Population Change and Changing Educational Attainment of Ethnic Groups in the United States,

Popul Res Policy Rev DOI 10.1007/s11113-011-9204-7 Population Change and Changing Educational Attainment of Ethnic Groups in the United States, 1980 2000 Franklin D. Wilson Uzi Rebhun Salvador Rivas Received: 22 May 2010 / Accepted: 21 March 2011 Ó Springer Science+Business Media B.V. 2011 Abstract This study assesses the effect of population change on decade changes in the educational attainments of country of origin populations in the United States. Our data are derived from decennial censuses, NLMS, the World Bank, and INS. We find that changes in the share of country of origin populations with one or more years of post-secondary schooling are associated with selected components of population change during the 1980 1990 and 1990 2000 decades. The specific components include survivors during the decade, in-migration, and emigration of the foreign-born. Likewise, intra-generational mobility is found to be an important determinant of changes in educational attainment. The discussion addresses limitations of the data and suggests directions for future research as well as policy implications. Keywords Country of origin Education Generational change Migration Multi-level analysis Survivors United States Collaboration for this project began when Uzi Rebhun was a Visiting Fellow and Salvador Rivas was a Post-Doctoral Fellow at the Center for Demography and Ecology (CDE), University of Wisconsin Madison during the 2004 2005 academic year. A previous version of this paper was presented at the annual meeting of the Population Association of America, Los Angeles, March 30 April 1, 2006. This paper also benefited from comments and suggestions by colleagues and students at the CDE seminar, UW Madison; and at a seminar presented at the Department of Sociology and Anthropology, The Hebrew University of Jerusalem. The authors thank the IPUMS Project at the Minnesota Population Center for providing access to the 1980, 1990, and 2000 U.S. censuses. F. D. Wilson University of Wisconsin Madison, Madison, WI, USA U. Rebhun (&) The Hebrew University of Jerusalem, Jerusalem, Israel e-mail: rebhun@mscc.huji.ac.il S. Rivas University of Luxembourg, Luxembourg, Grand Duchy of Luxembourg

F. D. Wilson et al. Introduction Most research on educational attainment and changes therein focuses on the roles of individual characteristics and area-specific contexts in shaping the educational experiences of individuals as they move through their life course (Mare 1995; Gamoran 2001; Entwisle et al. 2005). These studies are concerned mainly with such determinants as cognitive ability (Herrnstein and Murray 1994; Gamoran 2001), demographic characteristics including age and gender (Myers 2005), parentage or family and socioeconomic background (Featherman and Hauser 1978; McLanahan 1985; Gamoran 2001), place-specific opportunities such as those provided by neighborhood and school environment, and public policies (Leventhal et al. 2005). Special attention has been paid to the educational attainments of immigrants and their descendants for the assessment of their incorporation into mainstream society (Alba and Nee 2003; Hirschman 1983; Jensen 2001; Portes and Zhou 1993). These studies provide important insights into the origins and sources of differences in educational attainment among individuals and groups. The existing theoretical and empirical literature on the demography of education, however, suffers from the lack of an important complementary dimension: the macro processes of population change that are responsible for shifts in the educational attainment distribution of the U.S. population as a whole. Hence, the analysis in this paper describes and accounts for relative changes in the educational attainment distribution of the U.S. population stratified by country of origin of immigrants and their descendents. The aim is to advance previous work by giving greater attention to the contribution of population change to shifts in educational distribution. We focus specifically on the potential role of intra- and intergenerational changes and changes in the size of the foreign-born population on the changing educational attainment of the U.S. population. To this end, we focus on the period from 1980 to 2000. The availability of large samples of individual ethnic (ancestry) populations with varying proportions of immigrants and descendants provides a unique opportunity to estimate the sources of change in educational attainment. To pursue this objective, we employ a synthetic age-cohort approach to inter-censual estimates in order to decompose changes in educational attainment to components reflecting: (1) intergenerational change, largely through the impact of mortality on age cohorts distinguished by nativity; (2) intra-generational change, via educational mobility; (3) international migration (immigration and emigration); and (4) emigration of the native-born. We calculate estimates of the impact of each of these components on decade changes in population for five-year age groups stratified by country of origin and gender. We then estimate multivariate models to assess the impact of population change and origin-country characteristics on changes in educational attainment. The reminder of this article is structured as follows: Background and Hypotheses section provides a background review of possible ways in which sources of population change may contribute to macro-level changes in educational attainment; and identifies specific hypotheses to be evaluated. Data Sources and Sample Selection section reviews the data sources, sampling selection procedures, and operationalization of relevant variables to be used in subsequent analyses. Results section focuses on the results of descriptive and multivariate analyses.

