of state-specific factors that also encourage young adults born in those states to major in S&E, the instrument would be biased. Such factors need to be at the state level because the cohort fixed effects control for any national changes, such as the Internet bubble of the 1990s or changes in immigra- tion policy. The highly aggregated nature of the 16 immigrant origin groups helps alleviate concerns here. We group together immigrants from all East Asian countries. Mainland China, Japan, Korea, and Taiwan—major ori- gin countries of high-skilled immigrants who disproportionately work in S&E—are grouped together with Cambodia, Laos, and Vietnam. India is grouped together with Iran, Turkey, and other countries in South Asia and the Middle East. We also control for the relative attractiveness of STEM jobs within states at a point in time and the change in their relative appeal over time. The second instrument is also based on immigrants’ settlement pat-

terns but within states instead of across them. The instrument uses the

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Immigration and Majoring in Science and Engineering S95

A

share of immigrants in a given state from each of the 16 origin groups 10 years prior and the contemporaneous number of all immigrants from each of the 16 origin groups in all other states to create a measure of predicted immi- gration in that state.22 The measure is

Predicted immigrant sharest 5 o16

j51

Immigrants living in s j t210

Immigrants living in st210 Immigrants not living in sj

t

� �

Population st

:

ð3Þ The measure far exceeds the actual immigrant share in a given state because it is based on the number of immigrants living elsewhere, but it is none- theless well correlated with actual immigrant shares, as shown below. As with the across-state ðfirstÞ instrument, the immigrant shares for the within- state ðsecondÞ instrument—the fraction in the numerator in equation ð3Þ— are created using all immigrants from the previous census, not just young adults or college students. The second instrument has the advantage of not being based on the num-

ber of immigrants in a given state relative to other states. Instead, it is based on the composition of immigrants within a state. This avoids concerns about long-lasting state-specific shocks that affect both immigrant shares and the likelihood that US natives major in S&E. It does require assum- ing that shocks that affect the distribution of immigrants across origins within a state that also affect whether US natives major in S&E do not persist for 10 years. The highly aggregated origin groups we use reduce the applicability of this concern, as do the labor market controls, state and cohort fixed effects, and state-specific linear time trends. In addition, re- sults using this instrument were robust to using the distribution of immi- grants across origin groups within a state in 1960 instead of 10 years prior. Shocks would have to last for a long time—up to 40 years—for that in- strument to be biased.

IV. Results

A. Ordinary Least Squares Results

Table 3 reports the OLS results for the relationship between majoring in S&E and the two measures of the immigrant share. We show three specifications by sex: with state and cohort fixed effects but without the labor market controls, with the labor market controls, and also with state- specific linear time trends. Without the labor market controls or the time trends, whether men majored in S&E is not related to the immigrant share while they were in high school ðrow 1, col. 1Þ. Women, in contrast, were

22 Smith ð2012Þ and Wozniak and Murray ð2012Þ similarly use the number of immigrants in all other states to create an instrument.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Table 3 OLS Regression Estimates for Relationship between Majoring in S&E and Immigrant Share

Males Females

ð1Þ ð2Þ ð3Þ ð4Þ ð5Þ ð6Þ % foreign born while in high school 2.0037 2.0039 .0008 2.0104 2.0113 2.0118

ð.0036Þ ð.0029Þ ð.0043Þ ð.0029Þ ð.0027Þ ð.0040Þ % foreign born while in college 2.0123 2.0105 2.0015 2.0134 2.0130 2.0089

ð.0044Þ ð.0037Þ ð.0057Þ ð.0039Þ ð.0037Þ ð.0042Þ Labor market controls included No Yes Yes No Yes Yes State linear time trends included No No Yes No No Yes

NOTE.—Each estimated coefficient is from a separate linear probability regression. Regressions also include controls for age and its square, race/ethnicity, and state and cohort fixed effects. Standard error ðin parenthesesÞ are robust and clustered on state � cohort. For males, the number of observations i 300,187 for high school specifications, and 299,895 for college specifications; for females, the number of ob servations is 340,891 and 340,625, respectively.

