The Distributional Impact of High-Skilled Immigration: A Task-Based Approach

Transcription

1 The Distributional Impact of High-Skilled Immigration: A Task-Based Approach Gary Lin Cornell University This Version: May 8, 207 Abstract While the literature has identified an increased sorting of global talent into a relatively small number of regions, we still do not fully understand the distributional impact of this on native-born workers. In this paper I use a task-based approach to examine the impact of high-skilled immigration on local labor markets. The empirical strategy exploits a prominent U.S. immigration policy the Immigration Act of 990 as a plausibly exogenous shock in the supply of high-skilled immigrant workers in science, technology, engineering, and mathematics (STEM) occupations between 990 and 200. First, I find evidence to suggest that an increase in the employment of foreign-born workers in STEM fields displaces native-born workers in abstract occupations, who are reallocated into routine occupations. Second, I find suggestive evidence of native-born workers in abstract occupations experiencing positive wage gains. Taken together, these findings imply that there are countervailing effects of high-skilled immigration: One one hand, there is a supply effect, whereby native wages are lower as the supply of worker increases, and on the other hand, there is a productivity effect, whereby the least productive native workers in these highly skilled occupations are displaced, leading to positive wage gains for the remaining native workers. Third, I find some significant differences in the empirical results, both quantitative and qualitative, when I categorize workers based on broad occupation groups instead of conventionally used skill groups. Lastly, I find suggestive evidence that highskilled immigration contributes to the polarization of native wages, with the effect remaining statistically significant even after controlling for the impact of technological change. Keywords: H-B Visa; High-Skilled Immigration; Occupation; Wage Polarization; Technological Change JEL Classification: J2, J6, J62 Preliminary draft, please do not cite or circulate without author s permission

2 Introduction Despite the growing recognition among economists of the importance of high-skilled immigration in recent years in particular, the importance of foreign-born workers in science, technology, engineering, and mathematics (STEM) the literature is still largely undecided on the overall labor market impact of this selective group of workers. The increase in spatial concentration of global talent and the rise in supply of well-educated workers from developing nations suggest the topic of high-skilled immigration, if trends continue, is likely to play an integral part in public policy discussions. As such, it is important to understand the labor market impact as well as the distributional consequences of high-skilled immigration. Studies that examine the local labor market impact of high-skilled immigration tend to follow the convention of distinguishing native-born workers by skill or educational level (e.g. Peri et al. 205a). Few studies have examined the impact of high-skilled immigration, in particular the impact of foreign-born STEM workers, from the lens of a tasked-based approach to labor, such as the one proposed by Acemoglu and Autor (20). In such a framework, workers are grouped into broad task categories, proxied by broad occupation groups. Current literature appears to have settled on three categories: abstract, routine, and manual occupation groups; each of which approximately compromises one-third of the total employment in The taskbased approach has proved to be quite successful in explaining the changes in the employment and wage structure brought upon by the information and communication technology (ICT) revolution and trade (e.g. Autor et al. 2003; Autor and Dorn 203; Grossman and Rossi-Hansberg 2008). In this paper, I bridge the gap in the literature by adapting the approach to explore the labor market impact and the distributional consequences of the concomitant slowing in demand for abstract tasks and the steady rise in supply of foreign-born STEM workers. A vast literature documents the recent rise of STEM occupations in the United States and the growing influence this is having on the U.S. economy. Aided in part by the IT boom that began in the 980s and the explosion of wages for skilled workers, the United States has been a preferred destination for highly skilled immigrants, particularly those specialized in engineer- Their model is part of a broader class of assignment models (e.g. Costinot and Vogel 200). 2 The first occupation group consists of individuals working in managerial, professional, and technical occupations. These high-paying jobs tend to demand more training and cognitive ability. The second occupation group includes production, clerical, and administrative support and sales occupations. These jobs have become particularly susceptible to automation due to the highly routine content of the tasks. Lastly, the third group includes mechanics, craft and repair occupations, agricultural occupations, and low-skilled service occupations. These low-paying jobs involve high manual content but demand little education. See Autor and Dorn (2009) for more detail and intuition behind the partitioning.