Population Change and Changing Educational Attainment of Ethnic Groups Finally, Discussion section summarizes the findings and discusses them further with respect to their theoretical and policy implications. Background and Hypotheses The educational distribution of the U.S. population is dynamic, constantly changing in response to demographic, socioeconomic, cultural, and political forces that alter the educational aspirations and attainment of individuals as they move through their life course. Specific identifiable factors that affect the educational distribution of the population include intra- and intergenerational mobility, mortality, international migration to and from different places of origin, and period-specific socioeconomic and public policy changes that have varying impacts on age cohorts. From 1940 to 1990, the average educational attainment of the U.S. population changed dramatically (Fischer and Hout 2006). The median educational attainment of the age 25? population increased from 8.7 to 12.8 years and the percentage of those completing high school or more increased from 24.5 to 77.6 (U.S. Bureau of the Census 2008). An important component of these changes is intergenerational in character, with the educational attainment of offspring exceeding that of their parents (Featherman and Hauser 1978; Mare 1995). Mortality plays a direct role in facilitating intergenerational change. Thus, as older generations with lower educational attainment levels exit the population via mortality, the subsequent average level of educational attainment of the population increasingly mirrors that of the younger generation. Thus the first hypothesis to be evaluated is: Changes in the representation of individual age cohorts should increase the average educational attainment of the U.S. population, because the impact of mortality will be greater on the cohorts with the lowest average education. In addition, mortality is selective with respect to education, i.e., Poor and poorly educated people still die at higher rates than those with higher incomes and better education, and the disparity increased between 1960 and 1986 (Pappas et al. 1993; see also Sorlie et al. 1995); this also raises average education levels. Thus, the second hypothesis simply states that higher educational attainment is associated with lower mortality independent of age, and independent of generational change. Average educational attainment may also change as a result of intra-generational change. The average schooling of the members of an age cohort observed at two or more points in time may reveal marked differences as a consequence of intragenerational mobility. The magnitude of the differences depends strongly on the age at observation: the younger the age, the greater the observed differences. Schooling, although continuous through the completion of secondary school, may be disrupted afterwards for one or more years. This may occur for a number of reasons, including household responsibilities (child-bearing and child-rearing, caring for an older relative, etc.), financial circumstances, military service, and institutionalization. In addition, even if no disruption occurs, the pursuit of post-college schooling may increase the age of school completion. While most individuals complete postcollege education by the age of 30, some may not do so until later years (Mare 1995). Since intra-cohort changes in educational attainment simply reflect

F. D. Wilson et al. progression toward completion of education, we hypothesize that its impact on changes in average education attainment will be positive. Period-specific changes may affect the educational distribution of a nation in several ways. One of the most obvious is through shifts in the demand for labor with specific skills and education. For example, the shift in demand in favor of human capital/cognitive skills starting in the late 1970s dramatically altered the educational requirements for securing good-paying jobs in an economy that had become more service-oriented and that functioned in an increasingly global marketplace. Differentials in returns to schooling widened in favor of workers with college degrees. Younger age cohorts were better positioned to respond to these changes because their schooling was not yet completed and/or because their occupational careers had not reached a mature stage that would lock them into a particular trajectory. Second, since education is to a certain extent a public good, government policies altering opportunities for schooling such as veterans benefits; government-sponsored grants, fellowships, and loans; changes in education-based criteria for admission to the U.S.; changes in curricula and requirements for school completion (e.g., the No Child Left Behind Act); and tax polices may expand the enrollment of nontraditional students and alter the criteria and conditions for degree completion. Since there have been no reported changes in average educational attainment associated with public policy prescriptions, we present no hypothesis for testing. International migration may affect a country s average educational attainment via the size, duration, and educational composition of migration flows. A migration flow that is large in size and persists for a decade or more may increase the origin share of the host population; which in turn, through the educational composition of the flow, affect the educational distribution of the host population. This may increase or decrease average educational attainment, depending on whether the educational composition of the flow is higher or lower than that of the host population. One would expect higher or lower educational attainment at origin to be reflected in the composition of the migration flow, unless the stream is selective with respect to education in response to differential opportunities at destination. Currently, migration flows to the U.S. are bi-modal, attracting individuals with both limited and high educational attainment (Chiswick and Sullivan 1995; Chiswick and DebBurman 2004; Massey 1999; Feliciano 2005). Notably, however, a sustained flow linking an origin and destination is rarely one-directional. Invariably, a counter-flow develops as some individuals experience adjustment problems at destinations, while others return to origin after having attained a goal that initially motivated them to migrate to a destination or due to family considerations. We make no explicit prediction regarding the impact of immigration on changes in average educational attainment, largely because the average educational attainment of the individual countries of origin from which migrants are drawn vary considerably. However, we do include average education attainment for country of origin as a control. In this regard, it is reasonable to assert ceteris paribus, that education selection from origin could raise (if migrants are highly educated) or lower (migrants are less educated) educational attainment at destination. As for emigration, we hypothesize that the exit of the U.S. foreign-born will increase average education attainment, because those who leave have lower education, and