S96 Orrenius/Zavodny

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM All use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicag

s s -

less likely to major in S&E the higher the immigrant share while they were in high school ðrow 1, col. 4Þ. Adding labor market controls ðcols. 2 and 5Þ and then time trends ðcols. 3 and 6Þ has little effect on the estimated co- efficients. Both men and women were less likely to major in S&E the higher the

immigrant share while they were in college, as the second row of table 3 shows. The only exception is among men when both labor market con- trols and time trends are included ðcol. 3Þ, our preferred specification. The implied marginal effects are very small. In our linear probability

model with the log of the immigrant share, an estimated coefficient of 0.01 suggests that a 10% increase in the immigrant share reduces the probabil- ity that a US native majors in S&E by 0.1 percentage points. Evaluating this at the sample means, a 10% increase in the immigrant share—going from 3.8% to 4.18% in high school or from 5.9% to 6.5% while in college— corresponds to a decrease in the share of women majoring in S&E from 9.1% to 9%. However, the changes in the immigrant share that actually occurred were much bigger: the immigrant share in high school increased by 270% over 1970–2000, and the immigrant share in college by 300%. Evaluating the OLS estimates at those magnitudes implies that the share of women majoring in S&E would have been 13.8%–14.3% instead of about 11% in the 2000 cohort—a nontrivial difference—if immigrant shares had remained at their 1970 levels and all else had remained the same. In results not shown here, the estimated coefficients on the other var-

iables are largely as expected. Women in more recent cohorts were more likely to major in S&E. Hispanics are less likely than non-Hispanic whites to major in S&E, and Asians more likely. Black women are more likely than white women to major in S&E while the opposite is true among black men. The probability of majoring in S&E tends to be positively related

o.edu/t-and-c).

Immigration and Majoring in Science and Engineering S97

A

to the fraction of college graduates working in STEM occupations and to the income ratio of college graduates in STEM occupations to those in non-STEM occupations but not significantly related to the 10-year changes in those variables.

B. Instrumental Variables Results

Endogeneity of the immigrant share variables may bias their coeffi- cients either positively or negatively, as explained above. Table 4 presents the first-stage regression results for each of the instruments. For sim- plicity, we present only results for the pooled sample and control for sex in the regressions; results stratified by sex are similar. Both instruments do well, with F-test statistics above 20 with only one exception. Table 5 presents the instrumental variables results. The top panel pre-

sents estimates using the predicted immigrant share based on the distri- bution of immigrants across states in 1960. The bottom panel presents estimates using the predicted immigrant share based on the distribution of immigrants across origins within states 10 years prior. Each coefficient is from a separate IV regression. The IV estimates tend to be closer to zero than the OLS estimates,

and few of the estimated coefficients are statistically significant. The stan- dard errors are large enough that we cannot rule out finding no difference between the OLS and the IV results within or across specifications. None- theless, most of the estimated marginal effects remain very small.

Table 4 First-Stage Regression Estimates for Immigrant Share

1960 Distribution of Immigrants across States

Lagged Immigrant Composition within States

ð1Þ ð2Þ ð3Þ ð4Þ ð5Þ ð6Þ % foreign born while in high school .609 .634 .865 .757 .820 .611

ð.104Þ ð.091Þ ð.139Þ ð.134Þ ð.112Þ ð.108Þ F-test 34.09 48.86 38.84 31.96 53.77 31.81 % foreign born while in college .553 .564 .798 .403 .477 .450

ð.109Þ ð.098Þ ð.136Þ ð.093Þ ð.089Þ ð.087Þ F-test 25.97 33.14 34.43 18.96 29.77 27.00 Labor market controls included No Yes Yes No Yes Yes

State linear time trends included No No Yes No No Yes

NOTE.—Each estimated coefficient is from a separate OLS regression. Regressions also include con- trols for age and its square, sex, race/ethnicity, and state and cohort fixed effects. Standard errors ðin parenthesesÞ are robust and clustered on state � cohort. The number of observations is 641,078 for high school specifications, and 640,520 for college specifications.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Table 5 IV Regression Estimates for Relationship between Majoring in S&E and Immigrant Share

Males Females

ð1Þ ð2Þ ð3Þ ð4Þ ð5Þ ð6Þ A. 1960 Distribution of Immigrants across States

% foreign born while in high school 2.0072 .0003 2.0042 2.0040 .0006 2.0055

ð.0070Þ ð.0058Þ ð.0095Þ ð.0071Þ ð.0072Þ ð.0086Þ % foreign born while in college 2.0064 .0009 .0037 2.0070 2.0011 2.0245

ð.0107Þ ð.0092Þ ð.0120Þ ð.0119Þ ð.0124Þ ð.0133Þ B. Lagged Immigrant Composition within States