3 ing and computer-related fields. 3 Figure 3 shows that the share of labor supply from immigrant STEM workers have experienced disproportionate growth in abstract occupations relative to native-born STEM workers, who as a group eperienced negative growth since 980. It is worth noting that college-educated native STEM workers experienced positive growth in labor share until the tech bust in 2000, as seen in the dip from 2000 to 200, and has only recently recovered to its previous level in 205, while the share of immigrant STEM workers continued to increase within abstract occupations over this period. 4 These trends are particularly interesting considering that the overall growth in demand for abstract tasks has markedly plateaued since the tech bust in 2000 (see Autor 205 and Beaudry et al. 205). 5 In addition, Langdon et al. (20) notes that college-educated workers with STEM degrees often find work outside STEM fields. 6 For instance, the group of STEM degree holders that is most likely to pursue non-stem occupations is the group of life and physical science majors: A surprising 8% of these college graduates work in non-stem fields. 7 Although the report does not distinguish by country of origin, one can extrapolate from the findings that there is no shortage of nativeborn STEM workers. The rising employment of high-skilled immigrants, stagnant demand for high-skilled workers, and excess supply of college graduates with STEM degrees lead one to wonder whether native-born workers are being crowded out from these high-skilled occupations, and, if so, whether the adjustments take place through employment or through wages. Figure 4 displays the share of foreign-born workers in labor supply by occupation and education attainment. We can readily observe that the disproportionate representation of immigrant workers in STEM rises with advanced degrees: Approximately 45% of total hours worked by STEM workers with a graduate degree in 205 came from foreign-born workers, almost tripling its initial share in 980. This is consistent with Hunt (205) who finds that immigrant workers on average are better educated than native-born workers in computer and engineering-related fields. It is notable, however, that the foreign-born share of the labor supply in STEM fields has been trending upwards since 980, regardless of education attainment. In view of the observation that many STEM degree-holders do not work in STEM fields, the basic employment question remains: Do immigrant workers in STEM fields crowd out native workers? 3 See Kerr and Kerr (206) and Hanson and Slaughter (206) for surveys of recent developments in the area. 4 The graph also makes clear that most immigrant workers in STEM have at least a college education. 5 Note that in Figure 3 the STEM labor supply share is calculcated as a share of abstract occupations and thus is unaffected by this change. 6 College-educated workers are defined as workers with a bachelor s degree or have completed four years of college. 7 Using the 2009 American Community Survey, the authors estimate almost two-thirds of the 9.3 million collegeeducated workers with a STEM degree work in a non-stem field. It is worth mentioning that the definition of STEM occupations used in the report is broader than the one used in the paper. For example, they include data administrator as a STEM occupation, which is not considered here. 2

4 To answer this question, I take a local labor market approach while focusing on the economic outcomes of native-born workers by occupation. The reasons are two-fold. First, a local labor market approach captures the spatial variation in high-skilled immigration, linking the economic outcomes of native workers to spatial sorting of skill. Second, and more important, by focusing on occupation rather than skill or education, I can examine the reallocation of labor across tasks. In adopting the task-based approach to labor, or the Ricardian approach, after Acemoglu and Autor (20), I emphasize on distinguishing task (a choice) from skill (an endowment). Central to the Ricardian framework is the tenet that an individual, regardless of skill, can perform any task in the economy; however, to which tasks she chooses to supply her labor depends on her comparative advantage in performing them. For example, a high school dropout could choose to program computer software, and a computer programmer could choose to perform manual labor. We do not often observe such matches because in equilibrium each would find it in her best interest to offer her skill to the occupation that generates the highest return, i.e. manual labor for the high school dropout and computer programming for the computer programmer. Two existing studies (e.g. Peri and Sparber, 2009, 20) apply the idea to explore the complementarity and the substitutability between native and immigrant workers by examining the relative task content of labor supplied. They find that native workers, who the authors posit have comparative advantage in communication skills, tend to specialize in tasks that demand more interaction while immigrant workers specialize in tasks that are more quantitative and analytical. Although the authors include a discussion of highly skilled workers, they do not specifically examine the impact of immigrant STEM workers on local labor markets. Notwithstanding a couple exceptions mentioned above, much of the existing literature on immigration groups workers by skill level, i.e. education attainment. This convention of sorting workers follows from the canonical model of labor and the neoclassical concept of human capital, proxied by years of schooling, as a primary input in the production function of the economy. Herein, workers of distinct skill levels inelastically supply labor with endowed skills to produce imperfectly substitutable goods. Nevertheless, it has been widely documented that the canonical model fails to capture the non-monotonic changes in the wage and employment structure since 990. This insight leads one to wonder whether the Ricardian lens of labor is also applicable to the analysis of high-skilled immigration, particularly when high-skilled immigrants tend to be employed in certain occupations, such as abstract-intensive fields. Panels in Figure 6 display the trends for the share of total hours worked by college-educated workers within abstract, routine, and manual occupation groups. The difference between the solid line and the dashed line is the share of labor supply by college-educated immigrant work- 3

5 ers. Panel A and B demonstrate that both abstract and routine occupations have experienced similar gains in the share of college labor from 980 to 205, indicated by the slopes of the marked lines, while Panel C shows that skill-upgrading in manual occupations was minimal during this period. It can gleaned from Panel A that a significant share of the increase in the labor share of college-educated workers in abstract occupations came from foreign-born workers, shown by the visible widening of the shaded area since In Figure 7, I overlay panels A and B from Figure 6 and restrict to the time period after 990. The slope that corresponds to all abstract workers is roughly parallel to the slope of all routine workers, indicating similar increases in skill-upgrading. But perhaps what is more surprising is that even though the share of college labor in routine occupations has experienced a similar increase, unlike the steady rise in abstract occupations since mid-990s, most of the increase in routine occupations came only after the tech bust in Moreover, the gap between the dashed line and the dotted line only widened slightly, suggesting that most of the gains came from the increase in labor supply from college native workers. Taken together, these figures suggest that there has been a reallocation of college native labor from abstract to routine occupations a phenomenon that may be masked when viewed through the lens of the canonical model, which distinguishes workers by education attainment. Moreover, because the partition of native-born workers in immigration studies is also often left up to the researcher s discretion, it is important to explore whether the empirical analyses using the task-based approach and the canonical approach reach the same conclusions. In an effort to investigate the empirical distinctions between the two models, I present the labor market effects estimated using both approaches. In addition to the definition of worker groups, the second key ingredient of the spatial approach is the definition of a local labor market. I employ a definition that is under-utilized in the immigration literature: Rather than focus on cities or states, I use commuting zones (CZs) based on the 990 definitions in Tolbert and Sizer (996). Commuting zones are geographic regions that cover the entire U.S. labor market, both rural and urban, and arguably they better approximate local labor market conditions as they are delineated according to employment and commuting ties, as opposed to population size (e.g. metropolitan statistical areas) or political governance (e.g. states). My analysis presents evidence of statistically significant displacement effects of high-skilled immigration on American workers in abstract occupations, showing displaced native workers reallocate from abstract occupations to routine occupations. The negative employment effects are robust to different model specifications as well as to controlling for a rich vector of local labor market conditions. In terms of wage adjustments, I find the routine-manual native wage gap 4