Population Change and Changing Educational Attainment of Ethnic Groups came to achieve short-term goals or experienced adjustment problems. On the other hand, emigration of the native-born will decrease average educational attainment, because on average and depending on the size of the stream, exits are more positively selected. Our data, described below, allow us to measure presumed components of change effects quite directly. In the Discussion section we link the working hypotheses previously stated with the empirical results. Given that this is, to the best of our knowledge, a pioneering analysis of this sort, the robustness of the hypotheses have important theoretical and policy implications and may lay the foundation for future research of the role of macro-level population processes in changes in educational attainment. Data Sources and Sample Selection This study uses data from decennial censuses, the National Longitudinal Mortality Study (NLMS), the World Bank, and the Immigration and Naturalization Service (INS) to analyze decade changes in population and educational attainment during the 1980 1990 and 1990 2000 periods. Specifically, we use Public Use Microdata Sample (PUMS) files culled from the 1980 (1 and 5%), 1990 (1 and 5%), and 2000 (1%) decennial censuses. From these data, we generate population counts for five-year age cohorts stratified by ethnicity, ancestry, sex, nativity, immigration status, and educational attainment. We restrict the data to individuals who do not reside in group quarters. 1 The 1980 and 1990 samples are much larger than the 2000 sample for two reasons. First, since information on ancestry (or country of origin) is not available in the 2000 5%-sample, we cannot use it for our purposes. Second, because many of the immigrant/ethnic groups are of recent origin arrived in the United States since 1965 the larger sample sizes for 1980 and 1990 are necessary for an adequate sample size for these groups. Moreover, the 1% sample for 2000 is about 25% larger than that of the two previous censuses; it is sufficiently large for the purposes of this study. We use NLMS data to estimate mortality rates for five-year age groups during the 1979 1989 period stratified by ethnicity, sex, nativity, and educational attainment. The NLMS is a CPS-based data file with supplemental data from the National Death Registry identifying respondents in the CPS samples who died during 1979 1989. These data were subsequently merged with census-based tabulations on the basis of ethnicity, sex, age, nativity, and education as linking variables. After the merge, educational attainment was collapsed into two categories: individuals with less than 1 year of post-secondary schooling and those with one or more years of postsecondary schooling. Since more than three-quarters of the adult population of the U.S. population aged 25? have at least completed high school, we believe that one or 1 We exclude residents of group quarters because we cannot distinguish among residents of college dormitories, prisons, and other institutions. Most studies focus on the civilian non-group-quarter population and we follow this approach as well. Given the heterogeneity of this population, we assume that the inclusion of group-quarters populations in our estimates would have weakened the effects found. Moreover, such populations are typically less likely to provide information, thereby amplifying measurement error.

F. D. Wilson et al. more years of post-secondary schooling should provide more variation in educational attainment among country of origin groups in our sample for analytic purposes. Although the NLMS file does not report ancestry (country of origin), we were able to assign mortality indicators using ethnicity, which we then assigned to ancestry groups based on ethnic-group identification. The most important implication of this re-classification is that ethnic groups classified as belonging to the same racial category (i.e., non-hispanic white, black, Asian, or American Indian) are assigned the same mortality estimates. Finally, since the NLSM provides only estimates of deaths that occurred between 1979 and 1989, the survival probabilities applied to the 1980 and 1990 populations are the same. We admit that applying the 1980 1990 mortality schedule to 1990 2000 introduces an additional source of error, the magnitude of which is unknown. Thus, while decade changes in the probability of survival are not a factor responsible for observed differences, they are an important and necessary component of population change to include and adjust for. We assigned country of origin designation by following a particular order based on available information for ethnic/racial identity, country of birth, and ancestry. We first classify persons of Hispanic origin since they can be of any race, followed by non- Hispanic Asians, American Indians, persons of African Origin, and whites. We classified individuals by detailed Hispanic origin. If a specific origin was not identified, we used place of birth, followed (if necessary) by first ancestry mentioned. Next, we used the detailed information available for Asians (second order) and American Indians (third order), followed by place of birth and ancestry if necessary. For blacks, followed by whites, we relied on information for place of birth then first reported ancestry. The overwhelming majority of non-hispanic whites are of European/Middle-Eastern origin. Before we could apply this procedure to the 2000 census, we had to first create a single racial/ethnic classification based on a particular ordering of ethnic groups, from the most historically stigmatized to the least stigmatized (see Bobo et al. 2001). We assigned a single racial classification to individuals who identified themselves as being black (exclusively or selected more than one race in combination with black), followed by a single racial classification for Asians, then American Indians (if they indicated a tribal affiliation), and, finally, whites. While these steps captured most people in our samples, they did not capture everyone. In particular, for households in which individuals did not report a specific ancestry designation, if available, we assigned them to the ancestry reported by someone else in the household under the assumption that families and people that live together are likely to be of similar ancestry background. In cases where individuals reported multiple ancestries, we used only the one reported first. The 1980 PUMS has several multi-ancestry group classifications that we decided to break up for the purposes of standardizing these categories across the remaining PUMS data files. Finally, we assigned individuals for whom no ethnic classification was possible to the Other American category. This is a very heterogeneous category consisting primarily of persons who are native-born who chose not to identity themselves using a standard ethnic/racial identity (see Lieberson and Waters 1988). The reporting of ethnic/racial identity, and ancestry are problematic in most official data sources. Ethnicity and ancestry are not fixed attributes as each can