% foreign born while in high school .0012 .0029 .0074 2.0016 .0001 2.0149

ð.0065Þ ð.0057Þ ð.0115Þ ð.0060Þ ð.0056Þ ð.0120Þ % foreign born while in college .0090 .0095 .0121 2.0061 2.0014 2.0432

ð.0149Þ ð.0123Þ ð.0179Þ ð.0149Þ ð.0129Þ ð.0202Þ Labor market controls included No Yes Yes No Yes Yes

State linear time trends included No No Yes No No Yes

NOTE.—Each estimated coefficient is from a separate IV regression. Regressions also include controls for age and its square, race/ethnicity, and state and cohort fixed effects. Standard errors ðin parenthesesÞ are robust and clustered on state � cohort. For males, the number of observations is 300,187 for high school specifications, and 299,895 for college specifications; for females, the number of observations is 340,891 and 340,625, respectively.

S98 Orrenius/Zavodny

A

The one exception is the relationship between whether women major in S&E and the immigrant share while they are in college, shown in col- umn 6 of table 5. Both IV estimates are negative, significantly different from zero, and larger ðin absolute valueÞ than most of the other estimates. The estimate using the across-states instrument suggests that a 10% in- crease in the immigrant share reduces the probability that a woman majors in S&E by 0.25 percentage points, and the within-states instrument sug- gests that the probability falls by 0.43 percentage points. Evaluating those estimates at the actual changes that occurred and assuming that nothing else changed, the across-states instrument suggests that the share of women majoring in S&E would have been 18.6% instead of 11% in the 2000 cohort if the immigrant share had remained at its 1970 level. The within-states in- strument suggests an even bigger counterfactual: that the share of women majoring in S&E would have been 24%. This is equivalent to the share of men majoring in S&E in the 2000 cohort. Taken at face value, the estimated coefficients in the specifications with time trends thus suggest that immi-

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Immigration and Majoring in Science and Engineering S99

A

gration can explain all of the gender gap in S&E. However, the other speci- fications do not indicate that the immigrant share has a significant nega- tive effect on whether women major in S&E.

C. Robustness

Because California is the center of the high-tech industry and is home to many immigrants, it may drive the results. The immigrant share in Cal- ifornia for the 2000 cohort was over 18% in high school and 22.5% in college. Almost 9% of the sample was born in California. IV results with- out US natives born in California continue to indicate a sizable negative relationship between the probability that women majored in S&E and the immigrant share while they were in college. The other estimated coeffi- cients are also similar to those reported in the equivalent specifications in table 5. We also found similar results when dropping Texas, another large immigrant state, or Washington, DC, from the sample. The results also were robust to creating the within-state instrument

using 1960 census data instead of data from the 10-years-prior census. As noted above, this suggests that any shocks that bias the results would have to be very long lasting. The results were robust to controlling for the log of the fraction of US natives in a given cohort who remained in their state of birth, which helps control for the possibility that US natives moved in response to immigrant inflows.23

D. Consequences

If women were less likely to major in S&E because the immigrant share rose, what did they ultimately major in? We examine this using the same methodology we used to examine whether US natives were less likely to major in S&E but with the dependent variable equal to one if a college graduate majored in the humanities, the social sciences, psychology, busi- ness, medical and health sciences and services ðnursingÞ, or education. We estimated separate regressions for each of those majors. The results indi- cate that women were more likely to major in education and in psychol- ogy as the immigrant share while in college increased. Higher immigrant shares while in college also may have affected which

women chose to major in S&E. Women with the highest or the lowest ability in that area may have disproportionately shifted to other majors. The ACS does not have any measure of ability that would allow us to directly examine this, such as SAT scores or high school grades. However, we can look at whether women who majored in S&E were less likely to

23 For specifications with the immigrant share while in high school, we used the share of all US natives aged 18–22 who lived in their state of birth; for specifica- tions with the immigrant share while in college, we used the share of US natives aged 18–22 attending college who lived in their state of birth.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

S100 Orrenius/Zavodny

A

go on to graduate school, which may be a signal of ability. The ACS has a measure of completed education, which we use to examine whether US natives who majored in S&E had an advanced degree. The ACS does not ask the field of the advanced degree, so the data would also capture MBAs and law degrees. Table 6 presents the results for linear probability regressions on whether

S&E college graduates earned an advanced degree. The regressions include state-specific linear time trends as well as the other variables noted earlier. Column 1 reports results for men from OLS and IV regressions while column 3 reports results for women. The OLS results suggest that the probability that men earned an advanced degree is negatively related to the immigrant share while in college, and the probability among women

