The primary impact of immigrant inflows to a country is an expansion in the size of its economy, including the labor force. Per capita effects are less predictable: An injection of additional workers into the labor market could negatively impact some people in the pre-existing workforce, native- and foreign-born, while positively impacting others. The wages and employment prospects of many will be unaffected. The direction, magnitude, and distribution of wage and employment effects are determined by the size and speed of the inflow, the comparative skills of foreign-born versus native-born workers and of new arrivals versus earlier immigrant cohorts, and the way other factors of production such as capital adjust to changes in labor supply. Growth in consumer demand (immigrants also buy goods and services), the industry mix and health of the economy, and the nation’s labor laws and enforcement policies also come into play.
The primary determinant of how immigration affects wages and employment is the extent to which newly arriving workers substitute for or complement existing workers. As laid out theoretically in Chapter 4, wages may fall in the short run for workers viewed by employers as easily substitutable by immigrants, while wages may rise for individuals whose skills are complemented by new workers. For example, suppose foreign-born construction workers enter the labor market, causing a decrease in construction workers’ wages. Firms will respond by hiring more construction workers. Since additional first-line supervisors may be needed to oversee and coordinate the activities of the expanded workforce, the demand
and hence the wages of these complementary workers could receive a boost. On the other hand, where immigrants compete for the same jobs, whether as construction workers or academic mathematicians (Borjas and Doran, 2012), employment opportunities or wages of natives are likely to suffer.1 Further, where the availability of low-skilled immigrants at lower wages allows businesses to expand, total employment will rise. Wage and employment effects are predicted to be most pronounced in skill groups and sectors where new immigrants are most concentrated.
Given the potential for multiple, differentiated, and sometimes simultaneous effects, economic theory alone is not capable of producing decisive answers about the net impacts of immigration on labor markets over specific periods or episodes. The role and limitations of theory were assessed by Dustmann et al. (2005, p. F324):
Economic theory is well suited to help understand the possible consequences of immigration for receiving economies, and the theoretical aspects of the possible effects of immigration for the receiving economies’ labour markets are well understood. That is not to say that predictions of theory are clear-cut, however. It is compatible with economic models that changes in the size or composition of the labour force resulting from immigration could harm the labour market prospects of some native workers; however, it is likewise compatible with theory that immigration even when changing the skill composition of the workforce has no effects on wages and employment of native workers, at least in the long run. Economic models predict that labour market effects of immigration depend most importantly on the structure of the receiving economy, as well as the skill mix of the immigrants, relative to the resident population.
Empirical investigation is therefore needed to estimate the magnitude of responses to immigration by employers, by native-born and earlier-immigrant workers and households, by investors, by the public sector, and in housing and consumer-goods markets (Longhi et al., 2008, p. 1). Dynamic conceptual approaches are needed to assess some of the impacts of immigration, particularly those that require long periods of time to unfold.
In the context of the U.S. experience, immigrants have historically been most heavily represented in low-skilled occupations. This has prompted an extensive body of empirical work investigating whether immigration has had a negative effect on the wages and employment of low-skilled
1 Detailed discussion of when immigrant labor complements and when it substitutes for native employment can be found in Foged and Peri (2014) who analyzed relative employment effects using longitudinal employer-employee data for Denmark covering the period 1991-2008. Mouw et al. (2012) and Rho (2014) also examined this question using evidence from the Census Bureau’s Longitudinal Employer Household Data on worker displacement in high-immigration industries.
natives and earlier immigrants. However, a substantial and growing share of immigrants is highly skilled. In part because of this change—and also because of the possibility of positive spillovers from the highly skilled to other workers and to the economy more generally—this group is receiving increased attention. The panel’s summary of the literature in this chapter reviews both these strands of research: After reviewing the pivotal influence of substitutability among different labor inputs in Section 5.2, the focus of Sections 5.3 and 5.4 is predominantly on empirical analyses of low-skilled markets. Section 5.5 reports on a cross-study comparison of the magnitude of immigrants’ impacts on wages. Section 5.6 examines some of the research findings about the highly skilled, including the impact of immigration on innovation.
Given the complexity of mechanisms through which immigration shapes the economy, it is not surprising that the empirical literature has produced a range of wage and employment impact estimates. The basic challenge to overcome in empirical work is that, while wages before and after immigration can be observed, the counterfactual—what the wage change would have been if immigration had not occurred—cannot. A range of techniques has been used in the construction of this counterfactual, and all require assumptions to facilitate causal inference (i.e., identifying assumptions). The different approaches can be judged in part by the plausibility of these assumptions.
The panel has organized this review of empirical studies primarily in terms of methodological approach, using three labels common in this literature. We first describe and present results from spatial studies, which compare worker outcomes across geographic areas. Next, we review results from analyses that use aggregate (nationwide) data, including skill cell studies, which compare worker outcomes across groups defined to have similar education and experience, and structural studies, which implement the skill cell approach with a closer connection between theory and empirical estimation. Much of the discussion in these sections is concentrated on studies of the overall labor market and the low-skilled labor market. Later in the chapter, we turn our attention to evidence about high-skilled labor markets, including the effect of skilled immigration on innovation and entrepreneurship.
Spatial studies define subnational labor markets—frequently, these are metropolitan areas—and then compare changes in wage or employment levels for those with high and those with low levels of immigrant penetration, controlling for a range of additional factors that make some destination locations more attractive than others. As immigrants are likely to settle in those metropolitan areas that have experienced positive economic shocks, econometric methods are used to identify spatial variation in immigrant penetration that can be considered “exogenous”—that is, not determined
within the system being studied—with respect to the outcome that is modeled, which is typically the wages or employment of native-born workers. To illustrate, suppose an analyst is interested in identifying the impact of immigration on wages of the native-born in local labor markets. If immigrants settle predominantly in areas that experience the highest wage growth, then this will induce spurious correlation contaminating estimates of the causal effect of immigration; wage growth (or dampened wage decline) will be erroneously attributed to the increase in labor supply. An econometric solution to this problem presents itself if immigrants choose areas not just on the basis of economic conditions but also on the basis of non-economic factors, such as proximity to others with similar backgrounds. These noneconomic factors can help the analyst create variation in immigrant penetration that is independent of wage growth and that is not correlated with unobserved factors that determine wage growth. A subset of these studies has obtained identification by taking advantage of “natural experiments” created by unusual immigration events, such as the Mariel boatlift injection of more than 100,000 Cuban workers into the Miami labor market in 1980 (Borjas, 2016b; Card, 1990; Peri and Yasenov, 2015).
Another potential problem with the spatial approach, noted by Borjas (2014a), is that natives may react to an influx of immigrants by leaving affected areas, thus dissipating the labor market impacts of migration across the national economy. However, whether such responses by natives are indeed an empirical problem is controversial in the literature on immigrant inflows and native outflows (the panel considers this issue below in the review of research, e.g., Borjas, 2006; Card, 2001; Card and DiNardo, 2000; Kritz and Gurak, 2001). A more intractable problem with the spatial approach, also noted by Borjas (2014a), is that trade in goods between locales or movement of capital can also work to disperse the impacts of immigration nationally. In fact, an important insight of economic theory is that flows across localities, whether in labor, capital, or goods, will tend to diffuse the impact of immigration across the national economy, potentially making spatial comparisons less informative. To the extent that existing spatial studies have not been able to address all possible mechanisms through which local labor markets adjust, it is possible that they underestimate any impact of immigration on labor market outcomes at the national level. At the same time, economic theory also implies that domestic impacts of immigrant inflows are reduced to the extent that the United States trades with the rest of the world and that capital flows into and out of the United States (see Chapter 4).2
2 The extent to which trade serves to reduce the effect of immigration on an individual country has received attention theoretically, and these insights may apply to cross-city analyses. The classic factor price equalization model (Samuelson, 1948) holds that, if a country produces
As noted previously, the second broad category of research reviewed in this chapter focuses on aggregate (national level) data and entails dividing labor markets by skill, typically defined by years of education and experience. Borjas (2003) pioneered both the skill cell and structural approaches that comprise this line of work. In the skill cell approach, estimation relies on variation, not between geographical areas as is done in spatial analyses but between skill groups. The idea is to relate differences in immigrant inflows across the range of skill cells to differences in wage outcomes of native-born workers—just as the spatial approach relates differences in immigrant inflows across places to differences in wage growth. The drawback of this approach is that it does not estimate the entire impact of immigration. While it captures the effect on native-born workers of immigrants who have similar skills, it does not capture the effect on the native-born of immigrants who have dissimilar skills. It is unknown whether omission of these cross-group effects leads to an overestimation or underestimation of the wage impact of immigrants.
The structural approach involves assuming a particular production function describing the relationship between output and inputs (the factors of production), estimating the parameters that characterize the production technology (most notably the elasticities of substitution between factors of production), and then simulating the impact of changes in labor supply on relative wages of, say, native-born workers based on the estimated parameters and the assumed functional form of the production function.3 While, as noted earlier, all empirical approaches require identifying assumptions, structural models require particularly strong assumptions, and some of those assumptions build in specific numerical answers for the wage impact. Apart from the functional form assumptions for the production technology, as detailed in Section 5.3, results may be sensitive to assumptions about the feasibility and extent to which different inputs, such as more- and less-skilled workers or immigrants and native-born, may be substituted for one another. These assumptions are, however, necessary to reduce the dimensionality of these models in a way that makes them tractable.
Another issue for a structural approach is that predictions based on these models ignore general equilibrium effects, such as how different kinds
multiple goods that are each traded internationally, changes in relative supplies of labor of varying skills within that country need not have any effect on the relative wages by skill level within that country, provided the country is small relative to the rest of the world. On the other and, shifts in labor supplies by skill, say due to immigration, may affect relative wages if there is a significant nontraded sector or if a country specializes in one traded good (Dustmann et al., 2005; Kuhn and Wooton, 1991; Samuelson, 1948). See Blau and Kahn (2015) and Borjas (2014a) for a more extended discussion.
of workers interact with each other and how investment, consumption, and other responses in the economy play out. Finally, this approach, like the skill cell approach, assumes that the analyst is able to assign immigrants and native-born workers to cells within which their education and potential labor market experience are equivalent (see Dustmann and Preston, 2012).
Not all studies fall neatly into the taxonomy described above. Both spatial analyses and aggregate skill cell and production function studies may divide workers into skill groups, and a spatial study by Peri et al. (2015a) uses city-specific production functions to estimate total factor productivity growth of U.S. cities attributable to the addition of foreign-born science, technology, engineering, and mathematics (STEM) workers. Borjas (2014a, p. 127) prescribes a strategy for future research that would combine the findings from spatial approaches—where average wage effects are estimated directly from the data—with the restrictions implied by factor demand theory to estimate cross-group effects. Though there may be some overlap and gray areas across approaches, the panel follows this categorical organization in the detailed discussion below of empirical results and then considers the lessons derived from the literature in the concluding discussion (Section 5.7).
The foregoing discussion of economists’ approaches to analyzing the impact of immigration, as well as the Chapter 4 description of relevant theory, highlights the importance of some basic concepts in determining the effect immigrants may have on native-born workers. In particular, it is clear from a theoretical perspective that the expected impact of immigration is larger in the short run than in the long run, at least if the immigration is unanticipated. In addition, whether immigrants are substitutable for natives (and how closely) or complementary with them is important for determining the direction (negative or positive) as well as the magnitude of the immigrant effects. While the theoretical concepts are reasonably clear, empirically testing them is less so. Below, the panel considers some of the empirical issues that have arisen.
The Short Run Versus the Long Run
The standard distinction between the short run and the long run in microeconomic theory is that in the short run the capital stock is fixed and cannot adjust to changes in the demand for capital. Meanwhile, in the long run, capital is completely variable and adjusts fully to changes in demand for it. With immigration, the return to capital initially rises then falls over the adjustment period, eventually returning to its original level.
Macroeconomic theory further distinguishes between a short run in which technology and education (human capital) of workers are fixed and a long run in which they adapt to changing economic circumstances. This latter conception of the long run is the focus of the panel’s discussion of immigration in an endogenous growth context in Chapter 6.
These distinctions are murkier in the real world, since these concepts do not map one-to-one with time periods of specific, consistent length. One guide to the speed at which capital adjusts is a study by Gilchrist and Williams (2004) showing that in (West) Germany and Japan, both of which suffered a large loss of capital during World War II and large population inflows immediately afterwards, the return to capital fell to world levels by the 1960s. This suggests that, for U.S. immigration purposes, capital is likely to adjust fully in considerably less than 20 years and in some cases may even be built up in anticipation of immigration. In studies of the United States, Lewis (2011a) found immigration-induced changes in the adoption of manufacturing automation equipment in a 5-year span from 1988 to 1993, while Beaudry et al. (2010) found immigration-induced changes in the adoption of computers between 1990 and 2000. These studies show that there is at least some adjustment of U.S. capital and possibly technology over 5-10 years, though it is unknown whether the adjustment observed was complete. Moreover, it might be argued that the notion of complete adjustment in the face of ongoing immigration is not clearly defined, in that there is no theory and little empirical evidence on the effect of anticipated immigration on firm behavior.
Among the various approaches reviewed in this chapter, the structural approach deals most explicitly with the distinction between the short and long run. Though the structural models are static and do not model changes over time, they yield separate short- and long-run estimates of the impact of immigration based on explicit assumptions regarding the elasticity of the supply of capital. However, technology is held fixed, and the response of worker human capital is not dealt with explicitly. Results from the spatial approach and the simple skill cell approach are more difficult to characterize along a time dimension. Presumably, estimating the effects of a large, sudden, unanticipated increase in immigration—as occurred with the Mariel boatlift—in the year or two following the inflows captures the short-run effect of immigration. More generally, the estimated effect depends on the spacing of data (e.g., decennial or yearly), the exact specification of the regressions, and the timing of immigrant inflows between the observation points; certain specifications could reflect a mixture of short- and long-run effects (Baker et al., 1999). While the panel acknowledges these ambiguities, we follow an extensive literature in continuing to use the terms “short run” and “long run,” and we grapple with the distinction as it arises in our discussion of differences in magnitudes across studies in Section 5.5.
Substitution Between Inputs and Issues in Defining Skill Groups
Economic theory points to the importance of substitutability and, conversely, complementarity between different kinds of workers in determining the impact of immigration on the wages and employment of natives.4 Where immigrants and natives are substitutes, adverse wage and employment effects may result; the more closely immigrants’ skills and abilities match those of natives, the more adverse these effects are expected to be. This raises the issue of how empirical researchers measure skill and identify groups that are potentially in competition, as well as how they model the extent of substitutability between them. Thus, we consider these issues before delving into the empirical findings on the impact of immigrant inflows on natives and prior immigrants.
Substitutability between two groups—say native workers (N) and immigrant workers (I)—is measured by the elasticity of substitution. The elasticity of substitution between natives and immigrants gives the percentage change in the ratio of immigrant workers to native workers (I/N) employed in response to a given percentage change in the wages of natives relative to immigrants (wN/wI). So, for example, an elasticity of 2 would indicate that an increase of 1 percent in the wage of natives relative to immigrants would result in an increase of 2 percent in the ratio of immigrants to native workers employed. A very high value of this elasticity implies that as the relative wage of natives rises (so natives become more expensive compared to immigrants), employers would make a more sizable switch to hiring immigrant workers—suggesting that it would be easier to make the switch. A low value of the elasticity would suggest that a similar rise in the relative wage of natives would not lead to a very large increase in the relative number of immigrants employed, suggesting that employers find it difficult to replace natives with the immigrants. If the elasticity were equal to zero, a rise in the relative wage of natives would not change the number of immigrants employed at all, suggesting that employers find it impossible to replace natives with immigrants because the two groups are not substitutable.
Substitutes may be divided into perfect substitutes and imperfect substitutes. Two groups of workers that are perfect substitutes are so nearly identical for purposes of production that an employer will be indifferent between hiring a worker from one group or the productivity equivalent number of workers from the other. One somewhat confusing aspect of this terminology is that one might be tempted to assume that perfect substitutes
4 For simplicity and also due to policy concerns, the panel frequently refers to immigrant versus native-born workers. In reality, immigrant inflows may affect the wages not only of natives but of earlier immigrants as well. Some studies have looked explicitly at the impacts of new flows of immigrants on earlier immigrants, as well as on the native-born.
are equally productive—but that need not be the case. As long as the two groups’ relative wages reflect any productivity difference between them, employers will be indifferent between hiring one or the other. The elasticity of substitution between perfect substitutes is infinite. In such a case, if the relative wage of one group were to rise, the employer would shift entirely to the other group. Imperfect substitutes are, as the name implies, substitutable in the eyes of employers but not perfectly so. The magnitude of the elasticity indicates how closely substitutable the two groups are.
In implementing this concept of substitutability, an issue that arises is how to define skill groups. As we have noted, the large representation of less-educated individuals among immigrant inflows into the United States has focused attention of researchers on the wage and employment consequences of this inflow for less-skilled natives. But how is skill to be measured? This question arises across all the approaches this report surveys and has been answered in various ways. No approach is free from some level of disagreement about this issue. In general, studies employing the spatial methodology have used education level as the metric of skill (e.g., Card, 2005), although in a few cases occupations have been used to distinguish skill groups (Card, 2001; Orrenius and Zavodny, 2007). Aggregate skill cell and production function studies generally define skill by taking into account both experience (using age as a proxy) and education to form experience-education cells (e.g., Borjas, 2003). Finally, a recent alternative for defining skill in a way that groups immigrants and natives who are competing in the labor market assumes that two individuals with the same percentile ranking in the wage distribution are viewed as close substitutes in the eyes of employers; Dustmann et al. (2013) applied this approach for the United Kingdom.
One issue that has arisen in spatial studies, as well as in aggregate production function analyses, is how to delineate educational categories. Often, four educational categories are created: (1) did not complete high school, (2) completed high school only, (3) some college, and (4) completed college. Sometimes (e.g., Borjas, 2003, 2014a) the “completed college” group is further divided into college graduates and postgraduates, yielding five categories. Some research has focused on a subset of categories—for example, examining how the inflow of low-skilled immigrants affects the wages of low-skilled natives. Recently, however, questions have been raised as to whether each educational category should be viewed as a separate factor (that is, as imperfect substitutes). Based both on his review of recent aggregate time series studies and his own analysis of spatial data, Card (2009) argued that evidence supports the conclusion that high school dropouts are essentially perfect substitutes for high school graduates. In a production function context, Ottaviano and Peri (2012) also combined the two groups, providing evidence from their data that the elasticity of
substitution is quite high, even infinite in some estimates. The treatment of these two educational categories can have significant implications. As Card (2009) pointed out, immigrants have a much higher share of high school dropouts than natives, but a fairly similar share of “high school equivalent” workers (dropouts and graduates combined, accounting for differences in productivity). Thus, the change in the skill distribution caused by an inflow of immigrants, and the resulting impact of immigration on relative wages, is smaller if the high school dropout and high school graduate categories are aggregated.5 However, aggregating the two groups is not without controversy. Borjas et al. (2012), in particular, take issue with the justification for doing so, namely the evidence on the elasticity of substitution.
