Recent trends in the skill composition of legal U.S. immigrants.

Simpson, Nicole B.

1. Introduction

Most new immigrants to the United States originate in developing countries, and many of these are unskilled; however, there are still a significant number of educated workers entering the United States each year. Measuring short-term fluctuations in the skills of new U.S. immigrants is a challenging task. Education systems vary across countries, making it difficult to compare educational levels, and wage data are not collected for new U.S. immigrants on an annual basis (i.e., as they arrive in the United States). Thus, there is no clear understanding of the year-to-year fluctuations in the overall characteristics of newly arrived U.S. immigrants during the last few decades.

Much can be gained by attempting to understand the trends in the skill composition of legal U.S. immigrants, as discussed in Greenwood and McDowell (1999). The skill level of immigrants is important in determining both their economic cost and benefit. For example, the productivity effects of immigrants depend on their skill level, as does their utilization of welfare services. Using characteristics of immigrant populations in their decision making, policymakers could predict skill shortages and adjust immigration quotas accordingly. Current U.S. admission criteria are based mostly on family reunification; however, the United States could adopt the policy followed by many other countries, that of putting more emphasis on immigrants' skills. Immigrants with specialized education and work experience could be admitted to minimize shortages of workers in specific industries. Many new immigrants are young and in the early stages of their working lives, therefore targeted immigration could also be used to support long-term fiscal programs, such as Social Security.

Considerable disagreement exists within the literature as to the trends in recent U.S. immigrants' skills, with results that depend upon the source and frequency of the data studied and the comparison group (natives or previous immigrants). Most use Census or Current Population Survey (CPS) data to study immigration. For instance, Chiswick (1986) and LaLonde and Topel (1991) use data from the 1970 and 1980 Census and report a decline in the earnings of immigrants during the 1970s. Borjas (1995, 1999) does essentially the same, but includes data from the 1990 U.S. Census and finds a similar, but slowing decline in the skills possessed by incoming immigrants through the 1980s. Using annual 1979-1989 CPS data, both Funkhouser and Trejo (1995) and Cohen, Zach, and Chiswick (1997) indicate that the decline in the skill level of immigrants in the 1960s and 1970s ended in the 1980s. However, Betts and Lofstrom (2000), using the 1970-1990 Census data, discover an absolute increase in the educational attainment of immigrants over time but a relative decrease in the average immigrant education level compared with that of the native population. Unlike other studies, Barrett (1996) and Jasso, Rosenzweig, and Smith (2000a) use annual data from the U.S. Immigration and Naturalization Service (INS), but they also find an improvement in the skills of newly arrived immigrants in the mid-1980s.

According to Schultz (1998), there is no evidence suggesting that the improvement in the average immigrant's skill level during the 1980s continued into and through the 1990s. The most recent studies by Greenwood and McDowell (1999) and Jasso, Rosenzweig, and Smith (2000a) end with 1991 and 1995 INS data, respectively. We extend their work and the work of Barrett (1996), using INS data through 1999. Whereas previous studies using Census and CPS data have analyzed immigrants' skills with data that include illegal immigrants, nonimmigrants, (1) and immigrants who are unemployed, we study the skills of legal immigrants who are joining the labor force.

The INS data set is unique in four important ways: (i) It contains information on every new U.S. immigrant at the time of entry, (ii) it includes only legal immigrants in the sample, (iii) it reports the data on an annual basis, and (iv) it reports several individual characteristics, including country of origin, year of admission, and occupation. However, the INS data lack important individual characteristics, including the educational attainment and income of each immigrant. To compensate for this deficiency, we match the occupations in the INS data with educational attainment data from the CPS to generate a proxy for skill.

The paper is organized as follows. In section 2, we summarize recent U.S. immigrant trends. section 3 includes a description of the data and characteristics of the sample data set. Section 4 outlines the methodology used to measure immigrant skill. The results of the analysis are presented in section 5, and section 6 concludes.

2. U.S. Immigration Trends

According to annual INS data, the number of immigrants entering the United States has increased steadily since 1972. In the 1970s, approximately 400,000 immigrants legally entered the United States each year; this number was close to 750,000 per year during the 1990s (INS, 1972-2000).