Population Change and Changing Educational Attainment of Ethnic Groups change as a result of individuals having multiple ethnic/country of origin identities, in which the salience of each may change over time. Whatever identity individuals may select reflect personal/family circumstances and/or societal forces which may favor one ethnicity or ancestry over another during a particular period in time (see Alba 1990; Lieberson and Waters 1988). Unfortunately, we have no way of correcting for inter-censual changes that result from changes in ethnic/ancestry identity, nor do we even have a way of estimating the magnitude of errors in ethnic/ ancestry classifications using extant aggregate data. We used data from the INS public use data files for 1980 1989 and 1990 1999 to obtain information by country of origin on visa declarations for the foreign-born population that reached the U.S. during a given decade. The visa declarations of particular interest included those related to family, employment, non-immigrant, refugees, and asylum-seekers. Finally, educational attainment data for the countries of origin included in our sample were obtained from the World Bank (http://devdata. worldbank.org/edstats/td10.asp) as developed by Barro and Lee (2000). Results Country of Origin and Educational Attainment: Bivariate Analysis Table 1 provides summary statistics on the average percentage of one or more years of post-secondary schooling for twelve major world regions including three major native-born American groups. These twelve places of origin were calculated by grouping 73 country of origin populations for 1980, 1990 and 2000. These percentages are standardized by age and sex for each decade, using the total population as a standard. The descriptive statistics reported in this table (as well as in Tables 2, 3 and 4) are presented as a convenient way to summarize the variation in educational attainment for the 73 country of origin grouping. Thus, Table 1 illuminates considerable variation among country of origin groups in educational attainment. The groups with the highest percentages of one or more years of postsecondary schooling across decades include (in declining order) Middle Eastern, Asian, European, and Canadian; the lowest rank includes Mexican, American Indian, Hispanic, and African American. These rankings are consistent with those reported by others (Hirschman 2001; Mare 1995). In general, we find substantial increases in the percentage of country of origin populations with one or more years of post-secondary schooling, particularly during the 1980 1990 decade. The native-born American groups (African American, American Indian, and Other American), followed by Mexicans, experienced the largest increases over the two decades. Notably, however, these changes, while they narrowed the differences between these groups and the others, did not appreciably alter the ranking of the groups. Individuals from Oceania began the 1980 1990 decade with the highest percentage of persons with post-secondary schooling but experienced the smallest amount of change in educational attainment during the 20-year period.

F. D. Wilson et al. Table 1 Percentage country of origin population with one or more years of post-secondary education Country of origin 1980 1990 2000 Decade change (2)/(1) (3)/(2) (3)/(1) (1) (2) (3) (4) (5) (6) European 32.91 48.87 58.12 1.49 1.19 1.77 Mexican 12.35 20.07 22.23 1.63 1.11 1.80 Hispanic 22.61 30.73 35.31 1.36 1.15 1.56 Caribbean 27.04 37.19 41.49 1.37 1.12 1.53 Middle Eastern 42.44 55.37 59.68 1.30 1.08 1.41 Sub-Saharan African 36.14 49.48 46.48 1.37 0.94 1.34 Asian 42.21 52.44 58.69 1.24 1.11 1.39 Oceanian 44.29 44.90 49.07 1.01 1.09 1.11 Canadian 32.79 45.69 55.76 1.39 1.22 1.70 African American 19.48 31.14 38.24 1.59 1.22 1.96 American Indian 17.65 28.31 37.27 1.60 1.35 2.22 Other American 20.83 29.61 39.35 1.42 1.32 1.89 Source: 1980, 1990, 2000 PUMS Population 15 years of age and over Percentages standardized by age and sex. Total population used as standard Table 2 Observed and expected population with one or more years of post-secondary education: 1980 cohorts by country of origin (percentage) Country of origin 1980 cohorts 1980 observed 1990 expected (age 25?) Ratios Age 15? Total Survivors only (2)/(1) (3)/(1) (1) (2) (3) (4) (5) European 32.91 34.04 33.94 1.03 1.03 Mexican 12.35 12.94 13.07 1.04 1.06 Hispanic 22.61 24.52 23.79 1.09 1.05 Caribbean 27.04 29.61 28.40 1.10 1.05 Middle Eastern 42.44 46.68 44.54 1.10 1.05 Sub-Saharan African 36.14 45.86 38.49 1.27 1.07 Asian 42.21 47.02 43.90 1.11 1.04 Oceanian 44.29 47.77 45.66 1.08 1.03 Canadian 32.79 36.92 34.24 1.13 1.04 African American 19.48 20.89 20.89 1.07 1.07 American Indian 17.65 18.62 18.62 1.06 1.06 Other American 20.83 21.60 21.60 1.04 1.04 Source: 1980 PUMS Column (3), Total includes survivors and foreign-born arrived during decade Percentages standardized by age and sex. Total population used as standard

Population Change and Changing Educational Attainment of Ethnic Groups Table 3 Observed and expected population with one or more years of post-secondary education: 1990 cohorts, by country of origin (percentage) Country of origin 1990 cohorts 1990 observed 2000 expected (age 25?) Ratios Age 15? Total Survivors only (2)/(1) (3)/(1) (1) (2) (3) (4) (5) European 48.87 50.75 50.72 1.04 1.04 Mexican 20.07 20.32 21.45 1.01 1.07 Hispanic 30.73 32.40 32.14 1.05 1.05 Caribbean 37.19 37.97 38.87 1.02 1.05 Middle Eastern 55.37 57.66 57.70 1.04 1.04 Sub-Saharan African 49.48 53.77 52.44 1.09 1.06 Asian 52.44 56.26 54.19 1.07 1.03 Oceanian 44.90 48.01 46.05 1.07 1.03 Canadian 45.69 50.27 47.67 1.10 1.04 African American 31.14 33.21 33.21 1.04 1.04 American Indian 28.31 29.93 29.93 1.06 1.06 Other American 29.61 30.86 30.86 1.04 1.04 Source: 1990 and 2000 PUMS Column (3), total includes survivors and foreign-born arrived during decade Percentages standardized by age and sex. Total population used as standard The period-specific changes in educational attainment observed for all groups in 1980 2000 reflect the influence of several factors. First, the entrance of new cohorts with above-average educational attainment at the bottom and the exit, through death, of less-educated cohorts at older ages increased the overall educational attainment of country of origin groups across the decades. Second, intragenerational mobility may be a factor. Recall that the youngest age cohort, 15 24, may have increased its educational attainment levels in the 10 (1980 1990, 1990 2000) and/or 20 (1980 2000) years after their first observation in the census, respectively. Third, the entrance and exit of individuals via international migration may modify average educational attainment. A final source of change is inter-census errors of various types, including enumeration and reporting errors (discussed below). Observed and Expected Educational Attainment Tables 2 and 3 provide a different perspective on changes in the educational attainment of country of origin groups. These results do not reflect the influence of the entrance of new cohorts, educational upgrading, and inter-census errors. Column (1) in both tables mirrors the percentages reported in Table 1 for 1980 and 1990, respectively. Columns (2) and (3) provide expected post-secondary education percentages, which were derived by applying a mortality schedule to the 1980 and 1990 populations and adding the foreign-born population that arrived in 1980 1990