Table 6 Regression Estimates for Relationship between Postcollege Outcomes among S&E Majors and Immigrant Share

Males Females

Advanced Degree ð1Þ

Work in STEM ð2Þ

Advanced Degree ð3Þ

Work in STEM ð4Þ

A. OLS

% foreign born while in high school 2.0144 .0076 .0300 2.0349

ð.0119Þ ð.0095Þ ð.0164Þ ð.0141Þ % foreign born while in college 2.0333 .0198 .0175 2.0375

ð.0129Þ ð.0110Þ ð.0193Þ ð.0159Þ B. IV: 1960 Distribution of Immigrants across States

% foreign born while in high school .0072 2.0038 .0456 2.1009

ð.0266Þ ð.0232Þ ð.0416Þ ð.0280Þ % foreign born while in college .0264 .0133 .1102 2.1258

ð.0399Þ ð.0279Þ ð.0559Þ ð.0403Þ C. IV: Lagged Immigrant Composition within States

% foreign born while in high school 2.0302 2.0201 .0203 2.118

ð.0429Þ ð.0291Þ ð.0500Þ ð.036Þ % foreign born while in college 2.0620 .0041 .0935 2.199

ð.0632Þ ð.0395Þ ð.0734Þ ð.065Þ NOTE.—Each estimated coefficient is from a separate regression. Samples are conditional on having an

S&E major. Regressions also include controls for age and its square, race/ethnicity, STEM labor mar ket conditions, state and cohort fixed effects, and state linear time trends. Standard errors ðin parentheses are robust and clustered on state � cohort. For males, the number of observations is 73,243 for advanced degree specifications, and 71,068 for work in STEM specifications; for females, the number of observation is 31,361 and 28,893, respectively.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicag

- Þ s

o.edu/t-and-c).

Immigration and Majoring in Science and Engineering S101

A

is positively related to the immigrant share while in high school. The across-states IV results indicate that a higher immigrant share while in college increases the probability that female S&E majors earn an advanced degree. The IV results thus suggest that women with higher ability in S&E, as signaled by earning an advanced degree, may be the ones who remain S&E majors as a result of immigration, but this finding is not conclusive given the limitations of the ACS data. Whether immigration affects the likelihood that US natives work in

STEM is ultimately more important than whether it affects the likelihood that US natives major in S&E. We therefore look at the relationship be- tween whether a US native who majored in S&E is working in a STEM occupation at the time of the ACS and the immigrant shares when he or she was in high school and college.24 Columns 2 and 4 of table 6 report the results for men and women, respectively. The immigrant share does not have a negative effect on whether men who majored in S&E are work- ing in STEM, but it is consistently negatively related to whether women are working in STEM. Interestingly, the results are negative—although smaller in absolute value—for the immigrant share while in high school as well as in college even though the IV results did not indicate a negative effect of the immigrant share in high school on whether women majored in S&E. The results ðnot shownÞ are also negative for women—but not for men— if the sample is not conditional on majoring in S&E. We note that our goal here is not to explain gender differences in working in STEM—which would require a considerably more rigorous approach—but rather to focus on whether immigration appears to affect the choices men and women make. These results also serve as a check on the validity of our instruments.

If the instruments reflect a tendency of immigrants to settle in areas that experienced long-lasting positive trends in STEM fields, we would expect the IV estimates in columns 2 and 4 to be more positive than the OLS re- sults; the same factors that attracted immigrants would presumably also make college graduates more likely to work in STEM. Instead, the IV re- sults for women are consistently more negative than the OLS results. The IV results for men are also smaller than the OLS results, albeit not signif- icantly so.

24 As in Hanson and Slaughter ð2013Þ, we classify engineering occupations, soft- ware developers and programmers, computer and information analysts, database ad- ministrators and network architects, life and physical scientists, mathematicians, and computer scientists as STEM occupations. Because there may be selection in employment status, particularly among women, we include all S&E graduates re- gardless of current employment status. STEM occupation is based on the current occupation reported in the ACS by people who are currently employed; for people who are currently unemployed or not in the labor force but have worked within the last 5 years, the ACS reports the occupation of the last job. People who have been unemployed or out of the labor force for more than 5 years are not included in this part of the analysis.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