The composition of immigrants from different regions of the world has changed in the last few decades. The fraction of legal U.S. immigrants from Latin America increased from 31% in the 1980s to 35% recently. Also, the fraction of immigrants from Asia went down from 41% in the 1980s to 35% between 1990 and 1999. Meanwhile, fewer immigrants come from Western Europe (from 14% in the 1970s to less than 8% in the 1990s) while more have come from Eastern Europe and Russia in recent years (11% in the 1990s, up from 3% in the 1970s).

Changes in U.S. immigration policy have led to fluctuations in the type of immigrants allowed to legally enter the United States. The majority (65%) of new immigrants have entered under family preferences between 1972 and 1999. However, beginning in 1992, approximately 13% of all new immigrants enter under employment preferences. In addition, refugees constitute approximately 15% of all new legal immigrants between 1972 and 1999.

The U.S. immigrants of today are older than those in the past. In the early 1970s, the average new U.S. immigrant was approximately 26 years old; in the late 1990s, the average immigrant was over 30 years old. However, there has been little change in the male to female ratio. Females constitute a slight majority (52%) of all new immigrants between 1972 and 1999.

The INS distinguishes between new legal immigrants (i.e., those obtaining a green card) as being either "new arrivals" or "adjustments." New arrivals are immigrants who come to the United States from their home country and are granted permanent residency. Adjustments are people who entered the United States as nonimmigrants and obtain an adjustment to their visa status. In the early 1970s, only 20% of all new U.S. immigrants were adjustments; in recent years, that number has risen to approximately 40%. It is likely that these figures are understated: Massey and Malone (2002) find that approximately two-thirds of new legal immigrants (in 1996) had some U.S. experience prior to arriving in the United States. Thus, our discussion of "new" immigrants includes many immigrants with previous U.S. experience, including (but not limited to) adjustments.

3. Data

Historically, social scientists have used U.S. Census or CPS data to analyze the skill composition of immigrants. However, there are some shortcomings in trying to assess immigrants' skills over time with these data sets. (2) The Census and CPS do not typically capture immigrant characteristics at their time of entry. (3) Also, CPS and Census respondents include illegal immigrants, who represent a significant portion of total immigrants. Warren (1999) estimates that approximately 20% of the foreign-born population in the United States is illegal, while approximately 30% of the immigrant flow is illegal each year, based on INS calculations. (4)

Because the focus of this paper is to address the quality or skill composition of recent legal immigrants, we use INS data that contain information regarding every person that received a U.S. green card between 1972 and 1999. Unlike the CPS and Census, the INS data are reported annually (by year of admission) and contain information about all new legal immigrants at the time of entry.

An important drawback with using INS data to analyze the skill composition of immigrants is that they do not contain information on income or educational attainment, the standard quantifiable measures of skill. However, they report the occupation of each immigrant. For immigrants who qualify for immigration based on their job skills (i.e., employment-based preferences), the reported occupation describes the employment they will be performing in the United States. For all other types of immigrants, the data set may list the occupation he or she had in their last country of residence.

The INS data contain approximately 16.9 million U.S. immigrants who received green cards between 1972 and 1999. However, our sample is restricted to immigrants in the working population because we use an immigrant's occupation to infer labor market quality. In the INS data set, approximately 31.6% were children (less than 16 years of age) and students, 14.8% were homemakers, 9.9% were retirees and unemployed workers, and 8.2% did not report an occupation. Thus, our sample consists of approximately six million new legal U.S. immigrants, both male and female, who were either employed in their home country or planned to be employed in the United States.

Obviously, we are missing a large population of immigrants whose educational profile is important to our analysis. Many immigrants who do not report their occupation at the time of entry will undoubtedly work in the United States, especially homemakers, unemployed immigrants, and students. Therefore, our analysis of the trends in immigrant quality depends critically on the sample selection of those who report occupations over the time series. In Figure 1, we plot the fraction of working-age immigrants (i.e., between the ages of 16 and 65) who report an occupation in the INS data set for the entire working-age population and by immigrant type (new vs. adjustment or refugee). (5) Other than the periods 1980-1983 and 1998-1999, the fraction of the entire working-age population that reports an occupation is relatively constant, at approximately 47%. (6) Thus, across most of the time series, our sample is representative of immigrants who will work in the United States.

