Educational Outcomes and Intergenerational Mobility of Second Generation Migrants from the Middle East to the United States: a Comparative Study.

Transcription

1 Educational Outcomes and Intergenerational Mobility of Second Generation Migrants from the Middle East to the United States: a Comparative Study. Hisham S. Foad June 2013 One of the key issues in the immigration debate is the extent to which second generation immigrants (native born, but to immigrant parents) are able to assimilate in terms of education and income. Many of these studies have focused on the diculties and educational achievement gap faced by 2nd generation immigrants from the Middle East (MENA) living in Europe. My studies consider 2nd generation immigrants from the MENA living in the United States. These two groups originate in the same place, but could not be more dierent in terms of their educational outcomes. Whereas 2nd generation MENA migrants in Europe tend to have less education than their native peers, MENA migrants in the US have higher levels of education and are more likely to have undergraduate and graduate degrees. In this study I estimate several models examining the determinants of education. I nd that for the overall US population, parents' education, parents' income, and residence in an ethnic enclave all have positive eects on educational achievement. The impact of parents' education and income is no dierent for MENA migrants, suggesting that the education premium accruing to this group is likely to persist. However, any advantage of living in an ethnic enclave appears to disappear for MENA migrants, suggesting that these enclaves are failing to live up to their potential as incubators of immigrant human capital. Keywords: intergenerational mobility, education transmission, ethnic enclaves Department of Economics and Center for Islamic and Arabic Studies, San Diego State University. This work is preliminary and incomplete, please do not cite. All questions and comments may be directed to the author at hfoad@mail.sdsu.edu. 1

2 1 Introduction In recent years, the debate on immigration has shifted from issues surrounding new migrants to those issues facing existing populations of immigrants and their children. One of these issues is the extent to which second generation immigrants (native born, but to immigrant parents) are able to assimilate in terms of education and income. For example, numerous studies have shown that Middle Eastern migrants in Europe tend to be less educated, have lower wages, and worse employment outcomes than their native born, ethnic European counterparts. 1 But does the immigrant wage and education penalty persist across generations? If parent's education is a strong predictor of human capital, then we would expect education gaps between ethnic groups to persist. However, if intergenerational mobility is high (i.e. parent's education has less predictive power), then these gaps should not last for as many generations. Numerous studies have documented a gap between the educational outcomes second generation immigrants and natives. Dustmann and Frattini (2012a) compare test scores of children of Turkish immigrants in several OECD countries to native children in these countries and to Turkish children whose parents have not immigrated. They nd that the children of Turkish immigrants tend to have lower test scores than natives in the OECD countries sampled. However, their test scores are higher than those of Turkish children in Turkey, an observation they attribute to the higher school and peer quality found in these OECD countries. Ekberg et. al. (2010) present evidence that the gap between immigrants and natives may not be closing across generations. They look at earnings of 1st, 2nd, and 3rd generation immigrants in Sweden. Find that while the immigrant wage penalty does fall from the 1st generation to the 2nd and third, it never completely disappears, suggesting limited economic assimilation. Another study documenting the persistence of this gap is Colding et al (2009), who nd that in Denmark, dropout rates from vocational upper secondary education are much higher among children of immigrants. 1.1 Explaining Dierences in Education Chiswick (1988) oers three explanations for why second generation immigrants may have dierent educational outcomes than their ethnically native peers. First, some communities may have a stronger preference for schooling. In these communities, education is valued more highly and greater investments in children's human capital are made, holding other factors constant. Dierences in preferences could be attributed to cultural factors, native country experiences, or even the fact that immigrants often represent distinct group from both natives and their peers in their countries of birth. These immigrants have already made the choice to leave their native countries behind, and in the case of economic migration, they may have a stronger preference for education. Second, children from some communities may be discriminated against with regard to access to schooling or later in life with fewer job opportunities and advancement. For example, if there is a discrimination based wage penalty for being an immigrant, the return to education is reduced and the quantity of human capital investment is lowered. This is particularly relevant if immigrants face higher barriers in high skill jobs. 1 See Dustmann and Frattini (2012b) for a comprehensive look at this issue 2