S102 Orrenius/Zavodny

A

E. Why Do Women Respond Differently to Immigration than Men?

The above results indicate that the immigrant share of the age cohort has a more adverse effect on women than on men. This difference is par- ticularly important since women are less than half as likely as men to major in S&E. Considerable attention has been devoted to examining why women are less likely than men to major in S&E, but no single factor appears to be the smoking gun.25 Xie and Shauman ð2003Þ report that most female STEM baccalaureates decided to major in STEM while in college; in contrast, most male STEM baccalaureates already planned on majoring in that area when they entered college. This is consistent with our finding that the immigrant share in college has a more adverse effect than the immigrant share in high school on whether women major in S&E and on whether they ultimately work in S&E. The mechanism via which immigration affects women in college is un-

clear, however. One possibility is cohort crowding. If most male S&E majors are already planning on majoring in S&E but most female majors decide when they get to college, women’s experience in first-year courses may be critical. Women who do not get into S&E gateway courses as freshman may choose other majors. Immigration also may affect where US natives end up in the grade distribution in S&E courses, and women who earn lower grades may be less likely to persist in S&E majors ðOst 2010Þ. However, grades are unlikely to be the whole story. Women are less likely than men to persist in natural science, engineering, and eco- nomics majors at Duke University even after controlling for grades and SAT scores ðArcidiacono et al. 2012Þ. Peer effects may play a role in other ways as well.26 Zafar ð2013Þ finds

that female undergraduates at Northwestern University are less likely than males to major in S&E because they believe that they will not en- joy the courses. Why women believe they will not enjoy S&E courses is unclear; it does not appear to be related to beliefs about ability but may be related to the small number of other women in S&E courses. Women appear to be more likely to major in male-dominated fields such as S&E

25 For example, Turner and Bowen ð1999, 308Þ report that differences in aca- demic preparation help explain gender differences in choice of major, but differ- ences in SAT scores “capture much less of the dynamics of change over the past 35 to 40 years than do the panoply of residual forces, including differences in pref- erences, labor market expectations, gender-specific effects of the college experience, and unmeasured aspects of academic preparation.” Zafar ð2013Þ attributes gender differences in major to differences in beliefs about enjoying coursework and dif- ferences in preferences, but the source of those differences is unclear.

26 Professor gender also appears to matter. Carrell, Page, and West ð2010Þ show that women at the US Air Force Academy are more likely to major in STEM if they have female instructors in introductory math and science courses.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Immigration and Majoring in Science and Engineering S103

A

if they attend a single-sex college instead of a coeducational institution ðSolnik 1995Þ. Another possibility is that higher levels of immigration result in some

women attending institutions in which they are less likely to major in S&E. As discussed above, research does not show that immigration ad- versely affects the likelihood that US natives go to college ðBetts and Lof- strom 2000; Jackson 2011Þ, but some evidence suggests that immigration may change the type of institution US natives attend ðHoxby 1998Þ. Per- sistence in and returns to S&E vary across types of institutions, and which students persist in S&E differs across types of institutions as well ðArci- diacono 2004; Griffith 2010; Luppino and Sander 2012; Arcidiacono et al. 2013Þ.

V. Discussion and Conclusion

The National Academy of Sciences ð2007bÞ warned that the United States is losing its competitive edge because of insufficient investments in STEM education and research and the small share of US citizens entering STEM fields. Along with urging increased funding for STEM education and re- search, it recommended that the United States adopt immigration poli- cies that would encourage immigrant scientists and engineers to study and work in the United States. Immigration policy is important since immi- grants comprise 21% of US STEM workers with a bachelor’s degree, 41% of those with a master’s degree, and 58% of those with a PhD ðHanson and Slaughter 2013Þ. Because they are more likely to have studied S&E, college-graduate immigrants have higher rates of patenting, publishing, and creating businesses than college-graduate US natives ðHunt and Gauthier- Loiselle 2010; Hunt 2011Þ.27 The foreign born are also a large and grow- ing share of undergraduate and graduate students in the United States, especially in S&E fields. This study examined the link between these trends and policy recom-

mendations by examining whether the immigrant share of high school and college students affects US natives’ decisions to major in S&E at the under- graduate level, a crucial first step toward a career in STEM. The results sug- gest that, when controlling for state-specific linear time trends, women are less likely to major in S&E the higher the immigrant share in their college cohort. In addition, the higher the immigrant share while in high school and college, the less likely female college graduates are to later be working in a STEM occupation. Instrumental variables results do not indicate neg- ative effects among men, in contrast. This gender difference is striking given that women are much less likely to major or work in S&E. Being less likely to major in S&E and other lucrative fields plays a major role in the gender