6 has decreased. The point estimates on the abstract-routine and manual-abstract native wage gaps are less precise, but they all have positive signs. Examining the changes in wage levels, I find positive wage effects for both abstract and manual natives and negative wage effects for routine natives, though the estimates are mostly statistically insignificant. The noisy point estimates on the wage effects, particularly for abstract natives, could be attributed countervailing effects of high-skilled immigration. While the increase in high-skilled immigration has a supply effect which lowers wages of natives in competing occupations, it also has a productivity effect as high-skilled immigrants displace the least skilled natives in these occupations and only the most productive ones remain. The findings show that the Ricardian approach can be useful: Displacement effects and wage polarization effects attenuate in magnitude and sometimes lose statistical significance when workers are categorized by education attainment instead of occupation. Finally, I isolate the effects of high-skilled immigration from other contemporaneous changes in the labor market, specifically the increased automation of routine tasks. The U.S. labor market underwent substantial structural shifts between 990 and 200 that coincided with the rise in high-skilled immigration. Although the results presented so far are robust to different model specifications, one might still be concerned about the local demand shocks generated by the structural shifts, such as the tech boom, whose omission could lead to biased results. To address this concern, I extend the analysis to disentangle the economic impact of these market forces. I show that even after controlling for technological progress, the impact of high-skilled immigration remains significant, both in the statistical and in the economic sense. The plan of the paper is organized as follows. Section 2 presents a simple framework for the local labor market impact of high-skilled immigration. Section 3 describes the collection of data and the construction of the worker and wage samples. Section 4 discusses the empirical strategy and the potential threats to identification. Section 5 presents the main results, and Section 6 extends the discussion to include technological change. Section 7 concludes. 2 Conceptual Framework In this section, I present a simple extension to the task-based model in Acemoglu and Autor (20). The choice of this particular model rests on its transparency in the assignment of skill to task and its ease in generating comparative statics results for different worker groups. Both features prove to be particularly useful in aiding the analysis of the employment and wage adjustment mechanisms caused by the increase in the supply of immigrants. I now describe the model in detail. 5

7 Consider an economy which produces a final good Y by combining a continuum of intermediate goods i, or tasks, indexed by the unit interval r0, s. The tasks are ordered such that they are increasing in complexity. The production of the final good follows a Cobb-Douglas functional form Y = exp ln y(i)di 0 where y(i) is the production level of task i. The price of the final good P is chosen to be the numeraire. Markets are competitive and labor is the only input in the production of tasks. The labor market is made up of native- and foreign-born workers. Furthermore, native-born workers are distinguished by three skill types, medium (M), lower-tier high (H l ), and upper-tier high (H h ) types. There is only one type of foreign-born workers (F). In equilibrium, as I will show, each skill type is uniquely assigned to perform a set of non-overlapping tasks. These tasks, or occupations, allow me to identify each worker type uniquely by the primary task supplied. In later sections, together lower-tier and upper-tier high-skilled native-born workers represent natives employed in abstract occupations while middle-skilled native represent natives in routine occupations. To fix ideas, one can think of a upper-tier high-skilled native worker as a senior-level computer programmer; a lower-tier high-skilled native as a junior-level computer programmer; and a middle-skilled native as a data entry worker. Readers familiar with the literature may notice the absence of native-born manual workers in the current setup: The convention in the literature considers three skill types: low, medium, and high, which correspond to manual, routine, and abstract workers, respectively. 8 As I will show empirically in Section 5, the impact of high-skilled immigration is largely concentrated among native-born workers in routine and abstract occupations. Thus, for clarity of exposition I have chosen to leave out manual native-born workers in the theoretical analysis. Their inclusion in the model does not change the theoretical predictions of the impact of high-skilled immigration on routine and abstract native workers. Central to task-based models is the assumption, which I adopt, that workers are perfectly substitutable across tasks but differ in their productivity in each task. Each task is produced according to a linear production function: () y(i) = A M α M m(i) + A Hl α Hl (i)h l (i) + A Hh α H h h (i) + A F α F f (i) (2) where A i is a factor-augmenting technology and α i is the task productivity schedule for the 8 See Acemoglu and Autor (20). 6

