The Determinants of International Enrollment in U.S. Higher Education: Labor market openness, and the unintended consequences of H-1B policy.

Transcription

1 The Determinants of International Enrollment in U.S. Higher Education: Labor market openness, and the unintended consequences of H-1B policy. Kevin Shih University of California at Davis August 16, 2014 Abstract International students have long comprised an important part of U.S. higher education. While much speculation exists, there is little empirical work that identifies the important economic factors that encourage thousands of students across the world to seek U.S. post-secondary education each year. This paper explores the relationship between foreign post-secondary enrollment and various home country and U.S.-specific factors that affect the demand and supply of higher education. Results reveal a positive and robust relationship between the college age population in a country and the number of students it sends to U.S. universities. In addition, the openness of the U.S. labor markets to college-educated foreigners, as proxied by H-1B visa policy, is also positively related to international enrollment. I explore the causal effect of labor market openness by exploiting a dramatic fall in the H-1B visa cap, from 195,000 to 65,000 per year, that occurred in Triple difference estimates suggest that the fall in the cap lowered foreign enrollment by 9%. JEL Codes: F22, I21, J11 Keywords: International enrollment, Foreign students, Determinants, Globalization, Higher Education. This research was conducted while the author was a Research Associate of the Institute of International Education and a Doctoral candidate at the UC Davis Economics Department. The author thanks Christine Farrugia, Rajika Bhandari, Giovanni Peri, and Chad Sparber for insightful discussions and suggestions. This research does not reflect the views of the IIE. The author is accountable for all errors contained herein. 1

2 1 Introduction International students have long comprised a large portion of higher education enrollment in the U.S. Today over 800,000 individuals from across the world study at U.S. colleges and universities on temporary student visas. 1 However, remarkably little is known about the factors that encourage thousands of students to apply and enroll at U.S. post-secondary institutions each year. Strong growth in foreign enrollment over the last half century (see figure 1), combined with unprecedented recent surges in undergraduates from abroad beckons the need to identify the determinants of international student flows. Additionally, the presence of foreign students carries strong implications for U.S. higher education. Prior research suggests that foreign students may impact the educational attainment of natives. Foreign students might crowd natives out of higher education (Borjas 2004; Hoxby 1998) by increasing competition for seats. Alternatively, international students may bring positive externalities which could offset the increased competition (Hunt 2012; Jackson 2012). Furthermore, because foreign students often pay full sticker price tuition they contribute to expanding university resources. 2 Understanding the forces that affect foreign enrollment in the U.S. has important implications for both the education of U.S. natives and the U.S. higher education sector. Fluctuations in foreign enrollment may also have implications for the U.S. labor market. Although student visas do not allow international students to remain in the U.S. following completion of their studies, a large number of international students do remain by transferring to a work visa. 3 Thus, inflows of highly educated foreigners into the U.S. labor market 1 Open Doors (2013) puts the actual number at 819, A recent study (NAFSA 2013) estimated that over the academic year international students contributed over $17 billion to U.S. higher education through tuition and fees. 3 Ruiz (2013) reports that 35% of all H-1B visas awarded in 2010 were to individuals transferring from an F-1 student visa. Furthermore, among those F-1 visa holders transferring to an H-1B visa, nearly 75% were 1

3 may have strong impacts on the wages and productivity of U.S. natives (e.g. Borjas 2009; Peri et al. 2013). Finally, very recent research has identified foreign students and high skill workers as key contributors to advances in science, engineering, and technological innovation in the U.S. (Kerr & Lincoln 2010; Chellaraj et al. 2008; Stuen et al. 2012). Innovation and technological advances, in turn, are the main drivers of economic growth. Therefore, identifying the forces that govern foreign student mobility may be an important policy tool for the U.S. to remain at the technological frontier. There have been many descriptive studies regarding trends in international student mobility (e.g. Bound, Turner, & Walsh 2009; Bound & Turner 2010; Verbik & Lasanowski 2007; Guruz 2011; Findlay 2011). These works have importantly recognized the rising global movement of students across borders for higher education, and particularly into U.S. universities. Additionally, there has been a great amount of effort in identifying the economic factors that influence international migration (e.g. Hanson & McIntosh 2010; Hatton 2005; Hatton & Williamson 2005; Mayda 2010; Ortega & Peri 2013, 2014). In contrast, however, there is much less research on the determinants of international student flows. In theory the number of students coming from abroad each year is the equilibrium result of competition between supply and demand in the market for U.S. higher education. Factors, such as changes to the relative benefits of a U.S. degree, the relative costs of receiving a U.S. education, or preferences for study abroad in the U.S., may affect foreign demand for U.S. higher education. Alternatively, factors such as changes to student visa policy or fluctuations in federal higher education funding may alter the supply of higher education (the number of seats) available for foreign students. individuals with graduate degrees. 2