The second issue of importance is whether immigrants and natives within skill groups are perfect substitutes. This issue is potentially quite important in that, for cases in which natives and immigrants are imperfect substitutes, any negative wage effects resulting from immigrant inflows will be more concentrated on previous immigrants, who are usually the closest substitutes for new immigrants, lessening the adverse impact on natives.6
Various research findings lend support to the notion that immigrants are imperfect substitutes for natives with similar measured characteristics.7Chiswick (1978) found a lower return to experience and education among new immigrants than among natives—with this experience and education presumably primarily acquired abroad. In line with Chiswick’s findings, Blau and Kahn (2015) found, for a sample of newly legalized immigrants, that education acquired abroad had a lower return than education acquired in the United States, while Akee and Yuksel (2008) found that the gap between the return to foreign versus U.S. experience is larger than that for foreign versus U.S. education. “Downgrading”8 of immigrant skills is also suggested by Akresh’s (2006) finding that, in comparing the jobs immigrants held prior to and after migrating, they typically experienced downward occupational mobility. Also relevant is Kossoudji and Cobb-Clark’s (2000) evidence of occupational upgrading of immigrants upon legalization, which suggests downgrading of unauthorized immigrants skills relative to native-born workers. Blau and Kahn (2007a) reported higher unemployment rates of Mexican immigrants (the largest single group of immigrants) relative to native-born workers with similar age and education—again suggest-
5Card (2009) advocated the formation of just two skill groups: high school equivalent and college equivalent labor. This two-group structure has frequently been used in recent aggregate time series studies.
ing imperfect substitution between the two groups. Finally, evidence from Smith (2012) that an inflow of immigrants with a high school degree or less reduced the employment (measured in hours worked) of native teens suggests that newly arrived adult immigrants may be closer substitutes to native teens than to their adult counterparts.9
Other work highlights the role of English-language fluency, a factor largely unaccounted for in aggregate analyses, in producing imperfect substitutability between immigrants and native-born with similar observed characteristics. Using census data on immigrant-native wage gaps for immigrants who were fluent compared with immigrants with no English, Lewis (2011b) analyzed how native-immigrant differences in language skills contribute to occupational specialization. He found that native-born workers are more represented in occupations where communication is important, which suggests that within education level, immigrants and natives may be imperfect substitutes. As the length of time spent by immigrants in the United States increases, their English improves and immigrants and native-born with comparable education become closer substitutes. In a similar vein, Somerville and Sumption (2009) found that immigrant concentration in particular industries induces natives to shift into higher paying industries where language and other native skills come into play. Likewise, Peri and Sparber (2011) investigated the role of communication skills in producing immigrant/native-born differences in occupations requiring graduate degrees. They found that the foreign-born specialize in fields demanding quantitative and analytical skills and the native-born specialize in fields where interactive and communication skills are highly valued.
Additional evidence suggesting imperfect substitution between immigrants and the native-born was provided by Ottaviano and Peri (2012). Using a structural production function approach, they estimated substitution elasticities, whose values indicate that immigrants and natives were imperfect substitutes within the typical categories used, especially among the less skilled. The production function approach they employed enabled them to take this imperfect substitutability into account in estimating wage effects. Borjas et al. (2012) challenged these findings and presented evidence that the results are sensitive to assumptions made in the estimation process.10 Moreover, while Dustmann and Preston (2012) agreed that the usual approach groups together dissimilar immigrants and natives, they
9Orrenius and Zavodny (2008) found a similar result: Minimum wage increases resulted in higher employment rates among adult immigrants while rates fell for native-born teens. The evidence therefore suggests employers switched to older foreign-born workers in lieu of native-born teens once labor costs rose.
Spatial studies potentially have methods for handling imperfect substitutability between immigrants and natives as well. As an example, Altonji and Card (1991) estimated the link between the fraction of immigrants in the population and the wages and employment of less-skilled natives. Their specification allows any impact that immigrants with higher observable skills may have on the low-skilled native group (due to the immigrants’ imperfect substitution with higher-skilled natives) to be captured as well. It is also possible to build in adjustments to realign the way new arrivals are sorted into skill cells in these models. Orrenius and Zavodny (2007) examined the impact of immigrant penetration separately by occupational category, to allow immigrant substitutability to differ by skill. They argued that substitutability of immigrants for natives should be greater for less-skilled occupations and found results consistent with this hypothesis. In contrast, in their production function study referenced above, Ottaviano and Peri (2012) hypothesized, and found evidence, that among the highly educated, foreign-born workers are more highly substitutable for native-born workers. While these results differ, both studies found evidence of imperfect substitutability between immigrants and natives that appears to differ by skill level.
Other evidence supportive of imperfect substitutability between immigrants and natives comes from studies examining the impact of immigrant inflows on natives and prior immigrants separately. The idea here is that, if immigrants and natives are imperfect substitutes, the impact of immigrant inflows on prior immigrants should be larger than on the native-born, since immigrants are likely to be closer substitutes for each other than for natives. Many studies focus only on the native-born component of the pre-existing workforce, but when both groups are examined, larger negative wage and employment effects for previous immigrants than for the native-born are generally found (e.g., Card, 2001; Ottaviano and Peri, 2012).
Support for the view that immigrants downgrade upon arrival comes from the study noted above by Dustmann et al. (2013) for the United Kingdom. Although immigrants to the United Kingdom have typically had
11 Specifically, Dustmann and Preston (2012) argued that a key assumption in the Ottaviano and Peri (2012) approach is that immigrants and natives can be allocated to age-education cells within which their potential experience and education are comparable. This may, however, not be the case, as immigrants may—at least initially—downgrade, which means they compete with natives in segments of the labor market other than where one would expect them based on their observed education and potential experience. This will cause a bias in the estimates of the elasticity of substitution between immigrants and natives. Due to downgrading, immigrants and natives may appear to be imperfect substitutes even though, if correctly classified, they are not.
more education on average than native-born workers, they have fallen disproportionately at the lower end of the wage distribution. This finding, the authors claimed, has serious consequences for approaches that rely on preassigning immigrants to skill cells based on their observed age and education, within which they are assumed to be equivalent in production to natives.
Data from the Current Population Survey (CPS) indicate that downgrading is also an issue in the United States, although to a lesser extent. Figure 5-1 (from Dustmann and Preston, 2012) shows the predicted position, based on age and years of schooling, and the actual position of recent immigrants relative to the native-born wage distribution. The short-dashed line in the graph (labeled “actual”) indicates that recent immigrant workers are more concentrated in the lower quintiles and less concentrated in the
higher quintiles of the native wage distribution than would be predicted by their age and education profiles (the horizontal line is the reference indicating the nonimmigrant wage distribution; the long-dashed line is where one would predict immigrants to be located along the distribution of native wages if they received the same return on labor for their observed education and experience as natives did). Elsewhere in their paper, Dustmann and Preston (2012) showed that downgrading is strongest just after arrival (the period reflected in the graph); they found that over time, immigrants to the United Kingdom catch up to the occupations and wage levels predicted by their education.
Based on observations like these, Dustmann et al. (2013) argued against estimators that require the preallocation of immigrants to skill groups, arguing that this may not lead to meaningful estimates because immigrants may compete with native-born workers at other parts of the skill distribution than those to which one would assign them based on observed characteristics. Using a spatial approach, they proposed an estimator that does not rely on preallocation of immigrants to skill groups but instead regresses skill-group-specific native wages (in their approach, defined as percentiles of the wage distribution) on the overall inflow of immigrants. The resulting estimates have a straightforward interpretation and are not affected by downgrading.
While there is indeed suggestive evidence that immigrants and natives may be imperfect substitutes within skill groups defined by measured characteristics, there remains controversy regarding whether this is an important issue for empirical analyses and how it should be dealt with. The panel considers this issue further, along with the appropriateness of aggregating high school dropouts and high school graduates, in the context of the studies reviewed below.
In the pioneering work by Grossman (1982) on the “substitutability of immigrants and natives in production”—a paper that influenced much of the subsequent research—labor market boundaries were defined as metropolitan areas. Intuitively, since immigrants choose some destinations with greater frequency than others, comparing wage and employment trends across metropolitan areas should yield evidence about the impact of their arrival. As described above, the methodology involves testing whether native wage growth and employment rates in the high-immigration areas are lower than those in the low-immigration areas.12 The earliest studies relied solely on cross-sectional variation, while later work, beginning notably with Altonji
and Card (1991) and including most of the studies summarized here, recognized and attempted to deal directly with the endogeneity problem inherent in this approach: The magnitude of immigrant flows into an area is likely to be correlated with its economic vitality and wage growth.
Studies relying on geographic labor market variation are listed and compared in Section 5.8, Table 5-3. In considering the results of these studies, a useful starting point is the assessment of evidence presented 20 years ago in The New Americans (National Research Council, 1997). For the literature surveyed in that report, with the exception of Altonji and Card (1991), the estimated coefficient indicating the sensitivity of native-born wages to an increase in immigrants in a given local labor market was closely clustered around zero. The New Americans reported:
The evidence also indicates that the numerically weak relationship between native wages and immigration is observed across all types of native workers, white and black, skilled and unskilled, male and female. The one group that appears to suffer significant negative effects from new immigrants is earlier waves of immigrants, according to many studies. (National Research Council, 1997, p. 223)
As documented below, however, continued study of this issue over the past two decades has led to greater variation and detail in estimates of the wage impacts of immigration obtained from the local labor market approach.
Comparing the experiences of high-immigration and low-immigration geographic areas has a great deal of intuitive appeal. The concept is easy to understand. Blau and Kahn (2015, p. 813) outlined the advantages of the approach:
. . . the empirical work directly ties the key explanatory variable, immigration, to the outcomes of interest. No assumptions about how labor and other inputs combine in production processes need be made. In particular, one need not assume or try to estimate the degree to which immigrants and natives of equal observed skills substitute for each other, although such a relationship will influence the parameter estimates. In addition, using the area approach will provide more potential observations than using national aggregates, producing more efficient estimates.
The analytic challenges to spatial studies have to do with the endogenous factor flows and trade flows that potentially bias the estimates of cross-area wage differentials.13Borjas (2014a), Blau and Kahn (2015), and
13 This is also an issue for aggregate skill cell and production function models, discussed in Section 5.4, albeit possibly a lesser one. As explained by Llull (2015), immigrants to the United States do not display random experience levels (ages) and education.
others, as noted below, identified these challenges: (1) Immigrant flows are not randomly distributed across metropolitan area labor markets. As noted above, new arrivals are likely to select areas at least near those that are thriving economically14—that is, those experiencing wage and employment growth (e.g., California or Florida in the mid-to-late 1990s). This area-selection bias creates spurious, positive correlations between immigration to an area and that area’s employment conditions and relative wages. (2) Local labor markets are not closed, which means that natives (or earlier immigrants) are free to relocate their labor (and capital), which may at least partially equilibrate prices and quantities across markets defined by geographic areas. As possible evidence of this problem, Borjas et al. (1997) showed that, for the 1980-1990 period, the correlation between inflows of low-skilled immigrants and the wages of low-skilled natives was more negative, the larger the geographical area demarcated (regions versus states or states versus metropolitan areas). Similarly, Borjas (2003) included analyses by geographical areas (i.e., states) that reveal smaller negative effects on a skill group’s earnings from an immigrant inflow than did the national level estimates. (3) Trade in goods between areas will tend to equalize factor prices, including wages, across areas, in a process known as factor price equalization. Finally, models for which the key independent variable (immigration) is measured for small geographic areas with small samples are susceptible to measurement errors, greatly attenuating the measured impact of immigration (Aydemir and Borjas, 2007).
Endogeneity of Change in Immigrant Share and Labor Market Performance
The above complications associated with estimating cross-area wage and employment effects make it difficult to establish causal links. Regarding the endogeneity challenge, the question is: To what extent do immigrant inflows affect wages and employment and to what extent do wage and employment conditions influence immigrant inflows? Either could explain an observed correlation, and both probably occur to some degree in any given case. Indeed, Cadena and Kovak (2016) showed that low-skilled immigrants have settled in those cities that offer the highest wages, leading to a positive correlation between wage growth and immigrants’ location decisions. If new arrivals migrate to strong economies that are already experiencing high or rising wages, measured negative effects of immigration
14 Mainly due to housing, immigrants are often priced out of the most economically thriving neighborhoods within a metropolitan area (Saiz, 2008). For this reason, analyses at, for example, the census tract level may produce quite different results from those at the Metropolitan Statistical Area (MSA) or state level.
will be understated unless this counterbalancing influence is accounted for. Conversely, immigration may decline in response to relatively slow wage growth in areas that are economically depressed. Monras (2015) found that, during the Great Recession, “fewer people migrated into the locations that suffered more from the crisis.” This relative shrinking of the labor supply in the most hard-hit metropolitan areas would have alleviated some of the negative wage effects associated with the crisis by spreading the local recession shocks across regions or nationally.
As noted above, this endogeneity problem may be overcome by isolating the variation in immigrant inflows across areas that is neither determined by outcome variables (such as area wages) nor affected by the same unobserved factors that influence wages. The common approach to doing this is to find a variable (or a set of variables) that (1) is correlated with the inflow of immigrants to an area, but (2) is not correlated with factors that determine the growth of wages, other than through the inflow of immigrants. Such variables are called “instrumental variables” (IVs) or just “instruments.” While (1) is an empirical question, and can be tested, (2) is untestable and has to rely on the plausibility of the assumptions under which it is valid. The quality of the study depends therefore on the degree to which the assumptions underlying (2)—called exclusion restrictions—are plausible.
It can be difficult to find instruments that are highly correlated with the inflow of foreign-born workers into a local labor market yet uncorrelated with the other factors that determine wages or job growth in that area. The most common IV strategy, introduced by Altonji and Card (1991) and further developed by Card (2001), relies on the observation that immigrants tend to locate where there are already settlements of their co-nationals (see Bartel, 1989). Reasons suggested for this tendency include the possibility of drawing on preexisting networks, informational advantages, and access to cultural goods that are difficult to obtain without access to co-nationals. While past concentrations of individuals from one’s own country are likely to be correlated with future inflows to a particular area, they are at the same time unlikely to be correlated with future area-specific shocks that affect wages and employment. Based on this line of reasoning, the approach then allocates the overall inflow of immigrants from a particular country to spatial areas based on historical settlement patterns. For example, suppose the United States consisted of a Southern part and a Northern part only; assume further that, in 1980, 10 percent of all immigrants from Mexico lived in the North, while 90 percent lived in the South. Now suppose that 100,000 Mexicans arrived between 1999 and 2000. Based on the historical settlement pattern in 1980, this approach would assign 10,000 to the North and 90,000 to the South. Doing the same assignment process for all immigrant groups and summing up for each region results in an estimate
of the area-specific inflow of immigrants between 1999 and 2000 that is solely based on historical settlement patterns and is unlikely to be correlated with contemporaneous (i.e., 1999-2000) area-specific shocks to wages and employment.
One possible problem with this approach is that economic characteristics that initially made an area attractive to immigrants may persist over time. For example, if traits of the economy driving both economic growth and migration in gateway locations such as California or New York have systematically differed from other regions over many years, the downward impact of immigration on wages may still be masked. However, as Blau and Kahn (2015) noted, the finding by Blanchard and Katz (1992) that the wage effects of local employment shocks die out within 10 years provides some support for the interval, employed in most of these studies in the construction of the instrument, of 10 or more years between the previous immigrant concentrations used to derive the allocation and the current inflows.15
Due to concerns about whether local labor market conditions during the analysis period are, or are not, directly related to conditions for the period from which the instrument is constructed, researchers have begun exploring alternative instruments. For example, an IV constructed to deal with endogeneity of location choices may be based on a characteristic such as the distance between origin and destination countries. In a skill cell study based on cross-national comparisons, Llull (2013, p. 2) used variation in “push factors . . . interacted with distance to the destination country in order to construct an instrument based on variation over time and across destination countries.” So, for example, violence in Guatemala would be expected to increase migration to Mexico or the United States at a greater rate than to Europe. Llull further broke out variation by skill level, based on the assumption that destination choices will be more constrained for low-skilled workers because, compared with high-skilled workers, they have fewer resources to travel long distances.
Native Response to Immigration, Trade, and Technology Adjustments
Mobility of labor, capital, and goods between areas gives rise to a second analytic challenge for spatial studies. Cities and states are not closed economies, meaning that labor and capital flow from one to another, and these flows have the capacity to equalize prices.16 If immigrants were to
16 Price equalization pressure would also happen in the presence of trade even if labor and capital were immobile—see below and the theory discussion in Chapter 4. This is important because sometimes papers find that labor is not that mobile and mistakenly conclude that therefore prices are not equalizing.
arrive in disproportionate numbers in a city (or neighborhood, or whatever spatial unit defines the labor market), it is possible that some workers previously there may respond by moving elsewhere, which would diffuse the downward pressure on wages across cities:
. . . natives may respond to the wage impact of immigration on a local labor market by moving their labor or capital to other cities. These factor flows would re-equilibrate the market. As a result, a comparison of the economic opportunities facing native workers in different cities would show little or no difference because, in the end, immigration affected every city, not just the ones that actually received immigrants. (Borjas, 2003, p. 1338.)
In such a scenario, a comparison of wages across cities would reveal little, if any, wage effect.
While predicted by theory, evidence of the equilibrating hand of factor input mobility—specifically, native migratory response to increased job competition—is mixed. On one side, Card (2001), Card and DiNardo (2000), Kritz and Gurak (2001), and Peri (2007) found, for the U.S. context, either no relationship between the entry of immigrants and the exit (or failure to enter) of the native-born or that both immigrants and the native-born moved to the same cities and probably for the same reason: economic opportunity. Economically healthy cities, for example, likely attract inflows of both international and domestic migrants. These results suggest that outflows of natives may not significantly contaminate estimates of immigrant effects based on regional variation.
The evidence on the other side, for factor input mobility, includes Borjas (2006), who used Decennial Census data for the period 1960-2000 to show that internal migration decisions by natives are sensitive to immigrant-induced increases in labor supply. Specifically, high-immigration areas were associated with lower native in-migration rates and higher native out-migration rates. Native migration responses, in turn, “attenuate the measured impact of immigration on wages in a local labor market by 40 to 60 percent, depending on whether the labor market is defined at the state or metropolitan area level” (Borjas, 2006, p. 221). Some heterogeneity in responses has also been detected. For example, Kritz and Gurak (2001) found minimal overall connection between in-migration of foreign-born and out-migration of native-born, but they also found that the results varied by state and by group. They found a positive relationship between immigration and native out-migration for California and Florida and also found that, in states that have experienced the highest immigration, foreign-born men were more likely to out-migrate than were native-born men. That is, prior immigrants were more mobile than natives. Partridge and Rickman (2008) found out-migration responses to immigration to be more significant
in rural counties. In addition, they found that previous interstate movers (immigrant or native-born) were more likely to move from states with high recent immigration than either immigrants living in their state of first settlement or natives living in their state of birth.
A similar masking of cross-area impacts could occur due to intercity and interstate trade. Card (2005, p. 10) noted that, in the presence of trade across cities, “relative wages may be uncorrelated with relative labor supplies, even though at the national level relative wages are negatively related to relative supplies.” If low-skilled international immigrants move to Los Angeles, for example, the production of goods intensive in low-skilled labor will increase there. However, the prices of these goods in Los Angeles may not change compared to other cities because free trade within the United States ensures prices are equalized across cities and regions, and so are wages (which is the factor price equalization theorem). This means that so long as technology does not change, relative wages of low-skilled workers in Los Angeles compared to other cities will not change either. This logic holds as long as the inflow of immigrants is not so large that Los Angeles ceases to produce goods intensive in higher-skilled labor and comes to specialize in low-skilled intensive goods;17 in this case, relative wages of low-skilled workers in Los Angeles could indeed fall compared to other cities. These results are also contingent on there not being a significant nontraded sector and on Los Angeles producing just a small share of low-skilled intensive goods produced nationally.
In sum, any type of labor market response to immigration—whether along the margin of labor flows, capital flows, or flows of goods—can serve to diffuse the impact of immigration from the localities directly affected to the national economy. This kind of diffusion implies that even though one may not observe adjustments along a particular margin, there may be other unexamined and unexplored margins along which such adjustments can take place. Any such adjustments imply that spatial correlations between wages and immigration may underestimate the national wage impact of immigration.
The adjustments described thus far in this section explain why spatial studies may underestimate any national wage impact of immigration. However, the same reasoning implies that there are other adjustments—international trade in goods and services and capital flows across countries—mitigating the wage effect of immigration at the national level. Imports and exports of goods and services together represented 30.0 percent of U.S. gross domestic product (GDP) in 2014, indicating that the United States is well integrated in world trade. Along with large capital flows between
the United States and foreign countries, this trade may prevent or limit any wage response to immigration, though this is difficult to study empirically.