8 corresponding type of worker across different tasks. Each of the terms m(i), h l (i), h h (i), and f (i) corresponds to the number of middle-skilled native workers, the number of lower-tier high-skilled native workers, the number of upper-tier high-skilled native workers, and the number of immigrant workers used in the production of task i. From the lens of a task-based model, immigration is represented by the trading of a set of tasks" to foreign-born workers. Certainly, the allocation of tasks would depend on the comparative of both native- and foreignborn workers across tasks. Even though there is no definitive evidence to support the claim that high-skilled immigrants are the best and brightest, Hanson and Slaughter (206) and Hunt (205) provide some evidence to suggest that immigrants have a revealed comparative advantage in STEM occupations, which fall under abstract occupations. Even so, it may be the case that senior positions in those occupations may be more available to native-born workers because those positions demand particular social skills and cultural knowledge, at which native-born workers typically have a comparative advantage. These observations imply that foreign-born workers are more likely to have comparative advantage in abstract tasks that lie somewhere in the intermediate range of complexity. It also seems innocuous enough to assume that more skilled workers are relatively more productive at more complex tasks while less skilled workers are relatively more productive at less complex ones. These observations lead to the following structure of comparative advantage among workers: Assumption. α M (i)/α Hl (i), α Hl (i)/α F (i), and α F (i)/α Hh (i) are continuously differentiable and strictly decreasing. The assumption ensures that middle-skilled native workers have a comparative advantage in the production of the least complex tasks, lower-tier high-skilled natives and immigrants in the intermediate tasks, and upper-tier high-skilled natives in the most complex tasks. To complete the model, it remains to specify market-clearing conditions for the labor market. Again, to keep matters simple, I assume that the supply of labor is exogenous, which means market-clearing conditions are characterized by: 9 M = 0 m(i)di, H l = 0 h l (i)di, H h = 0 h h (i)di, F = 0 f (i)di. (3) Having fully characterized the economy, I now derive the equilibrium of the model. First, let J = {M, H l, H h, F} denote the set of worker groups. Markets are competitive so that every worker who performs the same task is paid the same wage: It must be the case that w j = 9 Note that I have slightly abused notation, using both capital letters to indicate worker type as well as supply of worker. 7

9 p(i)a j (i)α j (i) for j J and i E j, where E j is the set of tasks performed by worker type j. Furthermore, the price differential between any two tasks for a worker must be completely offset by her productivity differential. In other words, for any worker j and for any two tasks i, i E j, there must be exactly a unique price index P j such that P j p(i)α j (i) = p(i )α j (i ). Because the production function is Cobb-Douglas, the expenditure share across worker types must be equalized and hence it must be the case that p(i)y(i) = p(i )y(i ) for any two tasks i, i E j. 0 Recall the number of type j worker performing task i is j(i). Then for two tasks i, i E j, p(i)a j α j (i) j(i) = p(i )A j α j (i ) j(i ) (4) p(i)α j (i) j(i) = p(i )α j (i ) j(i ) (5) j(i) = j(i ). (6) That is, for each worker type j J it must be the case that the number of workers performing each task in E j is equalized, i.e. j(i) = j/ E j if i E j where j denotes the supply of worker of type j and E j denotes the size of the interval of tasks performed by a worker of type j. Furthermore, for two tasks performed by different workers, i.e. for i E j and i E j and j, j J and j = j, equalization of expenditure share across worker types implies p(i)a j α j (i) j(i) = p(i )A j α j (i ) j (i ) (7) j p(i)a j α j (i) E j = p(i )A j α j (i ) j E j P j A j j A j j = P j E j E j where the last equation follows from P j = p(i)α j (i) if i E j. In words, the relative price of tasks depends on the relative effective supply of workers performing the tasks and the relative supply of tasks. Next, an implication of Assumption is that there exist marginal tasks such that in equilibrium the firm is indifferent between using different skills in the production of these tasks. In other words, at these task thresholds, the cost of producing the tasks are equalized across skills, i.e. there is no arbitrage across skill types (see Figure ). Formally, let I L, I M, and I H 0 This follows from the first-order conditions of max y(i) PY 0 p(i)y(i)di. (8) (9) 8

10 denote the thresholds as ones such that the following equations are satisfied: A F α F (I H )F I H I M A Hl α Hl (I M )H l = A H h α H (I H )H h I H (0) = A Fα F (I M )F () I M I L I H I M A M α M (I L )M = A H l α Hl (I L )H l (2) I L I M I L In addition to the no arbitrage across skill condition, the unit cost of workers must also be equal to the value of marginal product and any differences in the productivity of worker types must be fully compensated by differential wages; that is, in equilibrium we must observe for each j, j {M, H l, H h, F}: w j = P j A j (3) w j w j = P ja j P j A j j j = E j E j. (4) Lastly, the solution to the maximization problem of the firm producing the final good and the choice of numeraire together imply exp 0 ln p(i)di = or equivalently 0 ln p(i)di = 0. Substituting in the definition of the price index for each task set P j = p(i)α j (i) for each i E j leads to: (ln P j lnα j (i))di = 0. (5) j J i E j To summarize, the unique equilibrium of this economy is given by a vector of task thresholds, task prices, and wages (I L, I M, I H, P M, P Hl, P F, P Hh, w L, w Hl, w F, w Hh ) that are characterized by Equations (9), (0), (), (2), (4), and (5). 2 Having characterized the unique equilibrium, we are interested in the mechanisms by which employment and wages of native-born workers adjust when there is an increase in the supply of high-skilled immigrants. It can readily be shown that the model yields the following predictions See Lemma in Acemoglu and Autor (20). 2 Uniqueness of the equilibrium follows from the fact that I L, I M, and I H are unique, a construct of the assumption on the structure of comparative advantage. The uniqueness of the equilibrium thresholds then lead to the uniqueness of task prices P j and wages w j for each j {M, H l, H h, F}. 9