3 Third, some communities may simply be over-represented in the most disadvantaged socio-economic groups. Thus, dierences across ethnic groups may simply be picking up dierences in socio-economic status. While we can control for some of these socio-economic factors like income, we cannot control for everything, making identication dicult. Several studies have examined the question of whether or not culture matters and most have concluded that once we control for non-cultural aspects of dierent communities, culture does not appear to matter. Van Ours and Veenman (2003) look at the educational attainment of Antillean, Moroccan, and Turkish students in the Netherlands.They nd that there are dierences between these groups and they each have worse educational performances than their ethnic Dutch counterparts. However, once the educational level of parents is controlled for, the education gap between these groups (both between dierent immigrant groups and between immigrants and ethnic Dutch) disappears. Schnitzlein (2012) looks at intergenerational mobility amongst rst and second immigrants in Denmark coming from Germany, Morocco, Pakistan, and Turkey. Controlling for a range of socio-economic factors, there are no signicant dierences in intergenerational mobility between these groups. Whereas culture has met with limited support as an explanation for education dierences between ethnic groups, considerable evidence has been found for discrimination both in education and labor markets. Bisin et. al. (2011) look at the labor market outcomes of non-eu immigrants in Europe. Find that rst generation immigrants are 17% less likely to be employed than natives. However, 2nd generation immigrants are no less likely to be employed than natives. That said, 2nd generation immigrants who maintain a strong ethnic identity have a signicantly lower chance of employment than natives. Carlsson (2010) nds additional support for labor market discrimination based on ethnic identity using a clever experiment in Sweden. He sent out resumes for job openings that were identical in every way except for the fact that one was from an immigrant from the Middle East, one was from a 2nd generation Middle East immigrant (born in Sweden to immigrant parents), and one was a native Swede. The native Swede had the highest chance of being invited to a job interview, while there was no dierence in the likelihood of being interviewed between the 1st and 2nd generation migrants. Further support is given by Meunier (2011), who argues that up to a quarter of the observed dierence in test scores between 2nd generation immigrants in Switzerland and ethnic Swiss can be attributed to the lower expected returns to education for immigrant students. Finally, Ludemann and Schwerdt (2010) look at discrimination in educational opportunities in Germany. They nd that 2nd generation immigrants have lower grades and worse teacher recommendations than would be predicted by their scores on student achievement and intelligence tests. Another factor that could drive a gap between immigrants and natives in terms of education and labor market outcomes is prociency with the local language. Domingues Dos Santos and Wol (2011) high school graduation rates for French children of North African descent are signicantly lower than those of ethnic French children. As the French language prociency of their parents increases, however, graduation rates move closer to the native level. Di Paolo and Raymond (2012) consider how knowledge of Catalan aects earnings of immigrants in the Catalan speaking part of Spain. They nd that knowledge of this language increases immigrant wages by 18%, a remarkable result considering that this a second language in the region (i.e. most native Catalonians speak both Catalan and Spanish.) Casey and Dustmann (2008) look at parental language prociency in the US. They nd that as as parental language uency increases, labor market outcomes for 2nd generation women improve. 3

4 1.2 The Impact of Living in an Ethnic Enclave Many immigrants live in ethnic enclaves, communities in which immigrants from the same country are overrepresented. What impact does living in an ethnic enclave have on educational and labor market outcomes? On the positive side, living in an ethnic enclave should reduce the cost of immigration and limit the discrimination in labor markets that many immigrants face. Furthermore, the returns to pre-immigration human capital should be higher (i.e. native country language skill may be more valuable when surrounded by people who speak that language.) However, there are several reasons to believe that living in an enclave could have negative eects. The need to assimilate is reduced, which could lower human capital investments. The skill set that is acquired for life in the enclave may also not be as transferable to life outside the enclave. For example, a child living in an ethnic enclave will not have as much pressure to learn English, which is ne within the enclave, but harmful if that child were to ever leave. Finally, there may be signicant neighborhood eects that could have both positive and negative impacts. If those living in the ethnic enclave tend to be well educated and have high incomes, then living in the enclave should have positive eects. If the enclave is of low socio-economic status, then it could have a negative impact on the educational and labor market outcomes of its residents. Most studies of ethnic enclaves have focused on the eect of these enclaves on earnings. 2 Early studies of this issue were plagued by simultaneity bias. Does living in an ethnic enclave aect earnings, or are immigrants choosing to live in those enclaves because wages may be higher there? Edin et. al. (2003) overcome this endogeneity problem by looking at refugees in Sweden as a natural experiment. These refugees were placed in a particular location by the Swedish government rather than by their own choice. They nd that for these refugee communities, living in an ethnic enclave increases low skilled wages by 13%. The quality of the enclave also makes a dierence, with high income enclaves having a stronger eect on wages than low income enclaves. Whereas many studies have considered the eect of living in an ethnic enclave on earnings, very few have evaluated their impact on the educational outcomes of immigrant children. Grönqvist (2006) looks at immigrants in Sweden and nds that for second generation immigrants, living in an ethnic enclave reduces the probability that they will graduate from high school. Further evidence of the negative eect of enclaves on educational outcomes is given by Chiswick and Miller (2001), who nd that living in an ethnic enclave is negatively related to English language prociency for immigrants to Canada. However, Gang et. al. (2000) nd that ethnic network size has a positive eect on educational attainment for 2nd generation immigrants in Germany. 1.3 What this Study Contributes Much of the existing literature on intergenerational mobility of migrants from the Middle East (henceforth MENA) focuses on immigration to Europe. This makes sense as MENA migrants to Europe represent a very large group and considerable attention has been paid to the diculties and successes of these groups in assimilating into European society. However, considerably fewer studies have focused on the educational outcomes of these migrants living in the United States. I argue that studying this group is important for two reasons. First, MENA migrants to the US are a fundamentally dierent group than their counterparts in Europe. Since the cost of migrating to 2 See Xie and Gough (2009) for an exhaustive list of these studies. 4