The ability of firms to change their technology is another factor possibly dampening negative wage impacts of immigration. The basic idea is that firms adjust technology to absorb workers who become more abundant through immigration (see Section 4.5). Similar to the situation with trade, this adjustment can lead to a situation where an immigration-induced labor supply shock is absorbed without changes in wages.18Hanson and Slaughter (2002) were among the first to compare the trade- and technology-induced adjustments to labor supply shocks on the industry level, while Dustmann and Glitz (2015) extended this literature by investigating adjustments at the firm level and considering the role of firm births and deaths in the adjustment process. Both papers found that technology-induced changes in factor intensity are more important for the absorption of immigration than trade-induced changes in the mix of outputs (see also Lewis, 2013). Lewis (2011a) focused on the technology explanation and examined how investment in automation machinery by U.S. manufacturing plants over recent decades has substituted for different kinds of labor. He concluded that “these investments substituted for the least-skilled workers and complemented middle-skilled workers at equipment and fabricated metal plants.” He found that metropolitan areas that experienced faster growth in the relative supply of less-skilled labor as a result of immigration “adopted significantly less machinery per unit output, despite having similar adoption plans initially [implying that] fixed rental rates for automation machinery reduce the effect that immigration has on less-skilled relative wages” (Lewis, 2011a, p. 1029).
Illustrative Results from Spatial Studies
Table 5-3 in Section 5.8 summarizes the results from spatial studies of the labor market effects of immigration, most of which employed IV methods to address the endogeneity of immigrants’ locational choices. While these studies are not uniform—they use different data, look at different time periods, and examine varying magnitudes of immigrant inflows—their results suggest that the impact of immigration on the group most likely to be affected, low-skilled workers, ranges from negligible to at least modestly negative. A more precise comparative assessment of the literature is provided in Section 5.5 below. As noted above, some groups such as prior
18 In terms of a standard model of production, this interpretation refers to a change in relative inputs due to a technology-induced rotation of the isoquant around a fixed isocost line, while the trade explanation above refers to a situation where relative inputs (i.e. shares of low-skilled to high-skilled labor) change due to the isocost line rotating around a fixed isoquant.
immigrants—for example, the Hispanic immigrants and Hispanic native-born studied by Cortés (2008)—appear to experience somewhat larger negative wage impacts. One contributing factor to the differential wage impact experienced by Hispanics, identified by Warren and Warren (2013) and Massey and Gentsch (2014), is that these groups are often competing in labor markets characterized by a rising share of unauthorized workers who are under increasing enforcement pressure. This may reduce their bargaining power and create downward pressure on wages in those labor markets. Employment impacts, measured in various ways discussed below, are also modest but perhaps vary more broadly across metropolitan areas.
Spatial studies commonly designate the skill group of natives, and sometimes immigrants, according to education level, although some use occupation as the skill dimension. Given the composition of immigrants relocating to the United States historically, the focus has generally been on their impact on low-skilled or other disadvantaged groups. The important study by Altonji and Card (1991) is an example. The IV approach used in most subsequent studies had its beginnings in this study and was later further refined in Card (2001). In Altonji and Card (1991), the 1970 share of immigrants in the population was used to construct the IV for immigrant inflows over the 1970-1980 period. As discussed above with regard to the possible imperfect substitutability of immigrants and the native-born with similar measured characteristics, focusing on the total immigrant share implicitly allows cross-effects to be examined. However, it does not allow an analysis of which immigrants are having the largest impact and instead measures the average effect.19
Overall, Altonji and Card (1991) found that immigration had a negative effect on wages, with a 1 percentage point increase in the immigrant share of the population reducing wages of low-skilled, native-born workers by 1.2 percent. They also found that a 1 percentage point increase in a city’s foreign-born share predicted a reduction in the earnings of black males with a high school degree or less by 1.9 percent, black females with high school or less by 1.4 percent, and smaller—and statistically insignificant—reductions in earnings for whites with a high school education or less. The only other spatial study that found negative wage effects of similar magnitude is Borjas (2014a); the panel discusses below why these results might differ from those of other studies. Regarding employment (as opposed to wage) effects, Altonji and Card (1991) found that immigration
19 For example, two cities may have the same share of immigrants but in one city immigrants may be predominantly high skilled and in another predominantly low skilled. As explored in Section 4.5, the estimated effect of the immigrant share variable may be smaller than if the effect of immigrant shares of low-skilled and high-skilled immigrants on their native counterparts were separately examined.
over the 1970-1980 period in low-wage industries led to modest displacement of low-skilled natives from those industries; but they found no statistically significant reduction in low-skilled natives’ weeks worked or the employment-to-population ratio.
LaLonde and Topel (1991) examined the impact of recent immigration on different arrival cohorts of prior immigrants. Their results are notable for identifying a negative relationship between new inflows and the earnings of recent prior immigrants—an effect that appeared to diminish with the amount of time prior immigrants had spent in the United States. In addition, they characterized the estimated effect of immigrants on the wages of nonimmigrants as “quantitatively unimportant” (Lalonde and Topel, 1991, p. 190). While they did not instrument for immigrant inflows, potentially underestimating the negative effect of immigrants, their findings are consistent with evidence discussed above of imperfect substitution between immigrants and native-born workers.
Since immigrants were disproportionately (relative to native-born workers) in the low-skilled category in the time periods examined, researchers expected larger impacts of immigration on the wages of low-skilled native-born blacks than whites because among low-skilled workers, native-born blacks are less skilled and otherwise disadvantaged compared to native-born whites. As noted above, Altonji and Card (1991) found adverse wage effects that were larger for blacks than whites. LaLonde and Topel (1991) also reported a negative effect for young (and hence inexperienced) native-born blacks, finding that a doubling in the number of new immigrants would decrease wages by a very modest 0.6 percent for young native-born black workers. Other studies for this period (e.g., Bean et al., 1988; Borjas, 1998) did not detect an effect for native-born black workers. Original analysis of Decennial Census data in The New Americans suggested that one reason for this minimal measured impact was that—as of the mid-1990s—immigrants and the black population still largely resided in different geographic locations and therefore were not typically in direct competition for jobs (National Research Council, 1997, p. 223). Until recently, large proportions of the nation’s immigrants were concentrated in relatively few geographic areas, making the distinction between high- and low-immigration areas somewhat intuitive. However, relative to 20 or even 10 years ago, immigrants are now much more spatially diffused, so one should not assume that these historical relationships continue to hold.
Returning to the question of the impact of immigration on the wages of less-skilled natives, subsequent studies by Card (2001, 2005, 2009) concluded that—in line with previous findings other than Altonji and Card (1991)—the impact of immigration on the wages of less-skilled natives was modest for the various time periods considered in these studies. The Card studies all use instrumental variables to address endogeneity of immigrant
inflows, and in Card (2001, 2009) the issue of native out-migration was addressed and found not to play a role. Card (2005, 2009) raised the possibility that high school dropouts and high school graduates are perfect substitutes as an explanation for these small wage effects. As noted above, if this is the case, then the skill distribution of immigrants is quite similar to that of natives and hence large negative wage effects on low-skilled natives are not expected.
While most studies in the spatial literature use education to define skill, it is noteworthy that Card (2001) and Orrenius and Zavodny (2007) focused instead on occupation. The former separated the labor market into different metropolitan areas and, within metropolitan areas, into different occupation groups. Immigrants’ inflows into cells defined by occupation and metropolitan area were predicted for each immigrant source country based on (1) the share of earlier immigrant cohorts from the source country living in the metropolitan area and (2) the national share of immigrants from the source country in each occupation. Card then summed over source countries to obtain the instrumental variable for immigrant inflows into these occupation-metropolitan area cells.20 The basic finding of this study was that immigration during 1985-1990 reduced real wage levels by at most 3 percent in low-skilled occupations in gateway U.S. metropolitan areas characterized by the highest immigration levels. Results varied by group: a 10 percent labor supply increase due to immigration (implying a much larger percentage increase in the number of immigrants) was associated with a wage decline of 0.99 percent for male natives and 0.63 percent for female natives, a decline of 2.5 percent for earlier female immigrants, and a change indistinguishable from zero for earlier male immigrants. It is notable that the largest negative effects were for an immigrant group. On the employment side, Card (2001, p. 58) found “relatively modest” effects of recent immigrant inflows on workers in the bottom of the skill distribution in “all but a few high-immigrant cities.” A 10 percent labor supply increase was found to have reduced the employment rate by 2.02 percentage points for male natives, by 0.81 points for female natives, by 0.96 points for earlier male immigrants, and by 1.46 points for earlier female immigrants.
Orrenius and Zavodny (2007) used a panel model with instrumental
20 That is, what Card termed the “supply-push component” of immigrant inflows for group g into occupation group j and city c (SPjc) is:
SPjc = ∑gτgjλgcMg,
where Mg represents the number of immigrants from source country g entering the United States between 1985 and 1990; λgc is the fraction of immigrants from an earlier cohort of immigrants from country g who live in city c in 1985, and τgi is the national fraction of all 1985-1990 immigrants from g who fall into occupation group j.
variables to estimate wage impacts of immigration on natives, also by occupation group. The authors found a small negative effect on the wages of low-skilled natives and no wage effect in more-skilled labor markets. A variable quantifying “immigrants who are admitted to the United States in a given year as the spouse of a U.S. citizen by occupation group, area, and year” works as the instrument because it is correlated with the rate of immigration into a given Metropolitan Statistical Area (MSA) and occupation but is uncorrelated with unobserved factors that drive wage growth (Orrenius and Zavodny, 2007, p. 11).
Smith (2012) examined spatial variation in employment for a narrowly defined group of workers under the hypothesis that new immigrant workers often compete in very specific labor markets. Also employing an IV model based on the geographic preferences of previous immigrants, he found that low-skilled immigration since the late 1980s had negatively impacted youth employment more than less-educated, native-born adult employment. He estimated that a 10 percent increase in the number of immigrants with a high school degree or less reduced the “average total number of hours worked in a year by around 3 percent for native teens and by less than 1 percent for less-educated adults.” This finding adds a new detail to the previous research that generally found modest negative or no relationship across states or cities between intensity of immigration and adult labor market outcomes across metropolitan areas or states (e.g., Card, 1990, 2001; Lewis, 2003). Smith (2012, p. 55) suggested that two factors were at work, “There is greater overlap between the jobs that youth and less-educated adult immigrants traditionally do, and youth labor supply is more responsive to immigration-induced changes in their wage.” His empirical analysis also suggests that, despite modest increases in schooling rates of natives in response to immigration, there is little evidence of higher earnings 10 years later in life. Smith concluded that it is possible “an immigration-induced reduction in youth employment, on net, hinders youths’ human capital accumulation.”
Other recent studies also suggest larger negative effects of immigrant inflows on earlier immigrants than on natives, consistent with LaLonde and Topel’s (1991) earlier findings and the notion of imperfect substitution between the two groups. Cortés (2008) examined the impact of immigrant inflows over the 1980-2000 period in immigrant-intensive predominantly service industries, following Card’s approach of instrumenting immigrant inflows using previous settlement patterns. Similar to Card, she found that low-skilled immigration does not have an effect on low-skilled native wages overall. She did, however, find a modest negative impact on the wages of low-skilled previous immigrants and low-skilled native-born Hispanics, especially those with poor English. Complementary findings by Lewis (2013) indicate that among immigrants, the wages of those with poor English skills are more
sensitive to immigrant inflows than the wages of those with good English skills. This evidence suggests that language skills may be a significant factor influencing substitutability between immigrants and natives with the same observed characteristics.
Sometimes “natural experiments” arise that provide unique opportunities to deal with the endogeneity problems inherent in spatial analysis. Such experiments also provide an opportunity to study the short-run effect of abrupt, unexpected immigration episodes, which should yield the most negative impacts on natives. An example is the pioneering work by Card (1990), who took advantage of one such case—the 1980 Mariel boatlift, which brought thousands of predominantly low-skilled Cuban immigrants (referred to as “Marielitos”) to Miami, expanding that area’s labor force by about 7 percent in just a few months. This circumstance allowed for a well-controlled analysis: Card was able to estimate the impact of this immigration episode by comparing wage and employment changes after the influx in Miami with wage and employment changes in otherwise similar metropolitan areas that did not experience this influx. The endogeneity problem confronting spatial analyses was avoided altogether because the arrival of the Marielitos to Miami had nothing to do with selection of a high wage destination. Card’s study was one of the first to use the identification strategy that became known as the “difference in difference” approach: comparing differences in wages or employment between Miami and other metropolitan areas, and over time. However, it still entails an important assumption—that, in the absence of the Mariel boatlift, wages and employment in Miami would have developed in similar fashion as in the comparison metropolitan areas (the “common trend assumption”).
Using this approach, Card (1990) found that, while the unemployment rate among black workers rose in the 2 years after the Marielito influx into the labor market, the rise was not significantly different from that experienced in four comparison cities (Atlanta, Houston, Los Angeles, and Tampa-St. Petersburg, chosen because of similar racial profiles and employment trends). One explanation Card provides is the flexibility of the Miami labor market in absorbing low-skilled workers by expansion of industries that produce goods that use low-skilled workers more intensively. In this study, the comparison cities substitute for the missing counterfactual: namely, what would have happened if the immigration had never taken place.
First, Borjas (2016b) and Peri and Yasenov (2015) have recently reappraised the Mariel boatlift immigration episode, carefully matching the skills of the arrivals with those of the pre-existing workforce. The skill-matching technique led them to focus on the impact on non-Hispanic (Borjas)
and non-Cuban (Peri and Yasenov) high school dropouts because high school dropouts represented about 60 percent of those arriving on Florida’s shores as a result of Castro opening the port of Mariel. The available data do not permit natives and immigrants to be distinguished, but Miami had few non-Hispanic immigrants at that time. Both papers were motivated in part by the development of a new technique (Abadie et al., 2010) to select comparison cities more systematically than did Card. Despite this methodological similarity, the authors reach very different conclusions. Peri and Yasenov concurred with Card, finding no detectable negative effects on wages of non-Cuban workers. Borjas found that a drop, in the range of a 10 to 30 percent decrease, in the relative wage of the least educated Miamians occurred between 1979 and 1985, representing a shock that took the better part of a decade to absorb. The divergent results in the two studies are due in large part to the composition of the samples and data sources examined to analyze wage trends in post-Mariel Miami and the comparison cities.
The misalignment of the study results described above suggests that differences in the implementation of a methodology can result in quite different estimates of the impact of immigration. Consideration of these studies also underlines that what occurred to the wage structure in Miami was a very unusual event—one that can be characterized as a true short-run shock occurring in a compressed time period, as opposed to more-anticipated immigrant flows that typically occur over longer time periods. The decade-long absorption of the supply shock in the Miami labor market was a unique episode and may not be fully informative about the dynamics of how labor markets in general adjust to immigration.
Monras (2015) exploited a different natural experiment. The Mexican peso crisis caused that country’s GDP to contract by 5 percent in 1995, leading to a surge in Mexican immigration to the United States for reasons unrelated to changes in the U.S. economy. This event allowed Monras to estimate a short-run effect by comparing wage data for 1994 and 1995 using the CPS. Unlike in the Mariel boatlift case, this natural experiment did not direct immigrants to a particular location in the United States, so Monras used the usual IV for immigrant location based on the 1980 settlement pattern of Mexicans. He found that a 1 percent increase in labor supply due to the immigration of Mexicans with an education of high school or less reduced the wages of pre-existing non-Hispanic workers with an education of high school or less by 0.7 percent. The pre-existing workers in this sample include non-Hispanic immigrants. The observed effect is less negative than that observed by Altonji and Card (1991) but more negative than those observed by Card (1990) and Cortés (2008). Monras found that internal migration caused most of the effect to dissipate within 10 years.
Using a natural experiment approach in the study of immigration is quite attractive, although, as one can see in our discussion of the impact
of the Mariel boatlift, the results are still not free from disagreement. It would certainly be of considerable interest to have a number of such studies for the United States. But, by its nature, this type of exogenous inflow of immigrants is a rare occurrence. While the panel’s review in this chapter is focused on empirical evidence for the U.S. experience, in this case, given the paucity of data for the United States, it is worth noting evidence from other countries where natural-experiment situations have arisen.
Blau and Kahn (2015) surveyed not only Card’s (1990) analysis of the Mariel boatlift but also studies of four other natural-experiment events: (1) the repatriation of French-Algerians following the end of colonial rule in Algeria in 1962 (Hunt, 1992); (2) the repatriation of Portuguese residents from former Portuguese colonies in Africa in 1974 (Carrington and de Lima, 1996); (3) the migration of Jews from the former Soviet Union to Israel after the loosening of emigration restrictions in 1990 following the fall of Communism (Cohen-Goldner and Paserman, 2011; Friedberg, 2001); and (4) the repatriation of ethnic Germans from Eastern Europe and the former Soviet Union following German reunification (Glitz, 2012). In each case, the immigrant inflows were relatively sudden and quite sizable. Blau and Kahn concluded, from the evidence of these studies, that “while the studies are not unanimous, there is at most weak evidence . . . that these episodes had important effects on the level or distribution of native wages, despite the size of the immigration shocks” (Blau and Kahn, 2015, p. 828).
The spatial studies described in Section 5.3 rely on variation in the immigrant density across metropolitan areas or states to infer differential wage and employment impacts. Skill cell studies, such as the pioneering study by Borjas (2003), exploit variation in the density of immigrants across groups of workers categorized by their work experience (typically using age as a proxy) and education, the principal (observable) determinants of skill. Sorting into these skill cells allows for a comparison of outcomes (typically wages) of workers presumed to compete in approximately the same labor market. Labor supply changes, in the form of new immigration, permeate various skill groups unevenly; for example, recent immigrants have been represented disproportionately at very low and very high education levels. The methodological approach is to compare the changes in natives’ outcomes in skill cells that experienced larger increases in immigrant density with the changes in natives’ outcomes in skill cells that had smaller increases; the comparison allows the impact of immigration to be inferred. Specifically, the approach measures the wage effect on natives of inflows of immigrants of similar skill, averaged across all skill levels.
Skill cell studies have typically (but not always) been conducted at the
national level, which alleviates the problem in spatial models of diffusion of any national impact across geographic areas.21 However, the problem remains that incoming immigrants with particular skills may be responding to changes in demand for workers of different skill types, thus leading to spurious correlations between wage growth by skill type and the change in immigrant density by skill type. Another problem with this approach is that the experience and education of immigrants—as reported in survey data—may not be as highly valued by employers as are their equivalents in native-born workers, meaning that immigrants may be allocated by the model to skill cells different from the ones in which they are actually competing with native workers. As noted above in this chapter, Dustmann and Preston (2012) discussed the role of skill downgrading in the sorting of immigrants into occupations. Relatedly, because surveys suitable for the study of immigration do not contain information on actual experience, necessitating the use of age as a proxy, the classification into skill cells is considerably less accurate for women than for men. For this reason, the reported studies using this approach have all limited themselves to analyzing the impact of immigration on males.
A quite distinct set of studies employs the methodological approach referred to as the structural approach. Structural studies of immigration typically divide workers into skill cells at the national level, but the hallmark of the approach is the imposition of theory-based relationships (structure) on the data. An attractive feature of the structural approach is that estimates can be used to simulate economic outcomes associated with different immigration scenarios. For example, a structural model can project the impact of visa policy proposals, such as to increase high-skilled immigration or to create programs allowing unauthorized immigrants credentials to work. However, the technical difficulties associated with this approach require the use of simplifying assumptions that influence the estimated outcomes. This section reviews in turn the published studies corresponding to the two methodologies.
Aggregate Skill Cell Analyses
Borjas (2003), the first paper using this approach, created skill categories based on four education groups—did not complete high school, completed just high school, attended some college, and completed college—and eight
21 National-level estimates do not eliminate this measurement problem to the extent that markets for human and financial capital are global rather than national, as they are increasingly becoming.
experience levels: 1-5 years, 6-10 years, and so on, up to 36-40 years.22Borjas (2014a) further divided “completed college” into “college graduate” and “post-graduate” based on evidence that workers with advanced degrees are often not competing closely for jobs with those who have just a college degree. The skill cell approach assumes that workers within each cell, whether foreign- or native-born, are perfect substitutes while workers across cells are imperfect substitutes. The wage impact of immigration on male natives is typically estimated by regressing cell-specific outcomes on the immigrant share in the respective education-experience group (skill cell).