11 in terms of native employment adjustment: 3 di H d ln F > 0, d(i M I L ) d ln F < 0, di L d ln F < 0. (6) In words, as a result of immigration, there is a contraction of native employment in tasks that were previously performed by lower-tier and upper-tier high-skilled native workers, i.e. an overall contraction of tasks that were previously performed by abstract native workers. There is also an expansion of tasks that were previously performed by middle-skilled native workers, i.e. routine native workers. The reasoning is as follows: First, the employment adjustment occurs by reallocating some of the tasks that were previously performed by lower-tier and upper-tier high-skilled native workers to immigrants. As a consequence of the displacement of lower-tier high-skilled native workers, a second adjustment takes place as some of the these workers move down the occupation ladder and move into tasks that were previously performed by middle-skilled native workers, leading to an expansion in the employment of native workers in routine occupations (see Figure 2). Similar reasoning implies that if manual native workers were to be incorporated into the model, the model would predict an occupational downgrading of some native workers in routine occupations to manual occupations, leading to an expansion of native employment in manual occupations. In addition to the employment results, the model also generates predictions on native wage adjustments. As a first step, it will be useful to consider the changes in the wage gap between two worker groups under broader conditions: () when the set of tasks performed by the two groups forms a convex set, i.e. when worker groups perform tasks that are adjacent to each other on the task distribution, and (2) when the task groups are not adjacent. As seen in Equation (4), the relative wage only depends on the relative supply of worker and the relative allocation of task. Moreover, it is demonstrated in the Appendix that the wage gap can be rewritten in terms of the ratios of factor-augmenting technologies and the task productivity schedules. As a simplest case, I first consider the wage gap between lower-tier high-skilled natives and middle-skilled natives, who perform tasks adjacent to each other. Taking logs and substituting in Equation (9), I can rewrite the wage gap as ln(w Hl /w M ) = ln A Hl ln A M β L (I L ) where β L (i) lnα M (i) lnα Hl (i). By assumption β L (i) < 0 for all i r0, s, it immediately follows d ln(w Hl /w M )/d ln F = β L (I L)dI L /d ln F < 0, that is, the wage gap between H l and M decreases as immigration increases. This example highlights that when native groups perform tasks adjacent to each other on the task distribution in equilibrium, the sign of the change in the wage gap between the two groups only depends on the sign of the change in 3 All derivations are relegated to the Appendix. 0

12 task threshold. Next, I consider the wage gap between native groups where tasks performed by the groups in equilibrium are not adjacent to each other. As a concrete example, let us consider ln(w Hh /w Hl ). Total differentiation with respect to ln F yields d ln(w Hh /w Hl ) d ln F ½ if β H (I H)( I H ) ½ (I M I L )β M (I M) k for some fixed k > (7) where k is the constant defined in the Appendix. In this case, whether wage gap increases or decreases depends on the slope of log comparative advantage of each worker group at the marginal task, i.e. the strength of comparative advantage at the marginal task. An interpretation of this result is that when two native workers are not close substitutes when the comparative advantage of workers leads them to perform tasks that are far away from each other on the unit interval then the distributional impact of immigration depends on the strength of each native worker s comparative advantage, which determines how skills are reallocated across tasks, near the equilibrium task threshold. If the two native workers are close substitutes, i.e. workers are performing adjacent tasks, then the native worker less similar in skill to the immigrant worker experiences a relative wage increase. A direct implication of this observation is that because manual native workers, though they are not modeled here, are less similar to high-skilled immigrant workers than routine native workers, and thus the routinemanual native wage gap unambiguously decreases as a result of an increase in high-skilled immigration. Because we do not observe H l and H h in practice, we need to consider the wage gap between high-skilled and middle-skilled native workers, i.e. the abstract-routine native wage gap. Let us denote the average wage of an abstract native worker as w H = λ H w Hh + λ M w Hl where λ i is the share of abstract natives with skill level i. To derive the change in the wage gap between abstract and routine native workers w H /w M, note that taking logs and totally differentiating with respect to ln F yield d ln(w H /w M ) d ln F = d pln M ln H + ln( I H + I M I L ) ln I L q d ln F (8) d( I = H + I M I L ) di L I H + I M I L d ln F I L d ln F. (9) Because there are two types of workers wedged in between upper-tier high-skilled natives and middle-skilled natives on the task distribution, namely the lower-tier high-skilled natives and the immigrants, the sign on d ln(w H /w M )/d ln F is ambiguous without additional assump-