5 the US is higher than it is to Europe (longer distance, fewer established ethnic enclaves), only those immigrants that anticipate a signicantly high return to migration will make the move to the US. As a result, MENA migrants to the US tend to be much better educated, earn higher incomes, and more likely to permanently reside in the US. The socio-economic prole of MENA migrants to Europe most closely resembles that of Mexican migrants to the US. Thus, we have two groups (MENA migrants to the US and Europe) that are culturally similar, but have dierent socio-economic characteristics. Comparing intergenerational mobility between these two groups allows for better identication of these socio-economic factors as well as institutional factors in Europe and the US. The second reason for studying MENA migrants to the US is that this is a rapidly growing immigrant community. According to Camarota (2002), the immigrant community from the MENA in the US had a seven-fold increase between 1970 and 2000, growing from fewer than 200,000 to over 1.5 million. Contrast this with the only three-fold increase in the general immigrant population over this period. Given that MENA migrants to the US tend to have higher fertility rates than the average US population (CDC National Vital Statistics System, 2011), the MENA diaspora in the United States is only expected to grow in the coming years. Given the rapid growth of this community, it is important to understand the degree to which the children of these immigrants are able to economically assimilate into American society. In this study, I empirically show that the MENA ethnic group has higher educational outcomes than the general US population. Overall, educational outcomes are increasing in parents' education, parents' income, and residence in an ethnic enclave. The eects of parents' education and income are no dierent for 2nd generation MENA migrants, suggesting that the education premium of being a MENA migrant will persist across generations. However, I nd that the advantages of living in an ethnic enclave disappear for MENA migrants, suggesting some fundamental dierences between these enclaves (and thus the people living in them) and the enclaves inhabited by other ethnicities. The rest of the paper proceeds as follows: Section 2 presents the data and variable construction, section 3 outlines the empirical methodology, section 4 discusses the results, and section 5 concludes. 2 Data 2.1 Dataset and Variable Construction To assess intergenerational mobility of MENA migrants in the United States, I use individual level data from the 2011 American Community Survey, public use microdata sample. As my study is concerned with second generation immigrants, I restrict my sample to only those people that were born in the United States. Second generation immigrants are dened as having been born in the United States to at least one parent born abroad. 3 2nd generation MENA migrants are dened as native born children who have at least one parent born in one of twenty on countries in the MENA. 4 3 A growing strand of literature dierentiates between 2nd generation and 2.5 generation immigrants. In the former case, both parents were born abroad, while in the latter, one parent is an immigrant and the other is native born. Ramakrishnan and Karthick (2004) argue that we need to distinguish between these two groups as they dier signicantly in terms of racial identication, education, and income. A future draft of this paper will assess the dierence between these groups. 4 The 21 countries are Algeria, Bahrain, Egypt, Iran, Iraq, Israel, Jordan, Kuwait, Lebanon, Libya, Morocco, Oman, Palestine, Qatar, Saudi Arabia, Sudan, Syria, Tunisia, Turkey, UAE, and Yemen 5

6 For each observation, I collect data on age, years of education, county of residence, parents' education, parents' income, parents' English speaking ability, and parents' birthplace. I restrict the sample to individuals aged 25 years and older. While this limits the sample size for ethnic groups that are relatively recent arrivals to the US (Somali refugees for example), it is necessary to not underestimate educational outcomes for these groups. From this dataset, I create the variables dened below in Table 1. Deleting all missing values (mostly related to parental variables), I end up with a sample of 56,442 observations. Table 1: Variable Denitions Ed i Years of education. This variable is computed from the ACS variable "educd" where, for example, completing 1st grade is equivalent to 1 year of education and completing 6th grade is equivalent to 6 years of education. High school graduates have 12 years of education, college graduates have 16 years of education, and those with a professional degree have 20 years of education. In the empirical analysis, I take the natural log of this variable. HS i College i P rofessional i MomEd i DadEd i MomLang i DadLang i Enclave i ME i A dummy variable equal to 1 if the observation has at least a high school diploma A dummy variable equal to 1 if the observation has at least a bachelors degree A dummy variable equal to 1 if the observation has completed a professional degree Similar to Ed above, but for the observation's mother Similar to Ed above, but for the observation's father A dummy variable equal to one if the observation's mother self reports her ENglish langiage ability as "speaks only English," "speaks very well," or "speaks well." Equal to zero if the mother rates her ability as "does not speak English" or "speaks English, but not well." Similar to MomLang, but for the observation's father. A dummy variable equal to one if the observation lives in a county that is considered to be an ethnic enclave. An ethnic enclave is dened as a county in which immigrants from a particular country are at least three times as likely to live in as the average American resident. a A dummy variable equal to 1 if at least one of the observation's parents was born in the Middle East. a The results are not too dierent if I lower this threshold to 2 or raise it to 4 6