4 This paper provides an empirical analysis of the determinants of foreign enrollment at U.S. colleges and universities in two parts. The first part uses a panel data on 137 sending countries and examines a wide range of demand and supply factors. Using a panel fixedeffects regression model, I assess the importance of various factors including tertiary age population and real GDP per capita in sending countries, bilateral USD exchange rates, imports from and exports to the U.S. Additionally, I evaluate the importance of U.S. labor market opportunities by using measures of expected wages (average wages of immigrants from sending countries) and labor market openness (H-1B visas issued to sending countries). 4 The results show that tertiary age population in sending countries is a strong and robust determinant of international enrollment. Further, labor market openness/access, as proxied by H-1B visa issuances, is also positively related to international enrollment. The second part explores the role of U.S. labor market openness in greater detail by exploiting a policy shift that greatly decreased access to the U.S. labor market for highly educated foreigners. Specifically, in 2004 the cap on H-1B visas was reduced from 195,000 per year to 65,000 per year. Interestingly, however, this reform did not represent a great reduction in labor market access for all sending countries. By 2004, a group of countries (Australia, Canada, Mexico, Chile, Singapore) had established preferential trade agreements with the U.S., which created alternative visas for their highly educated nationals. Furthermore, not all foreign students saw decreased access. Foreign graduate students were less affected than undergraduates because Congress had set aside an extra 20,000 H-1B visas for holders of graduate degrees, and exempted applicants seeking employment in Universities from the cap. 5 I assess this natural experiment using a triple difference framework, and find that the 4 The H-1B visa is the main visa for highly educated foreign-born individuals to work in the U.S. The H-1B visa is described in more detail in sections 5 and 7. 5 Since many foreign graduate students go on to become Professors or researchers within Universities, the exemption meant that the fall in the cap represented a much smaller decrease in U.S. labor market openness than for undergraduates. 3

5 2004 drop in the size of the H-1B program reduced international enrollment by 9%. Section 2 briefly reviews the existing literature on the determinants of international higher education enrollment in the U.S. Section 3 provides some descriptive statistics on international students in the U.S. to help provide context to the analyses. Section 4 provides a theoretical context to how various economic factors affect foreign enrollment in the U.S. Section 5 outlines the empirical model and data used to assess the importance of various determinants of international enrollment at the graduate and undergraduate levels. Section 6 presents the main results, and provides some extensions and robustness checks. Section 7 describes the H-1B policy experiment and discusses the results. Section 8 concludes. 2 Literature Review A large and active body of research has attempted to identify the determinants of international migration. These papers generally use cross-sectional or panel data on several countries to relate immigrant stocks or flows to a range of economic variables, such as GDP, exchange rates, distance, language, trade, and income. For example, Mayda (2010) and Ortega & Peri (2013) examine determinants of bilateral migration flows between OECD countries and find that increases in income per capita in destination countries tend to attract more immigrants. Clark et al. (2007) focus specifically on the U.S. and find that policy plays an important role in regulating the flow of immigrants. 6 In contrast, there is much less empirical work exploring the forces that cause millions to seek higher education abroad each year. 7 This research question bears special importance 6 Other examples include Hanson & McIntosh (2010), Hatton (2005), Hatton & Williamson (2005), and Ortega & Peri (2014). 7 UNESCO estimated that in 2012 over 3 million students at the tertiary education level were studying outside their home country. See 4

6 for the U.S., whose colleges and universities lead the world in hosting over 15% of the stock of internationally mobile students. 8 Furthermore, the determinants of international student mobility may be quite different than the determinants of international migration. Because U.S. student visa policy does not permit stay beyond completion of studies, prospective foreign students may be more myopic than prospective migrants. 9 Alternatively, the determinants of foreign student mobility and immigration may be quite similar, and thus, changes to U.S. immigration policy may result in unintended consequences for the composition of students at U.S. colleges and universities. A small number of papers have used empirical methods to identify the determinants of foreign student flows. These papers implement empirical designs similar to those used to study the determinants of international migration, often relating international enrollment to a variety of factors (e.g. home country GDP, population and higher education quality, exchange rates, trade with the U.S., etc. ). 10 Many studies have focused on international student mobility within the European Union (Brezis & Soueri 2011; Gonzalez et al. 2010; Van Bouwel & Veugelers 2013) and to the United Kingdom (Naidoo 2007; Jena & Reilly 2013). While the U.S. has been the world leader in hosting international students over the past half century (Freeman 2010), only a handful of studies have examined international enrollment in the U.S. Early work by McMahon (1992) examined the inflow of students to U.S. universities from 18 developing nations in the decades after World War II, and found trade linkages established over the 1960 s and 70 s were positively related to international enrollments. 8 See the OECD report: how-is-international-student-mobility-shaping-up_5k43k8r4k821-en. 9 For example, short-term fluctuations in exchange rates may strongly affect international student flows, but have no effect on immigration. 10 Kato & Sparber (2011) is a notable exception in that they use quasi-experimental difference-in-difference methods to examine how H-1B policy affects the quality of foreign students applying to U.S. universities. 5