13 tions and must be assessed empirically. Nevertheless, it is shown in the Appendix that when β L (I L) is sufficiently larger than the ratio β H (I H)/β H (I H) + β M (I M), the inflow of highskilled immigrant workers will drive up the abstract-routine native wage gap. In summary, we have the following proposition, which I empirically verify in Section 5: Proposition. Consider an equilibrium (I L, I M, I H, P M, P Hl, P F, P Hh, w L, w Hl, w F, w Hh ). In terms of employment adjustments, an exogenous increase in the supply of foreign-born workers leads to a contraction of tasks that were previously performed by native-born abstract workers and expansions of tasks that were previously performed by native-born routine workers and by native-born manual workers. In terms of wage adjustments, an exogenous increase in the supply of foreign-born workers decreases the native-born routine-manual wage gap but has ambiguous effects on the relative wages of other native-born worker groups. If β L (I L) is sufficiently larger than the ratio β H (I H)/β H (I H) + β M (I M), then the abstract-routine native wage gap increases. 3 Data, Definitions, and Measurements I now summarize the construction of the sample, with details relegated to the Appendix. For my purposes, it is not only important to have a time-consistent panel of local labor markets to adequately capture the movement of labor; it is also important that there are sufficiently many markets to provide enough spatial variation for the construction of valid instruments. This makes commuting zones particularly well-suited for to my study. Unlike metropolitan statistical areas, whose definition often change between decades, the definition for commuting zones, which is based on counties, has remained relatively stable. For these reasons, I construct a time-consistent panel of 722 commuting zones that covers the entire U.S. mainland (excluding only Alaska and Hawaii) and span the three decades from 990 to Data on occupations, employment, wages, and worker demographics are collected from the Ruggles et al. (205) Integrated Public Use Microdata (IPUMS). To construct the worker and wage samples, I collect data from Decennial Census sample files from 990, 2000 and, a merged American Community Survey (ACS) sample file from The worker sample, which makes up the sample of working-age population throughout the text, is constructed from non-institutionalized individuals between the ages of 8 and 64 (inclusive) and have worked positive number of weeks in the previous year. Unpaid family 4 The construction of commuting zones follows the instructions provided on David Dorn s website. 5 The three-year ACS sample file is referred to as the 200 ACS sample for convenience. 2

14 workers and individuals in the military profession are dropped. A native-born worker, or simply a native, is defined to be an individual born inside U.S. territories or born abroad to an American parent. A foreign-born worker, or an immigrant, is defined to be an individual born outside U.S. territories with an identifiable birthplace. Individuals born at sea are dropped. All wage variables are constructed by including individuals from the working-age population who have earned positive wage and salary income in the previous year. Self-employed workers are dropped. Nominal wages are adjusted to 999 dollars using the consumer price index. Weekly wages are calculated as yearly wages and salary income divided by the number of weeks worked in the previous year. 3. Definition of STEM Occupations A comprehensive list of STEM occupations and STEM college majors is provided by the U.S. Department of Commerce (USDOC). 6 Occupations range from highly skilled jobs that require many years of training (e.g. aerospace engineers) to occupations that do not require a bachelor s degree (e.g. database administrators). Since the literature generally restricts STEM occupations to jobs that demand a high level of technical and mathematical skills, for the analysis I focus on a subset of STEM jobs from the list: computer scientists and computer-related occupations, engineers, life and physical scientists, and other science-related fields (e.g. mapping scientists). A complete list of STEM occupations can be found in the Appendix. 4 Identification First and foremost, teasing out the effects of high-skilled immigration along the task distribution requires the appropriate categorization of natives. Following the suggestions of Autor and Dorn (203) and Autor et al. (205), I partition workers into three broad occupation groups consisting of abstract, routine, and manual workers. Evidently, most occupations demand a combination of all three task inputs; therefore, to estimate the changes in task prices, it would be necessary to construct a measure of task intensity for each of the three inputs for each occupation. Nevertheless, given that my interest lies in estimating the impact of high-skilled immigration on the average native worker of each of the three types, I abstract away from the details and proxy each native type by one of the three broad occupation groups and the primary task supplied within of each group. Naturally, with this definition, all of the STEM occupations fall under the umbrella of abstract tasks. 6 See USDOC website. 3