7 2.2 Prole of Educational Outcomes across Ethnic Groups Using this methodology, I am able to decompose educational outcomes across several dierent ethnicities. Table 2 presents the average years of education and the fraction of the sample that has a high school diploma, college degree, and professional degree. I consider these outcomes between native born and immigrants across the entire sample and between immigrants and native born whites, Middle Easterners, Mexicans, Latin Americans (non-mexican), Sub-Saharan Africans, South Asians, and East Asians. Natives from each group are dened as people who were born in the US to either parents from one of these regions or who claim a rst ancestry from these regions. Thus, I am mixing together 2nd and higher generation immigrants in these groups, though the majority are 2nd generation (native born, but to immigrant parents.) Table 2: Educational Attainment across Immigrant and Ethnic Groups Years Education H.S. College Professional All % 28.0% 10.9% Foreign Born % 28.6% 12.3% Native Born % 27.9% 10.6% White Immigrants % 31.9% 14.6% White Natives % 27.4% 10.3% MENA Immigrants % 46.8% 21.4% MENA Natives % 51.0% 22.1% Mexican Immigrants % 5.2% 1.4% Mexican Natives % 14.3% 4.2% Latin American Immigrants % 16.6% 5.8% Latin American Natives % 23.6% 8.3% SS African Immigrants % 39.0% 17.4% SS African Natives % 19.7% 17.4% S. Asian Immigrants % 66.2% 36.4% S. Asian Natives % 74.3% 39.9% E. Asian Immigrants % 44.5% 16.8% E. Asian Natives % 50.7% 17.4% Years Education refers to the average years of education attained for each group. 12 years of education is equivalent to a high school degree, 16 years a college degree, and 20+ years a professional degree. H.S., College, and Professional refer to the fraction of each group that have at least a high school diploma, bachelors degree, and professional degree repsectively. Immigrants are dened as people who were born in a country other than the U.S., while Natives were born in the U.S. with reported ancestries to the specied group. Across all ethnicities, immigrants have on average 1.3 fewer years of education, with 12 years on average for immigrants and 13.3 years for natives. A much lower percentage of immigrants have completed high school, with under 70% of immigrants holding a high school degree compared to nearly 90% for natives. Interestingly, those immigrants who have completed high school are much more likely to have completed college and professional degrees. In my sample, about 28% of native born Americans have 7

8 at least a bachelors degree and 10.6% have a professional degree. For immigrants, these shares are 28.6% and 12.3%. That immigrants tend to have higher rates of tertiary education attainment could simply be a reection of US immigration policy favoring skilled migration. There are signicant dierences across ethnic groups. migrants from the MENA tend to be better educated than the general US population. MENA migrants average 14 years of education and a signicantly higher fraction of these immigrants have college and professional degrees. Ethnic Middle Easterners born in the US are even better educated than their immigrant forebears. MENA natives average nearly 15 years of education, over half have at least a bachelors degree and 22% hold a professional degree, averages well above average American resident. The same pattern that holds for MENA migrants and natives also holds for the East and South Asian diasporas in the US. For each group, both immigrants and natives tend to be better educated than the average American, and the high education of immigrants is transferred over to future generations with increased educational outcomes for natives. Average education tends to be lowest for Mexican immigrants, but there are signicant increases when going from Mexican immigrants to natives. This suggests that the average education gap for Mexican immigrants will disappear over a few generations. The only group that shows a decline in educational outcomes across generations are sub-saharan Africans. Immigrants average 13.6 years of education and greater fractions of college and professional school graduates than the average American. However, 2nd generations have worse outcomes than the average American. Table 3 presents parent's educational outcomes across my sample. In every case, the average years of education for parents are lower than average years for their children, indicating an increase in the average education over time for both immigrants and natives in the US. Contrasting mother's and father's education, the biggest gender gaps are for South Asians (fathers have two more years of education), and for the MENA (fathers have 0.9 more years on average). Interestingly, the largest education gaps are also for the most educated groups, and even for these groups, average years of mother's education are the two highest in the sample. At dierent levels of education, the largest gender gaps are at the college and graduate levels. This pattern is not unique to any ethnic group, however, and is more likely due to traditional gender roles and childbearing decisions. 5 Given the diversity in educational outcomes across groups, it is useful to further break these gures down by specic MENA countries. Table 4 presents these outcomes across the seven countries with the largest diasporas (immigrants and native born with ancestral ties) in the US. In every case, 2nd (and higher) generation immigrants have more education than immigrants, a remarkable fact given that immigrants themselves are coming in with already high levels of education. Looking across groups, the highest levels of education are for Egyptian and Iranian immigrants. Both of these two groups had large waves of migration in the 1960's and 1970's. The older diasporas like Lebanon and Syria (most of whom arrived in the 1920's-1940's) tend to slightly lower years of education, though still higher than the average US population. The largest gap between immigrants and natives is for the Iraqi diaspora. Iraqi immigrants have a relatively low 11.9 years of education, compared to 15.7 years for US born Iraqis. Much of this gap can be attributed to the recent wave of Iraqi refugee migration. Overall, though, the trends across individual countries mirror those for the MENA as a whole and the rest of the analysis will treat these dierent diasporas as a single group. 5 Interestingly, this trend is expected to reverse in the future given the larger fraction of women in American universities than men. 8