7 Rosenzweig (2006) expanded the analysis to include 125 sending countries, relating student visas issuances with various economic variables including wages in sending countries. Although the regressions only rely on cross-country variation, the results suggest that higher wages in home countries are negatively correlated with the number of international students in the U.S. Recent studies have begun to improve upon these early papers by using panel data. Using data from , Liu & Wang (2009) find that Federal support for higher education and the population of young individuals in sending countries are positively linked with international enrollment. Bird & Turner (2014) separately analyze factors affecting undergraduate and graduate enrollment from abroad. They also find a similarly strong relationship between exchange rates and undergraduate enrollment that is primarily driven by China. Additionally, they find that real GDP per capita and the college age population in sending countries have strong positive relationships with both undergraduate and graduate enrollment in the U.S. This paper empirically explores the determinants of international student flows to the United States, building from these prior works. I contribute to the existing literature in three key ways. First, I compile panel data from a variety of sources to assess a larger number of economic factors than previously considered. Second, I develop and test two key measures of the strength and accessibility of U.S. labor markets, which have not previously been considered in relation to international student flows. Lastly, I leverage a unique natural experiment to identify the causal impact of decreased U.S. labor market access via a policy change that reduced the number of H-1B visas available. Before continuing further, the next section presents some basic descriptive statistics on international students in order to better understand and frame the results. 6

8 3 Summary Statistics on International Students International students currently comprise 4% of total higher education enrollment (Open Doors, 2013). This statistic, however, masks tremendous heterogeneity in foreign student representation within different areas of higher education. Importantly, the share of international students varies largely by academic level. Table 1 shows that during the academic year, although international students received only 3.3% of bachelors degrees, they received nearly 12% of masters degrees, and almost 30% of doctoral degrees. Furthermore, the significant presence of foreign students in U.S. graduate education is not a recent phenomenon Bound et al. (2010) document that during the two decades between 1936 and 1956, international students were already receiving more than 10% of all doctoral degrees awarded in engineering, physical sciences, life sciences, and economics. Understanding the forces that affect enrollment of international students in the U.S. is crucial for educators, administrators, and policymakers to properly plan for future demand shocks to higher education. 11 Even greater heterogeneity in international student representation exists when looking across fields of study. Table 1 displays the top 5 fields of study by academic level, ranked by the percent of degrees awarded to international students. In about half of all engineering, computer/information science, and mathematics & statistics PhDs were awarded to noncitizens. Indeed, the tendency of international students to study such Science, Engineering, Mathematics, and Technology (STEM) based fields 12 have made them an important 11 There is some evidence that increased students from abroad may crowd natives out of higher education. Borjas (2004) finds that increased graduate students from abroad is associated with lower enrollment of white males in U.S. graduate schools. Similarly, Hoxby (1998) finds that increased foreign students at the undergraduate level is associated with fewer native minorities attending college. In contrast, several studies find no evidence of crowding out (e.g. Jackson, 2009; Hunt, 2012). 12 Note that engineering, computer/information science, and mathematics & statistics fields receive the highest share of international students at the bachelors, masters, and doctoral levels. 7

9 contributor to research productivity at U.S. universities and U.S. STEM innovation. 13 An additional important fact regarding international students concerns their countries of origin. While U.S. universities host students from almost every country in the world, a small number of countries account for the majority of international students. In 2012, the top 30 sending countries comprise nearly 90% of all international enrollment. The top 10 Asian countries alone account for 66% of total enrollment from abroad. Importantly, approximately half of all international students come from just two countries: China and India. Chinese students account for 33% of all international post-secondary students, while India provides nearly 20% of all enrollment from abroad. 14 A final noteworthy feature of international students concerns their sources of funding. The F-1 Visa, the main student visa for foreign nationals to attend accredited post-secondary institutions, requires applicants to demonstrate the ability to pay full sticker price tuition. Further, international students are generally not available for federal financial aid. Over 63% of international students report personal and family sources as their primary source of funding (Open Doors 2013). Thus, because international students generally pay full sticker tuition price which is funded by personal and family sources, they generate large revenues for the higher education sector. A recent study (NAFSA, 2013) estimated that over the academic year international students contributed over $17 billion to the U.S. higher education through tuition and fees. Further, NAFSA (2013) estimates that international students supported 313,000 jobs and contributed a total of $24 billion to the U.S. economy via spending for tuition, housing, and local services. Thus, the factors that affect international student enrollment also directly affect the resources available to U.S. 13 Chelleraj et al. (2008) show that international students increase innovative activity through increased patent applications and grants. Black & Stephan (2007) document that international students and postdoctoral scholars contribute tremendously to scientific research publications. 14 See International-Students/All-Places-of-Origin/