15 The structural equation of interest is the following: y Native, G c,t = η y,g Foreign, STEM Ec,t + X c,tγ + ϕ t + ϕ r + ε s,t (20) which is a first difference specification and where y c,t is either the decadal change in employment share or decadal log change in mean weekly wages for an occupation group G {Abstract, Routine, Manual} or a skill group G {College, Noncollege} in commuting zone c and year t. 7 The term X c,t corresponds to a vector of CZ covariates. I include start-of-period CZ covariates that the literature has shown to be important determinants in shifting labor demand: the share of manufacturing employment, female labor force participation rate, and, perhaps most important of all, the share of college-educated native-born workers. Because of possible human capital externalities (see Moretti 2004) and endogenous amenities driven by the clustering of skilled workers (see Diamond 206), it is essential to isolate local demand shifts due to the agglomeration forces driven by native-born workers from the forces that are driven by foreign-born workers. Likewise, it is also important to isolate the impact of high-skilled immigration from low-skilled immigration; thus, I include start-of-period share of noncollegeeducated immigrant workers. Lastly, to account for any remaining changes due to differential changes in regional productivity or labor demand shifts due to local industrial mix, I include Bartik (99) employment and wage instruments wherever possible; the only regressions where Bartik instruments are excluded are the regressions concerning the log wage ratios. The construction of these control variables can be found in the Appendix. The terms ϕ t and ϕ c are year fixed-effects and region fixed-effects, which are included to capture national trends and state-specific trends that are approximately linear in time. Because immigrant workers are likely to move to areas with employment opportunities, the shift in supply of foreign-born workers is likely to be endogenous and therefore an instrument is required. The research design follows the instrumental variable approach developed in Peri et al. (205a), where identification relies on a variant of the shift-share instrument. The instrument predicts the growth of immigrant STEM workers in each commuting zone based on two sources of variation: One comes from the plausibly exogenous variation in the supply of immigrant STEM workers, driven by the fluctuations in the number of annual H-B visas from 990 to 200; 8 the second is the initial distribution of immigrant STEM workers in each commuting zone in 990. The predicted inflow and outflow of immigrant STEM workers are calculated 7 College-educated workers are workers with a bachelor s degree or have completed four years of college. 8 Following their approach, I also include TN visas which grant temporary work authorizations for Canadian and Mexican workers under NAFTA. The annual number of TN visas are small relative to H-B, less than about % until 2004 and less than about 5% to

16 based on two different methods: () the total number of H-B visas issued by nationality or region, and (2) the total number of annual H-B visa issuance. 9 The first construction is based on the immigrant-enclave instrument pioneered by Altonji and Card (99) and Card (200) whereby immigrant flows are predicted based on the historical pattern of immigrant settlements. It exploits the tendency of immigrants to cluster by ethnicity. The second construction follows the intuition of Kerr and Lincoln (200) in that the instrument predicts immigrant flow based on local labor market s historical dependency on immigrant STEM workforce. 4. Immigration Act of 990 and the H-B Program In this section, I briefly describe the main features of the H-B visa program, which is used in the construction of the instruments. A controversial change in the Immigration Act of 990 introduced an annual cap on the number of H-B visas, essentially limiting the supply of foreignborn workers. Figure 5 plots the annual H-B visa cap and the total number of annual H-B visas issued, which is used to predict the supply-driven changes in the inflow of highly skilled foreign-born workers in the construction of the instrument variables. As seen in the figure, the cap has remained quite steady at 65,000 per year since its inception. The few years of exception are the years between 999 and 2003, when the cap was raised as high as 95,000 visas to meet the demand of high-tech firms during the peak of the tech boom. After the tech bust in 2000, there was an excess supply of H-B visas and the cap was returned to 65,000, and it has remained there to date. Beginning in 2004, an additional 20,000 visas have been granted annually to foreign-born workers with at least a master s degree. There is no cap on the number of visas sponsored by American nonprofit organizations and institutions such as universities. At the time of applying for an H-B, firms are required by law to prove that () the hiring of a foreign-born worker does not displace American workers and (2) foreign-born workers are paid the prevailing wage for the job. Classified by the United States Citizenship and Immigration Services (USCIS) as a non-immigrant visa, the H-B visa has become a main portal for highly skilled foreign-born workers to enter the United States. Initially, H-B visas were designed for employers who seek temporary foreign-born workers in specialty occupations for a period of three years, and, with renewal, up to six years. Specialty occupations are broadly defined, but according to Kerr and Lincoln (200) approximately 60% of successful applicants go into science, engineering, and computer-related occupations. Therefore, we should expect the instruments to be highly correlated with the growth of foreign-born workers in STEM. 9 The 0 regions or nationalities are Africa, Canada and Oceania, China, Eastern Europe, India, Mexico, Other, Rest of Americas, Rest of Asia, and Western Europe. 5

17 4.2 Validity of Instruments The causal interpretation of η requires the changes in the supply of foreign-born STEM workers to be uncorrelated with unobserved CZ-level labor demand shocks, which will hold if the instruments are valid. The validity of the instruments is based on two factors: () the visas disproportionately go to foreign-born workers in STEM occupations, and (2) the annual cap reflects the demand of these workers at the national level and not driven by changes in local demand. Although the latter assumption cannot be tested, it seems plausible given that the visa cap is set nationally and the distribution of immigrant STEM workers is fixed at 990 levels. 20 Table presents the 2SLS first-stage results for the excluded instrument across the various specifications later used in the paper. The F-tests are all relatively strong, even after the inclusion of a full vector of CZ covariates. In the Appendix, I additionally provide first-stage results by time period, showing strong correlation in both decades, albeit less so in the last decade. 5 Results The literature is still undecided on the overall labor market impact of high-skilled immigration. For instance, a recent study by Doran et al. (204) adopts a random assignment research design that matches administrative tax data to administrative data on H-B visa lottery participants and finds evidence to suggest H-B workers substantially crowd-out natives in firms that win the lotteries. On the contrary, two recent studies by Peri et al. (205a,b) suggest H-B immigrants generate large productivity spillovers to similarly-skilled native workers. Through a calibration exercise, Bound et al. (205) provide an even more nuanced view of the impact of high-skilled immigration, suggesting that, though immigrant computer scientists may have crowded out American computer scientists, total computer science employment would have been lower if the employment of immigrant computer scientists could not be increased from their 994 levels. In the this section, I empirically assess these opposing claims through the examination of the medium- to long-run labor market effects of high-skilled immigration. First, I examine the labor reallocation of native-born workers across tasks, and second I assess the wage adjustments by task groups. In my empirical analysis, I verify the predictions of the theoretical model from Section 2 in that I find a reallocation of native-born workers from abstract to routine occupations and a decrease in the native-born routine-manual wage gap. In addition, I find 20 A previous version of the paper constructed the instruments using 980 as the base year. The results are comparable using both approaches, but 990 is preferred because there were considerably fewer foreign-born STEM workers in 980 which led to many zeros in the predicted immigrant STEM employment. 6