[FIGURE 1 OMITTED]

Unlike the popular Census occupational coding, the occupation variable in the INS data set has been highly aggregated since 1983. Between 1972 and 1973, the INS used the 1960 Census detailed codes; between 1974 and 1982, the INS used the 1970 detailed Census codes; between 1983 and 1999, the INS used a 25-category classification based on the 1980 detailed Census codes; since 1999, it has used a 50-category classification. We map all of the different occupational coding systems into the 25-category system used by the INS between 1983 and 1999, following Jasso, Rosenzweig, and Smith (2000a). (7) The INS occupational codes are listed in Table 1. In the following section, we consider the role of each recoding and how it affects the trends in immigrant quality.

There is another potential problem with the occupation variable. It is quite possible that immigrants who report their home country occupation may not obtain employment in that occupation in the United States: Several occupations do not allow skills to be perfectly transmitted across countries. For example, a physician in Mexico may not be able to immediately obtain employment as a physician in the United States because of differences in licensing. Nevertheless, the occupation that the immigrant pursued in their home country is a signal of the immigrant's ability. If there were an influx of foreign physicians in the United States, regardless of whether they were able to obtain employment in the United States as physicians, the immigrant pool would be more educated than if there were an influx of foreign farmers.

To see if the occupations in the INS data represent the actual occupations of immigrants after arrival, we compare the distribution of occupations from the INS data with the distribution of immigrant occupations in the March files of the CPS data set. The only years of overlap of the two data sets are 1994-1999, and to be consistent with the INS data, the CPS data are sorted to include only new immigrants by year of entry. (8) Figure 2 shows that the occupational distributions are quite similar across the two data sets. If we take the CPS data as true, Figure 2 suggests that the INS data understate the fraction of service workers, postsecondary teachers, and math and computer scientists. Alternatively, the INS data slightly overstate the fraction of laborers, executives, secretaries (ASP), and engineers. However, there seems to be no systematic under- or overstating of skill levels.

[FIGURE 2 OMITTED]

The INS fiscal year is different from year of entry. (9) We reorganize the data by actual year of admission to allow us to compare new immigrants in the INS data with new immigrants in the CPS. We realize that the INS data do not accurately depict year of entry for all immigrants. As mentioned previously, many immigrants are adjustments and enter the United States well in advance of receiving their green cards.

Greenwood and McDowell (1999) highlight two other drawbacks of the INS microdata: (i) The INS is missing approximately 15% of the 1979 sample, and (ii) some of the observations between 1980 and 1983 have missing variables. However, they do not find a significant amount of bias from either omission. (10)

Since our sample is restricted to immigrants in the working population, the sample characteristics are different from the characteristics of the entire INS data set, especially with respect to sex, age, region of origin, and entry classes. Many female immigrants are classified as homemakers and will not enter the United States as labor market participants. However, increasing numbers of women enter the United States with intentions to work; women constitute 41% of the 1990s cohort in the sample, up from 35% in the 1970s. (11) To date, almost all studies on this topic limit their analysis to immigrant men; exceptions include Duleep and Sanders (1993), Greenwood and McDowell (1999), and Betts and Lofstrom (2000).

The average age of immigrants in the sample is much higher than in the entire INS data set since we consider only those immigrants who plan to enter the U.S. workforce (i.e., older than 16 years of age). Also, the trends in region-of-origin representation in the sample fluctuate between 1972 and 1999: There are periods when the distribution of immigrants in our sample is quite different from the distribution in the entire INS data set. Lastly, while the distribution of immigrant entry classes (employment, family, refugee, etc.) is quite similar across our sample and the entire INS data set, our sample overstates new arrivals, especially in 1998 and 1999. The overall characteristics of the sample will be important when sorting out the reasons behind changes in immigrants' skills over time; we return to these issues in section 5.