Purely correlational (i.e., ordinary least squares, or OLS, regression) estimates based on Decennial Census data in Borjas (2003) and Borjas (2014a)23 revealed a negative correlation, for male workers, between wage growth and the share of immigrants by skill group. This relationship is illustrated in Figure 5-2.
The scatter diagram data suggest that, at the national level, male wages should fall by 3 to 4 percent if immigration increases the number of male workers in a skill group by 10 percent due to immigration (approximately the effect of immigration on labor supply cumulatively from 1980 to 2000) (Borjas, 2003). Most of this effect is driven by observations at the low end of the education spectrum. As summarized later in this chapter, the national skill cell studies find larger negative wage effects on native-born workers from immigration inflows than do other approaches (i.e., spatial and structural studies).
Two papers by different authors expand on the skill cell work of Borjas. Llull (2015) addressed the endogeneity of immigrant density by skill cell and observed that the characteristics of arriving immigrants are not random but determined in part by both the labor demand and wages for a given skill cell in the United States. He developed a new instrument based on a cross-country analysis of the determinants of migration. The number of immigrants of each skill type expected in the United States is predicted based on events abroad: events that are very unlikely to be correlated with the return to education and experience in the United States. His results are striking: using this instrumental variable almost triples the negative effect found by Borjas (2003), yielding the most negative wage effect of any published study (equal to Altonji and Card’s  impact on wages of low-education black men). The panel speculates below as to why this might be.
22 Experience, sometimes termed “potential experience,” was calculated based on the estimated number of years that had elapsed since the individual finished school.
are sensitive to the form of the regression used. They found that changes to the way the statistical relationship is estimated and a change in the way immigration is captured each leads to less-negative estimates of the impact of immigration on wages and renders estimates at different levels of geographic aggregation more similar. The issues raised by the sensitivity of the Borjas results to the Card and Peri robustness tests, particularly as they relate to the measure of immigrant inflow, are potentially relevant for a number of immigration studies using a similar approach.24
It should be noted that estimates produced using the spatial and nonstructural skill cell approaches are not conceptually comparable. Whereas
24 See, for example, Borjas (2003, 2006, 2009), Bonin (2005), Bratsberg et al. (2013), and Steinhardt (2011). Card and Peri (2016) argued that their immigration measure (immigrant induced labor supply changes) is preferred because it is not biased by endogenous native flows; Borjas (2003, Ch. 4, fn. 8) argued that his measure (the fraction of immigrants in the skill group, including labor-market-specific fixed effects) is preferable because of nonlinearities between wages and measures of the immigrant supply shock.
the skill cell approach identifies the average direct effect of increasing the number of workers in the various skill groups on wages of (male) workers in these skill groups, spatial studies often estimate different parameters (depending on the specification), many of which also capture indirect effects induced by complementarities between immigrants and native workers at other parts of the skill distribution. These indirect effects may come about because an increase in workers in one skill group may decrease wages of workers in that group but increase wages of complementary workers across skill groups (e.g., the case where immigrants compete and harm the wages of construction or kitchen workers but enhance the opportunities and wages of first-line supervisors or wait staff). Further, there must be sufficiently low substitution between age-education cells to allow for estimation of the standard skill cell model. And, as with any methodology, data must be sufficient to allow the analyst to correctly allocate immigrants into skill cells defined by high degrees of substitutability within a cell.
A strong assumption in the skill cell approach—discussed in Section 5.2—is that immigrants and natives with the same measured education and the same age (or potential experience) are very close substitutes. Immigrants’ education and labor market experience are often not comparable to that of natives, and immigrants therefore earn less than observationally similar natives, particularly when they first arrive in the host country. This downgrading can be dramatic, as Dustmann et al. (2013) illustrated for the case of the United Kingdom. As a result, immigrants compete most closely with natives in other skill cells than those to which they would be assigned, based on education and experience observables. As an example consider an Iranian surgeon who practiced for 15 years in Iran but upon arrival in the United States speaks little English and is not comfortable with the U.S. operating theatres or technology. This individual’s labor market experience in Iran may hold little value in the United States. As a result, the immigrant may initially work in a lower position, perhaps as a nurse, and then possibly move to a physician’s position as the individual gains English proficiency and acquires experience and the requisite medical licenses. Thus, although arriving with high measured skills, this immigrant competes with individuals in another skill cell than the one to which the immigrant would be assigned, based on observables.
It is possible to build in adjustments to realign the way new arrivals are sorted into skill cells in these models. For example, by using occupation as the indicator of skill, Orrenius and Zavodny (2007) bypass the estimation problem created by skills downgrading in more restrictive models.
Much of the research described above, including the cross-area (spatial) analyses and simple skill cell correlations, impose little structure on the econometric models from which wage and employment impacts are estimated. In contrast, structural models build on theoretical relationships to simulate labor market responses to immigration. In these models, identification (i.e., establishing the differences between a situation with immigration and one without) is achieved by using the model structure, which imposes a relationship between labor supply and wages, the magnitude of which depends on the estimated parameters that characterize the production function (i.e., the relationship between output and inputs of the factors of production). Typically, simple variants are used such as the constant elasticity of substitution (CES) production function,25 to derive these relationships—specifically, the elasticities of substitution between different skill groups—and describe them with a small number of parameters. These estimated parameters may then be used to simulate the impact of changes in labor supply due to immigration on the relative wages of native-born workers.
The implementation of structural models raises a number of issues. For one, there is the need to select a production function; this imposes functional form assumptions that may be restrictive. As noted above, beginning with Borjas (2003), the literature has used a nested CES framework. Decisions must also be made about which cross-group substitution elasticities to estimate, which can have a strong effect on the findings from structural models. The number of such cross-group effects that may be estimated is limited because, as that number grows, the empirical exercise quickly becomes intractable. For example, Borjas (2003) separated the labor force into 32 skill groups defined by education and work experience. In order to estimate all cross-group elasticities, 1,024 (or 32 × 32) effects would have to be estimated. Borjas (2003) instead estimated the extent of substitution across education groups and across experience groups, then calculated the skill-group elasticities from this smaller set of starting estimates. Later researchers—for example, Ottaviano and Peri (2012), discussed below—have modified some of these assumptions.
Structural model simulations may be performed for either short-run or long-run scenarios. As discussed above, short-run analyses measure the wage impact of immigration before there has been sufficient time to adjust capital inputs; that is, in the short run capital is fixed. The long run is a time frame that by definition is sufficiently long to allow firms to adjust the
25 As its name suggests, under this production technology assumption, there is a constant percentage change in factor (e.g., capital and labor) proportions at all output levels. A formal presentation of the CES version of the structural model can be found in Borjas (2014a, pp. 106-112).
amount and type of physical capital (e.g., by purchasing new machines or building new plants) used in response to factor shocks.26 If, for example, there is an immigration-induced decline in the wages of relatively low-skilled production workers, this may lead to an increase in investment in industries using more of this type of labor, potentially cushioning the decrease in their wages (see Section 4.5). The simulations conducted under these two alternative assumptions may be regarded as bounding the wage effects associated with an immigration shock (at least the wage effects estimated using this approach). Borjas (2003), along with a number of other studies, performed simulations of specific labor supply shocks. These studies assume that the entire immigration that occurred over a certain period (such as 1990 through 2010) happened all at once, and then the simulation projects the impact of this supply shock in the short run and in the long run. Borjas (2003), in particular, emphasized the short run and assumed the stock of physical capital is fixed. One rationale for adopting this assumption is the lack of evidence with respect to how long it takes capital to adjust in different situations. Ottaviano and Peri (2012), on the other hand, emphasize the long run.
An important point is that in the empirical literature, temporal distinctions between the short and long run do not necessarily map precisely with the theoretical concepts. In the real economy, there is variation in how long it takes capital to adjust (the defining characteristic of the long run). Indeed, if capital adjusts quickly, the long run could be quite short in calendar time; if it adjusts slowly it might be quite protracted in calendar time. In terms of the structural models, what is really meant by “the short run” is that capital is perfectly inelastic in supply while “in the long run” capital is perfectly elastic in supply.27
Another important point is that while structural models can estimate changes in relative wages across groups in the short or long run, the assumptions necessary to estimate the model require that the average wage cannot be affected by immigration in the long run. The production function specification dictates that a 10 percent immigration-induced increase in supply have a 0.0 percent impact on average wages in the long run and must lower the average wage by 3.0 percent in the short run (Borjas, 2014a, p. 109).28 This technical assumption cascades to all other estimates of the
27 The panel also notes these are static models whereas a full modeling of the long-run/short-run distinction would specify a dynamic model.
28 See Section 4.2 (or Borjas, 2014a) for a formal explanation of the underlying production function theory behind these numbers. Again, the intuition is that, in the short run and with other inputs to production fixed, additional workers will compete for a limited number of
wage impact of immigration using this framework. As a result, since the average wage cannot change in the long run, adjustments to immigration occur only in relative wages: The groups that received disproportionately large numbers of immigrants may experience a long-term relative decline in their wage, while the wage of the groups that received very few immigrants may see a relative increase in the long run. It is important to keep these mathematical restrictions in mind when interpreting any wage impact estimated from the structural approach.
As with any theoretical approach, the simplifying assumptions entailed in the aggregate production function approach come at a cost (Blau and Kahn, 2015, p. 812):
. . . Specifically, one must decide how to disaggregate labor into skill groups and also what types of substitution/complementarity relationships to allow. As examples of the latter, recall Lewis’s (2011b) model allowing skilled and unskilled labor to have asymmetric relationships with capital or Ottaviano and Peri’s (2012) models allowing differing substitution relationships between different pairs of education groups. Moreover, researchers must also decide whether to allow immigrants and natives within a skill group to be imperfect substitutes, and if so, whether the immigrant/native substitution parameter should be the same for all skill groups (Lewis, 2011a).
The relative wage and employment impacts predicted by these models hinge crucially on estimates of the elasticities of substitution between native-born and foreign-born workers overall, and the separate elasticities between education and experience groups or between skill groups. The less interchangeable different kinds of workers are, the less they compete and the less downward pressure inflow of one group can exert on wages of another.
An important early paper using the aggregate production function approach in this area was Borjas et al. (1997). These authors compared the actual supplies of workers in particular skill groups to what they would have been in in the absence of immigration and then used results from previous studies on the elasticity of substitution among skill groups to compute the impact of the immigrant supply shock on the relative wages of skill groups. The study, which focused on the 1980-1995 period, examined two
jobs, which exerts downward pressure on wages. In the long run, once firms have had time to adjust capital stocks, the demand for labor increases along with the size of the economy and wages will be pushed back upward toward initial levels. The elasticities of substitution between immigrant workers and different types of established workers in the labor market dictate which workers’ pay will change by more than −3 percent and which workers’ pay will change by less than −3 percent.
relative wage comparisons: (1) the wages of high school dropouts relative to those with at least a high school degree and (2) college graduates relative to high school graduates (where all workers were aggregated into “high school equivalents” and “college equivalents”). The authors found that immigration accounted for a 3-6 percent decline in the wages of high school dropouts relative to high school graduates between 1980 and 1995—in the range of 27-55 percent of the total decline for that group over the period. In contrast, they found that immigration did not explain much of the increase in the college wage premium (i.e., the college versus high school equivalent comparison). These findings reflect the fact that, for these larger educational group aggregates, immigration did not substantially affect relative supplies of workers in each skill category.
Although the results from Borjas et al. (1997) are intriguing, there were limitations to the study. The underlying production relationships (parameters) were obtained from outside sources and the relative wage effects of immigrant supply shifts were mechanically predicted from these elasticities of substitution. Furthermore, each specification (the wage group comparisons in (1) and (2) above) distinguished (compared) just two types of labor.
These and other issues were addressed by Borjas (2003), who focused on the impact of immigration on relative wages in the United States over the 1980-2000 period using a nested CES production function approach. Borjas assumed—similar to Card and Lemieux (2001)—that workers within the same education category but who differ in their labor market experience are not perfect substitutes in production. As in the analysis by Borjas et al. (1997) of the shorter post-1980 period, Borjas (2003) found substantial negative wage effects of immigration with capital held fixed, particularly on the low skilled. He estimated that the immigrant inflow from 1980 to 2000, equal to an increase in the labor supply of working men of about 11 percent, lowered the wages of male native high school dropouts by 8.9 percent and those of male college graduates by 4.8 percent.
As noted earlier, Borjas disaggregated skill groups by work experience (proxied by age) as well as education levels, forming 32 education-experience cells. His addition of the experience dimension built on the insight from human capital theory that workers enhance their skills not only through investments in formal schooling (i.e., education) but also by accruing skills through labor market experience. He thus assumes that not only are workers with different education levels imperfect substitutes but workers with the same education but different experience levels are imperfect substitutes. In the real world, immigrant inflows vary across education-experience cells, and the extent of that variation changes over time. This variation helps allow the impact of immigration on the labor market to be identified. Borjas assumed that, within education-experience cells, immigrants and natives are perfect substitutes. In contrast to the study by Borjas
et al. (1997), which used outside information to obtain the parameters of the production function, Borjas (2003) directly estimated parameters of the production function and then simulated the wage impacts based on the estimated elasticities.
Even the highly disaggregated approach proposed by Borjas (2003) involves some simplifying assumptions. Recent work suggests that results using the structural approach are sensitive to these assumptions. We illustrate this point with findings from Ottaviano and Peri (2012), a study of the relative wage effects of immigration over the 1990-2006 period based on Census Bureau data (from the Decennial Census and the American Community Survey [ACS]), which used the same broad framework as Borjas (2003) but changed some of the assumptions. A key distinction is that Ottaviano and Peri make different assumptions than Borjas about the supply of capital. Whereas Borjas (2003) assumed that capital supply is inelastic (does not have time to react to growing labor supply), Ottaviano and Peri assumed that it is perfectly elastic.
In addition, Ottaviano and Peri made two important changes in how substitution across groups is specified.29 First, in contrast to Borjas (2003), Ottaviano and Peri allowed immigrants and natives to be imperfect substitutes. We have already discussed how, given language differences and other factors, it might be reasonable to assume that immigrants and natives are imperfect substitutes. Further, they split the sample in order to allow the substitutability between immigrants and natives for the less educated (high school dropouts and high school graduates) to differ from that for the more highly educated (those with some college and college graduates). The intuition underlying this assumption is that language and other barriers are less prevalent among highly educated foreign-born workers than among less educated foreign-born workers, allowing highly educated foreign-born workers to be closer substitutes for their native-born counterparts.30 Their estimated elasticities are consistent with imperfect substitutability that differs in magnitude by education category: They obtain a native-immigrant elasticity of substitution of 11.1 for the less educated and 33 for the more highly educated (indicating that workers in the latter category are more interchangeable). Allowing for imperfect substitution between immigrants and natives is potentially important because the less closely immigrants
29Manacorda et al. (2012), writing in parallel with Ottaviano and Peri (2012) on the United Kingdom, also developed the same approach based on the idea of immigrants and natives being imperfect substitutes within age-education cells.
30 The results from Peri and Sparber (2009) offer some support for the imperfect substitutability idea; they found that low-skilled foreign-born workers are employed disproportionately—highly so in some cases—in occupations such as construction, kitchen work, etc., that demand more physical effort and less communication skill.
substitute for natives, the smaller the effect immigrants will have on the wages of natives with the same observable skills.
Second, while Borjas (2003) imposed the same elasticity value for all adjacent education groups, Ottaviano and Peri (2012) specified the elasticity of substitution between education groups as being different (independent of native/immigrant status). They posited that, in the current economy, high school dropouts can fill many of the same kinds of jobs as workers with just a high school diploma; in other words, they hypothesized that high school graduates and dropouts often compete in the same labor market. This would be consistent with Card (2009), who found high school graduates and high school dropouts to be virtually perfect substitutes. (Recall that the economists’ designation of perfect substitutes means that, for instance, high school dropouts and high school graduates can be traded at a constant rate, but that rate does not have to be one-to-one.) At other skill levels—for example between those in the labor force with some college and those with a graduate degree—the degree of substitution may be lower. Consistent with this reasoning and with Card (2009), Ottaviano and Peri found that the elasticity of substitution between high school dropouts and high school graduates is at least 10 and is infinite in some estimates, while the elasticity of substitution at higher skill levels is much lower. Since most immigrants to the United States are low skilled, the wage impact of an increase in immigrant supply will be lower if high school dropouts and high school graduates are combined, since the immigrant supply shock will constitute a smaller percentage of the same skill-group labor force in the larger aggregate.
In contrast to Borjas et al. (1997) and Borjas (2003), Ottaviano and Peri (2012) found that immigration had only a very small effect on native wages within skill groups. Using the more detailed set of parameters reflecting imperfect substitutability between natives and immigrants within an education-experience cell, they found that the effect of immigration over the 1990-2006 period was to reduce the wages of native-born high school dropouts in the range of 0.6-1.7 percent. Averaged across all skill categories, the study found that U.S.-born workers experienced a slight increase in wages as a result of immigration.
The Ottaviano and Peri (2012) specification is not without controversy. Borjas et al. (2012) presented evidence that the estimates of their two key substitution elasticities—that between immigrants and natives and that between high school dropouts and high school graduates—are sensitive to the type of data used and to what regressors are included in the underlying production function models.31 As Blau and Kahn (2015, p. 821) noted,
31Dustmann and Preston (2012) presented evidence that downgrading of immigrants may lead to finite estimates of the elasticity of substitution between immigrants and natives even if the true elasticity of substitution in infinite.
“The varying results in the estimates of the substitution elasticities illustrate a potential drawback of this type of approach to estimating the impact of immigration.”
The contrasting findings between the Borjas (2003) and Ottaviano and Peri (2012) studies suggest that results from structural models are influenced by crucial assumptions, some of which involve unobserved and untestable issues. However, these two studies also differ along a number of dimensions, ranging from the time period studied to whether the results are obtained under the assumption of capital being inelastic or perfectly elastic, that make them difficult to compare. To abstract from the impact of extraneous factors and to focus on the importance of substantive decisions, the panel extends an analysis presented in Borjas (2014b). Table 5-1 summarizes wage simulations associated with alternative specifications (“scenarios”) for a consistent time period, 1990-2010, treating all immigration between 1990 and 2010 as if it constituted a single supply shock.32,33 The table includes the following scenarios for both the short run and the long run (“GB” refers to Borjas, 2003; “OP” refers to Ottaviano and Peri, 2012; variables are defined and discussed below):
- Scenario 1: Immigrants and natives in a skill group are perfect substitutes (σMN = ∞), and high school dropouts and high school graduates are different groups—similar to GB.
- Scenario 2: Immigrants and natives in a skill group are imperfect substitutes (σMN = 20.0, as in OP), and high school dropouts and high school graduates are different groups (as in GB).
- Scenario 3: Immigrants and natives in a skill group are perfect substitutes (σMN = ∞, as in GB), and high school dropouts and high school graduates are perfect substitutes (σHS = ∞, as in OP).
- Scenario 4: Immigrants and natives in a skill group are imperfect substitutes (σMN = 20.0), and high school dropouts and high school graduates are perfect substitutes (σHS = ∞)—similar to OP.
In Table 5-1, the term σMN is the elasticity of substitution between immigrants and natives with the same measured skills. This term equals infinity if the two groups are perfect substitutes (the assumption in Borjas ) or equals 20 for the “preferred” estimate in Ottaviano and Peri (2012). The term σHS is the elasticity of substitution between high school
32 For an analysis spanning 20 years, one might reasonably argue that—to the extent immigration is less a “shock” than a somewhat predictable flow—investment patterns reflect some level of anticipation of the expansion of the workforce and population generally.
33 In contrast to the macro literature, in all these scenarios the elasticity of substitution between labor and each of the different types of capital is assumed to be identical, precluding the capital skill complementarities discussed in Chapter 4.
|High School Dropouts||High School Graduates||Some College||College Graduates||Post-College||All Education Groups|
|Percentage Supply Shift||25.9||8.4||6.1||10.9||15.0||10.6|
|A. Short Run|
|Scenario 1*: σMN = ∞|
|Scenario 2: σMN = 20.0|
|Scenario 3*: σMN = ∞ and σHS = ∞|
|Scenario 4: σMN = 20.0 and σHS = ∞|
|B. Long Run|
|Scenario 1:* σMN = ∞|
|Scenario 2: σMN = 20.0|
|Scenario 3:* σMN = ∞ and σHS = ∞|
|Scenario 4: σMN = 20.0 and σHS = ∞|
*Because, in this scenario, native-born and foreign-born workers are perfect substitutes, it is unnecessary to differentiate between the two; hence, only one row for “all workers” is shown.