F. D. Wilson et al. Table 4 Average ratios of components of change by country of origin Country of origin Survived Died Recent migrant a N_Born residual F_Born emigrant F_Born residual 1980 1990 decade European 0.37973 0.33618 0.51176 0.02632 0.50974 0.35658 Mexican 0.14451 0.10102 0.13122 2.03848 0.15269-4.03205 Hispanic 0.26892 0.23799 0.30322 1.18711 0.31570 0.37150 Caribbean 0.32023 0.27790 0.38002 0.34200 0.37871 1.03450 Middle Eastern 0.42826 0.38554 0.45005 0.58226 0.42990 0.75121 African 0.36015 0.29132 0.54954 0.65907 0.57600 0.09238 Asian 0.38238 0.35864 0.49707 0.95882 0.46862 0.43771 Oceania 0.46303 0.41882 0.40256 3.09474 0.45735 0.43232 Canadian 0.33580 0.29028 0.52356 0.44729 0.50921 1.24190 African American 0.21681 0.15954 1.00000 0.92473 1.00000 1.00000 American Indian 0.19778 0.15591 1.00000 0.76068 1.00000 1.00000 Other American 0.22832 0.19211 1.00000 4.25441 1.00000 1.00000 1990 2000 decade European 0.51828 0.47808 0.59186 0.8187 0.57626 0.30343 Mexican 0.21153 0.16679 0.15305 0.9244 0.09825 0.11428 Hispanic 0.33207 0.27750 0.38529 0.5412 0.32690 0.29763 Caribbean 0.40358 0.34811 0.48161 0.0267 0.40459-0.09179 Middle Eastern 0.53744 0.50098 0.55873 18.1242 0.50204 0.57152 African 0.47872 0.40723 0.70164 0.3853 0.53948 0.51471 Asian 0.47462 0.45882 0.57165 1.9 0.52341 0.84247 Oceania 0.44446 0.41806 0.43434 0.7477 0.45624 0.52474 Canadian 0.44901 0.40312 0.65674 0.8259 0.60483 0.66530 African American 0.32209 0.25593 1.00000 0.5830 1.00000 1.00000 American Indian 0.28775 0.24705 1.00000 1.6761 1.00000 1.00000 Other American 0.31750 0.26688 1.00000 0.3689 1.00000 1.00000 The ratios refer to the number of individuals with one or more years of post-secondary education to the total population, calculated over country of origin, sex and age a Recent migrant refers to the foreign-born who arrived during the decade or 1990 2000, respectively. Column (2) contains both survivors during the decade and the recently arrived foreign-born; Column (3) contains survivors only. The survivor column reflects intergenerational change as well as period specific effects of mortality. Although the changes reported in these tables are smaller than those reported in Table 1, they clearly indicate that both mortality and the in-migration of the foreignborn contributed to decade changes in educational attainment. Comparing columns (4) and (5), one can determine the relative impact of survivors versus migration on changes in educational attainment. Table 2 indicates that for Europeans, Mexicans, and Hispanics, survivors account for most if not all of the changes. Note, however, that migration actually reduced the educational attainment of Mexicans, implying

Population Change and Changing Educational Attainment of Ethnic Groups that recent arrivals had lower educational attainment than those already present. Among the other origin groups that have foreign-born components, there are three instances (Caribbean, Middle Eastern, and Oceanian) in which the impact of survivors and migration is identical and three others (Sub-Sahara African, Asian, and Canadian) for which the impact of migration is greater, implying that inmigration raised the average educational attainment of these groups. For the 1990 cohorts (Table 3), the amount of change is smaller and survivor/ migration differences are smaller as well. In six of nine groups (European, Mexican, Hispanic, Caribbean, Middle Eastern, and Sub-Saharan African), survivors have a greater impact on educational attainment. In the Mexican and Caribbean cases, migration actually lowered educational attainment, suggesting that the inflow of migrants comprised a disproportionate share of individuals with lower educational attainment levels. Since these groups have the longest histories of migration to the U.S., it is likely that the educational selectivity of migration diminished to a greater extent. The diminishing of educational selection is also apparent among the other groups, but not as emphatically. This pattern is expectable, i.e., education selectivity should diminish over time as the flow from origin to destination continues. The analyses that follow seek to account for the decade changes in educational attainment observed in Tables 2 and 3 by applying a multi-level model of decade changes in educational attainment. Multivariate Models We begin by discussing a model of the components of population change in the 1980 1990 and 1990 2000 decades. The model is presented as: j¼65 74!75 84 X i¼15 24!25 34 h ðp tþn P t Þ ¼ P t S t;tþn þ Pt 1 S t;tþn þ IM fb t;tþn i þresid nb t;tþn þ EMfb t;tþn þ RESID fb t;tþn where P t and P t?n denote the population at the beginning and the end of a decade, respectively; S t, t?n denotes the probability of survival during a decade; IM fb t;tþn denotes foreign-born population arriving in the U.S. during a decade; RESID nb t;tþn denotes the residual for the native-born population; EM fb t;tþn denotes the foreign-born population emigrating from the U.S. during a decade; and RESID fb t;tþn denotes the population remaining after other components for the foreign-born are derived. This formula does not provide a detailed specification of the components of change; instead, it reports only those components that the extant data will allow. This formula for decade change is applied to individual origin groups (n = 73) by sex (2), age (12), and educational attainment (2) in each decade. Estimates for the native- and foreignborn were derived separately and subsequently combined. The twelve five-year age groups begin with ages 15 19 and end with ages 70 74, each relating to the beginning of a decade. The NLMS provides no education data on persons under 15 years of age, and mortality among persons aged 75? is too great for our purposes. ð1þ