10 higher education institutions. 4 Theory: Supply and Demand for U.S. Higher Education U.S. policymakers and educators often speculate about the economic and institutional forces that affect international student s decisions to study in the states. For example, Victor Johnson, a senior adviser for public policy for NAFSA speculates: If people are coming here for a couple of days to do nothing but buy a new wardrobe, it would be strange if the exchange rate didn t affect their educational decisions. 15 Alternatively, an article describing a recent surge in Chinese students also presents anecdotal evidence that the ability to explore different majors is an attractive feature of U.S. higher education, as the rigid Chinese system forces students to choose majors prior to enrolling in college (Haynie, 2013). Basic economic theory suggests that the number of foreign students enrolled at U.S. universities is the resulting equilibrium of supply and demand in the market for higher education. International student demand for U.S. higher education may be affected by a multitude of factors that alter the costs and benefits to studying in the states. Factors that increase the net benefit are likely to increase foreign enrollment. These include, for example, higher expected returns (wages) to a U.S. degree, greater probability of employment in the U.S., or better quality universities than at home. In contrast factors that increase the costs of enrolling in a U.S. university are likely to reduce foreign enrollment. For example, rising tuition, depreciation of home currency relative to the US dollar, or increasing wages at home all directly increase the monetary or opportunity costs of studying abroad in the U.S. Alternatively, the supply of higher education seats available for foreign students is de- 15 See Schworm (2008). 9

11 termined by the costs and benefits to universities of hosting international students. 16 For example, if international students benefit universities through paying higher tuition than natives, admissions committees may increase the seats available to international students. Increases in federal higher education funding or declining enrollment of native-born U.S. students may also increase the capacity of universities to host international students. Alternatively, more stringent student visa policy might raise the administrative costs of hosting international students. 17 While theory is a useful starting point, it is unclear which factors should matter the most. Furthermore, theory cannot shed light on how large of an increase/decrease in enrollment from abroad should be expected from fluctuations in factors that affect supply or demand, unless strict assumptions are imposed on the elasticities of supply and demand curves. Nevertheless, empirical analysis can provide valuable insight into these issues. 5 Empirical Methods & Data This paper contributes to existing literature by compiling a wider array of economic determinants than previously considered. The principal source of data on international student enrollment in the U.S. comes from the Institute of International Education (IIE). 18 I collect yearly undergraduate and graduate enrollment counts by country of origin from the Open Doors reports. These data list total enrollment counts by country of origin and by academic level. 16 Importantly, U.S. student visa policy has historically had no caps. Thus, if an international student is accepted at a U.S. university, and passes various security checks, he/she may receive a student visa. 17 There is evidence that heightened restrictions on visa issuances after 9/11 increased the burden for university administrators to process visas (Alberts 2007). 18 The IIE was founded in 1919 and has published yearly statistics of international students in U.S. higher education in volumes called Open Doors since

12 I supplement this data with a variety of sources that measure economic factors which might affect the supply or demand for higher education in the U.S. To assess the impact of opportunity costs in the home country, I use Real GDP per capita as a proxy. Real GDP by country comes from the Penn World Tables (Feenstra et al., 2013). Specifically, I use Expenditures-side Real GDP, which better captures real living standards and is more suitable for analysis across countries and over time. 19 I then calculate real GDP per capita by dividing expenditure-side real GDP by population, also available in the PWT. While increases in Real GDP may reflect rising wages in the home country and hence increasing opportunity costs, they may also represent rising income which could enable many more students to afford study in the U.S. Exchange rates have often been speculated to be an important factor in the decision to study in the U.S. To assess their importance, I use exchange rates denominated in homecurrency per USD, also available in the PWT. Holding prices constant, fluctuations in exchange rates should raise or lower the cost of attending college in the U.S. 20 For example, an appreciation (increase) in the Euro/USD exchange rate raises the cost of attending a college in the U.S. for students from Europe. Thus, exchange rates should be negatively correlated with foreign student enrollment. As the focus of this analysis is on the quantity of foreign students, it is likely that demographic shifts in sending countries may affect the number of internationally mobile students. To assess this channel I gather data on the population of tertiary education/college age individuals by sending country from UNESCOs Institute for Statistics. 21 The college 19 Expenditure-side real GDP is calculated at chained Purchasing Power Parities (PPPs), to compare relative living standards across countries and over time. Using PPPs adjustments are important to capture the real costs of living which differ across countries. See Feenstra et al. (2013) for detailed description of this variable. 20 These costs may include tuition, living costs, transportation costs, and application fees. 21 Tertiary age differs depending on the country, but generally cover individuals aged who attend educational levels from undergraduate to graduate education in the U.S. See for more information. 11