4. Measuring the Skill Composition of Immigrants

Barrett (1996, 1998) and Jasso, Rosenzweig, and Smith (2000a) use annual earnings to measure immigrants' skills. Specifically, Jasso, Rosenzweig, and Smith (2000a) calculate average earnings of all native U.S. men in 1979 (age 21 through 65 in the labor force) for each of the 25 occupational categories using Census data. (12) Based on the occupation reported, these earnings are assigned to each respondent in the INS data set.

There are several disadvantages with using annual earnings in 1979 as a proxy for skill: it is possible that 1979 was not a representative year for earnings in all occupations. For example, the difference in earnings between a lawyer and a mechanic was much larger in 1989 (in real terms) than in 1979 because of structural changes in the U.S. economy. A fixed price index would not capture these differences. In addition, educational attainment in the United States has been steadily rising in recent decades, while real income has fluctuated. We find education to be a more reliable proxy for skill than income in this context. In addition, preliminary evidence from a new data set, the New Immigrant Survey, suggests that new legal immigrants (in 1996) were as well-schooled as the native-born in the United States (Jasso, Rosenzweig, and Smith 2003). Thus, using native education levels as a proxy for new legal immigrants seems reasonable.

Using CPS data, we compute the predicted average educational attainment of the U.S. population for each occupation over the period 1972-1999. We focus on CPS respondents who are at least 16 years of age and report occupation, educational attainment, age, sex, and marital status. (13) For each occupation classification, we estimate the average educational attainment (educ) of the U.S. population between 1972 and 1999, controlling for age, sex, and marital status. Sex and marital status are represented by dummy variables (female and married). Specifically, we run the following ordinary least squares (OLS) regression for each occupation i:

(1) [educ.sub.i] = [[beta].sub.0,i] + [[beta].sub.1,i] age + [[beta].sub.2,i] female + [[beta].sub.3,i] married + [[epsilon].sub.i]

and obtain 25 sets of coefficients {[[beta].sub.0,i], [[beta].sub.1,i], [[beta].sub.2,i], [[beta].sub.3,i]}; one for each occupational classification. Using the estimated coefficients, we compute the predicted educational attainment of immigrants based on their individual characteristics (from the INS data). After controlling for individual characteristics, an immigrant in occupation i will have the same educational attainment as a member of the U.S. population in occupation i.

Our method controls for the increasing educational attainment that characterized the U.S. labor force during this period. We assign the same predicted educational attainment to immigrants that are exactly alike (in age, sex, marital status, and occupation) regardless of their year of entry into the United States. If we find an upward trend in the predicted educational attainment of immigrants, our methods guarantee that it is caused only by changes in the overall characteristics of immigrants (i.e., the distribution of immigrant occupation). As additional support, we compare the actual and predicted educational attainment of CPS immigrants in the Appendix and find no evidence of an upward bias in our proxy.

We realize our measure of educational attainment is not perfect since U.S. data are used to estimate the coefficients. First, immigrants and natives in the same occupation may not have the same level of education. Immigrants in professional jobs often have more education than natives, while immigrants in manufacturing jobs are typically less educated. Thus, the predicted level of immigrant education may differ from actual educational attainment. Still, a group of immigrants with more professionals, ceteris paribus, would result in an increase in the actual, as well as the predicted, educational attainment of immigrants. Our proxy would accurately represent this trend. (14) The second problem with our skill proxy is the systematic differences in educational levels by occupation across source countries. That is, farmers from Mexico may have very different educational attainment than farmers from Canada. Since we use U.S. data to impute education, Mexican and Canadian farmers would have the same amount of education (after controlling for other individual characteristics). Nonetheless, we provide evidence in the appendix that supports our estimate of educational attainment as a reliable proxy to analyze trends in the skills of different immigrant populations.

5. Results

First, we describe the general trends in immigrant quality that we observe using our skill proxy. Then, we discuss the various mechanisms that are driving the trends.