9 Table 3: Parent's Educational Endowments across Ethnicities All White MENA Mexican Lat. Am. SS Africa S. Asia E. Asia Years Education Mother Father High School Mother 72.9% 80.3% 81.6% 45.7% 65.5% 75.5% 81.2% 79.6% Father 78.4% 81.4% 80.4% 45.0% 64.6% 78.9% 91.3% 83.4% College Mother 14.1% 14.1% 31.5% 4.6% 11.1% 22.3% 44.1% 30.6% Father 20.3% 20.7% 42.6% 6.0% 12.1% 25.8% 64.7% 34.9% Professional Mother 4.9% 5.2% 12.0% 1.3% 3.9% 9.8% 17.5% 7.3% Father 8.4% 8.5% 21.5% 2.4% 4.6% 16.3% 38.2% 12.3% Yrs Education refers to the total years of education completed by the observations mother and father. For example, if mother's years of education is 12, then the observation's mother has completed high school. High School, College, and Professional are dumy variables equal to 1 if the observation's parent completed high school, college, or graduate school respectively. English is a dummy variable equal to 1 if the observation's parent rates their English speaking ability as "speaks well" or "speaks very well" and equal to zero if the parent rates their ability as "does not speak English" or "does not speak well." 2.3 Dening Ethnic Enclaves An interesting issue raised in the literature is the extent to which living in an ethnic enclave aects intergenerational mobility and investment in human capital. On one hand, living in an ethnic enclave may increase human capital investment since immigrants will face lower migration costs when moving to an enclave and their children will face less labor market discrimination in these enclaves, increasing the return to education. On the other hand, living in an enclave reduces the need for a migrant to assimilate, limiting English language acquisition and thus reducing educational attainment. Furthermore, if people within the enclave have lower socio-economic indicators, then there may be negative neighborhood eects on educational outcomes. I use a unique methodology to dene an ethnic enclave. I construct an agglomeration index (AI) for each ethnic group and US county. The agglomeration index for ethnic group j living in county i is dened as the fraction of ethnic group j living in county i divided by county i s share of the US population: AI j i = P opj i /P opj P op i /P op (1) For example, 3.815% of the MENA diaspora lives in Wayne County, Michigan. This county represents 0.674% of the US population. The agglomeration index for the MENA diaspora in Wayne County, Michigan is thus AIW MENA aynecounty,mi = 3.815%/0.674% = 5.7. This number can be interpreted as how much more or less likely a member of a particular ethnic group is to live in a county than the average 9

10 Table 4: Educational Attainment across MENA Countries and Ancestries Years Education H.S. College Professional Mena Immigrants % 46.8% 21.4% Mena Natives % 51.0% 22.1% Egyptian Immigrants % 64.7% 22.8% Egyptian Natives % 74.6% 38.2% Iranian Immigrants % 54.1% 26.4% Iranian Natives % 72.6% 37.5% Iraqi Immigrants % 28.7% 8.2% Iraqi Natives % 69.2% 38.5% Jordanian Immigrants % 33.6% 14.2% Jordanian Natives % 34.3% 11.4% Lebanese Immigrants % 41.4% 22.3% Lebanese Natives % 50.6% 20.9% Syrian Immigrants % 40.0% 20.5% Syrian Natives % 39.7% 17.1% Turkish Immigrants % 55.1% 28.5% Turkish Natives % 46.7% 22.2% Years Education refers to the average years of education attained for each group. 12 years of education is equivalent to a high school degree, 16 years a college degree, and 20+ years a professional degree. H.S., College, and Professional refer to the fraction of each group that have at least a high school diploma, bachelors degree, and professional degree repsectively. Immigrants are dened as people who were born in a country other than the U.S., while Natives were born in the U.S. with reported ancestries to the specied group. American county. 6 Thus, ethnic Middle Easterners (both immigrants and native born) are 5.7 times more likely to live in Wayne County, Michigan than the average American. An agglomeration index less than 1 means that a member of the diaspora is less likely to live in that county than the average American, while an index over 1 suggests that they are more likely. An advantage of this measure is that it is scale free. For example, would expect there to be a large number of ethnic Middle Easterners in New York City, simply because there are a lot of people that live in New York City. But how signicant is the MENA community? Is it tightly organized with services oered for its residents such as Arabic language schools and ethnically owned businesses? The agglomeration index measures the relative importance of the community in a particular county and is thus more likely to be correlated with the features of an ethnic enclave that would aect educational attainment. Table 5 presents the top 15 and bottom 10 counties in the US for the MENA diaspora based o the agglomeration index (conditional upon there being at least one ethnic Middle Easterner in a county). The results are fairly consistent with anecdotal evidence on where the largest MENA communities are 6 For Western European ancestries, I set the agglomeration index equal to 1. This implicitly assumes that the geographic distribution for people of Western European descent mimics the general US population (of which they are a big part). 10

11 Table 5: Agglomeration Indexes for Middle Eastern Migrants State County Agglomeration Index Top 15 Counties Michigan Wayne County 5.7 Michigan Macomb County 5.5 California Stanislaus County 5.4 New Jersey Hudson County 5.4 Michigan Oakland County 5.3 Tennessee Davidson County 5.1 California Los Angeles County 4.6 New Jersey Passaic County 4.6 New Jersey Bergen County 4.5 Virginia Alexandria city 4.5 California San Mateo County 3.9 California San Diego County 3.9 Virginia Henrico County 3.9 New York Richmond County 3.8 Texas Collin County 3.8 Bottom 10 Counties Oregon Douglas County 0.06 Texas Johnson County 0.06 Connecticut Middlesex County 0.05 North Carolina Alamance County 0.05 Illinois Tazewell County 0.05 Utah Weber County 0.04 Idaho Canyon County 0.03 Colorado Weld County 0.03 Texas Cameron County 0.03 Louisiana Orleans Parish 0.02 The Agglomeration Index is computed as the fraction of all migrants from the Middle East living in a particular county divided by that county's share of the US population. It can be interpreted as how much more or less likely a Middle Eastern migrant is to live in a particular county. Thus, an Agglomeration Index of 3.0 implies that a MENA migrant is three times as likely to live in that county as the average American, while an index of 0.5 means that the migrant is only half as likely to live there. located. As expected, several counties around Dearborn, Michigan represent the largest ethnic enclaves, reecting the very large Middle Eastern community in this region. Stanislaus County in California also shows up as a signicant enclave, mostly due to the very large Iraqi Assyrian community there. While there are some expected results in this top 15, there are a few surprises, such as Garrett County, Maryland, which is a fairly rural county in western Maryland. 7 7 This may be an outlier as the ACS only reports 857 ethnic Middle Easterners in this county, which ends up being a large agglomeration index given the relatively small population size of this county. I have estimated the empirical model excluding counties with less than 100,000 people and the results do not qualitatively change, however. 11