SOURCE: Borjas (2014a, Tables 5.2, 5.4, 5.6). The simulations use the nested constant elasticity of substitution (CES) framework, set σE = 5.0, and assume that the aggregate production function is Cobb-Douglas, implying σKL = 1.0.
dropouts and high school graduates. It is equal to infinity if the two groups are perfect substitutes.
The above scenarios summarize how the key differences in the structural studies literature can be linked back to the studies’ modeling assumptions. Allowing capital to adjust (i.e., moving from a short-run to a long-run scenario) reduces the estimated negative effects across the board—that is, for all workers as well as for relative wage effects within each education group. As the elasticity of substitution between native-born and foreign-born is changed from the two groups being perfect substitutes (σMN = ∞) to imperfect substitutes (σMN = 20.0), the impact on the wages of “all workers” (natives and immigrants) for any given skill group is unchanged, but within each skill group, imperfect substitutability is associated with a larger negative wage impact on earlier foreign-born workers (prior immigrants) and a smaller negative wage impact on native-born workers. This makes sense for the following reason: In cases where foreign-born and native-born are close substitutes, one would expect an immigration shock to have a more equal impact on the two groups; imperfect substitutability between the two groups insulates natives from negative effects to some degree. Comparing otherwise similar scenarios in which high school dropouts and high school graduates are imperfect substitutes (Scenarios 1 and 2) versus scenarios in which they are perfect substitutes (Scenarios 3 and 4), one can see that the impact of allowing high school dropouts and high school graduates to be perfect substitutes has the effect of reducing the negative wage impact for high school dropouts. This makes sense, as any negative impact from the inflow of unskilled workers is now diluted across a larger portion of the labor supply (high school dropouts plus high school graduates). A portion of the negative wage impact is averaged into the value for high school graduates, which becomes slightly more negative. The simulations also show that allowing for imperfect substitution between immigrants and natives does not greatly attenuate the wage impact of immigration on high school dropouts. There is still a 2 to 5 percent wage loss, depending on whether one looks at the long run or short run. The scenario that does lead to a much lower negative or even positive impact of immigration on the lowest skilled workers is the one that also incorporates the possibility that high school dropouts and high school graduates are perfect substitutes.
When comparing simulated effects across education groups within a scenario, it is useful to remember that all structural simulated effects reflect a combination of the estimated parameters relating relative wages and relative labor supply across skill groups and the simulated amount of immigration-induced labor supply by skill group. Unlike in spatial and skill cell studies, the impacts cannot be separated into the amount due to the responsiveness of the skill group to changes in labor supply and the magnitude of the group’s simulated labor supply change. However, the pattern
across columns in Table 5-1 does mirror qualitatively the magnitudes of the labor supply changes by education over 1990-2010, the values used in the simulation. Negative effects for natives tend to be larger for high school dropouts, the group with the largest immigration inflow, followed by those with post-college education, a group that also experienced relatively large inflows. Native high school graduates and those with some college tend to experience smaller negative effects and, indeed, under most scenarios, slightly positive effects in the long run, consistent with relatively small immigrant inflows over the period. The impacts on college-educated natives are very similar to the mean effects across education.
Key takeaway points from this simulation are that the assumptions about capital—fixed short run versus adjusted long run—and substitutability among skill groups have large effects on estimates of wage impacts. Wage effects (overall and within skill groups) are more negative in the short run than in the long run, when they are sometimes even positive. And, for both the short-run and long-run scenarios, the largest negative effects on native less-skilled workers are for the scenarios in which immigrants and natives are perfect substitutes and high school dropouts and graduates are imperfect substitutes (Scenario 1). The smallest negative effects on native less-skilled workers are for the scenarios in which immigrants and natives are imperfect substitutes and high school dropouts and graduates are perfect substitutes (Scenario 4). Indeed, under this scenario, all native groups except the postcollege-education group benefit from immigration in the long run.34
As is apparent from the literature review above, the results of a given study of the impact of immigration on wages or employment are typically directly comparable to only a handful of others. Sometimes two studies are not directly comparable because the underlying methodology is fundamentally different. For example, skill cell studies estimate the effect of immigrants on the most similar natives, omitting the effect of immigrants on less similar natives that is captured in most spatial studies, while structural studies build in the assumption that average wages are unchanged by immigration in the long run and hence are essentially studies of relative wages. But often, even within a methodology, studies are not immediately
34 Recall, also, the all-important point that the absolute wage impact numbers are dictated by production function assumptions; only the relative wage impacts across skill groups are driven by the data. And, here too, the magnitude of the relative wage impacts is tied to the relative size of the immigrant inflows by the assumptions of the model.
comparable because of differences in the way the number of immigrants is captured. For example, the study may focus on immigrants as a share of the labor force or the share of the labor force that is of a particular skill (instrumented by the predicted immigrant inflows of that skill type). For this reason, in Table 5-2, the panel presents in terms of a common metric the results of several prominent spatial and skill cell papers discussed in this chapter, along with the largest and smallest structural impacts for all natives and for native high school dropouts, based on the results in Table 5-1. For each study, the table shows calculations of the wage effect on the indicated group of natives of an increase in immigrants that raises labor supply of the state, occupation, skill cell, or education group by 1 percent. Wage effects in bold are the coefficients reported in the source study; other coefficients were calculated by the panel as outlined in the Technical Notes in Section 5.9.
For most spatial and skill cell studies, the calculations to convert their results into the common metric are straightforward (see Technical Notes in Section 5.9), though they do involve using a particular value of the share of immigrants in the labor force. To make the studies as comparable as possible, the same value of the share of immigrants in the labor force should be used in all the calculations, even though a given study’s average share will depend on the exact years of data used. To calculate the underlined values in Table 5-2, the panel set the immigrant share of the labor force at its 2000 value of 12.6 percent for those studies requiring harmonization. However, the harmonization approach for spatial and skill cell papers does not lend itself as readily to the structural studies, which involve several parameters rather than a single parameter. Nevertheless, it is useful to make a more crude adjustment to see whether the structural results are broadly in line with those of other studies, setting aside the issue of relative versus absolute wage changes. For the structural studies, the simulations reported in Table 5-1 above may be thought of as the result of a simple increase in the share of immigrants in the labor force from 1990 to 2010, rather than the result of more complex changes in different types of labor over the period. The wage effects reported in the simulations may then be divided by this increase in immigrant share to get the effect of a percentage point increase in immigrant share, a figure that may then be converted to the effect of a 1 percent increase in the labor supply, as was done for the spatial and skill cell studies. A similar exercise may be performed for Borjas (2016b) and Peri and Yasenov (2015). (See the last two subsections of Section 5.9 for Technical Notes on the panel’s calculations for Borjas (2015), Peri and Yasenov (2015), and the structural studies.)
Table 5-2 confirms that there is a wide range of estimated elasticities and makes clearer than do unharmonized results which estimated elasticities are most negative and what patterns exist in the size of elasticities. Consider first the results for all natives and native dropouts (i.e., excluding
results for minorities). There is considerable variation in the findings, with results ranging from a set clustered around zero (including small positive values) to a set in the −0.8 to −1.0 range within each of the three approaches (with the exception of three studies noted below). Results close to zero are obtained in the spatial studies of Card (2001) for native men and women and of Cortés (2008) for native dropouts, in Card and Peri’s (2016) skill cell regressions for all native men, and also in the long-run structural models for all natives (whose results are close to zero by assumption). Results in the −0.8 to −1.0 range are those of the Altonji and Card (1991) spatial study and the structural short-run calculation for dropouts (Scenario 1, in which capital is fixed and immigrants and natives are perfectly substitutable). Two much more negative estimates (again excluding the elasticities for minorities) are Borjas’s (2016b) upper bound for native non-Hispanic men (−1.4) and Llull’s (2015) skill cell analysis for all native men (−1.7). On the other hand, the considerably more positive estimate of 0.3 is from Peri and Yasenov’s (2015) study of the same Mariel Boatlift immigration episode studied by Borjas (2015).
Some notable patterns emerge. Confirming expectations based on economic theory about which groups are most negatively affected by immigration, native dropouts tend to be more negatively affected than better-educated natives (as indicated by comparing results for dropouts with the overall results for all workers or all men or women). The results in the table also suggest that this negative effect may be compounded for native minorities. Altonji and Card (1991) found more negative results for low-education blacks than low-education whites: The coefficient for black males reported in the table is the most negative effect they reported. Cortés examined a number of groups and found the largest negative effects for Hispanic dropouts with poor English, as well as larger negative effects for Hispanic dropouts than for all dropouts. This could be because native dropout minorities are the closest native substitutes for immigrants. As the results in panel C, Structural Studies, of Table 5-2 show, the closer substitutes immigrants and natives are assumed to be (the higher σMN), the more negative the effect of immigration on natives. While not reported in Table 5-2, structural estimates that distinguish between the effects on (prior) immigrants and on natives found larger negative effects on immigrants (Table 5-1), and the relatively large negative effects found by Monras (2015) are for dropouts, including non-Hispanic immigrants.35
Although theory predicts larger negative effects on native wages of immigrant inflows in the short run than in the long run, this pattern
35 The Borjas (2016b) study’s large negative effects are for male non-Hispanic dropouts, including non-Hispanic immigrants (although the latter are likely few in number and perhaps no more similar to immigrants than they are to natives).
|Study||Wage Effect (%)||Which Natives|
|A. Spatial Studies|
|Altonji and Card (1991)||−1.7||Dropouts, black men|
|Borjas (2016b)||−1.4||Dropouts, non-Hispanic men|
|−0.5||Dropouts, non-Hispanic men|
|Monras (2015)||−0.7||High school graduates or less, non-Hispanic, including immigrants|
|Cortés (2008)||−0.6||Dropouts, Hispanic with poor English|
|Peri and Yasenov (2015)||0.3||Dropouts, non-Cuban|
|B. Skill Cell Studies|
|Card and Peri (2016)||−0.2||Men|
|Card and Peri (2016)||−0.1||Men|
|C. Structural Studies|
NOTES: Panel C, “Structural studies,” refers to the results in Table 5-1: the maximum and minimum values for the effect on all natives (except the long-run minimum value for Scenarios 1 and 3, which is zero by assumption) and on native dropouts are reported. “Dropouts” refers to high school dropouts; HS to high school or less. “10-year differences” refers to analysis relating the 10-year change in wage to the 10-year change in immigration; “fixed effects (10-yearly data)” indicates that levels rather than changes were used. All nonstructural coefficients are from instrumental variables estimates except Borjas (2003) and Card and Peri (2016), where they are the ordinary least squares (OLS) coefficients from the nonstructural estimation, and Borjas (2016b).
Altonji and Card’s (1991) black native dropouts had less than 13 years of education, while the dropouts of all races had less than 12 years. The Cortés (2008) sample is of dropouts in immigrant–intensive nontraded sectors. Monras’s (2015) natives included earlier non–Hispanic
|Which Immigrants||Short Run?||Note|
|Dropouts||Yes||Upper bound, Mariel boatlift|
|Dropouts||Yes||Lower bound, Mariel boatlift|
|HS or less, Mexican||Yes||1-year difference|
|Dropouts||—||Fixed effects (10-yearly data)|
|Dropouts||—||Fixed effects (10-yearly data)|
|Dropouts||—||Fixed effects (10-yearly data)|
|All||—||5-year difference, wage level|
|All||—||5-year difference, wage level|
|All||—||IV, ﬁxed effects (10-yearly data)|
|All||—||OLS, ﬁxed effects (10-yearly data)|
|All||—||OLS, 10-year differences|
|All||—||OLS, 10-year differences|
|All||Yes||Scenario 1: σMN = ∞|
|All||Yes||Scenarios 1 and 3: σMN = ∞|
|All||—||Scenario 1: σMN = ∞|
|All||Yes||Scenario 4: σMN = 20|
|All||Yes||Scenarios 2 and 4: σMN = 20|
|All||—||Scenarios 2 and 4: σMN = 20|
|All||—||Scenario 4: σMN = 20|
immigrants. Natives and immigrants cannot be distinguished in Borjas’s (2016b) data. The elasticity of substitution between immigrants and natives is σMN; when it is infinite, the two groups are perfect substitutes.
Bolded figures are coefficients reported directly from the cited study; underlined figures are the result of the panel’s calculation using the paper’s coefficient and an immigrant density of p = 0.126, the national value for the 2000 labor force. See Section 5.9 for technical notes on these calculations and those for the structural cases and a number of other papers that do not involve p = 0.126 and are implicitly evaluated at a different p (though a very similar one in the case of the structural papers).
For column 5, the “short-run” designation indicates effects found over a less than 5-year span, or structural calculations with capital held fixed. The length of time required for capital, technology, and other factors to respond to unexpected or expected immigration inflows, and hence the distinction between short and long run, cannot be rigorously determined.
does not come through unambiguously in the table. The pattern is pronounced for the structural studies (see also Table 5-1), where the short run is imposed in accordance with theory by fixing the capital stock at its initial value. But the pattern is less clear in nonstructural studies. Monras’s (2015) study of the 1-year effect of unanticipated inflows is clearly capturing the short run and does find a relatively large negative effect (−0.7), while Borjas’s (2016b) negative elasticities based on the first 7 post-arrival years after an unanticipated immigrant inflow (−0.6 to −1.4) are also likely to be capturing a short-run effect. However, Peri and Yasenov (2015) estimated the most positive elasticity of any study (0.3), based on the first 3 years after the Mariel Boatlift examined by Borjas (2016b). This elasticity is statistically insignificant, and the authors characterize their paper as finding no negative effect rather than finding a positive effect, but their result would rule out, statistically, effects as negative as the lower bound finding of −0.6 in Borjas (2015).
Studies examining the relation between 10-year changes in immigration and 10-year changes in wages (“10-year differences”)—Altonji and Card (1991) and Card and Peri (2016)—capture exactly 10 years of adjustment and hence probably also capture considerable capital adjustment. The same is true for studies using data spaced 10 years apart but not differenced (“fixed effects”), such as Llull (2015), Borjas (2003), and Cortés (2008), which capture adjustment over at least 10 years.36Card’s 2001 paper examining the effect of flows over 5 years is more difficult to categorize in terms of capital adjustment. The contrasting results of studies examining the same frequency of effects suggest the importance of other factors in determining the elasticity estimated, including the methodological approach.
There appear to be some differences in elasticities by approach not accounted for by the share of studies in each looking at the long versus short term, at dropout natives versus all natives, and minority natives versus all natives. On balance, the skill cell studies find the most negative wage impacts and the structural studies the least negative, with the spatial studies in the middle; differences between approaches are of about the same order of magnitude as the variation among studies using the same approach. Below, the panel revisits some of the methodological differences discussed in Section 5.3 to see if this ranking is expected, with particular attention to the medium- to long-run time frame probably captured by most nonstructural studies.
Spatial studies can be biased either to find a positive effect (if instrumental variables do not correct adequately for immigrant location choice) or to find zero effect (if trade in goods and flows of capital and labor
distribute the effect nationally), but they will also incorporate changes in technology, technique, or sector that genuinely mitigate negative wage impacts, and they will include cross-effects of immigrants on less similar natives. The skill cell studies avoid the biases of the spatial studies, which makes them more likely to find negative effects, as they do, but they do not include cross-effects, whose overall direction (negative or positive) is unknown. Thus, it is not certain that the larger effects from skill cell studies compared to spatial studies are to be expected. All skill cell studies to date have examined the impact of immigration on men only, unlike studies using other approaches. However, other studies do not paint a clear picture of whether women or men are more vulnerable to immigration impacts, leaving it unclear as to whether the gender focus of skill cell studies explains why their estimated wage effects on natives are more negative.
The structural studies preclude any effect on overall wages in the long run, due to the choice of a production function that is assumed to remain constant over time. This rules out any overall shift up in wages due to increasing returns to scale or any immigration-induced skill-neutral technological progress, but it also precludes any overall shift down in wages due to decreasing returns to scale. Moreover, it rules out any downward pressure on dropout wages if the induced technological progress complements high-skilled workers and substitutes for low-skilled workers (though given large inflows of low-skilled immigrants, this would probably not be expected). It is therefore not easy to trace the ranking of the impact by approach back to the methodological characteristics of each.
It is useful to discuss possible reasons for the variations in estimated elasticities within each of the approaches (i.e., spatial, skill cell, and structural studies). The reasons for the variation within the structural approach are transparent: Short-run effects are larger, and effects with natives and immigrants assumed to be perfect substitutes are larger than those where they are not.37 Further, as discussed above with respect to Table 5-1, assumptions about substitutability across education groups, particularly whether or not high school graduates and high school dropouts are perfect substitutes, also influence the results, with the assumption of perfect substitutability resulting in smaller estimated negative effects. Thus, as suggested by our discussion of the simulation results in Table 5-1, the value of the wage elasticities from the structural estimates in the bottom panel of Table 5-2 depends on the particular scenario being considered. One general conclusion is that the value of the wage elasticity is not as greatly affected when one only allows for imperfect substitution between natives and immigrants with the same level of education. The value of the wage
elasticity for high school dropouts, however, becomes much less negative or even positive when one adds the assumption that high school dropouts and high school graduates are perfect substitutes.
The differences among studies within the spatial approach seem fairly consistent with differences in the immigrants and natives studied and whether the impact estimated is short or long run. The results of the Altonji and Card (1991) study do, however, appear more negative than expected on this basis. They may be affected by the use of an earlier and less sophisticated historical settlement-pattern instrument than was used in later studies. Additionally, some spatial studies (and some skill cell studies) investigate time periods that are long enough to capture long-run adjustments in capital and technology and in natives’ human capital accumulation and to reflect increasing aggregate demand as a result of immigration. Spatial and reduced-form skill cell studies potentially capture some adjustments that the structural analyses rule out and, if the instrumental variable is ineffective, some that are unintended, such as equalization of wages spatially as a result of trade or of capital and labor mobility.
The variation within the skill cell studies may reflect both economic and econometric issues. Llull (2015) may have found a very large negative effect because the novel instrument used picks out the impact of immigrants fleeing turmoil, an immigrant category that may possibly have a more negative impact on native wages (see discussion below). The variation among the other three OLS studies appears to reflect econometric issues. Card and Peri (2016) indicated that the original Borjas (2003) skill cell elasticity of −0.6 is sensitive to changes in the form of regression used.38 The OLS results should be the same whether the fixed effects or 10-year differences method is employed. Card and Peri showed that the results are considerably less negative for differences (−0.2), suggesting a problem of omitted variables or possibly that the regressions are capturing different frequency (short versus long run) effects.39 Furthermore, when Card and Peri (2016) changed the immigrant variable from the (change in the) immigrant share of the labor force to the change in the number of immigrants divided by the initial labor force, the elasticity becomes close to zero (−0.1). They argued that the latter immigrant variable is superior, as it is unaffected by changes in the native labor force that might be driven by the same factors as immigration.40 It is unclear to what degree the Llull instrumental variables elasticity is robust to these changes.
38Card and Peri (2016) tested the robustness of a slightly different Borjas (2003) specification from that in Table 5-2, so their results should be compared to a harmonized elasticity from Borjas (2003) of −0.5.
Llull’s (2015) skill cell study has the most negative elasticity of any study (−1.7), which raises a possibility not considered thus far: that the impact of immigrants may vary according to the reason for their migration to the United States. His addition of instrumental variables estimation triples the size of the OLS effect found by Borjas (2003), and his choice of instruments may help inform why instrumental variables raise the magnitude so much. All other instrumental variables studies in the table use an instrument based on historical settlement patterns, which means that the estimated effect is that of immigrants who chose their U.S. location to be close to their co-ethnic predecessors (this is called the Local Average Treatment Effect) and who may therefore be disproportionately composed of immigrants who have been encouraged to come to the United States by family ties. Llull used forced migration as an instrument for the share of immigrants in a skill cell. Economic or political turmoil or natural disasters in the origin country provide random variation in immigration that is not related to better employment opportunities in the destination. His estimates therefore reflect the impacts of immigrants fleeing acute problems and for whom family ties may be less important. This raises the possibility that such immigrants have a more negative effect on natives than do immigrants encouraged by family ties, particularly if their arrival is less likely to be anticipated. An alternative interpretation is that the traditional spatial studies’ instrument based on where earlier settlers settled is simply invalid because those earlier settlers settled in high-wage cities.