F. D. Wilson et al. Several terms in the formula require further explanation. The foreign-born population that arrived during a decade (IM) actually consists of several sub-groups including (1) legal immigrants persons who are permanent U.S. residents or have applied for this status; (2) legal non-immigrants persons admitted on employmentor school-related visas and their dependents as well as diplomats and their dependents; (3) the unauthorized persons who entered the U.S. illegally or who failed to leave once their visa expired; (4) return migrants, who emigrated but returned during the decade (this group may also include an unknown share of [1 3] who arrived in a previous period, emigrated, then subsequently returned); and (5) individuals who were incorrectly identified as having arrived in a decade. The foreign-born population that arrived during a decade (IM) is particularly difficult to track. While the composition of the recently arrived foreign-born is known, the actual size of this population and its components are not known. The difficulty is compounded by inaccurate reporting, resulting partly from reliance on proxy reports provided by individuals in the household who do not actually know when someone arrived in the country. A second problem is that there may well be a large discrepancy between when an individual arrived and when he or she applied for resident status. Apparently, many individuals report the date on which their status changed (e.g., from student visa to permanent-resident visa) and not the date on which they actually reached the country (Redstone and Massey 2004). Some foreign-born individuals delay reporting their presence in the U.S. until they obtain citizenship; many individuals do not consider themselves residents before citizenship. Ultimately, the unavailability of an accurate count of the recently arrived foreign-born and their share in the total population would seriously bias efforts to subdivide the recently arrived foreign-born into the components listed above. The residual component for the native-born ðresid nb t;tþnþ includes (or excludes) an unknown number of persons due to classification, reporting, and enumeration errors, as well as an unknown number of the native-born who emigrated from the U.S. during the decade. Similarly, biases plague the residual component for the foreign-born ðresid fb t;tþnþ since both residuals were derived by applying a forward survival estimation procedure, subtracting survivors (and emigrants for the foreign-born) during a decade from the observed population at the end of the decade. These biases are not present in the estimate of foreign-born emigration because the estimates (parsed by sex, age, and Mexican origin) were derived from a different data system (Van Hook et al. 2006). Table 4 reports estimates of components of decade change in population expressed as average ratios of the population with one or more years of postsecondary schooling to the total population of a region of origin category. The ratios are averages of those calculated for each ethnic group by sex and age within each region-of-origin category. The ratios attempt to capture the educational composition of the population represented by a given component and may be interpreted as the average share of a sub-group that has one or more years of post-secondary schooling. 2 For most of the region-of-origin groups across the components of 2 Appendix Table 6 estimates the components of changes for the total population of a region-of-origin category, expressed as a percentage of the mid-decade population of the region of origin.

Population Change and Changing Educational Attainment of Ethnic Groups population-change categories, educational composition favors those with highschool education or less. Most of the exceptions are associated with the recentmigrant and foreign-born-emigrant categories, particularly in the 1990 2000 decade. The ratios for the two residual categories require further explanation. In both instances, they reflect a range of values from positive to negative, resulting in ratios that may exceed one (1.00). For example, the Mexican native-born residual has a value of (2.038), which probably reflects a deficit of Mexicans with highschool education or less observed at the end of the census period as against a surplus of persons with post-high-school education similarly observed. We estimate a multi-level model predicting decade changes in educational attainment that incorporates the previously defined measures as well as other explanatory variables. The multi-level model may be specified as follows: Log PED t;tþn =PED t ¼ Aj þ Xk b j DEM þ Xk b j COC þ e j ð2þ j¼1 j¼1 A j ¼ c k þ Xl k c k COUNTRY þ l k ð3þ where PED denotes the proportion of a country of origin group with one or more years of post-secondary schooling at the beginning and the end of a decade, respectively; DEM denotes demographic variables including sex, age, a dummy variable for the 25 34 age group, and the share of an ethnic group s population that is foreign-born; and COC consists of the six components of population change measures. Recall that the components are expressed as the share of the total population of a specific component of population change with one or more years of post-secondary schooling stratified by country of origin, sex and age categories. By including these components in the model, we seek to determine whether the educational composition of a component is related to decade changes in educational attainment. The dummy variable for the beginning-of-decade 15 24 age group is included in order to capture the effect of intra-generational mobility. We assume that the vast majority of members of a cohort will have completed their schooling at age 25 34, and, in any event, by including this variable we attempt to separate the impact of educational mobility from changes in population associated with survival during the decade and migration of the foreign-born. Level 2 (Eq. 3) consists of eleven dummy variables representing region of origin. It is assumed that individuals from the same region of origin have similar characteristics and experiences in the U.S., including educational outcomes. The variable (A j ) represents the intercept values from Eq. 2, one for each of the 73 groups, and thus represents systematic variation in changes in educational attainment not accounted for by the variables represented in Eq. 2. Table 5 reports the results of the multi-level model estimation for the 1980 1990 and 1990 2000 decades. The primary question that we wish to answer on the basis of these results is whether changes in educational attainment are associated with the educational composition of a component. A positive coefficient for immigration, for example, would indicate that the educational composition of migrants during the preceding decade is associated with an increase in the percentage of the U.S.

