Spatial Sorting of Immigrants Across Urban and Rural Areas in the United States: Changing Patterns of Human Capital Accumulation Since the 1990s
利用1990、2000年人口普查和2005年美国社区调查数据，分析移民按教育程度在都市与非都市地区的空间分选模式，发现高技能移民更倾向都市，但随时间推移分选减弱，且收入溢价与区位选择相关。
Immigration policies in the United States influence the size and composition of the immigrant population, but once in the United States, migrants can freely choose where to settle.1 Traditionally, immigrants have located in large gateway cities such as New York, Chicago, Los Angeles, and Miami. In recent years, however, we have seen a growing number of newcomers choosing to live in nonmetropolitan areas (Kandel and Cromartie 2006). The influx of immigrants provides opportunities and challenges for local economies, including highly debated impacts on labor market outcomes (Borjas, Freeman, and Katz 1996), productivity improvements, increased innovation, ethnic entrepreneurship, and improved allocative efficiency (Poot 2007). It also places new demands on housing, health care, and education, which, in the absence of federal policies and support for immigrant integration, carry with them financial burdens for local and state governments. How drastically these impacts will manifest themselves at the local level depends on the number and the characteristics of the newcomers, in particular their level of education. Recent studies suggest that locational choices differ by educational attainment level, and immigrants contribute significantly to the resulting disparities between human capital accumulation in highly urbanized places and human capital depletion in rural areas (Waldorf 2008). To understand the spatial sorting of immigrants by educational attainment, this article focuses on two sets of questions. The first set deals with immigrants' locational choices. Do immigrants' locational preferences vary by education? Moreover, do the preferences for locating in urban and rural areas vary across time and space? Many rural counties have experienced a steady decline of their native-born labor force, creating a gap filled by low-skilled immigrants. In particular, the influx of immigrants into interior states is a relatively recent phenomenon that is predominantly fueled by low-skilled immigrants. The second set of questions focuses on immigrant incomes in rural versus urban areas. We expect urban incomes to exceed rural incomes. However, what is the magnitude of the urban location premium? Most importantly, do the premiums vary systematically over time, across space, and by educational attainment, and do they correspond to the revealed locational preferences? The analysis is carried out using data from the 1% Public Use Micro Samples (PUMS) and the 2005 American Community Survey (ACS). We use discrete choice models to estimate immigrants' locational preferences and a continuous regression model to estimate urban location premiums across space and time. Between 1990 and 2000, the foreign-born2 population in the continental United States increased from 20 to 31 million, accounting for 8% of the total population in 1990 and 11% in 2000. Traditionally, the foreign-born selected predominantly into metropolitan locations, where they accounted for 9.3% of the population in 1990 and 12.7% in 2000. However, the relative change was even higher in nonmetropolitan areas, where the foreign-borns' share increased from 1.7% to 2.9% of the total population. These trends continued during the first decade of the twenty-first century. Compared to the native-born population, migrants from abroad have a larger share of persons without a high school degree but also a larger share of persons with a graduate or professional degree. This almost U-shaped distribution is due to U.S. immigration policies that favor family unification but also make special provisions for high-skilled workers. Between 1996 and 2005, the annual share of persons obtaining permanent resident status (“green card”) on the basis of employment varied between 9% and 22% only, whereas a majority of green card recipients were relatives of U.S. citizens or permanent residents. These policy-induced channels of immigration are likely to perpetuate the aggregate characteristics of immigrants. They also have a clear spatial manifestation because newcomers will locate where their immigration sponsors reside. Consequently, the immigrant population at the local level is often quite homogeneous. Locational choices are realized through migration. Households make locational adjustments due to changing needs and preferences. Not surprisingly, age, sex, and marital status are salient predictors of migration propensities and destination choices. Migration and the associated locational choices are, however, also a form of human investment (Sjaastad 1962), and empirical support for educational selectivity is abundant. Compared to the less-educated population, college graduates are more likely to move to take advantage of economic opportunities (Krieg 1991) and to move long distances (Kodrzycki 2001). Moreover, they are attracted to large urban areas where the human capital stock itself becomes an amenity that attracts additional college graduates (Gottlieb and Joseph 2006) and abundant consumer goods and services, entertainment, recreation, education, employment, and social opportunities act as additional strong pull factors (Glaeser 1998; Florida 2002). While these regularities hold true for the population at large, we know little about migration selectivity and rural/urban preference differentials by educational attainment of the immigrant population. Instead, the literature has often treated the immigrant population as homogeneous and focused on, for example, immigrants' persistent preference for gateway cities (Hempstead 2007), their impact on the locational choices of the