An important point is that, while Table 5-2 suggests which native wages are more susceptible to a given immigration inflow, what the table does not show is that native groups differ in the magnitude of immigrant inflows they face. For example, since native dropouts experience a larger immigrant-driven labor supply increase than do natives overall (see Chapter 4), their greater susceptibility to immigration is compounded by higher inflows.
The results of these comparative exercises remain consistent with The New Americans (National Research Council, 1997) in suggesting that, particularly when measured over a period of 10 years or more, the impact of immigration on the overall native wage may be small and close to zero. However, estimates for subgroups span a wider range and suggest some revisions in understanding of the wage impact of immigration since the 1990s. At that time, the authoring panel’s conclusion that “immigration has had a relatively small adverse impact on the wage and employment opportunities of competing native groups” seemed to summarize well what the academic studies indicated. However, the intensive research on this topic over the past two decades, summarized in Table 5-2, displays a much wider variation in the estimates of the wage impact on natives who are most likely to compete with immigrants, with some studies suggesting sizable negative wage effects on native high school dropouts. In addition, the
recent literature is in agreement with The New Americans in finding larger negative effects for disadvantaged groups and for prior immigrants than for natives overall, when those effects are examined separately. (Results for prior immigrants are not shown in Table 5-2 but were reviewed earlier in Section 5.3.) Thus, the evidence suggests that groups comparable to the immigrants in terms of their skill may experience a wage reduction as a result of immigration-induced increases in labor supply, although there are still a number of studies that suggest small to zero effects.
Much of the research on the impact of immigrants focuses on the inflow of immigrants with low education and skills. Immigration patterns for the United States drive some of this emphasis because new arrivals are disproportionately represented in lower educational attainment segments of the population. As of 2011, ACS data show that nearly one-third of foreign-born individuals in the United States do not have a high school diploma and about 23 percent have a high school diploma and nothing beyond (Orrenius and Zavodny, 2014).41 It is therefore often presumed that the majority of immigrants will enter low-skilled labor markets, and this is where fear has been expressed that natives’ job opportunities will be lost. However, as discussed in Chapters 3 and 6, another sizable concentration of immigrants is in high-skilled educational categories, and in recent decades immigrants have become overrepresented in certain occupations—for example, computer software developers, medical scientists, registered nurses, teachers, accountants, computer systems analysts, and physicians—requiring high education and skill levels.42 ACS data also show that, as of 2011, 27 percent of the foreign-born have a college degree or higher, compared with just over 28 percent for natives, and 29 percent of workers in the U.S. economy with doctoral degrees are foreign-born. Reflecting these trends, researchers considering the overall impact of immigration on wages and employment have become increasingly interested in what is happening at the high end of the skills spectrum. Consideration of the impact on natives of high-skilled immigration raises some similar questions to those considered earlier for less-skilled immigrants. But in addition, new questions arise in the context of high-skilled immigration: High-skilled immigrants may innovate, or help natives innovate, and more generally may have positive spillovers on native productivity.
42 See Orrenius and Zavodny (2014). The fact that a large share of immigrants is highly skilled is not new. Immigrants have always had a bimodal distribution by education. That the high end is overrepresented relative to natives is, however, a new development.
Technological progress is a key driver of productivity growth and ultimately of economic growth (Griliches, 1992). If immigrants innovate and advance technology, they therefore increase the growth rate of native income in addition to raising its level. Jones (2002) estimated that 50 percent of U.S. total factor productivity (TFP)43 growth in recent decades is attributable to scientists and engineers. One way high-skilled immigrants could increase technological innovation is through a greater concentration than natives in science and engineering occupations. Immigrants are likely to be overrepresented in such occupations, since scientific and engineering knowledge transfers easily across countries; it does not rely on institutional or cultural knowledge, is not associated with occupations with strict licensing requirements like the practice of medicine, and does not require the sophisticated language skills of a field such as law (see Chiswick and Taengnoi, 2007; Peri and Sparber, 2008). High-skilled immigrants could also increase innovation if a combination of immigration policies and immigrant self-selection leads them to be more educated or of higher inventive ability. Even immigrants who do not innovate themselves may increase innovation by providing complementary skills to inventors, such as entrepreneurship. On the other hand, because natives are likely to respond to the arrival of immigrant innovators, any immigrant contribution to innovation is unlikely to be simply additive. Potential native innovators could be deterred by the additional competition or could be attracted by the possibility of collaboration.
These considerations make studies of immigrant innovation and entrepreneurship, and of skilled immigration more generally, of great interest and importance. In this section, after providing background on the visa pathways available to skilled immigrants, the panel examines the effect of high-skilled immigration on native wages and employment. We then review the effect of immigration on innovation followed by the effect of immigration on entrepreneurship. While research in this area is quite recent, there is very little to suggest that wages are driven down or that native workers are displaced in high-skilled occupations; the evidence is stronger, though still inconclusive, that the direction of any impacts is at least modestly positive. The innovation literature as a whole indicates that immigrants are more innovative than natives and increase innovation per capita, thus likely boosting economic growth per capita. Immigrants appear to innovate more than natives not because of greater inherent ability but due to their concentration in science and engineering fields.44
43 TFP is defined as that portion of output not accounted for by the amount of capital stock and (quality-adjusted) labor force used in its production.
Visa Pathways for High-Skilled Immigrants
Foreign-born workers with a bachelor’s degree or equivalent initially move to the United States either on a temporary visa or as a permanent resident. A worker who enters as a permanent resident may do so as a relative of a U.S. permanent resident or citizen, sponsored by an employer as an EB-1, EB-2, or EB-3 worker considered particularly qualified, or on an EB-5 investor’s visa. Permanent residents are free to change employer. Temporary work visas are issued to the foreign worker’s U.S. employer to hire him or her specifically: The worker is not free to choose an employer on arriving in the United States, faces barriers to changing employer after arrival, and may not become self-employed (or start a company) nor become unemployed. Those who enter on temporary visas may succeed in obtaining permanent resident status by marrying a U.S. citizen or through their employer’s sponsorship. Some foreign-born workers initially enter the United States as students or trainees on F-1 visas and take advantage of the Optional Practical Training period permitting up to a year and a half of work, and/or they obtain another status after graduation.45 Individuals who enter as dependents of temporary visa holders and may be unable to work initially gain permanent residence if their family member does.
Because those entering as permanent residents typically stay longer in the United States than do those entering on temporary visas, the initial visa composition of new entrants is different from that of the stock of workers at a given point in time. The National Survey of College Graduates shows that, in 2013, 38 percent of foreign-born, college-educated workers had entered with lawful permanent residence, 16 percent on a temporary work visa, 25 percent on a student or trainee visa, 11 percent as the dependent of a temporary visa holder, and 9 percent on other temporary visas.
The two most common entry work visas are the intracompany transferee visas (L-1A and L-1B), whose numbers are uncapped and are for 1-3 years, renewable for a maximum stay of 5-7 years, and the specialty worker (H-1B) visas, whose number is capped (in the for-profit sector) and which are issued for 3 years, renewable once. Both are “dual intent” visas, meaning the employer may sponsor the worker for permanent residence. Intracompany transferees have been transferred to the United States by an employer for whom they have worked abroad for at least a year. Some skilled workers also enter as a J-1 exchange visitor, although the number of J-1 holders who are skilled is not known.46 As discussed below, while
H-1B visa holders have been subject to much scrutiny, despite imperfect data, there has been much less analysis of L-1 visa holders.
Impact of High-Skilled Immigration on Wages and Employment
As noted previously, the impact of high-skilled immigration on native wages and employment has been the focus of less attention than the impact of low-skilled immigration. However, in part due to the substantial and rising share of high-skilled immigrants, as well as the possibility of positive spillovers from this group, increasing attention has focused on them. Much of this research employs the spatial approach. As elsewhere in the spatial literature, it is difficult to identify the causal effect of skilled immigration—again, reverse causality or unobserved common factors may confound results. For example, wage increases for natives may lead to increased growth of immigration by STEM workers—so again, the potential for results to be contaminated by locational choices persists.47 Analyses must address the possibility that cities with rapid productivity growth will experience wage growth and (for nonobservable reasons) will also attract foreign STEM workers.
A study by Peri et al. (2015a) devised an instrument to address the endogeneity problem, apportioning the changing national-level number of H-1B visas to cities based on the 1980 distribution of foreign-born STEM workers, thus combining the identification methods of Card (2001) and Kerr and Lincoln (2010).48 Their study period, 1980 to 2010, is especially dynamic because college-educated STEM workers grew from 2.4 percent of total employment to 3.2 percent over the period, and foreign-born workers were responsible for more than 80 percent of this growth. The authors found that a rise in foreign-born STEM workers by 1 percentage point of a city’s total employment (close to the total increase over the period) increases the real wages of college-educated natives by 7 to 8 percent and those of noncollege-educated natives by 3 to 4 percent (Peri et al., 2015a, p. 3). The effect on the native employment rate was not statistically significant. These results are consistent with a positive effect of inflows of foreign-born STEM workers on the wages of both college-educated and, to a lesser extent, noncollege-educated natives. However, the very large estimates of
47 Controlling for factors such as native response may be especially important in this context, given that high-skilled labor markets are likely to be national and even international in spatial scope.
48 To ensure that this is an effective instrumental variable, the authors tested to confirm that “the initial (1980) distribution of other types of foreign-born workers (e.g., less educated and manual workers), the initial industry-structure of the metropolitan area, and the subsequent inflow of non-STEM immigrants do not predict growth in foreign STEM workers” (Peri et al., 2015a, p. 3).
this wage increase raise the possibility that there may be additional factors driving the determination of wages in high-skilled labor markets that are not captured by this approach.
Other studies of the impact of high-skilled immigration on wages and employment analyze the impact on specific groups of native workers. Because around half of workers receiving H-1B visas in recent years have been hired to work in computer occupations, these jobs are the most likely to be negatively impacted by an inflow of skilled immigrants. To examine this, Peri et al. (2015b) took advantage of the fact that H1-B visas were allocated via lottery in 2007 and 2008. Some cities appeared to satisfy less of their firms’ demand for H-1B workers than did others, although the city demand for H-1B visas has to be proxied by firms’ preliminary expressions of interest, which are much more numerous than actual applications. The authors found that the more a city’s demand for H-1B workers outstripped the visas its firms won in the lottery, the lower the city’s employment and wage growth for native-born workers in computer occupations. They inferred that H-1B workers do not displace but rather complement natives in computer-related occupations.49
These positive estimated effects on native wages (Peri et al., 2015a, 2015b) and employment (Peri et al., 2015b) are consistent with high-skilled immigrants’ being complementary with natives, especially high-skilled natives; with human capital spillovers stemming perhaps from interactions among workers; or with skilled immigrants innovating sufficiently to raise the productivity of all workers. For example, highly educated hires may stimulate the productivity of natives—at least in the computer-related occupations studied—incentivizing firms to expand hiring. This type of mechanism is also explored in important research by Moretti (2010), who found that each job in the tradable high tech sector (making products that need not be consumed locally) generates between 0.5 to 2 additional jobs in the local economy. Immigrant innovation is considered in detail later in this section (Section 5.6).
However, not all studies find beneficial wage and employment effects of skilled immigrants. Borjas (2009) examined the correlation between immigrant share and the earnings of doctorate-holders by doctoral cohort and discipline. He estimated wage elasticities of −0.24 to −0.31, where these elasticities indicate the percentage change in earnings associated with a 1 percent change in labor supply due to immigration. The larger estimate (absolute value) was obtained when the elasticity was calculated using only the sample of foreign-born doctoral recipients that intended to stay in the United States. In addition, Borjas and Doran (2012) examined
the impact of the arrival of 336 Soviet émigré mathematicians after the collapse of the Soviet Union. They found that American mathematicians in subfields with active émigrés were published and cited less after 1992 and were more likely to move to lower-quality institutions and out of active publishing.
Do Natives Change Field or Occupation in Response to Skilled Immigration?
One reason many high-skilled natives may not be harmed by high-skilled immigration, especially over longer time periods, is that they may shift into other fields in which they have a comparative advantage due, for example, to qualifications or language skills or in which they are complementary to immigrants. Such shifts would increase the economy’s productivity via greater specialization and would constitute one of the benefits of immigration. Peri and Sparber (2011) provided evidence that inflows of highly educated immigrants cause natives to switch to more communication-intensive occupations. Cortés and Pan (2014) found that U.S. states with the highest flows of foreign-born nurses experienced decreased numbers of natives entering the profession and sitting for licensing exams; the researchers detected an offsetting increase of similar size in the numbers of natives entering teaching professions in these states. They used an instrument based on historical immigrant flows for foreign nurses. Borjas (2007) employed a fixed effects panel of universities over time to study the effect of foreign-born graduate students on native-born graduate student enrollment. He did not find evidence of a crowd-out effect for the typical native, but there was a strong negative correlation between increases in the number of foreign-born students enrolled at a particular university and the number of white native-born men in that university’s graduate program.
The possibility of natives changing occupation or field of study has been of particular interest in the context of immigrants’ effect on innovation. Consequently, a number of papers ask whether skilled immigration causes natives to leave or fail to enter STEM fields. Orrenius and Zavodny (2015) examined whether native-born bachelor’s students pick a science or engineering major. The covariates of interest measure the concentration of immigrants in college as well as the concentration of immigrants when the natives were of high school age, and the instruments are variants of the historical settlement pattern instrument. They found that the presence of immigrants deterred some native-born women from choosing a science or engineering major; this effect was not found for native-born men. Some evidence of native response to immigrants entering STEM fields was also found by Bound et al. (2015). Using a structural model, the authors estimated that native employment in computer science would have been
7.0-13.6 percent higher in 2004 absent increased immigration after 1994; they also found wages for computer scientists would have been 2.8-3.8 percent higher. However, they found that total employment in computer science would have been 3.8-9.0 percent lower. This is consistent with the possibility that immigration increased software innovation, although this is of course hard to measure.
Another reason that an increase in the numbers of high-skilled immigrants in the labor market may not lead to lower overall employment or wages of highly educated natives is positive productivity effects, or spillover effects whereby technological progress is spurred through the creation and diffusion of knowledge and innovation. This topic is discussed below.
Theoretical Considerations Relevant to Innovation
As explained in Borjas (2014a) and discussed above, it is high-skilled immigrants’ potential positive externalities, rather than the simple fact that they are more productive individually than low-skilled immigrants, that distinguish their impact on natives from that of other immigrants. Innovation is the channel through which immigrants could potentially have the largest positive externality. Innovation, whether by natives or immigrants, eventually enters the public domain and increases the productivity of workers not linked through the market to the original innovator. Immigrant innovators may also have a positive externality on native innovators, which could magnify the externality due to their own innovation.
However, the arrival of a certain number of innovative immigrants is not likely to boost the number of innovators in the country by the same number. Some innovative immigrants will not enter innovative work, while innovative natives will respond to the immigration. Some natives may leave innovative work to exploit the increase in their comparative advantage in language-intensive work (Peri and Sparber, 2009). Conversely, if immigrant innovators render native innovators more productive, the number of native innovators could rise. Studies of the effect of immigration on innovation must take these responses into consideration when judging whether immigration is likely to have boosted economic growth rather than simply having caused a one-time increase in efficiency.
Methodological Considerations for the Impact of Immigration on Innovation
Many of the methodological concerns relevant for the impact of immigrants on innovation are the same as those relevant for the impact on wages and employment, especially the endogenous pattern of immigrant density across the units of observation. One dimension along which studying inno-
vation is trickier is measurement: most studies proxy for innovation with patents, while some compute TFP50 and assess the effect of immigration on productivity. Patent counts measure inventions, a type of knowledge with the potential to increase TFP, but not all inventions are patented and not all innovation comes in the form of inventions. Innovative business practices are not inventions, for example, while innovative software became patentable in 1995 amid debate about whether a software innovation constitutes an invention (see Hall and MacGarvie, 2009). Furthermore, patents vary greatly in terms of quality, though future citations to patents provide a guide to quality. On the other hand, the measurement of TFP is fraught with difficulties such as the specification of the correct production function and further measurement issues such as the rate at which capital depreciates (Aiyar and Dalgaard, 2005).
Conversely, along a second important dimension, studying the impact on innovation is more straightforward than studying the impact on wages. While the presumption of factor price equalization (or at the least, factor price insensitivity) through interregional trade means that the effect of local concentrations of immigrants on wages is likely to be national in part, and not purely local, there is no equivalent of this constraint for patents; whereas the benefits of innovation diffuse across the country, the location of the original inventor does not. Nor would a response of capital flows equalize patenting across regions. The adjustment mechanism that does remain is geographic mobility of native innovators reacting to any immigration-induced changes in innovator wages. If immigrant innovators have negative effects on native innovator productivity and wages in their region, native innovators will avoid immigrant locations. This native relocation will lead a spatial identification approach to underestimate the benefit of immigration. Nevertheless, the forces for national diffusion of innovation responses are weaker than for the diffusion of wage responses. The implication is that using spatial variation in immigration to identify the effect on patenting is subject principally to the endogeneity problem of immigrants possibly choosing their location based on the outcome variable. Studies focusing directly on productivity, however, are subject to problems similar to wage studies: a bias toward finding no effect remains even if immigrant location is successfully instrumented, due to the forces equalizing labor market conditions across regions.
Are Immigrants More Innovative Than Natives?
Immigrants are most likely to increase innovation if they are themselves more innovative than natives, making an individual-level comparison of
immigrants and natives the logical first step. At least as measured by patents, immigrants do innovate considerably more than natives. Using the 2003 National Survey of College Graduates (NSCG), Hunt (2011) showed that among individuals with a bachelor’s degree or higher, immigrants are twice as likely to patent as natives, while Kerr (2007) documented the rapid rise from 1975 to 2004 of U.S. patents authored by U.S. residents with Indian and Chinese first and last names: from 2 percent to 9 percent of all patents for Chinese names and from 2 percent to 6 percent of all patents for Indian names. Kerr could not distinguish first and second generation immigrants, but this growth is nevertheless fundamentally fueled by immigration.
More specifically, the Hunt (2011) study showed that 0.9 percent of natives, compared to 2.0 percent of immigrants, had been granted a patent in the previous 5 years. One measure of the quality of these patents is whether they have been licensed or commercialized; 0.6 percent of natives compared to 1.3 percent of immigrants had licensed or commercialized a patent granted in the previous 5 years. All these differences were statistically significant. She also found that, conditional on having at least one patent, immigrants and natives had similar numbers of patents.
Hunt’s dataset is one of the few with visa information, and she found that the particularly innovative immigrants were those who entered on a temporary worker visa or a temporary student visa (especially as a graduate student or postdoctoral fellow). It seems that foreign-born workers or students chosen by a firm or university were more innovative than those who entered on a green card, most of whom were joining family in the United States and who patented at levels similar to natives. Hunt also investigated the source of the immigrant advantage and found that the immigrants’ edge was due to their being much more likely to have studied science or engineering as a highest degree and to a lesser extent to their having higher education than natives. Her comparison among immigrants and natives with similar fields of study and level of education did not yield any statistically significant differences in patenting.
Do Immigrants Increase Innovation?