13 age population may be an important determinant of foreign student enrollment in the U.S. A country that experiences a particularly large birth cohort may mechanically have a larger number of individuals from that cohort applying and attending college, and naturally some of this excess demand for higher education may spill over to the U.S. Trade linkages between countries may foster other types of interaction, including educational exchange (McMahon 1992). I compile bilateral import and export values from U.S. census data on import and export values by Harmonized System (HS) codes, U.S. state, partner country. These data are aggregated across states and HS codes to obtain a total value of imports and exports between the U.S. and sending countries. 22 Finally, I develop measures of U.S. labor market strength and labor market access that have yet to be included in analysis on the determinants of international student mobility. In particular studying in the U.S. may be attractive because of the potential to earn U.S. wages and/or find employment in the U.S. To proxy for expected U.S. wages to students from sending country c, I calculate average annual wages of immigrants aged from country c with a bachelors degree or higher from the 2000 U.S. Census and the American Community Surveys. 23 Average wages are available for 97 countries of origin, but unfortunately no surveys are available from 1998 and 1999 that provide enough observations to accurately estimate averages for each country. 24 To proxy for U.S. labor market openness, I gather data on H-1B visas issued by sending country from the State Department (Department of State, 2012). The H-1B visa is a 22 These data were kindly made publicly available by Peter Schott. See Schott (2008) for more details. Data were taken from Peter Schott s website:. 23 Immigrants are defined as individuals not born in the U.S. and who are not born abroad to U.S. citizen parents. The sample is limited to immigrants who worked a positive number of weeks in the previous year, reported earning positive wage/salary income, and have a bachelors degree or higher. 24 The Current Population Survey is available in 1998 and 1999, but the sample size was too small to accurately calculate average wages for the 97 countries. 12

14 three year work permit (renewable up to 6 years), generally for highly educated individuals. Only U.S. employers may apply for H-1B visas on behalf of foreign applicants, and foreign applicants must have an employment offer thus the H-1B visa is tied to employment in the U.S., and changes in the number of H-1B visas issued may reflect changes in employment opportunities. Additionally, the U.S. government has maintained numerical caps on the number of H-1B visas allowed in each year. Thus, fluctuations in H-1B visas issued also reflects policy openness to foreign nations and access to U.S. labor markets. The H-1B is the visa that international students would most likely transition to in order to work in the U.S. following completion of their studies. 25 In fact, Ruiz (2013) uses data from SEVIS to show that around 35% of H-1B visas issued in 2010 were to international students transferring from F-1 visas. Since more than a third of H-1B visas are awarded to foreign students, I use H-1B visas lagged three years to reduce the possibility of mechanical correlation due to graduating students transitioning to H-1B in the same year. 26 I combine these data to form a panel dataset consisting of 137 sending countries from and implement the basic empirical specification: log E ct = α+β 1 log RGDP ct +β 2 log P op college ct +β 3 log Bilateral ct +β 4 log LaborMkt US ct +γ c +γ t +ε ct (1) In specification 1, E ct represents the total number of international students from country c enrolled in U.S. universities in year t. 27 RGDP ct represents real GDP per capita of country c. P op college ct measures the population of tertiary education/college age (18-30) in sending countries. Bilateral ct include three variables that measure the economic relationships be- 25 For more background on the H-1B visa see Kato & Sparber (2013). 26 Thus H-1B visas issued in 1997 are related to enrollment in 2000, visas issued in 1998 are associated with enrollment in 2001, etc. Note this strategy is less than perfect because a mechanical correlation may still occur using lags if H-1B visa holders end up enrolling at U.S. universities at a later point in time. 27 Note that enrollment is measured for the academic year from fall of year t to spring of year t + 1. Thus enrollment in year 1998 reflects enrollment for the academic year

15 tween the sending country and the U.S. These include exchange rates, imports, and exports. LaborMkt US ct represents the two explanatory variables that capture the strength or openness of the U.S. labor market: average wages of college educated immigrants from country c, and H-1B visas issued to individuals from country c. 28 Lastly I include year effects, γ t, to control for shocks affect all international students, and country fixed effects, gamma c to control for time-invariant, country-specific factors. 6 Results To begin understanding the determinants of international student enrollment, Table 2 presents regression results of specification 1 on 137 countries from Columns (1) and (2) display results on total international enrollment, while columns (3)-(6) split enrollment by academic level. Columns (3) and (4) show results for international undergraduate enrollment, and columns (5) and (6) show results for international graduate enrollment. All variables are expressed in logs, unless otherwise specified. Regression coefficients are displayed with standard errors clustered at the country level in parenthesis. Columns (1), (3), and (5) include log exchange rates, log real GDP per capita, log imports, log exports, and log tertiary age population as explanatory variables. These specifications also control for year effects and country fixed effects. The results from column (1) show a positive and statistically significant relationship between total international enrollment in the U.S. and both real GDP per capita of sending countries and tertiary age population in sending countries, and U.S. exports to sending countries. These results suggest that 28 Expected wages and employment opportunities in the U.S. conceivably vary by county of origin due to selection of international students. A country that tends to send students from the upper tail of the ability distribution can expect its students to have higher wages/employment opportunities in the U.S. than a country whose students are drawn from the lower tail of the ability distribution, if there is a strong correlation between abilities at home and the U.S. These concepts were theoretically and empirically elucidated by Borjas (1987). 14