12 Table 6: Agglomeration Indexes for other Ancestries Mexico Latin America Imperial County, California 11.3 Hudson County, New Jersey 9.4 Hidalgo County, Texas 11.1 Queens County, New York 8.5 Webb County, Texas 10 Union County, New Jersey 8.3 El Paso County, Texas 9.1 Osceola County, Florida 6.9 Cameron County, Texas 8.2 Broward County, Florida 6.8 Yuma County, Arizona 7.2 Arlington County, Virginia 6.2 Tulare County, California 7 Essex County, New Jersey 5.7 Merced County, California 6.5 Suolk County, Massachusetts 5.2 Madera County, California 6.1 Alexandria City, Virginia 5.1 Santa Barbara County, California 5.9 Montgomery County, Maryland 4.7 Eastern Europe S.S. Africa Kings County, New York 10.8 Alexandria city, Virginia 24.6 Richmond County, New York 7.5 Montgomery County, Maryland 12.3 Queens County, New York 6.3 Hennepin County, Minnesota 9.8 Cook County, Illinois 6.0 Bronx County, New York 9.4 Bergen County, New Jersey 5.8 Ramsey County, Minnesota 9.1 Sacramento County, California 4.7 DeKalb County, Georgia 7.8 Yolo County, California 4.7 Montgomery County, Tennessee 7.6 Spokane County, Washington 4.7 Clayton County, Georgia 7.6 Passaic County, New Jersey 4.3 Arlington County, Virginia 7.3 DuPage County, Illinois 4.2 Essex County, New Jersey 6.9 South Asia East Asia Middlesex County, New Jersey 14.6 San Francisco County, California 10.7 Fort Bend County, Texas 9.7 Honolulu County, Hawaii 8.3 Hudson County, New Jersey 8.1 Santa Clara County, California 8.2 Queens County, New York 7.9 San Mateo County, California 6.6 Santa Clara County, California 7.8 Alameda County, California 6.4 Somerset County, New Jersey 7.8 Queens County, New York 5 Alameda County, California 6.7 Orange County, California 5 Harford County, Maryland 6.6 Fort Bend County, Texas 4.7 Collin County, Texas 5.5 King County, Washington 4.2 DuPage County, Illinois 5.1 Bergen County, New Jersey 4.2 The agglomeration index is computed as the fraction of country's migrants to the US living in the specied county divided by tht county's share of the US population. For example, the index for Mexican migrants living in Imperial County, California is 11.3, implying that migrants from Mexico are 11.3 times more likely to live in Imperial County than the average American resident. I also present the top ten ethnic enclaves for six other ethnicities in my sample in Table 6. The results are again consistent with anecdotal evidence on what communities are ethnic enclaves. For example, Mexicans are heavily represented in Southern California and South Texas, while Latin Americans are prevalent in South Florida. Sub-Saharan Africans are well represented in the Baltimore/D.C. area as 12

13 well as several counties around Minneapolis with large Eritrean and Somali refugee communities. South Asians have signicant communities around Silicon Valley, while East Asians are well represented in San Francisco (large Chinese and Korean communities) and Hawaii (signicant Japanese community). 3 Methodology It is common in the literature on intergenerational mobility to write the relationship between outcomes of parents and outcomes of children as y i,t = α + ρy i,t 1 + u i,t (2) y i,t and y i,t 1 measure outcomes such as education, earnings, or wealth for children (t) and their parents (t 1). For example, let y be equal to education. Then, the earnings of family i s child is determined by family i s parental education and other inuences u. The parameter α can be thought of as the average eect of these other inuences. If we assume that the variances of y i,t and y i,t 1 are the same, then ρ is the population correlation coecient between y i,t and y i,t 1. 8 The estimated coef- cient ˆρ will then measure the fraction of economic advantage (measured in terms of education in this case) that is transmitted across generations. This term is also known as the transmission parameter (Dustmann et. al., 2012). An estimate close to zero suggests high intergenerational mobility, while a coecient closer to 1 indicates low mobility. We can use this methodology to assess assimilation of immigrant populations over time. Dene equation 1 separately for the native and immigrant populations: Natives: y N i,t = αn + ρ N y N i,t 1 + un i,t Immigrants: y I i,t = αi + ρ I y I i,t 1 + ui i,t Allow the two transmission parameter to dier between these groups as: ρ I = ρ N + γ. The expected outcome dierential between natives and immigrants in generation t will then be given by: E(y N i,t) E(y I i,t) = α N α I + ρ N ( E(y N i,t 1) E(y I i,t 1) ) γe(y I i,t 1) (3) Suppose that the only dierence between native and immigrant educational outcomes is intergenerational mobility (i.e. α N = α I ). Then, equation 2 simplies to: E(y N i,t ) E(yI i,t ) = ρn ( E(y N i,t 1 ) E(yI i,t 1 )) γe ( y I i,t 1). Thus, the parameter ρ N will determine the rate at which the gap between immigrant and native educational outcomes disappears. Assuming that γ = 0, then if ρ N = 1, this gap will persist over time since there is no intergenerational mobility. If ρ N = 0, then the gap disappears from one generation to the next. Thus, assimilation of immigrants will be faster for smaller values of ρ N. If γ < 0, then the transmission parameter is larger for natives than for immigrants and the education gap will persist longer. Of course, this analysis assumes that yi,t N > yi i,t. If immigrants have higher educational outcomes than natives (as is the case with MENA migrants), then the story above will be reversed. A negative value 8 If the variances are dierent across generations, then the OLS estimator ˆρ measures ρ σ yt /σ yt 1 13