native-born population (Frey 1995), and spatial sorting across urban neighborhoods (Florax, de Graaff, and Waldorf 2005). Similarly, little is known about the differential economic impact of high-skilled and low-skilled immigrants in rural versus urban areas. In fact, so far the literature has by and large neglected immigrants in rural areas. We use a pooled sample of 75,938 foreign-born persons taken from the 1% PUMS of the 1990 and 2000 U.S. Censuses and the 2005 ACS. At the time of the survey, the selected persons were: at least twenty-five years old and in the labor force, had stayed in the United States for at most ten years, resided in the contiguous United States, and did not have U.S. citizenship. At the time of immigration, the selected persons were at least sixteen years of age and thus completed most of their education outside of the United States. If more than one person in a household met these selection criteria, one person was chosen at random. We used a series of binary choice models with the dependent variable specified as the immigrant's place of residence, categorized as metropolitan versus nonmetropolitan. Overall, 85% of the sampled immigrants live in a metropolitan area. The pivotal explanatory variables are educational attainment, time, and space. The education variable stratifies between three levels: no high school diploma, high school diploma (and possibly some college), and at least a bachelor's degree. Fixed effects that differentiate between the three survey years capture temporal trends. Spatial variations are controlled by location fixed effects that signify the place of residence in one of four regions.3 The regionalization singles out states with major gateway cities, that is, California, Texas, Florida, Illinois, and New York, with its neighboring states New Jersey and Connecticut. In total, 78% of the sampled immigrants reside in the gateway states. The remaining states are assigned to three regions, with “West” and “East” each accounting for slightly more than 6% of the sample, and immigrants from the “Interior” making up almost 9% of the sample (see table 1 for the assignment of states to regions). In addition, we control for a battery of personal characteristics. They include traditional predictors of migration and locational choices (age, sex, marital status, race, and ethnicity) as well as information that comes into play when dealing with an immigrant population, namely the country of birth, length of stay in the United States, English language proficiency, and linguistic isolation. To estimate metropolitan location premiums, we first estimate a continuous regression model of personal income and subsequently compare estimated incomes for select immigrant groups in metropolitan versus nonmetropolitan areas. We hypothesize that metropolitan location premiums vary jointly by space, time, and education. The first two columns of table 2 show the results for two logit models of locational choice, with the observations weighted by “person weights” provided by the Census Bureau. Surprisingly, some key demographic predictors of migration and locational choice (age, sex, ethnicity) do not influence immigrants' location choice and neither do variables describing the immigration context (length of stay in the United States, English proficiency, country of origin). What is important, however, is the immigrants' marital status (married immigrants are less likely to reside in a metropolitan than nonmarried immigrants) and race (whites prefer metropolitan areas more strongly than other races). The effect of education is allowed to vary across time in Model (1) and region in Model (2).4 In both models, education, time, and region emerge as powerful predictors of locational choice, and the interplay between the three variables yields a complex mosaic of immigrants' spatial sorting. Figure 1 shows the estimated probabilities of residing in a metropolitan area by educational attainment level over time and across the four regions. The general trends suggest that the chance of locating in a metropolitan area rises with increasing educational attainment level, grows over time, and is higher in gateway states than in the West, East, or interior states. There are some deviations, however. In 2000, those without a high school degree were slightly more likely to reside in a metropolitan area than those who completed high school. The time trend is nonmonotonic for college graduates. Their chances of residing in a metropolitan were highest in 2005 but slightly decreased between 1990 and 2000. The regional variation, however, is consistent. Independent of education and time, the chances of living in a metropolitan area is topped by the gateway states and followed closely by the West and is smallest in the interior states. Estimated conditional probabilities of residing in a metropolitan area by educational attainment level for three different years (1990, 2000, and 2005) and four different regions (Gateway, East, West, and Interior) The estimated gaps in the probability of locating in a metropolitan area between educational attainment groups are not constant over time or across regions. In 1990, the spatial sorting by educational attainment level is quite strong. This is especially the case for the interior states, where the probability of a highly educated immigrant (with characteristics as specified in figure 1) residing in a metropolitan area is more than 1.5 times higher than for a comparable immigrant without a high school education. In contrast, in the gateway states, the probability is less than 1.2 times higher for the highly educated immigrant group than for the poorly educated group. By the year 2005, the gaps had diminished substantially in all regions. In fact, in the West and in the gateway states, there is