16 living standards and demographic changes in the home country may influence the number of students that ultimately study in the U.S. Similar effects are found when splitting by academic level in columns (3) and (5), although exports lose statistical significance when examining graduate enrollment from abroad. Columns (2), (4), and (6) adds in very demanding country specific linear time trends to control for linear growth of home-specific factors, such as constant growth in educational quality. While the coefficients on real GDP per capita and exports lose statistical significance, note that the coefficients on tertiary age population remains robust to these demanding controls. The size of the coefficients suggest that a 1% increase in a country s college age population is associated with a 1% increase in undergraduate enrollment from that country, and a 0.7% increase in graduate enrollment from that country. Demographic shifts that increase the number of individuals in a birth cohort can increase the numbers of students studying in the U.S. when that cohort reaches college age through a variety of mechanisms. Increases in the size of birth cohorts may mechanically lead to increased numbers of students studying abroad if preferences for U.S. education are relatively fixed within countries and over time. For example, assuming U.S. higher education supply is not operating at capacity, if one out of every two college age individuals prefer studying in the U.S. instead of at home, increases in the college age population will mechanically increase the number of students from that country studying in the U.S. Alternatively, larger cohort size may lead to increased competition within that cohort for college admission. If universities in the home country cannot absorb this higher demand, students will forgo college education unless they can seek opportunities abroad Bound & Turner (2007) find that increased cohort size is associated with lower undergraduate degree attainment in the U.S. Bird & Turner (2014) find that the impact of college age population growth is lower in countries with many universities per population. 15

17 The strength of the relationship between college age population can be visualized in Figure 2, which displays a scatterplot of changes in tertiary age population against changes in graduate or undergraduate enrollment from , and from While the positive relationship identified in regressions is salient in the scatterplots as well, figure 2 reveals two possible outliers may drive the results. India dominated both population growth and growth in international student enrollment in the first half of the sample, while China has emerged as the leader in the most recent decade. It is quite plausible that India and China are the main drivers of the strong relationship between cohort size and international student enrollment in earlier regressions. Table 3 examines whether the relationship between enrollment and college age population remains after removing India and China from the regressions. Column (1) runs nearly the same regression as in column (2) of table 2, but also removes China and India. Columns (2) and (3) also remove China and India and replicate the regressions of columns (4) and (6) in table 2, respectively. Interestingly, the results show that the positive and significant relationship between population growth and enrollment is robust to the exclusion of India and China. Coefficients do not change much and remain highly statistically significant. 6.1 The impact of expected U.S. wages and labor market openness The regression analysis presented thus far has focused more on the impact of home country characteristics (Real GDP per capita, college age population), while excluding many important U.S. characteristics that may attract international students. An important factor in the decision to study abroad in the U.S. may be the possibility of future employment in the U.S., and/or the wages associated with working in the U.S. To test the importance of access to U.S. employment I add log of H-1B visas issued (lagged three years) to each country. 16

18 To evaluate the effect of potential earnings in the U.S. I use the log of average U.S. yearly earning of college educated immigrants from each country. Because the availability of wages by country are limited, the analyses which include log average yearly earnings of immigrants is limited to 91 countries from Table 4 shows regression results that add log H-1B visas issued four years prior, in columns (1), (3), and (5), and log average annual earnings, in columns (2), (4), and (6). Columns (1) and (2) present results for total enrollment, columns (3) and (4) show results for undergraduate enrollment, and column (5) and (6) presents results for graduate enrollment. The results show that college age population continues to have a strong positive relationship with enrollment from abroad. Interestingly, while average wages seem to have little relation to international enrollment, the number of H-1B visas issued is positive and statistically significant in all but one specification. These correlations suggest that when students evaluate whether or not to apply and enroll in a U.S. university, their decisions may be heavily influenced by the probability of securing employment in the U.S. after completing their studies. Therefore, changes in immigration policy and the availability of employment opportunities in the U.S. may be an important factor that determines the flow of students from abroad. Although the positive and statistically significant relationship between international student enrollment and both college age population and H-1B visas issued is compelling, the empirical design is not without flaws. Many factors may bias the results, thus preventing a causal interpretation. For example, increasing rates of educational attainment in countries may cause increases in both the number of H-1B visa recipients and the number of students enrolled at U.S. universities, leading the coefficient on H-1B visas issued to be biased upward. I thus present a basic robustness check that exploits the fact that H-1B visas limits do not apply to all countries, and later turn to a policy-driven analysis. 17