14 for γ implies that immigrant education will converge with native education, while a positive value for γ suggests that the immigrant education premium will persist longer. Of course, intergenerational mobility is not the only factor determining the gap between immigrant and native educational outcomes. Such things as language acquisition, dierences in the way educational systems treat immigrant children, discrimination, or dierent educational incentives due to living in an ethnic enclave could all contribute to the education gap between these groups. To account for these other factors, I build an augmented version of the model presented in equation 1. This model relates years of education to parents education, parent's income, parent's prociency with English, and whether the observation lives in an ethnic enclave. Ed i = α+ρ 1 MomEd i +ρ 2 DadEd i +β 1 ln P ary i +β 2 MomLang i +β 3 DadLang i +β 4 Enclave i +u i (4) Observation i s log years of education is regressed on mother's log years of education and father's log years of education. P ary i is dened as the observation's parent's family income, calculated as the maximum between mother's and father's income. 9 Language is measured as a dummy variable equal to 1 if the observation's parent can speak English either well or very well and equal to zero if the parent either doesn't speak English at all or has trouble with English according to the Census questionnaire. The variable Enclave is equal to 1 if the observation lives in a county with an agglomeration index greater than To evaluate how intergenerational mobility diers between the average American and ethnic Middle Easterners born in the United States, I create a dummy variable ME which is equal to 1 if the observation was born in the United States, but claims a rst ancestry from a country in the Middle East. I then interacted this dummy with each of the variables in equation 3: Ed i = α + ρ 1 MomEd i + ρ 2 DadEd i + β 1 P ary i + β 2 MomLang i + β 3 DadLang i +β 3 Enclave i + δ 0 ME i + δ 1 MomEd i ME i + δ 2 DadEd i ME i + δ 3 P ary i ME i (5) +δ 4 MomLang i ME i + δ 5 DadLang i ME i + δ 6 Enclave i ME i + u i The parameter δ 0 captures how much more education 2nd generation MENA migrants are predicted to have than the average American. Specically, MENA migrants are predicted to roughly δ 0 % more years of education. The interaction terms estimate how much more or less factors like parental income education, and English language ability aect educational attainment. For example, a 1% increase in mother's education is predicted to increase years of education by (ρ 1 + δ 1 )%, compared to ρ 1 % for the average American. If the sum of δ 1 + δ 2 is negative, then we predict that parental education has a smaller eect for MENA migrants than for the average American. This is equivalent to the case where γ < 0 above. 9 This implicitly assumes that in cases where mother's and father's family diers, that the richer parent is involved in the child's educational investment. Estimating the model using mother's and father's income separately leads to fairly negligible eects on the parameters. For most of the sample, mother's family income is identical to father's family income. 10 See the preceding section for a denition of how this variable was computed and its use. 14

15 Since MENA migrants have on average more education than the average American, then a negative value for δ 1 +δ 2 implies that over time, MENA migrants education levels will converge with the average American. If δ 1 + δ 2 > 0, then parental education has a larger eect for MENA migrants and the higher levels of education for 2nd generation migrants are likely to persist. In addition to the OLS model in equations 4 and 5, I also estimate probit models that predict the probability of a migrant obtaining a bachelors and professional degree. Dene Degree j i as a dummy variable equal to 1 if the observation has obtained a bachelors or professional degree (j = 1, 2). We can then evaluate the factors that inuence the probability of obtaining this degree with a probit model: 11 P rob(degree j i = 1) = Φ(MomEd i, DadEd i, P ary i, MomLang i, DadLang i Enclave i ) (6) Where Φ is the cumulative distribution function of standard normal distribution. I estimate equation 6 via maximum likelihood as well as a modied version of equation 6 where each variable is interacted with the MENA dummy. In this setup, I am estimating how these variables aect the probability that someone gets a bachelors or professional degree. 4 Discussion Table 7 presents OLS and Maximum Likelihood probit estimates of the models in equations 4-6. The six columns represent the OLS regression on the baseline model, the OLS model with interactions, the probit model for college graduation with and without interactions, and the probit model for professional degrees with and without interactions. First consider the OLS regressions in column 1. This baseline model shows that parental education has a signicant impact on educational attainment. A 10% increase in either mother's or father's education is predicted to increase child's education by almost 1%. Across my sample. parent's education has a mean around 12 years and a standard deviation of 4 years. Thus, a 1 standard deviation increase in parent's education is about a 33% increase over the mean, which would translate to a 3.3% increase in child's education. Parental education matters, but it's not the only factor that does. There is considerable variation in parental income, with a one standard deviation increase in parent's income translating to about a 75% increase from the mean. Thus, a 1 standard deviation increase in parental income is predicted to increase child's education by 75%*0.06 = 4.5%. So richer parent's tend to invest more heavily in their child's education. An interesting result holds when we consider parent's language ability, specically that children whose parent's have trouble speaking English grow up to have between 1% and 3% more education than those whose parents speak English well. This is exactly the opposite of the result that would have been predicted by theory and what the existing literature has found. This may be evidence that immigrants who struggle with English are more acutely aware of the value of an education and therefore strive to make sure that their children will not face the same diculties as they do. Finally, living in an ethnic enclave appears to have a positive eect on education, with people living in these enclaves obtaining about 4% more education on average. 11 I also estimated the relationship with a logistic regression and found no qualitative dierences, though the magnitudes of the parental education variables was larger 15