27 For a recent survey on immigrant-native differences in innovation and entre- preneurship, see Kerr ð2013Þ.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

S104 Orrenius/Zavodny

A

wage gap among college graduates ðe.g., Brown and Corcoran 1997; Wein- berger 1999Þ. Although the results suggest that immigration may make women less

willing to major in S&E, there are several caveats to this finding. First, we condition on being a college graduate. The fraction of the population that graduates from college has increased over time, particularly among women, and the selectivity of this pool may have decreased at the same time as immigrant shares have risen. Immigration may raise the bar in STEM fields, increase the selectivity of US natives into S&E majors, and ultimately have a positive effect on innovation. Research indicates that immigration via the H-1B temporary worker program does not reduce the level of patent activity or S&E employment among US natives ðKerr and Lincoln 2010Þ. Immigration by school-aged youth similarly need not necessarily have a negative impact on US natives’ innovative activities. Looking at whether immigration affects selectivity into S&E majors and longer-run outcomes, such as graduate school and employment in S&E, using data with a better measure of ability is an important area for future research. Even if immigration discourages US-born women from choosing S&E

majors, we caution against drawing strong policy implications. If foreign students are on average better at S&E fields or have a stronger preference for those fields and are therefore more likely to major in those fields, this frees up US natives to pursue other careers. Research shows that foreign students in S&E who remain in the country are at least as innovative and productive as similarly educated US natives ðHunt 2011Þ, and foreign grad- uate students appear to have positive effects on patent activity ðChellaraj, Maskus, and Mattoo 2008Þ. In addition, changes in immigration policy may lead to unexpected ef-

fects. Kato and Sparber ð2013Þ show that a binding cap on H-1B temporary worker visas, which many foreign students obtain after completing their US studies, leads to a decrease in the quality of foreign students applying to US undergraduate institutions. Changing immigration policy in an at- tempt to encourage more US natives to major in S&E may similarly have unanticipated consequences. To the extent that people choose a profession in accordance with their comparative advantage, the distribution of majors by nativity is optimal and targeted policy changes would not be warranted. Exceptions to this outcome might be situations where US citizenship is a condition for STEM employment, such as in the defense industry and at na- tional security agencies.

References

Altonji, Joseph G., Erica Blom, and Costas Meghir. 2012. Heterogeneity in human capital investments: High school curriculum, college major, and

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

A

A

A

B

B

B

B

B

—

—

B

B

—

Immigration and Majoring in Science and Engineering S105

All

careers. NBER Working Paper no. 17985, National Bureau of Economic Research, Cambridge, MA. rcidiacono, Peter. 2004. Ability sort and the returns to college major. Journal of Econometrics 121:343–75. rcidiacono, Peter, Esteban Aucejo, and V. Joseph Hotz. 2013. Uni- versity differences in the graduation of minorities in STEM fields: Evi- dence from California. NBER Working Paper no. 18799, National Bureau of Economic Research, Cambridge, MA. rcidiacono, Peter, V. Joseph Hotz, and Songman Kang. 2012. Model- ing college major choices using elicited measures of expectations of coun- terfactuals. Journal of Econometrics 166:3–16. arnett, Melissa D., Gerhard Sonnert, and Philip M. Sadler. 2012. More like us: The effect of immigrant generation on college success in math- ematics. International Migration Review 46:891–918. ettinger, Eric. 2010. To be or not to be: Major choices in budding sci- entists. In American universities in a global market, ed. Charles T. Clotfelter. Chicago: University of Chicago Press. etts, Julian R. 1998. Educational crowding out: Do immigrants affect the educational attainment of American minorities? In Help or hindrance? The economic implications of immigration for African-Americans, ed. Daniel S. Hamermesh and Frank D. Bean. New York: Russell Sage. etts, Julian R., and Magnus Lofstrom. 2000. The educational attainment of immigrants: Trends and implications. In Issues in the economics of immigration, ed. George J. Borjas. Chicago: University of Chicago Press. orjas, George J. 2000. Foreign-born teaching assistants and the academic performance of undergraduates. American Economic Review 90:355–59. ——. 2007. Do foreign students crowd out native students from grad- uate programs? In Science and the university, ed. Paula E. Stephan and Ronald G. Ehrenberg. Madison: University of Wisconsin Press. ——. 2009. Immigration in high-skill labor markets: The impact of foreign students on the earnings of doctorates. In Science and engineer- ing careers in the United States: An analysis of markets and employ- ment, ed. Richard B. Freeman and Daniel L. Goroff. Chicago: Univer- sity of Chicago Press. ound, John, Brad Hershbein, and Bridget Terry Long. 2009. Playing the admissions game: Student reactions to increasing college competi- tion. Journal of Economic Perspectives 23:119–46. ound, John, and Sarah Turner. 2007. Cohort crowding: How resources affect collegiate attainment. Journal of Public Economics 91:877–99. ——. 2010. Coming to America: Where do international doctorate stu- dents study and how do US universities respond? In American universities in a global market, ed. Charles T. Clotfelter. Chicago: University of Chicago Press.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