19 As stated earlier, changes in H-1B visas issuances generally reflect changes in access to U.S. labor markets. 30 However, a few countries, since 2004, have established preferential agreements with the U.S. that allow its college educated workforce to circumvent H-1B limits. Nationals of Australia, Canada, Chile, Mexico, and Singapore can acquire alternative work permits that are not bound by H-1B restrictions. 31 Thus, if a truly positive relationship between H-1B visas and foreign student enrollment exists because potential foreign students are attracted by access to U.S. labor markets, then it should also be true that countries whose access to U.S. labor markets are less bound by H-1B policy should show little to no relationship between enrollment of foreign students and visas issued. Put differently, for countries whose college educated nationals have several other alternatives to obtaining work permits in the U.S., changes in H-1B issuance should not be a very good signal of access to U.S. labor markets, and thus foreign students should respond very little. I formalize this robustness check by interacting log H-1B visas issued with a dummy variable that equals one if the country currently has substitutes to the H-1B visa via preferential trade agreements. 32 If the aforementioned theory is true, then it is necessary (but not sufficient) that the coefficient on the interaction term be negative. This is because the total effect of H-1B visas for countries not bound by H-1B policy the coefficient on log H-1B visas issued plus the coefficient on the interaction term should be less than the effect of H-1B visas for countries bound by H-1B policy, which is simply the coefficient on log H-1B visas. The results of this robustness check are shown in table 5. Columns (1), (2), and (3) present results for total, undergraduate, and graduate enrollment, respectively. In all cases 30 Additionally, changes in H-1B visas issued may reflect changing supply of college educated workers in home countries. Controlling for college age population partially accounts for this, but does not completely rule out a home country supply side story. 31 See Kato & Sparber (2013) for a complete description of the policies surrounding these nations. 32 I.e. the dummy variable equals one if the country is Australia, Canada, Chile, or Mexico. Singapore is excluded from all analysis in the paper because of a lack of data. 18

20 the coefficients on the interaction term are negative and statistically significant. 33 Further, a wald test of the equality of the log H-1B coefficient and the coefficient on the interaction term fails to reject the null hypothesis that they are equal. Thus the results are consistent with the hypothesis stated earlier availability of H-1B visas seems to be positively correlated with international student enrollment, except for countries who have alternative visa programs to the H-1B visa. 7 Unintended Consequences - an H-1B Policy Experiment The results presented thus far have revealed that U.S. labor market openness might be an attractive factor that prospective international students weigh when deciding whether to enroll in the U.S. undergraduate or graduate programs. The regression analysis uncovered a strong positive relationship between H-1B visas and foreign enrollment in U.S. universities. Importantly, however, these results still fall short from establishing a causal relationship. There may be various unobserved confounding factors that vary within countries and over time, that are correlated with both enrollment in the U.S. and H-1B visa issuances. For example, the strong presence of U.S. firms in a sending country may simultaneously foster knowledge diffusion that increases opportunities to secure H-1B visas and study at U.S. universities. Policy changes to the H-1B program offer unique natural experiments to estimate the causal impact of H-1B visas, and hence labor market access, on the decision to study in the U.S. The H-1B program began in 1990 with a congressionally mandated cap of 65,000 visas per year. The cap was increased in 1998 to 115,000, and again to 195,000 in To be precise, the combined effect of a 1% increase in H-1B visas issued on graduate enrollment from Canada would be the main effect of H-1B visas (0.143) plus the effect of the interaction (-0.211). 34 The expansion of the cap to 195,000 was a part of the American Competitiveness in the Twenty-First 19

21 A series of policy changes in 2004, in particular, provide a very good natural experiment to explore impacts on foreign enrollment. In 2004, the H-1B cap fell, from 195,000, to its original level of 65,000 per year, marking a dramatic decrease in foreign access to the U.S. labor market. Interestingly, however, this fall in the cap did not necessarily signal lower access to the U.S. labor market for all countries. Importantly by 2004, four countries Canada, Chile, Mexico, and Singapore had signed agreements with the U.S. that provided their highly educated workers alternatives to the H-1B visa. In 2005, Australia also signed a preferential agreement that provided an alternative for highly educated Australian nationals to work in the U.S. 35 Thus, while students from most countries experienced a dramatic restriction in access to the U.S. labor market (treated countries), students from 5 control countries did not. Comparing changes in foreign enrollment from treated countries against enrollment changes from the five control countries could provide causal estimates of the impact of H-1B policy on foreign enrollment. This difference-in-differences design is used by Kato & Sparber (2011) to examine the impact of H-1B policy on the quality of foreign students. They assess how SAT score reports to from international students to universities evolved after the 2004 fall in the cap. While they do not provide formal difference-in-differences analysis on international student quality, they do note that from 2001 to 2006, undergraduate enrollment from the 5 control countries remained relatively constant, while undergraduate enrollment at other countries declined by 14%. Century Act (AC21). 35 Specifically, Canada, Mexico, Chile, and Singapore signed free trade agreements with the U.S. These agreements created alternative visas which were very similar to the H-1B visa. In particular, TN visas were created in 1994 under the North American Free Trade agreement for citizens of Canada and Mexico. While there is no cap on TN visas, there is an approved list of occupations that is more restrictive than the H-1B. H-1B1 visa program was enacted in September 2003, which essentially set aside 1,400 of the available H-1B visas for citizens of Chile, and 5,400 for citizens of Singapore. Lastly, in May 2005 a bill was enacted establishing E-3 visas for Australian citizens. Capped at 10,500 visas per year, the E-3 visa provided Australian professionals a close substitute to the H-1B visa. For further details regarding these policy changes see Kato & Sparber (2011). 20