16 Table 7: OLS and Probit Estimates of Education Determinants Yrs Education Yrs Education College College Professional Professional (1) (2) (3) (4) (5) (6) M omed [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] DadEd [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] P ary [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] M omlang [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] DadLang [0.005] [0.003] [0.000] [0.000] [0.000] [0.000] Enclave [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] M E [0.047] [0.020] [0.768] MomEd ME [0.391] [0.480] [0.851] DadEd M E [0.608] [0.938] [0.761] P ary ME [0.768] [0.069] [0.450] M omlang M E [0.393] [0.096] [0.610] DadLang M E [0.081] [0.183] [0.545] Enclave M ena [0.023] [0.107] [0.452] Constant [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] The rst two columns give OLS estimates of equations 4 and 5 in the text. The dependent variable is log years of education. Mother's and father's years of education are also logged. Columns 3 and 4 present probit models where the dependent varible is whether or not the observation has a bachelors degree, while columns 5 and 6 present probit models in which the dependent variable is whether or not the observation has a professional degree. In these four regressions, mother's and father's years of education or NOT logged. For the probit regressions, the estimated reported coecients give the change in the z-score for a one unit change in each explanatory variable. In all cases, heteroskedsticity consistent p-values are given below the coecients (null hypothesis is that the coecient is equal to zero.) See Table 1 for full variable descriptions. 16

17 Column 2 presents estimates of how this model diers for 2nd generation MENA migrants. First o, we see that MENA migrants obtain about 10% more education than the average American. However, there are no signicant dierences between MENA migrants and the average American in terms of transmission of parent's education to their children. Given that MENA migrants have more education than their non-mena peers, the results in this study suggest that this gap will persist for at least the next generation. Income and mother's language ability do not appear to dier between these two groups, but there is some evidence that father's ability with English has a larger eect on MENA migrants than for other Americans. Finally, the positive eects of living in an ethnic enclave appear to disappear for MENA migrants. This is a very interesting result that suggests that the antidiscrimination and job market opportunities aorded to other groups living in enclaves are not accruing to 2nd generation MENA migrants. Explaining why we observe this dierence is an interesting avenue for future research with important policy implications. Looking at the probit model for college graduates, we see that in the baseline model, the qualitative results are consistent with the OLS model. Parents' education and income still have positive eects, as does living in an ethnic enclave. Parents' abilities to speak English still have a negative eects. The estimated coecients on the continuous variables in this model (parents' education and income) are not easily interpretable, but we can evaluate the impact of the dummy variables on the probability of graduating from college by looking at marginal eects. The marginal eect of mother's ability with English is -0.10, while that for father's English ability is This means that holding all else equal, people with mothers that speak English well are 10% less likely to graduate from college, while those with fathers that speak English are 6% less likely to graduate. The marginal eect of living in an enclave is 0.073, meaning that living in an ethnic enclave increases the probability of college graduation by 7.3%, holding all else constant. Looking at the probit model with interactions, the MENA dummy variable and the interaction between ethnic enclave and MENA are both signicant as in the OLS model. Using the estimated marginal eects of these dummy variables, a 2nd generation MENA migrant is 49% more likely to have graduated from college than the average American. However, the positive eect of living in an ethnic enclave essentially disappears when looking at MENA migrants, as the combined marginal eect of Enclave and Enclave M E is equal to = Thus, living in an ethnic enclave increases the probability of graduating from college by only 2% for MENA migrants, as compared to 7% for all other Americans. Interestingly, the interaction between mother's ability with English and MENA is signicantly positive in this example, suggesting that for MENA migrants, having a mother that can speak English well actually has a positive eect on educational outcomes. The probit model for obtaining a professional degree displays the same pattern for the baseline model: parents education/income and living in an ethnic enclave have a positive eect on the likelihood of obtaining a graduate degree, while parent's ability with English has a negative eect. The marginal eects of the dummy variables are all signicantly smaller than they were for obtaining a graduate degree. Having a parent who speaks English well is predicted to reduce the likelihood of obtaining a graduate degree by about 2.5%, compared to 7-10% for a college degree. Living in an ethnic enclave is predicted to raise the probability of getting a graduate degree by about 1.7%, compared to over 7% for a college degree. The same pattern holds for the continuous variables parent's education. Figure 1 presents the marginal eects of both mother's and father's education on the probability of obtaining a college and graduate degree evaluated at 1-20 years of education. Both mother's and father's education have a bigger impact on the likelihood of obtaining a bachelors degree than for obtaining a professional degree. Interestingly, in both cases, mother's education has a larger impact than fathers education for 17