General Trends

Figure 3 illustrates the predicted average educational attainment of immigrants in the INS sample between 1972 and 1999. As a benchmark, we plot the predicted average educational attainment of natives from the CPS. Note that these are not actual levels of education for natives: Predicted educational attainment is computed using our skill proxy. (15) In addition, our skill proxy overstates immigrants' actual education levels (as discussed in the Appendix); therefore, it cannot be used to directly compare immigrant and native labor market quality at a point in time. However, this series provides a point of reference when considering the trends in immigrant quality. According to Figure 3, the predicted educational attainment of natives is steadily increasing throughout our time series, indicating that natives are in occupations that require more education over time. This compares to the observed volatility in the predicted educational attainment of immigrants, suggesting that immigrant quality is much less predictable from year to year. For the remainder of the study, we analyze the trends in immigrant quality by comparing immigrants across cohorts. That is, we compare recent cohorts of immigrants to past cohorts.

[FIGURE 3 OMITTED]

The skill level of legal immigrants is quite high in 1972, with an average predicted educational attainment of 12.88 years. This high quality generally continues until 1978, when the skill level of new legal immigrants declines by 2.5%, according to our proxy. The decline of immigrant quality continues through the mid-1980s. By 1987, new legal immigrants have 12.43 years of education according to our calculations, a reduction of approximately 3.5% since 1972.

Immigrants' skills improve after 1987. Between 1987 and 1993, skill level increases by 3.6%, reaching a predicted educational attainment similar to that of the early 1970s cohort. This finding supports the work of Barrett (1996) and Cohen, Zach, and Chiswick (1997) who find that the decline in the skill level of immigrants slows in the 1980s.

We find a further fall in skill level from 1993 to 1997, which is shorter and less severe in scope than the 1972-1987 decline. For example, in 1997, new immigrants are only 1.3% less educated (in predicted levels) than new immigrants arriving in 1993.

Immigrants' skills improve once again in the late 1990s; in fact in 1998 newly-arrived immigrants have the highest level of predicted education in the entire sample. This strong recovery of immigrant skill compensates for the mild decline in skills earlier in the 1990s.

We next test to see if the changes in immigrant skill observed in Figure 3 are significant. We conjecture that the years 1987, 1993, and 1997 represent changes in trend; i.e., the slope in the trend changes at these points. We generate linear splines between the end points and test whether their slopes are equal using an F-test. We find that the slope of the trend line from 1972 to 1987 is significantly different from the slope of the trend line from 1987 to 1993; also, 1987 to 1993 differs from 1993 to 1997, which, in turn, differs from 1997 to 1999. (16) Thus, immigrant skill changes significantly over the time series.

Sources of Trends

The overall quality of new legal immigrants is driven by various sources, including the individual characteristics of immigrants, immigration policy, sample selection, primary regions of origin, and occupation recoding. In this section, we discuss the role of each of these mechanisms in affecting the observed trends in immigrant quality. We decompose the time series into four periods using the trend breaks developed above. The four periods we analyze are 1972-1987, 1988-1993, 1994-1997 and 1998-1999.

To depict the various mechanisms that affect the level of immigrant quality (as measured by our skill proxy), we compare the predicted educational attainment of various subgroups of our sample and relate it to the distribution of these subgroups over time. We compare immigrants by their type (new arrival vs. adjustment), across immigrant entry classes (family, employment, refugee, etc.), and by source regions. (17)

We also run a series of regressions to compare the importance of individual characteristics in explaining the levels of immigrant quality across cohorts. We compare the predictors of educational attainment across the four immigrant cohorts defined above. Using OLS, we regress the estimated educational attainment of immigrants on several independent variables, including immigrant type, region of origin, and entry class, all of which are represented by dummy variables. We run the following regression:

(2) [educ.sub.j] = [[beta].sub.0,j] + [[beta].sub.1,j] New + [[beta].sub.2,j] WE + [[beta].sub.3,j] EE + [[beta].sub.4,j] Asia + [[beta].sub.5,j] Otherreg + [[beta].sub.6,j] Family + [[beta].sub.7,j] Employ + [[beta].sub.8,j] Div + [[beta].sub.9,j] IRCA + [[beta].sub.10,j] Otherpref + [e.sub.j]

where [educ.sub.j] represents predicted educational attainment of immigrants that enter the United States in period j; New is a dummy variable for new arrival immigrants; and WE, EE, Asia, and Otherregion are dummy variables for immigrants from Western Europe, Eastern Europe, Asia, and the Other region, respectively; Family, Employ, Div, IRCA, Otherpref are dummy variables for immigrants who enter the United States under family preferences, employment preferences, diversity preferences, IRCA preferences, and other preferences, respectively; and e is the error term. Given this formulation, the comparison group for the regression results is Latin American immigrants that obtain an adjustment and enter the United States under refugee preferences.