22 The difference-in-difference design is difficult to implement empirically as only having 5 countries in the control group may afford too small of a sample size to detect any meaningful effects. Interestingly, however, another reform in 2004 provides an additional control group the government mandated that 20,000 additional H-1B visas, not included in the cap of 65,000, were to be reserved only for individuals with graduate degrees. Additionally, after 2000 foreign highly educated workers hired by non-profit organizations and universities were exempt from the cap. These two reforms meant that the H-1B cap reduction in 2004 represented much smaller declines in U.S. labor market access to students from abroad deciding whether to enroll in U.S. graduate programs. Thus, comparing the enrollment behavior of foreign undergraduates (treated students) relative to foreign graduate students (control students) adds another dimension of plausibly exogenous variation to identify the causal impact of H-1B policy on foreign enrollment. Thus I build on the insights of Kato & Sparber (2011) and formalize these various reforms in a classic triple difference regression framework to estimate the causal impact of the H-1B visa policy on international student enrollment. I gather data from 2000 to 2007 on 149 countries that send undergraduate and graduate students to the U.S. I consider the five countries Canada, Mexico, Australia, Chile, and Singapore to be control countries, whereas all other countries are treated. I define the pre-treatment period from , and the post-treatment period from I remove 2004 from the analysis as Australia was still considered to be treated in Enrollments are separated by undergraduate and graduate levels, and I consider undergraduates as treated students, and graduate students as control students. 21

23 7.1 Comparison of Means The top panel of Table 6 shows a simple comparison of average log undergraduate enrollment for treatment and control countries, before and after the 2004 reforms. The table is divided into four cells by time (pre-2004 reform vs. post-2004 reform) and country (treated vs. control). Each cell shows standard errors, clustered at the country level, in parenthesis, while the number of observations appear in brackets. Average undergraduate enrollment fell by around 5% for control countries, while falling 12.2% for treated countries. Thus, enrollment fell in treatment countries relative to control countries by -6.8%. This difference-in-difference estimate (DD T S ) suggests that the 2004 reduction in the H-1B visa cap lowered undergraduate enrollment by nearly 7%. However, the DD T S estimate is statistically insignificant, likely due to the small number of observations in the control group. Further, the validity of this difference-in-difference estimate requires that enrollment in treatment and control countries would have trended similarly in the fall in the H-1B cap. We can partially examine this assumption by observing the behavior of graduate enrollment over this period. As stated earlier, by 2004 several reforms had created alternative pathways for graduate students to gain access to the U.S. labor market. The bottom panel of Table 6 shows similar mean comparisons of log graduate enrollment. The simple differences in means show that graduate enrollment declined by a statistically insignificant 1.4% for control countries, while increasing by 0.9% in treated countries. Thus, the difference-indifference estimate for graduate students (DD CS ) suggests that the 2004 drop in the H-1B visa cap increased graduate enrollment by 2.3%. However, the DD CS estimate is statistically indistinguishable from zero. To better assess the fall in the H-1B visa cap of 2004, I turn to a triple difference (DDD) 22

24 framework. The validity of DDD estimates requires that there were no other shocks that coincided with the fall in the H-1B cap and also differentially affected undergraduates in treated countries. Further, DDD estimate removes trends within-country and within-academic level. The final row in Table 6 provides the triple difference comparisons in means. The DDD estimate shows that the H-1B reform of 2004 led to nearly a 10% fall in international enrollment in the U.S. 7.2 Robustness I examine the robustness of DDD results by using a regression framework. Importantly, treatment and control countries may differ in a variety of factors that may lead to bias. To this end I use data from the Penn World Tables on various country-level attributes, and control for these in the regressions. Formally, I use the following design: log E lct = α + γ c + γ l + γ t + β 1 (P T C) + β 2 (P T S) + β 3 (T C T S) + β 4 (P T C T S) + δx ct + ɛ lct (2) The dependent variable in specification 2 represents log enrollment, which varies by academic level (l = undergraduates or graduates), country (c) and year (t). I control for academic level, country, and year effects (γ l, γ c, and γ t, respectively) and all two-way interactions of P (a dummy equal to 1 for all years after the 2004 drop in the H-1B visa cap), T C (a dummy equal to one for treated countries), and T S (a dummy equal to one for treated students (undergraduates). The coefficient on the triple interaction, β 4, gives the DDD estimate. To explore the robustness of β 4, I include various controls (X ct ) that vary by country 23