B

B

C

C

C

C

C

C

C

D

F

G

G

H

H

S106 Orrenius/Zavodny

All

rown, Charles, and Mary Corcoran. 1997. Sex-based differences in school content and the male-female wage gap. Journal of Labor Eco- nomics 15:431–65. runello, Giorgio, and Lorenzo Rocco. 2011. The effect of immigration on the school performance of natives: Cross country evidence using PISA test scores. IZA Discussion Paper no. 5479, Institute for the Study of Labor, Bonn. ard, David. 2001. Immigrant inflows, native outflows and the local labor market impacts of higher immigration. Journal of Labor Economics 19: 22–64. ard, David, and John E. DiNardo. 2000. Do immigrant inflows lead to native outflows? American Economic Review 90:360–67. ard, David, and Alan B. Krueger. 1992. Does school quality matter? Returns to education and the characteristics of public schools in the United States. Journal of Political Economy 100:1–40. arrell, Scott E., Marianne E. Page, and James E. West. 2010. Sex and science: How professor gender perpetuates the gender gap. Quarterly Journal of Economics 125:1101–44. hellaraj, Gnanaraj, Keith E. Maskus, and Aaditya Mattoo. 2008. The contribution of international graduate students to US innovation. Re- view of International Economics 16:444–62. hin, Aimee, N. Meltam Daysal, and Scott A. Imberman. 2012. Impact of bilingual education programs on limited English proficient students and their peers: Regression discontinuity evidence from Texas. IZA Discussion Paper no. 6694, Institute for the Study of Labor, Bonn. ork, Daniel L., and Paul R. Voss, eds. 2006. Once, only once, and in the right place: Residence rules in the decennial census. Washington, DC: National Academies Press. ynarski, Susan. 2008. Building the stock of college-educated labor. Jour- nal of Human Resources 43:576–610. reeman, Richard B. 2009. What does global expansion of higher edu- cation mean for the US? NBER Working Paper no. 14962, National Bureau of Economic Research, Cambridge, MA. ould, Eric D., Victor Lavy, and M. Daniele Paserman. 2009. Does immigration affect the long-term educational outcomes of natives? Quasi- experimental evidence. Economic Journal 11:1243–69. riffith, Amanda L. 2010. Persistence of women and minorities in STEM field majors: Is it the school that matters? Economics of Education Re- view 29:911–22. anson, Gordon H., and Matthew J. Slaughter. 2013. Talent, immigra- tion, and U.S. economic competitiveness. Washington, DC: Compete America Coalition. ardoy, Inés, and Pål Schøne. 2013. Does the clustering of immigrant peers affect the school performance of natives? Journal of Human Capital 7:1–25.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