1972-1987

During the 1972-1987 period of decline, the INS recoded occupations in INS fiscal years 1974 and 1983. According to Figure 3, there is a slight decline (rise) in immigrant skill in entry year 1973 (1982). In addition, the period 1980-1982 includes many observations with missing variables. Figure 1 shows a significant fall in occupation reporting rates between 1980 and 1982 for the entire INS population. In addition, adjustments and refugees begin reporting occupations at lower frequencies than new arrivals in 1982. At least some of the volatility in overall immigrant quality prior to 1982 is due to these factors. However, since the 1982 recoding, we find a consistent decline in immigrant quality that lasts until 1987.

Much of the decline in quality between 1972 and 1987 is attributable to a significant decline in the predicted educational attainment of adjustments and refugees, as illustrated in Figure 4. A pronounced decline in adjustment and refugee quality occurs between 1982 and 1987, coinciding with the Refugee Act of 1980, which increased the number of refugees admitted to the United States. According to Figure 5, the fraction of adjustments and refugees in the sample increases in entry year 1982 and remains high thereafter. The quality decline occurred because refugees are of relatively low labor market quality (Figure 6). Even though their quality has not significantly changed during the period, the dramatic increase in the number of immigrants admitted under refugee preferences has brought the average down. Figures 6 and 7 show that the same is true for immigrants admitted under family preferences; their share increased in the 1980s, but their average quality did not change.

[FIGURES 4-7 OMITTED]

We find significant differences in the predicted educational attainment of immigrants from different source regions for the 1972 to 1987 period (Figure 8). This period is characterized by a dramatic fall in the quantity of high-skilled Western European immigrants (Figure 9). In addition, there is a substantial fall in the skill of Asian immigrants between 1972 and 1978, due primarily to an influx of 35,000 Vietnamese immigrants in 1978. This leads to an overall decline in immigrant skill that lasts until 1987.

[FIGURES 8-9 OMITTED]

The first column of Table 2 presents the regression results for the 1972-1987 cohort. We find significant differences in the labor market quality of new immigrants and immigrants who obtain an adjustment in their immigrant status. The coefficient for the new immigrant dummy variable is negative. The regression results also indicate that Asian immigrants are in occupations that require the most education of all new U.S. immigrants (with the largest positive region coefficient). Since all of the other region coefficients are positive, Latin American immigrants, the comparison group, are in occupations with the lowest educational requirements. Table 2 also reveals that immigrant entry class is an important predictor of labor market quality. Immigrants admitted under employment preferences (while small in number during this period) demonstrate very high predicted educational attainment while those admitted under refugee preferences (the control group) have the least.

In general, the steady decline in quality between 1972 and 1987 is the result of a significant reduction in the labor market quality of adjustment immigrants, an increase in the number of immigrants entering under family preferences, and a steady decline in the predicted educational attainment of Asian immigrants (who constitute one-third of all immigrants during this period).

1988-1993

Between 1988 and 1993, the predicted educational attainment of adjustment immigrants increased dramatically, with 13.5 years of predicted education in 1993 compared with 12.3 years in 1988 (as seen in Figure 4). Similar to the previous cohort, the coefficient for the new immigrant dummy variable is negative, as evident in column 2 of Table 2. The 1988-1993 cohort exhibits the largest differential between new immigrants and adjustments: Adjustments have 53% more years of predicted educational attainment than new arrivals. The improvement in the labor quality of adjustments coincides with a period during which fewer adjustments report their occupations (Figure 1). We do not know the extent of the sample selection bias and if the increase in predicted education of adjustments is over- or understated. However, it is likely that at least part of the observed improvement in overall immigrant quality during this period is due to the labor market improvement of adjustments.