F. D. Wilson et al. population with one or more years of post-secondary education. The reported results suggest that the answer to this question is affirmative, at least with respect to several components. First, note that the educational composition of survivors (representing intergenerational change), recent migrants, and foreign-born emigrants has strong effects on changes in educational attainment. In both decades, the greater the share of the recently arrived foreign-born and foreign-born emigrants with one or more years of post-secondary schooling, the greater the increase in average educational attainment. These findings suggest that new arrivals are more advantaged educationally while emigrants are less advantaged, thus raising the educational attainment of those who remain. Although the inverse association for survivors would appear counter-intuitive, one need only recall that survivors are more likely to be favorably endowed educationally. Thus, if a greater share of survivors is already highly educated, the rate of increase for this group would be inversely related to educational attainment observed at the beginning of the decade. The educational composition of those who died during the 1980 1990 decade had no statistical effect on changes in educational attainment but had a statistically significant negative effect for the 1990 2000 decade. This finding simply indicates that the odds of dying are greater for the less educated, as one would expect. The native- and foreign-born residuals had no effect on changes in educational attainment. The absence of statistically significant effects may reflect the fact that these components are derivative and their composition likely heterogeneous. Table 5 also highlights the effects of other variables on changes in educational attainment. First, as expected, changes in educational attainment are greater for men than for women and inversely related to age. The share of an origin population that is foreign-born is associated with smaller change in education, but only in the first decade. Finally, with respect to the Level 1 variables, educational upgrading is strongly and positively related to changes in educational attainment. Although this variable was included to capture intra-cohort changes in educational attainment, we should also caution that some of the effect of the educational upgrading might actually include the educational selectivity of migration. This is because the 25 34 age group is also at greater risk of migration. The region-of-origin variables have very few statistically significant effects. In fact, the average intercept value associated with the Level 1 equation (see Table 5) remained virtually unchanged with the addition of the Level 2 variables. Nevertheless, given their high share among all immigrants, we should not undermine the negative effect of being Mexican on educational attainment. This is even more salient if the desired mainstream trajectory remains that of old European immigrants and their descendants. The gap between these two groups has widened and become significant, especially over the last decade. In a previous analysis (not shown), specific characteristics of individual countries were used, including median education at the beginning of the decade for country of origin, changes in median educational attainment at origin during the decade, and four measures of visa status at entry during a decade. The latter measures include the proportion of all foreign-born individuals who entered during a decade based on family, employment, legal nonimmigrant status, and refugees/asylum seekers. None of the visa-related variables were

Population Change and Changing Educational Attainment of Ethnic Groups Table 5 Multi-level analysis of change in educational attainment Variables 1980 1990 1990 2000 Coefficient SE Coefficient SE Level-one Intercept 0.4313*** 0.0254 0.2223*** 0.0201 Sex (male) 0.0837** 0.0327 0.0356* 0.0179 Age (15 and over) -0.0210** 0.0071-0.0501*** 0.0042 Share FB, 1980(1990) -0.1942* 0.0939 0.1157 0.0655 Com. Pop. Chg. Survived -3.2623*** 0.4017-1.3012*** 0.1926 Died 0.3419 0.3869-0.4730** 0.1869 Recent migrant 1.1545*** 0.0868 0.1689*** 0.0429 N_Born residual 0.0005 0.0008 0.0000 0.0001 F_Born emigrant 0.3278*** 0.0511 0.0964* 0.0463 F_Born residual 0.0012 0.0031 0.0001 0.0004 Educat upgrade 0.4394*** 0.0640 0.1868*** 0.0382 Level-one intercept 0.4312*** 0.0342 0.2223*** 0.0257 Level-two Region of origin European 0.1910 0.1429 0.2373* 0.1053 Mexican 0.0420 0.2461-0.4704** 0.1802 Hispanic 0 1249 0.1551-0.1906 0.1141 Caribbean 0.2889 0.1752 0.1419 0.1291 Middle Eastern 0.4214** 0.1598 0.6 0.1181 African 0.0576 0.2688 0.0221 0.1986 Asian 0.2281 0.1619 0.0813 0.1210 Oceania 0.3421 0.2708 0.0076 0.2009 Canadian 0.0725 0.2694 0.1539 0.1982 African American -0.9659*** 0.2729-0.1224 0.2004 American Indian -1.0349*** 0.2729 0.0980 0.2005 Oth American (Omitted) * p \ 0.05; ** p \ 0.01; *** p \ 0.001 Level-one observations = 1,752; level-two observations = 73 statistically associated with changes in education, and median education at origin was significant only for 1990 2000. Discussion The results indicate clearly that selected components of population change are statistically associated with macro-level decade changes in educational attainment. Specifically, the educational composition of survivors (largely reflecting intergenerational change), foreign-born in-migrants and emigrants, and death (1990 2000