18 suggestive evidence of positive wage gains for abstract natives and an increase in the native abstract-routine wage gap. To check the robustness of model specification, I vary the model in several ways. Row provides point estimates for different native groups using the baseline specification that includes a full vector of controls; Row 2 restricts the regression analysis using the baseline specification to CZs that intersect a MSA; Row 3 restricts the baseline specification to CZs with population over 00,000 in 990 (the minimum CZ population that excludes CZs without immigrant STEM observation); Row 4 uses the aggregate H-B instrument in place of the immigrant-enclave instrument that is used in rows through Employment Effects Table 3 provides empirical support for the predictions of the model. The preferred specification in column of Table 3 demonstrates that a one percentage point increase in the share of immigrant STEM workers decreases native employment in abstract occupations by 2.9 percentage points, indicating a statistically significant displacement effect. To put this in perspective, the interquartile range of CZ exposure to high-skilled immigration, i.e. a movement from a CZ in the 25th percentile to a CZ in 75th percentile in terms of exposure, over the two decades from 990 to 200 is around 0.4 percentage points. 2 Thus, the crowding-out effect of native employment in abstract occupations is small ( = 0.4) but non-negligible. The table also verifies the prediction of the model that displaced native workers reallocate into routine occupations, showing a statistically significant point estimate of 2.84 in the preferred specification. The point estimate on native manual worker is small and statistically insignificant. Changes in employment shares may still mask the more subtle changes if employment adjustments take place on the intensive margin, i.e. through a reduction in weeks or hours of work. To examine this channel of adjustment, I now empirically assess the reallocation of labor supply defined as the total number of hours worked in the previous year among native workers across different tasks. Table 4 demonstrates that the results are qualitatively similar to the point estimates from employment shares. However, it is worth mentioning that the point estimate on native abstract workers is now slightly more negative ( 4.29 in the preferred specification compared to 2.93 from before) which suggests that in addition to the reallocation of native workers from abstract to routine occupations, there is also a reduction in the share of total hours worked in abstract occupations by native workers overall. The point estimates on routine natives are statistically significant quantitatively similar to previous estimates while the point estimates on manual natives remain small and statistically insignificant. 2 Exposure to high-skilled immigration is defined as the change in employment share of immigrant STEM workers. 7

19 5.2 Wage Effects The model in Section 2 also provides testable implications on native wage adjustments, which can be directly verified by examining the changes in relative wages among native worker groups. The model predicts that an increase in the supply of foreign-born workers leads to a drop in the wage gap between lower-tier abstract and routine native workers; the same mechanism leads to a decrease in the wage gap between routine and manual native workers; and finally, the effect on the overall abstract and routine native-wage gap is ambiguous and depends on the relative strengths of the comparative advantage of each worker groups. Because I do not observe lower-tier abstract natives in the sample, I calculate the log wage ratio between the average abstract native and routine native. As see in the preferred specification in Table 5, a one percentage point increase in the share of immigrant STEM workers decreases the wage gap between routine and manual native workers by.5 log points. On the other hand, the point estimates on the wage gaps between abstract and routine natives and between manual and abstract natives are both positive, though only statistically significant in one specification. Thus, the empirical findings provide partial support for the theory and suggests that high-skilled immigration may have contributed to wage polarization. Nevertheless, at this point I cannot rule out the potential confounding shock of routine-biased technological change: If capital-skill complementaries induced high-skilled immigrants to locate in regions where computer prices fell, then the point estimates will be biased. As a robustness check, in section 6 I show that the wage polarization results remain statistically significant and similar in magnitude, even after controlling for polarization effect of technology shock. In addition to the distributional concerns, a more general concern raised against the loosening of immigration policies, such as the H-B program, is that they depress native wage levels, particularly in immigrant-competing occupations. To examine the validity of this concern, I now turn to the impact of high-skilled immigration on native wages across worker groups. Column of Table 6 shows that native workers who are most likely to be directly affected by high-skilled immigration appear to have experienced positive wage gains, though the point estimates are noisy and mostly statistically insignificant. Given the displacement effects observed on abstract native workers, we should interpret the absence of a significant wage result as being driven by two competing effects. One is a productivity effect driven by the selection mechanism discussed in the model, whereby the least productive native workers in abstract occupations (low-tier high-skilled native workers) are forced down the occupational ladder and only the most productive highly skilled natives remain, which leads to positive wage gains. The second effect is a supply effect, lowering native wages in abstract occupations as a result 8