barely any difference in the probabilities of locating in a metropolitan area by educational attainment level, with the estimated chances of locating in a metropolitan area exceeding 90% for all groups. In the interior states and to a lesser extent in the East, the spatial sorting persists in 2005, but it has become much weaker than in 1990. Models (3) and (4) address the spatial and temporal disparities in migrants' personal income between metropolitan and nonmetropolitan areas, differentiated by educational attainment level. The demographic predictors have a substantial influence on immigrants' income. Income increases with age and is positively affected by being married, male, white, and non-Hispanic, and by higher education levels. Personal incomes also rise as immigrants extend their sojourn in the United States, while they are negatively influenced by linguistic isolation and poor English language proficiency. The country of origin is also influential, with those from developed countries having a higher income than others. Models (3) and (4) yield similar estimated metropolitan location premiums, defined as the ratio of estimated personal incomes of immigrants in a metropolitan area to those of corresponding immigrants located in a nonmetropolitan area. In general, immigrants' incomes in metropolitan areas exceed those in nonmetropolitan areas, yielding a metropolitan location premium greater than one for all groups. However, the magnitude of the metropolitan location premium varies substantially and significantly by educational attainment level, over time, and across the four regions (figure 2). In 1990, the premium increases with increasing educational attainment level in all regions. For all educational attainment groups, the premium in the gateway states exceeds those in the other regions. The premium is particularly small for the least educated in the interior states. This pattern corresponds closely with the locational preferences derived from the locational choice model. That is, the preference for metropolitan areas is particularly strong for those groups and regions that have the highest metropolitan location premium. In 2000, the metropolitan location premium for the least educated immigrants had slightly increased compared to 1990 but became substantially smaller for high school and college graduates. This again corresponds with the earlier discussed locational preferences in 2000, when the least educated were somewhat more likely to reside in a metropolitan area than ten years earlier, and those with more education had not experienced a shift in location patterns relative to 1990. The most drastic change occurred by 2005, when the group of least educated emerged as the one with the largest metropolitan location premium in all four regions, matching the group's sharp increase in metropolitan location probability. In contrast for college and high school graduates, the 2005 metropolitan location premiums also increased compared to 2000, but they did not reach the 1990 levels. Nevertheless, both groups experienced steep increases in their chances of residing in a metropolitan area, especially in the East and in the interior states. Estimated metropolitan location premiums for three educational attainment levels (less than high school, high school, and bachelor or higher) for three different years (1990, 2000, and 2005) and four different regions (Gateway, East, West, and Interior) The results suggest that spatial sorting of immigrants by educational attainment is a persistent feature of immigrants' location choice, with highly educated immigrants being more likely to reside in metropolitan areas than less-educated immigrants. These location patterns may exacerbate the already stark disparities between brain-rich metropolitan and brain-poor nonmetropolitan regions as metropolitan areas receive more than their fair share of highly educated immigrants. However, our results also show that immigrants' spatial sorting by educational attainment level weakened over time. By 2005, both high-skilled and low-skilled immigrants show a very strong preference for metropolitan areas throughout the United States, and the probability of residing in a metropolitan area is quickly approaching unity for all educational groups in all regions. As a corollary, we can conclude that the growing immigrant communities in nonmetropolitan areas, especially in the interior states, are not due to changing locational preferences but are more likely due to the growing size of the immigrant population overall and compositional shifts toward those more prone to locate in a nonmetropolitan area. As more immigrants select into interior states, where the preference for a nonmetropolitan location is still persistently stronger than elsewhere and where, until recently, very few immigrants settled, immigrant communities in nonmetropolitan places can indeed grow quite quickly. This may lead to new challenges for local policy makers if the affected communities have little prior immigration experience and may be exacerbated if the community is small so that immigrant communities are quite visible. Our results also suggest that spatiotemporal variations in metropolitan preferences among the less educated correspond closely to the magnitude of the metropolitan location premium. This is not always the case for the better educated though, thus suggesting that factors other than income fuel their strong metropolitan preference. Two salient issues should be addressed in future research. First, the regional division used in this analysis is broad, and the observed trends may not necessarily represent trends at a smaller spatial scale. Second, methodologically, the analysis should be extended to explore and tackle the possible endogeneity of locational choices and income.