F. D. Wilson et al. only) are related to decade changes in educational attainment. Additional variables including sex, age, share of foreign-born (1980 1990 only), and intra-generational mobility, are also statistically associated with changes in education. The strong positive effect of age specific educational upgrading attests to the importance of intra-generational change. However, as previously noted, we cannot be sure that all of the change captured by the upgrade variable actually reflects intra-generational change due to the particular manner in which this variable was operationalized. We were surprised to discover that the Level 2 variables had little or no effect on changes in educational attainment. In fact, the lack of changes in the intercept when the Level 2 variables are added indicates that a multi-level approach is not required in order to explain changes in educational attainment. Future research will be needed to determine whether another set of variables for Level 2 would be more appropriate. It would also be interesting to explore spatial variations in the effect of region of origin variables. The approach applied here to the decomposition of changes in the average educational attainment of the U.S. population focusing on the role of components of change in population is distinctly different from previous efforts. While it is generally known that generational change, mortality, and international migration affect the educational distribution of a society, we believe that this is the first attempt to empirically study and measure these sources of change systematically. It provides a comprehensive insight into the interplay between different complementary structural demographic factors, and their independent role in shaping a major social character of the American society. This approach should not be confused with approaches that seek to account for changes in the educational status of individuals linked to various social, economic, ethnic, and demographic strata. While this type of change has direct implications for aggregate change, it is not a substitute. Efforts to understand the nature of educational attainment over time are not new, but understanding how educational attainment is affected by demographic factors is. The population of any country is in constant flux due to the births and deaths of its people. Increasingly, however, in-migration and out-migration also play a role as countries experience push and pull factors that propel or attract people, respectively. These factors of course affect a multitude of social indicators including estimates of educational attainment. Education and, in turn, educational attainment remain not only important paths to social mobility but also important predictors of health outcomes and behaviors. Therefore, insofar as public policy efforts find it important to understand and improve the population s educational attainment, having the clearest possible picture of the education levels of all segments of the population native- and foreign-born across time becomes incredibly important for the purposes of reconciling efforts with goals. That is to say, if research finds that an apparent increase in the educational attainment of a particular group across time traces mainly to the group s immigrant population rather than its native-born segment, then policy efforts would have a better idea of how to target and tailor efforts to improve education among this group and, in turn, that of the population at large. This, of course, says nothing about the quality of education, which is also an important component of educational attainment; everything else held constant, however, the number of years of schooling is the appropriate outcome variable on which to focus. The substantive statistical import of observed decade changes in educational attainment, as captured by the intercept and major demographic components, appears

Population Change and Changing Educational Attainment of Ethnic Groups to have diminished between the 1980 1990 and the 1990 2000 decades. We decided against performing global statistical tests of the differences observed between decades as we suspect that the differences may be as much a result of serious methodological flaws in measurement and limited data as substantive. In fact, given the absence of more and better quality data it would be impossible to empirically adjudicate between alternative explanations of decade changes, whether within and/or between the methodological or substantive spheres. Nevertheless, this study is an important first look at the very dynamic process of educational attainment over time at the macro population level. In our current socio-historical moment it is important to understand and establish with some certainty how population dynamics affect this very policy relevant and resource intensive target-population level educational attainment. Our results elicit two important recommendations. First, educational upgrading adult post-secondary schooling appears to be an important contributor to educational change. Insofar as this is true, future research should strive to understand the mechanism by which this occurs. That is to say, when and how does educational upgrading occur? For example, is it between jobs or does it happens concurrently? These sorts of questions need to be answered to help policy-makers fine-tune opportunities for adults, especially during difficult economic times. The second recommendation concerns the availability and quality of the data. Clearly, to adequately decompose the effects that pertain to each component identified in this study, researchers need timely and high-quality data. In the case of the United States, current data limitations forced us to make several assumptions, especially in regard to the composition of age cohorts and the characteristics of undocumented immigrants, for whom official statistics are difficult to ascertain. With a population as dynamic as the United States, understanding how demographic shifts affect and, perhaps, obscure true group means is difficult; without adequate data, however, it is impossible. Accordingly, several limitations of this study must be noted. The most important is its inability to perform a complete decomposition of the components of population change. This includes the limited availability of information on the size and composition of the foreign-born population and the inability to account accurately for various sources of errors, including enumeration, recording, and classification errors. The limited availability of information in the census makes it impossible to separate these factors. A second limitation relates to the lack of a mortality schedule for the 1990 2000 period. By applying the schedule for the previous decade, we eliminate any possibility of determining whether decade differences in mortality were a contributing factor to observed changes in several of the components and in variations among ethnic groups. Nevertheless, the population decomposition approach adopted in this paper is ideally suited to the investigation of changes in educational attainment across two to three generations of cohorts that move through the life course, where it is possible to observe all vital events, migrations, and period-specific changes that affect people differently. Appendix See Table 6.