The superior innovative performance of immigrants, as measured by instruments such as rates of patenting, does not, however, necessarily imply that immigration increases innovation, since natives are likely to change their behavior in the face of immigration and could reduce their own innovation. One of several studies that tackled the more difficult issue of overall innovation is that of Hunt and Gauthier-Loiselle (2010), who used census and U.S. Patent and Trademark Office (USPTO) data to form a panel of states from 1950 to 2000. The key explanatory variable, the intercensal
change in the share of a state’s population that is skilled immigrants, is endogenous: High-skilled workers are more likely to migrate to states that are experiencing positive shocks to innovation, either narrowly or as part of more general skill-biased technological change, unobservable to the econometrician. Like many authors of wage impact studies, Hunt and Gauthier-Loiselle used an instrument based on historical immigrant settlement patterns: in this case, the 1940 settlement pattern.51
The results showed that influxes of high-skilled immigrants—those with either at least a bachelor’s or master’s degree or those working in science and engineering occupations—statistically significantly increased patenting per capita. A 1 percentage point increase in the immigrant college graduates’ population share increased patents per capita by 9-18 percent, with the larger effects resulting from the instrumental variables analysis. This means that the net result of the immigrants’ own innovation, any native movements in or out of innovative jobs, and any effect of immigrants on the productivity of native innovators was positive. The magnitudes are such that the increase in skilled immigration in the 1990s can account for one-third of the large patenting increase in that decade. In turn, this additional patenting may have increased GDP per capita by 1.4-2.4 percent by the end of the decade.
Immigrants may also increase native patenting, but because U.S. patents do not note the birthplace of the inventor, Hunt and Gauthier-Loiselle (2010) could not directly examine this question. However, calculations of the immigrant contribution based on the individual-level 2003 NSCG suggest that the state panel must reflect considerable positive effects on native patenting. But the standard errors for the calculations are large, and the individual-level immigrant contribution may not always have been at its 2003 level.
Most other papers study the effect of more specific groups of immigrants than Hunt and Gauthier-Loiselle (2010). Kerr and Lincoln (2010), for example, used USPTO patent data and CPS data to form a panel of cities for 1995-2008, but they examined the effect on patenting of increased numbers of workers entering the United States on H-1B visas. Hunt (2011) suggested that these workers are indeed very likely to patent. The difficulty is that the distribution of H-1B holders by states is unknown and must be proxied for by using the number of preliminary applications, Labor Condition Applications, or simply noncitizen immigrants, which introduces measurement error into the regression. Identification of the effect comes from variation in the initial share of the population that is on H-1Bs, interacted
51 A potential issue with this approach is that if controls do not account for state-specific patenting shocks that are very persistent and influence national inflows of particular immigrant groups, the instrument could be correlated with the error term. Though this does not seem likely, it cannot be ruled out, for example, that California has had serially correlated positive patenting shocks that caused low-skilled Chinese to settle there before 1940 but that have motivated high-skilled Chinese to move to the United States in more recent years.
with the change in the H-1B national cap—under the assumption that a state’s increase is greatest where the initial share is greatest.
Kerr and Lincoln (2010) found that an increase in the national H-1B cap statistically significantly increased patenting in cities with many H-1B holders compared to cities with fewer H-1B holders. A 10 percent increase in the cap was associated with a 0.3-0.7 percent increase in patenting for each standard-deviation change in a city’s share of H-1Bs. The magnitude of these results is not easily comparable with those of Hunt and Gauthier-Loiselle (2010). Kerr and Lincoln found that immigrants had little or no effect on the patenting of those with Anglo-Saxon names, who were disproportionately natives. This contrast with the Hunt and Gauthier-Loiselle findings could be attributable either to imperfections in one or both studies relevant to measuring this externality or to the focus by Kerr and Lincoln on short-term effects, whereas Hunt and Gauthier-Loiselle focused on long-term effects.
In contrast to the studies described so far, Doran et al. (2015) found no contribution to patenting from H-1B visa holders. Specifically, they found that relative to firms that lost the 2006 and 2007 H-1B lotteries, winning firms had no increase in the number of patents in the 9 years following their acquiring the H-1B workers. The use of a lottery makes the identification in this study methodology particularly clean.
The differing results across these studies may reflect immigrant heterogeneity generally and among H-1B workers in particular. A large share of the H-1B inflows consists of young computer programmers working for information technology software services firms; often both firm and worker are Indian.52 Such workers tend to stay only a short time in the United States53 and reflect U.S. participation in Mode 4 of the General Agreement on Trade in Services;54 these workers are not expected
52 For example, in FY 2006, about one-half (51%) of first-time H-1Bs were awarded to computer programmers; 69 percent of first-time H-1B visas were awarded to workers ages 25-34; and 54 percent of first-time H-1B visas were awarded to Indians. See U.S. Citizenship and Immigration Services data at https://www.uscis.gov/tools/reports-studies/immigration-forms-data [May 2017].
53Clemens (2010, p. 14) reported that for a large Indian software services firm making great use of the H-1B program, typical U.S. assignments last 6-15 months (though it is common for H-1B winners to return to India, then later take another H-1B assignment in the United States).
54 This trade agreement took effect in January 1995 and is binding on all members of the World Trade Organization, which was established on the same date. Mode 4 concerns the supply of a service by a service supplier of one member, through the presence of natural persons of a member. The United States committed to permitting temporary work permission for intracompany transferees from abroad (an unlimited number of L-1 visas) and for 65,000 specialty occupation workers (H-1B visas). See https://www.uscis.gov/tools/reports-studies/immigration-forms-data [May 2017].
to innovate.55 Such workers are a large share of the flows that drive the results reported by Doran et al. (2015), but they are a much smaller share of the stocks of immigrants who entered on temporary work visas and were found to be so innovative by Hunt (2011). A cross-section such as that used by Hunt (2011) implicitly weights immigrants according to the duration of their stay in the United States.
Two other papers examine fascinating cases of high-skilled immigration and its effect on innovation. Using a clever identification based on the different specializations of American and Soviet mathematicians, Borjas and Doran (2012) showed that American mathematicians’ research was reduced by the arrival of Russian mathematicians after the Cold War but that total U.S.-based mathematical research remained approximately constant. In contrast, Moser et al. (2014) showed that German Jews who fled to the United States in the 1930s greatly boosted patenting in chemical fields. They found that the German Jews increased native patenting by attracting to their subfields natives who would otherwise not have patented, while reducing the patenting of natives already in the field. As with Doran and colleagues, their instrument exploits differences in specialization—in this case between German Jews and American chemists. Both of these studies examined the impact of exceptionally skilled immigrants, and one would not necessarily expect to find similar impacts of immigration from, for example, recent immigrants in the H-1B program.
A quite different approach is to measure the effect of immigration on productivity directly. The advantage of this approach is that productivity is the economists’ ultimate interest, while the disadvantage is that productivity is difficult to measure and innovations improving productivity diffuse across the country. The measure of productivity most closely linked to innovation is TFP. Measuring TFP involves modeling output by selecting a production function for the economy—a difficult exercise—and measuring the values of inputs, which involves judgments on matters such as the rate of depreciation of capital. TFP is measured as the residual in the modeling exercise and is sensitive to modeling and measurement choices, so this type of evidence cannot provide conclusive proof of an immigration impact on productivity.
Peri (2012) measured state-level TFP for a panel of states and linked this directly to immigration, using as an instrument historic settlement patterns (similar to Hunt and Gauthier-Loiselle, 2010) or distance to the Mexican border. Peri et al. (2015a) calculated the effect of immigrant science
55 Computer science graduates in general do not patent more than do workers outside science and engineering (Hunt et al., 2013), but innovation in computer science may often be more akin to improved business organization than invention and hence poorly captured by patent counts.
and engineering workers (unusually broadly defined) on TFP by combining effects on wages and employment, described above, with the assumption that the capital to labor ratio is constant in the long run. Both papers (Peri, 2012; Peri et al., 2015a) found that immigration increases TFP.
Do Immigrants Foster Growth Through Entrepreneurship?
For inventions to speed growth, they must be brought to market. Inventiveness and business acumen are therefore complementary inputs to technology-spurred productivity growth. These inputs may be embodied in a single person or may be combined though collaboration among two or more people. New inventions are often best developed and marketed in new firms, making entrepreneurship a particularly important type of business acumen in this context.56Baumol (1993, p. 260) argued that, just as capital investment and human capital may be treated as endogenous to economic growth, “[t]o some degree, the same story can be told about the exercise of entrepreneurship, investment in innovation, and the magnitude of activity directed to the transfer of technology. These too, clearly, are influenced by past productivity growth achievements and they also, in their turn, influence future growth.” A link can be made between the literature on entrepreneurship and endogenous growth theory (Lucas, 1988) by recognizing that an expanded capacity for entrepreneurial ability is a form of human capital. Schultz (1980, p. 437) stated that “. . . the abilities of entrepreneurs to deal with the disequilibria that are pervasive in a dynamic economy are a part of the stock of human capital. . . . An innovation by a business enterprise (Schumpeter’s innovator) is an endogenous event.”57
Researchers interested in economic growth as well as in entrepreneurship and business formation frequently examine the rate at which immigrants open new firms. Fairlie and Lofstrom (2015) used ACS data from 2006 to 2010 to calculate that the 2.4 million immigrant business owners (defined simply as the self-employed, with or without employees) made up a slightly higher share of all business owners (18.2%) than their share of the total U.S. workforce (16.3%). This translates into slightly higher business ownership among immigrants than among natives: 11.0 percent of immigrants and 9.6 percent of natives owned a business in this dataset.
56Wennekers and Thurik (1999, p. 46) described entrepreneurship as the “manifest ability and willingness of individuals, on their own, in teams, within and outside existing organizations, to: perceive and create new economic opportunities (new products, new production methods, new organizational schemes and new product market combinations); and to introduce their ideas in the market, in the face of uncertainty and other obstacles, by making decisions on location, form and the use of resources and institutions.”
However, there are variations in entrepreneurship by immigrants’ country of origin, as well as by industry. Indian immigrants are the most entrepreneurial of any group including natives, and immigrant businesses represent more than a quarter of businesses in the transportation, accommodation, and recreation and entertainment sectors.58
Monthly business startup data constructed from matching respondents across months in the 2007-2011 CPS were used by Fairlie and Lofstrom (2015) to estimate that immigrants represented 24.9 percent of new business owners, a figure much higher than the 15.6 percent of the nonbusiness-owning population immigrants represent. This finding seems at odds with the fairly similar overall self-employment rates found for immigrants and natives. One possibility is that the more recent immigration cohorts were more entrepreneurial than either natives or earlier immigrants, which should eventually lead to a larger difference in the stock of self-employed. Figure 5-3, which shows that the immigrant self-employment rate has risen relative to the native rate since 2000, is consistent with this possibility. Alternatively, higher business startup rates could imply a higher failure rate for immigrant entrepreneurs. Consistent with immigrant businesses being younger, they are also smaller: Using data from the 2007 Survey of Business Owners, Fairlie and Lofstrom (2015) found that immigrant-owned firms had $434,000 in average annual sales and receipts compared with $609,000 for nonimmigrant firms.
Business owners’ level of education may also be used as a measure of the likely contribution of businesses to the economy. Fairlie and Lofstrom (2015) reported that, while immigrants are highly overrepresented among owners with less than a high school degree, at almost 45 percent, they also represent 15.7 percent of owners with a college degree. However, the latter share may overstate the value of immigrants’ contribution if immigrants turn to self-employment because their foreign education and experience are less valuable, or less valued, than American education and experience (Borjas, 1986; Fairlie and Meyer, 1996; Portes and Zhou, 1996), rather than because of an innovative business idea. For example, Akresh (2006) found that 50 percent of immigrants experienced occupational downgrading59 on arrival in the United States. While showing that immigrants contribute significantly to self-employment, data representative of the population thus do
58 For theoretical and empirical analysis of the clustering of immigrant and ethnic groups in particular types of self-employment, see Kerr and Mandorff (2015). Kloosterman and Rath (2001) focused on small business formation in nontradable sectors such as lower-end retailing and restaurants.
59 As characterized by the author, this term refers to transitions by immigrants into jobs for which they are overeducated or overqualified and which may entail a loss of occupational status or prestige relative to the job they held in their country of origin.
not show clearly the contribution of immigrant entrepreneurs to successful (in terms of size or growth) or innovative firms.
To pinpoint immigrant contributions to innovative entrepreneurship, better data are required. Currently, it can be difficult to distinguish between the self-employed who have a small number of employees and those that do not, or between businesses that are innovating and those that are not. The smaller average size of immigrant businesses may obscure a pattern in which immigrants disproportionately start small, noninnovative businesses that are less likely to grow as well as successful, eventually large, innovative businesses. There are hints that immigrants disproportionately start very successful businesses, suggested by high-profile examples of public U.S. companies with foreign-born founders, such as Google, eBay, Yahoo!, and Sun Microsystems. In a sample of 1,300 “high-impact” technology firms and 2,000 founders across the United States, Hart and Acs (2011) found that around 16 percent of firms have at least one immigrant founder. Wadhwa et al. (2007) found that immigrants started 25 percent of new high tech companies with more than $1 million in sales in 2006, while Anderson and Platzer (2006) found that immigrants represented 25 percent of founders of recent public venture-backed companies. Qualitative studies such as Saxenian (1999) also emphasize the large immigrant contribution to technology startups.
Using the 2003 NSCG, Hunt (2011) was also able to narrow the focus to fast-growing startups. Like many surveys, the NSCG includes questions on firm size and self-employment, which permit a distinction between the self-employed with more than 10 employees and the self-employed with fewer than 10 (including none). Hunt (2011) took advantage of unusual additional startup information in the NSCG to examine the probability of founding a firm that grew to more than 10 employees in 5 years. She found that, conditional on characteristics, immigrants are 30 percent more likely to found such firms than are similar natives. In an unconditional comparison between all immigrants and all natives, the result is the same in sign and magnitude, but statistically insignificant: The rarity of the outcome (0.6% of native respondents founded a firm that met the condition) makes standard errors large in all regressions and also precludes investigation of the startup’s industry or the founder’s patenting activity.
A literature overlapping with the immigrant business formation and entrepreneurship literature examines links between immigrants and their home countries. For example, Saxenian (2002) described a phenomenon she called “brain circulation.” As high tech entrepreneurs first migrate to the United States for some combination of education, business experience, and innovation experience, they found technology companies or affiliates at home while maintaining or increasing U.S. ties. Kerr (2008) quantified the positive links between immigrant patenting in the United States and labor productivity and manufacturing (especially high technology) output in immigrants’ home developing countries, while Foley and Kerr (2013) showed that an increase in a multinational company’s patenting by workers of a particular ethnicity was followed by greater investment by the company in the home country corresponding to the ethnicity. Studies of whether immigrants boost trade between the source and destination country also have implications for growth and entrepreneurship (Gaston and Nelson, 2013).
Immigrants may increase international trade in two ways. First, they may have a taste for goods available only in the home country, which stimulates demand for imports directly and also indirectly as natives acquire a taste for the same foreign goods. Second and more relevant for this section, immigrants know the markets in their home country and maintain business ties there founded on trust and social capital. These ties and knowledge can reduce the problems of incomplete contract enforcement and asymmetric information that constitute barriers to trade. The empirical literature in this area is less sophisticated in dealing with potential endogeneity than other literatures related to immigration. The most rigorous paper examining the United States, Bandyopadhyay et al. (2008), confirmed the results of the wider literature by finding that U.S. states that received an increased number of immigrants from a particular source country increased their exports
to that country. Andrews et al. (2015) confirmed these results for Germany using firm-level data. However, the possibility that a common third factor is increasing both trade and immigration, or that the causality runs both ways, cannot be ruled out in all cases.
The literature on immigrants and entrepreneurship is informative about the number of businesses formed by immigrants and the importance of ties between immigrant innovators and entrepreneurs and their home country, but it is only suggestive about whether immigrants causally stimulate trade or whether immigrants have a causal impact on U.S. growth through fast-growing or innovative startup companies. More research and more data with which to perform it are required to not only confirm the reported associations but also shed light on causation.
Economies respond to immigration through several mechanisms: adjustment of factor prices, shifts in output mix, and changes in the use of production technology. The extensive literature on the economic impacts of immigration primarily focuses on the marginal product of labor and the resultant wage and employment outcomes in receiving countries’ labor markets. The review in this chapter reflects this research emphasis. However, shifts in sectoral composition and adaptation of new technology are also discussed, both to fully understand immigration wage and employment dynamics and because they are interesting in their own right. The impact of immigration on capital accumulation and economic output, considered in Chapter 4, is relevant here in differentiating between short- and long-run changes in wages. The panel also considered the relationship between the immigration of high-skilled workers and innovation and how this relationship may generate changes in long-run economic growth; however, this topic is addressed more comprehensively in Chapter 6.
The empirical evidence reviewed in this chapter reveals one sobering reality: Wage and employment impacts created by flows of foreign-born workers into labor markets are complex and difficult to measure. The effects of immigration have to be isolated from many other influences occurring simultaneously that shape local and national economies and the relative wages of different groups of workers. Among the largest of these influences are changes in production technology, communications technology, and the global economy, which together promote international trade in goods and services (and hence offshoring), global supply chains, and foreign investment. Additionally, firm births and deaths occur, people retire, workers switch jobs, and a stream of young native-born job seekers come of age—all factors that affect the labor market. The inflow of the foreign-born at a given point in time is, under normal circumstances, a relatively
minor factor in the $18 trillion U.S. economy.60 That said, quantitatively significant labor supply shocks do occur, especially in localized markets, such as that which accompanied the 1980 Mariel boatlift in Miami (Borjas, 2016b; Card, 1990; Peri and Yasenov, 2015). Even then, the wage impacts may be difficult to detect.
The measurement task is further complicated because the impact of immigration on labor markets varies across time and place, reflecting the size of the inflow, the skill sets of natives and incoming immigrants, the local industry mix, the spatial and temporal mobility of capital and other inputs, and the overall health of the economy. Some of the processes that are set in motion take place immediately upon arrival of the foreign-born, while others unfold over many years. Aside from supplying labor, immigration (like population growth generally) adds to consumer demand and hence the derived demand for labor in the production of goods and services. This counterbalancing impact potentially plays a role in explaining why much of the empirical research finds small wage impacts associated with immigration. As noted above, the changes in wages and employment attributable to immigration can be difficult to identify because other factors tend to swamp the relatively small role that immigration typically plays in the overall labor market. In short, the uniqueness of immigrant inflows to time and place implies that it is difficult to use the lessons from one episode to predict the impact under different circumstances in the future.
Beyond these real world complexities, several additional measurement problems must be resolved. Primary among these (at least for some kinds of studies) is the endogeneity of immigrants’ locational choices—most notably, the interaction between the vibrancy of local economies and people’s location choices. Evidence suggests (Borjas, 2001; Somerville and Sumption, 2009) that immigrants locate in areas with relatively high labor demand and wages for the skills they possess and that immigrants are more willing than natives to relocate in response to changes in labor market conditions (Cadena and Kovak, 2016). If immigrants predominantly settle in areas that experience the highest wage growth, a spurious correlation arises: Wage growth (or dampened wage decline) will be erroneously attributed to the increase in labor supply. Additionally, correct identification of the wage and employment effects of immigration must account for the possible migration response of natives to the arrival of immigrants. Researchers have made great strides addressing these identification issues in recent decades; even
60 While the incremental flow of new immigrants appears to generate modest economic impacts, the stock of foreign-born individuals that has accumulated over time may be significant to long-run economic growth (see Chapter 6). Also notable is the fact that immigrants account for almost half the labor force growth in the United States since the mid-1990s (see Chapter 2).
so, the degree of success in dealing with them is still debated and methods are still being perfected.
Several analytic approaches have been developed to estimate wage and employment impacts associated with immigration, each with strengths and weaknesses. Spatial studies compare wage and employment trends in high versus low immigration areas, often defined by metropolitan areas, in order to identify the impact of immigration on wages and employment. A different set of studies examines the impact of immigration by exploiting variation in the density of the foreign-born across skill groups, typically defined by experience (age) and education groupings, instead of across geographic areas. Spatial studies must contend with the challenge of the endogeneity of destination locations, as described above. Meanwhile, skill cell studies, by focusing on the effect of immigrants on similar natives, may miss wage and employment effects induced by complementarities between immigrants and native-born workers at other parts of the skill distribution.