H

H

—

H

Ja

Je

K

K

K

L

L

M

N

—

N

—

Immigration and Majoring in Science and Engineering S107

All

oxby, Caroline M. 1998. Do immigrants crowd disadvantaged American natives out of higher education? In Help or hindrance? The economic implications of immigration for African-Americans, ed. Daniel S. Ha- mermesh and Frank D. Bean. New York: Russell Sage. unt, Jennifer. 2011. Which immigrants are most innovative and entrepre- neurial? Distinctions by entry visa. Journal of Labor Economics 29:417–57. ——. 2012. The impact of immigration on the educational attainment of natives. NBER Working Paper no. 18047, National Bureau of Eco- nomic Research, Cambridge, MA. unt, Jennifer, and Marjolaine Gauthier-Loiselle. 2010. How much does immigration boost innovation? American Economic Journal: Macro- economics 2:31–56. ckson, Osborne. 2011. Does immigration crowd natives into or out of higher education? Photocopy, Northeastern University. nsen, Peter, and Astrid Wurtz Rasmussen. 2011. The effect of immi- grant concentration in schools on native and immigrant children’s read- ing and math skills. Economics of Education Review 30:1503–15. ato, Takao, and Chad Sparber. 2013. Quotas and quality: The effect of H-1B visa restrictions on the pool of prospective undergraduate stu- dents from abroad. Review of Economics and Statistics 95:109–26. err, William R. 2013. U.S. high-skilled immigration, innovation, and entrepreneurship: Empirical approaches and evidence. NBER Working Paper no. 19377, National Bureau of Economic Research, Cambridge, MA. err, William R., and William F. Lincoln. 2010. The supply side of in- novation: H-1B visa reforms and U.S. ethnic innovation. Journal of Labor Economics 28:473–508. evine, Phillip B., and David J. Zimmerman. 1995. The benefit of addi- tional high-school math and science classes for young men and women. Journal of Business and Economic Statistics 13:137–49. uppino, Marc, and Richard Sander. 2012. College major competitive- ness and attrition from the sciences. Photocopy, University of Cali- fornia, Los Angeles School of Law. assey, Douglas S., ed. 2010. New faces in new places: The changing geography of American immigration. New York: Russell Sage. ational Academy of Sciences. 2007a. Beyond bias and barriers: Fulfill- ing the potential of women in academic science and engineering. Wash- ington, DC: National Academies Press. ——. 2007b. Rising above the gathering storm: Energizing and em- ploying Americans for a brighter economic future. Washington, DC: Na- tional Academies Press. ational Science Foundation. 2002. Science and engineering indicators 2002. Arlington, VA: National Science Foundation. ——. 2012. Science and engineering indicators 2012. Arlington, VA: National Science Foundation.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

S108 Orrenius/Zavodny

A

———. 2013. Science and engineering degrees: 1966–2010. Arlington, VA: National Science Foundation.

Neymotin, Florence. 2009. Immigration and its effect on the college-going outcomes of natives. Economics of Education Review 28:538–50.

Ost, Ben. 2010. The role of peers and grades in determining major per- sistence in the sciences. Economics of Education Review 29:923–34.

Rask, Kevin. 2010. Attrition in STEM fields at a liberal arts college: The importance of grades and pre-collegiate preferences. Economics of Edu- cation Review 29:892–900.

Regets, Mark C. 2001. Research and policy issues in high-skilled international migration: A perspective with data from the United States. In Innovative people: Mobility of skilled personnel in national innovation systems. Paris: Organisation for Economic Co-operation and Development.

Saiz, Albert. 2007. Immigration and housing rents in America. Journal of Urban Economics 61:345–71.

Schwartz, Amy E., and Leanna Stiefel. 2006. Is there a nativity gap? New evidence on the academic performance of immigrant students. Educa- tion Finance and Policy 1:17–49.

———. 2011. Immigrants and inequality in public schools. In Whither op- portunity? ed. Richard Murnane and Greg Duncan. Washington, DC: Brookings Institution.

Smith, Christopher L. 2012. The impact of low-skilled immigration on the youth labor market. Journal of Labor Economics 30:55–89.

Smyth, Frederick L., and John J. McArdle. 2004. Ethnic and gender differ- ences in science graduation at selective colleges with implications for ad- mission policy and college choice. Research in Higher Education 45:353–81.

Solnik, Sara J. 1995. Changes in women’s majors from entrance to grad- uation at women’s and coeducational colleges. Industrial and Labor Re- lations Review 48:505–14.

Turner, Sarah E., and William G. Bowen. 1999. Choice of major: The changing ðunchangingÞ gender gap. Industrial and Labor Relations Re- view 52:289–313.

Weinberger, Catherine J. 1999. Mathematical college majors and the gen- der gap in wages. Industrial Relations 38:407–13.

Wozniak, Abigail, and Thomas J. Murray. 2012. Timing is everything: Short-run population impacts of immigration in US cities. Journal of Urban Economics 72:60–78.

Xie, Yu, and Kimberlee A. Shauman. 2003. Women in science: Career processes and outcomes. Cambridge, MA: Harvard University Press.

Zafar, Basit. 2013. College major choice and the gender gap. Journal of Human Resources 48:545–95.

Zhang, Liang. 2004. Crowd out or opt out: The changing landscape of doctorate production in American universities. Cornell Higher Edu- cation Research Institute Working Paper no. 63, Cornell University.

This content downloaded from 129.015.064.251 on November 21, 2019 14:29:56 PM ll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).