An influential variant of the skill cell literature is the third general approach reviewed in depth in this chapter. This structural approach imposes a modeling structure that relies heavily on assumptions about the relationship between output and the inputs to production (including different kinds of labor). The underlying structure assumes that average wages are unchanged by immigration in the long run—a period of time long enough such that all inputs to production, including capital, may be adjusted by firms. This assumption limits such analyses to estimating relative wage impacts across different groups, such as across high school dropouts, those with a high school degree, those with some college, and those with a college degree. The technical assumptions are therefore not innocuous; the most significant ones concern the degree to which capital is adjusted by firms in response to new worker inflows, the degree to which immigrants and natives within the same skill group are substitutable, and the degree to which high school graduates and high school dropouts are substitutable.
While many studies conclude that, economy wide, the impact of immigration on average wages and employment is small, a high degree of consensus exists that specific groups are more vulnerable than others to inflows of new immigrants. Theory predicts that the workers already in the receiving labor market who are the closest substitutes for immigrants are most likely to experience immigration-induced wage declines. Prior immigrants are typically the closest substitutes for new immigrants, followed by native high school dropouts, who are more affected due to the large share of low-skilled workers among immigrants to the United States. For this reason and due to concern about the economic well-being of native high school dropouts, much of the empirical literature concentrates on low-skilled labor markets.
Empirical research in recent decades suggests that findings remain by and large consistent with those in The New Americans (National Research Council, 1997) in that, when measured over a period of more than 10 years, the impact of immigration on the wages of natives overall is very small. However, estimates for subgroups span a comparatively wider range indicating some revisions in understanding of the wage impact of immigration since the 1990s. As noted above, for example, some studies have found sizable negative short-run wage impacts for high school dropouts, the native-born workers who in many cases are the group most likely to be in direct competition for jobs with immigrants. Even for this group, however, there are studies finding small to zero effects, likely indicating that outcomes are highly dependent on prevailing conditions in the specific labor market into which immigrants flow or the methods and assumptions researchers use to examine the impact of immigration. The literature continues to find less favorable effects for certain disadvantaged workers and for prior immigrants than for natives overall.
For the larger group of studies of natives overall or of low-skilled natives, the panel compared the magnitude of estimated wage impacts after harmonizing (to the extent possible) the effects associated with an immigrant influx equivalent to a 1 percent increase in labor supply. Some notable patterns emerge. Consistent with theory, native dropouts tend to be more negatively affected by immigration than better-educated natives. Some research also suggests that, among those with low skill levels, the negative effect on native’s wages may be larger for disadvantaged minorities (Altonji and Card, 1991; Borjas et al., 2012) and Hispanic high school dropouts with poor English skills (Cortés, 2008). Since native dropouts experience a larger immigrant-driven labor supply increase than do natives overall, their greater susceptibility to a given immigrant inflow is compounded by higher inflows. Another regularity consistent with theory is that there are larger negative effects on native wages from immigrant inflows in the short run (i.e., in studies of the immediate impacts of abrupt immigrant inflows or in which inflows are observed over shorter periods of time, or in the case of the structural studies, when capital is assumed fixed). Estimated negative effects tend to be smaller (or even positive) over longer periods of time (10 years or more) or in the case of structural studies, when capital is assumed to be perfectly flexible.61
The results from our comparison of magnitudes also suggest that some of the differences in the estimated effects of immigration on natives are due to methodology, since they cannot be fully accounted for by whether the studies are looking at the long versus short term, at high school dropout
61 In the case of structural studies, when capital is assumed to be perfectly flexible, wage effects on natives are zero, although this result is built in by theoretical assumptions.
natives versus all natives, or minority natives versus all natives. The skill cell studies appear to find the most negative wage impacts and the structural the least negative, with the spatial studies in the middle. As noted earlier, the approaches are not fully comparable. The numerical value of some of the elasticities from the structural approach is often built in by the technical assumptions. The skill cell studies avoid the endogeneity biases of the spatial studies, which makes the former more likely to find negative effects. However, they do not include cross-effects, for example the impact of an inflow of immigrants in one skill group on the wages of natives in another skill group, whose overall sign is unknown. If positive, cross-effects would not reverse the sign of the reported net effects but would lessen their magnitude.
Most studies find little effect of immigration on the employment of natives. However, recent research (Smith, 2012) does find that native teen employment, measured in hours worked, but not the employment rate, is reduced by immigration. Moreover, as with wage impacts, there is evidence that the employment rate of prior immigrants is reduced by new immigration—again suggesting a higher degree of substitutability between new and prior immigrants than between new immigrants and natives.
The impact of high-skilled immigration on native wages and employment has been the focus of less attention than the impact of low-skilled immigration. The results of spatial studies are mixed, but some find a positive impact of high-skilled immigration on the wages and employment of both college-educated and less educated natives. If confirmed, such findings would be consistent with high-skilled immigrants being complementary with natives, especially high-skilled natives; with human capital spillovers stemming perhaps from interactions among workers; or with high-skilled immigrants innovating sufficiently to raise the productivity of all workers. However, other studies that examine the earnings or productivity of narrowly defined groups of high-skilled workers (such as doctorates in narrow fields or professional mathematicians) found that high-skilled immigration had adverse effect on the wages or productivity of these high-skilled natives.
Finally, immigrants influence the rate of innovation in the economy, which potentially affects long-run economic growth. While research in this area is very recent, literature on the topic as a whole indicates that immigrants are more innovative than natives; more specifically, high-skilled immigrants raise patenting per capita, which is likely to boost productivity and per capita economic growth. Immigrants appear to innovate more than natives not because of greater inherent ability but due to their concentration in science and engineering fields. With so much focus on the labor market, this critical issue—the relationship between immigration and long-run economic growth—is sometimes overlooked by researchers and in the public debate. We turn to this and other topics in Chapter 6.
As an aid for readers, Table 5-3 provides a summary comparison of the spatial (cross-area) studies and structural studies discussed in Sections 5.2 through 5.7. For each study, the author, the population sample analyzed, the methodology, and the key findings are listed.
|Study||Sample, Analysis Unit||Methods||Findings|
|Altonji and Card (1991)||U.S. men and women, 1970-1980, MSA||Spatial correlation, first differences, IV||1 percentage (pctg.) point increase in immigrant share lowers native wages by 0.3% to −1.2%; employment and participation effects negligible.|
|LaLonde and Topel (1991)||U.S. men, 1970-1980, MSA||Spatial correlation, first differences||Negative wage effects for new immigrants, effects die out for earlier immigrant cohorts, no effects for natives.|
|Card (2001)||U.S. men, 1990 cross-section; natives and earlier immigrants by MSA × broad skill/occupation/gender group||Spatial correlation, IV, analysis across cities and skill levels simultaneously to remove bias from omitted variables||Immigrants lower wages of less skilled natives—wages 0.99 pctg. points (male natives), 2.5 pctg. points (female earlier immigrants), 0 (other groups). 10% labor supply increase reduces employment rate 2.02 pctg. points (male natives), 0.81 pctg. points (female natives), 0.96 pctg. points (male earlier immigrants), 1.46 pctg. points (female earlier immigrants).|
|Cortés (2008)||U.S. men and women, 1980-2000, MSA||Spatial correlation, IV, country||Low-skilled immigrants don’t affect native wages overall. Previous immigrant and Hispanic wages lowered (1-1.5%).|
|Peri et al. (2014)||U.S. city × period (periods: 1990-2000, 2000-2005, 2005-2010)||Estimated H-1B–driven rise in STEM workforce, based on 1990 foreign STEM workforce by city and sending country and national-level distribution of H-1B visas by sending country||1 pctg. point increase in foreign share in STEM workers raises native STEM wages 7-8%.|
|Borjas (2003)||Education level × experience level × U.S. census survey, 1960-2000||Number of foreign-born workers in each education-experience-year group||~10% migration-induced labor growth in 1980-2000 cut wages for native non-high school completers 8.9%.|
|Camarota (1998)||Cross-section (1991) for U.S.||Wages of all workers||Wages: −0.5% overall; wages for workers in low-skilled occupations −0.8%.|
|Card (2001)||Cross-section (1990), IV for U.S.||Relative wages and employment of low-skilled natives||No effect on relative wages, small negative impact on relative employment.|
|Card (1990)||Miami men and women, 1980-1985||Spatial correlation; measured impact of increase in low-skilled labor supply shock associated with Mariel boatlift||No effect on wages or unemployment of unskilled workers.|
|Dustmann et al. (2005)||UK men and women, 1983-2000 (pooled cross-sections)||Spatial correlation, IV methods; first-differences with IV (1983-2000); participation rate, (un)employment rate, and hourly wages of the working population by education||Immigration has statistically insignificant effect on wage of each skill group.|
|Dustmann et al. (2013)||UK men and women, 1997-2005||Spatial correlation by wage percentile, IV method||Immigration lowers wages at 5th and 10th percentiles, raises average and above median wages.|
|Clemens (2013)||58 employment offices × 66 months, North Carolina, Feb. 2005-May 2011||Great Recession–caused unemployment jump, 2008-2009||Even after total unemployed in studied counties rose from 283,000 to 490,000, and with 6,500 job openings, only 7 native workers took and held farm jobs for the 2011 season—the rest were filled by migrants.|
|Study||Sample, Analysis Unit||Methods||Findings|
|Smith (2012)||U.S. youth and adults||Spatial correlation, IV||10% increase in immigrants with high school degree or less reduced average number of hours worked by 3-3.5% for native teens; less than 1% for less educated adults.|
|Kerr and Lincoln (2010)||U.S. cities × year, 1995-2007||Estimated number of H-1B holders in a city, by ethnicity, based on national-level H-1B ethnic breakdown and number of H-1B applications in 2001-2002||Among top quintile of cities in H-1B dependence, 10% increase in national H-1B population associated in same year with 6-12% increase in patent filing by people with Indian or Chinese names and 0-2% rise overall.|
|Peri (2012)||U.S. states × U.S. Decennial Census, 1960-2006||Distance to Mexican border; estimates of migrant stocks based on 1960 stocks by state and sending country; national-level growth rates by sending country||Immigration increases productivity (output per units of labor and capital input).|
|Borjas and Doran (2012)||U.S. and Soviet mathematicians who published in 1970-1989||Arrival of ~336 Soviet émigré mathematicians in U.S. just after collapse of Soviet Union||American mathematicians in subfields with active émigrés were published and cited less after 1992 and more likely to leave profession, indicating zero-sum displacement by new immigrants.|
|Moser et al. (2014)||166 chemistry subfields × year, U.S., 1920-1970||Starting in 1933, arrival of 26 Jewish émigré chemists from Nazi Germany & Austria, distinguishing their pre-departure subfields from ones with active German/Austrian researchers who did not leave||American inventors in subfields with émigrés recorded an extra 170 patents/year in 1933-1970 in total, 70% over pre-1933 level.|
|Structural (Aggregate Production Function) Approach|
|Grossman (1982)||1970 Decennial Census data (native, second generation, and foreign-born workers)||Uses cross-section data to estimate a trans-log production function to compute elasticities of substitution between immigrants and natives||Second generation and foreign-born workers are substitutes for 3rd generation and higher native-born.|
|Borjas et al. (1997)||U.S. 1980-1995, men and women||Applies estimated substitution elasticity of low-skilled for high-skilled workers to immigrant share||Wage elasticity is −0.322; immigrants lowered wages of high school dropouts relative to high school grads by 4.8%.|
|Borjas (2003)||U.S. men, 1980-2000||Nested production function, IV methods||Immigrants lower wages of dropouts by 8.9% and college graduates by 4.9%.|
|Borjas (2014a)||Education level × experience level × U.S. census survey, 1960-2010||Nested production function, IV methods; number of foreign-born workers in each education-experience-year group||10.6% migration-induced labor growth in 1990-2010 cut wages for native non-high school completers 6.2% (no capital adjustment) or 3.1% (after full capital adjustment).|
|Orrenius and Zavodny (2007)||Panel model with IV (1994-2000) (for U.S.)||Wages of natives by occupation groups||Wages of low-skilled natives negatively impacted by immigration (-0.26%); no effect for more-skilled labor.|
|Ottaviano and Peri (2012)||Education level × experience level × native/immigrant × U.S. census survey, 1990-2006||Nested production function, IV methods; total person-weeks of work in each education-experience-nativity-year group||Small effects on wages of dropouts. Disaggregated, a 10% migration-induced labor growth 1990-2006 raised wages for native non-high school completers 1.7%, cut them for foreign-born 8.1% (both after full capital adjustment).|
|Borjas (1987)||1980 Decennial Census data (white, black, Hispanic, and Asian natives and immigrants)||Generalized Leontief||Immigrants have small effects on native-born but sizable impact on earnings of immigrants themselves.|
This appendix explains the calculations behind the magnitudes of immigrants’ impact on wages reported in Table 5-2. Some papers report the impact of increasing the share of immigrants in the population or labor force by one percentage point, some of increasing the ratio of immigrants to natives by one percentage point, some of increasing immigrants by an amount that would increase the labor force (including natives) by 1 percent, and some the impact of particular episodes of immigration. The goal is to report what each paper implies about the percent change in wages in response to immigration that increases the labor force by 1 percent, , where w is the wage and L the labor force, and j indexes the unit of observation, which may be the labor force of a state, an occupation, or a skill cell or education group.
All papers use the log wage, logwj, as the dependent variable. In several papers, the independent variable is logkj, where , the share of employment or the labor force or population that is of education or occupation type j, a share which may be rewritten as a reminder that L is composed of immigrants N and natives M. The dependent variable is instrumented with predicted immigration, making the coefficient on kj, θ, the effect of a change in the share that is due to immigration: . For such cases θ is re orted bolded in the table
In other papers, the dependent variable is , the share of immigrants in the labor force, occupation, education group, or skill cell j. In this case, , where θ is the coefficient on mj.
In yet other papers, the dependent variable is , the ratio of immigrants to natives in the labor force or education group j. In this case, , where θ is the coefficient on mj.
To make comparable the magnitudes in papers whose dependent variable is pj or mj, one must choose a value of pj. The ideal for any given paper would be the average pj in the paper’s sample. However, the panel evaluated at a common pj value to ensure that the magnitudes of effects across papers do not differ simply because the average immigrant density differs across papers. For papers seeking to estimate the impact of all immigrants, we set p = 0.126, which is the immigrant share of the labor force in 2000. So 1 − p = 0.874; magnitudes calculated in this way are underlined in Table 5-1. Below, we explain how we treated results from each paper that can be rendered comparable in this way, and how we treated results from papers assessing particular episodes, including the structural results. While the implicit value of pj for the structural papers is close to 0.126, the results from other particular episodes may implicitly or explicitly be evaluated at a different pj.
- Independent variable is p.
- θ = −1.2 for all native-born high school dropouts. The most negative coefficient is θ = −1.9 for black native-born male high school dropouts (Altonji and Card, 1991, Table 7.7; Table 7.8, row 6, final column).
- If the number of immigrants rises sufficiently to raise labor supply by 1 percent, native-born high school dropout wages fall (0.874)(−1.2) = −1.0%; native-born black male high school dropout wages fall (0.874)(−1.9) = −1.7%.
Borjas (2003) Nonstructural Estimation
- Independent variable is p.
- θ = −0.637 for native men (own-wage coefficient; Table III, row 3 column 2, in Borjas, 2003). The impact on women was not studied.
- An increase in the number of immigrants sufficient to increase the labor force by 1 percent reduces wages by (0.874)(−0.637) = −0.56%.
- Borjas (2016b) reports on p. 27 that wages of (non–Hispanic) dropouts fell 10-30 percent as a result of the Mariel boatlift.
- His Table 2, p. 49, indicates that the boatlift increased the share of Marielitos among high school dropouts from 0 percent (by definition) to 17.5 percent (column 3). The denominator (column 1) appears to be all high school dropout workers including Hispanic non-Marielitos.
- Columns 1 and 2 of Table 2 (and the text on p. 27) indicate that the boatlift increased the labor supply of high school dropouts by 21 percent, assuming column 1 is indeed all dropout workers.
- Since a 21 percent increase in high school dropout labor supply due to immigration reduced wages 10 to 30 percent, a one percent increase reduces wages by −0.48 percent to −1.43 percent.
- Peri and Yasenov (2015) report in their Table 4 that wages rose 4.5 percent immediately after the Mariel boatlift.
- They report on page 7 that the Marielitos increased the number of high school dropouts by 15-18 percent.
- This means that a 1 percent increase in labor supply increased wages by 0.25-0.30 percent.
- Independent variable is log kj (log fj in Card’s notation).
- θ = −0.099 for native men, +0.063 for native women (Card, 2001, Table 10, lower panel).
- If the number of immigrants rises sufficiently to increase labor supply by 1 percent, native wages fall by θ (this is the local average treatment effect [LATE] interpretation of instrumenting the log share of a group in the labor force with immigration to the group).
- Independent variable is p (own-wage coefficient) Δ p or .
- θ = –0.237 (for Δ p; Table 2, 2nd row last column); θ = –0.124 (for ; Table 2, 3rd row last column).
- If the number of immigrants rises sufficiently to increase labor supply by 1 percent, native wages fall (0.874)(−0.237) = −0.21% (for Δ p; for , the magnitude may be read directly from the coefficient θ, so −0.12%).
- Independent variable is log kj.
- θ = −0.05 and is insignificant for all native dropouts (Cortés, 2008, Table 8). There are a variety of θ for other native groups (Table 10).
- If the number of immigrants rises sufficiently to increase labor supply by 1 percent, native high school dropout wages fall by θ (this is the LATE interpretation of instrumenting the log share of a group in the labor force with immigration to the group).
- Independent variable is (own-wage coefficient).
- θ = −2.0 (own-wage coefficient; Llull, 2015, Table 7 bottom row).
- If the number of immigrants rises sufficiently to increase labor supply by 1 percent, native wages fall (0.874)(−2.0) = −1.75%
- Independent variable is mj.
- The calculations below ignore the fact that the regressions in Monras (2015) also control for logLj (theory suggests controlling for logNj, a variable with a coefficient of 0.05).
- Controls for GDP and the size of the labor force mean that implicitly capital is held fixed. However, because the study looks at the short-run effect of an unexpected inflow, these controls may matter little.
- θ = −0.75 (Table 4 column 7, author’s preferred coefficient).
- p = 0.055
- If the number of immigrants rises sufficiently to increase labor supply by 1 percent, native wages fall by (−0.75)/(1 − .055) = −0.79%. Alternatively, one can rely on theory and an approximation based on small p to read the effect directly from the coefficient, to obtain −0.75 percent (for this to be correct, logNj must be controlled for in the regression).
Structural Wage Effects in Table 5-2
- The wage-effect values shown in the second column of the “Structural Studies” section of Table 5-2 simulate the effect of immigration from 1990 to 2010, when the share of immigrants in the labor force rose from 9.3 percent to 16.4 percent—a 7.1 percentage point rise.
- We focus on the most negative and the least negative (or most positive) scenarios in Table 5-1, ignoring the result that the long-run effect on all workers (Scenarios 1 and 3) is zero, since this is an assumption embedded in the model.
- For all natives in the short run, wage impacts in Table 5-1 range from −3.2 percent (Scenario 1) to −2.6 percent (Scenario 2).
- For all natives in the long run, wage impacts in Table 5-1 are 0.5 percent (Scenario 4) or 0.6 percent (Scenario 2).
- For native dropouts in the short run, wage impacts in Table 5-1 range from –6.3 percent (Scenario 1) to −2.1 percent (Scenario 4).
- For native dropouts in the long run, wage impacts in Table 5-1 range is from –3.1 percent (Scenario 1) to 1.1 percent (Scenario 4).
- So a 1 percentage point increase in the immigrant share would imply reductions in wages of these amounts divided by 7.1 (the observed percentage point rise in the immigrant share between 1990 and 2010), or:
- −0.37 to −0.45 percent for all natives in the short run
- +0.07 to +0.08 percent for all natives in the long run
- −0.89 to −0.30 percent for native dropouts in the short run
- −0.44 to +0.15 percent for native dropouts in the long run
- If the number of immigrants rises sufficiently to increase labor supply by 1 percent, native wages would fall by 0.874 times the values in the previous bullet. In Table 5-2, values for the wage effect are rounded to one decimal place, which results in several scenarios being reported as having the same effect.