This period is also characterized by significant changes in immigrant entry class due to changes in U.S. immigration policy. The proportion of immigrants entering under family preferences falls, especially from 1990 to 1993 (Figure 7). This is associated with a dramatic increase in employment-based immigrants due to the Immigration Act of 1990 (which allotted 140,000 visas annually to employment-based preferences). Almost 23% of all immigrants enter under employment preferences in 1993, up from less than 1% in 1991. The 1990 Act is also responsible for the diversity preference, which represented more than 8% of all immigrants in 1992 and 1993 in our sample. According to the regression results for this cohort (column 2 of Table 2), immigrants with employment and diversity preferences have the highest predicted educational attainment of all entry class types, resulting in an increase in the overall predicted educational attainment of immigrants during this period. This is despite the presence of relatively low-skilled immigrants admitted under the auspices of the Immigration Reform and Control Act after 1990 (see Figures 6 and 7). (18)

The 1988-1993 period witnessed an improvement in the labor market quality of immigrants from all regions of origin, as evident in Figure 8. However, Asia surpasses Latin America as the largest supplier of U.S. immigrants (Figure 9) and experiences a significant rebound in labor market quality in 1992 and 1993 (Figure 8).

The improvement in immigrant quality between 1988 and 1993 is due to several factors: The predicted educational attainment of adjustments increases, immigration policy favors more highly-skilled immigrants by increasing the amount of employment and diversity preferences, and immigrants from all regions of origin improve in labor market quality.

1994-1997

Most of the slight decline between 1993 and 1997 is due to a reversal of the situation from the previous period. Specifically, more adjustments enter the United States during this period (Figure 5) but they are of lower labor market quality than the previous cohort of adjustments (Figure 4). The 1994-1997 cohort, however, has the smallest differential in predicted educational attainment between adjustments and new immigrants, as evident in column 3 of Table 2.

Compared to the 1988-1993 cohort, immigration policy during the 1994-1997 period favors less-skilled immigrants. Figure 7 shows that the fraction of immigrants entering under family preferences increases between 1994 and 1997, while fewer immigrants enter under employment and diversity preferences.

Based on the regression results, all of the region coefficients are larger for the 1994-1997 cohort relative to the previous cohort, indicating that immigrants from all source regions are improving relative to the Latin American group (the control group). That is, immigrants from Latin America have slightly less predicted educational attainment in this cohort than the previous cohort, a fact confirmed in Figure 8.

The composition of source regions changes during this period: Immigrants from Latin America and the Other region increase in quantity but decline in skill, (19) while immigrants from Western Europe improve in skill but decrease in quantity. The combination of these effects leads to a decrease in the overall skill level of new U.S. immigrants between 1994 and 1997. Our results confirm that shifts in the primary source regions have a large impact on the overall labor market skill of U.S. immigrants. (20)

1998-1999

There are several driving forces behind the dramatic improvement in the labor market quality of immigrants in 1998 and 1999 observed in Figure 1. First, the 1998 recoding of occupations may be producing an upward bias in our skill proxy. To determine the extent to which the recoding contributed to this spike in immigrant quality, we apply the Oaxaca method. That is, we hold the distribution of source region and entry class type at their 1997 levels and calculate the predicted years of education in 1998 and 1999. This yields an estimate of 12.88 years of predicted educational attainment for 1998, a 1.26% increase from 1997; this compares with a 2.9% increase using the previous estimate for 1998. Therefore, the 1998 occupation recoding and other unobservable factors account for 43% of the increase. We did the same for 1999 and found that 88% of the increase is due to unobservable factors. Thus, the 1998-1999 improvement is overstated. Still, over half of the improvement in 1998 is due to observable factors (which are discussed further on), so that overall immigrant skills improved in 1998 relative to the 1994-1997 cohort. However, the results for 1999 are not reliable since the recoding is significantly distorting our skill proxy that year.

In addition to occupation recoding, some of the increase in the predicted educational attainment of immigrants in 1998 and 1999 is attributable to sample selection bias. According to Figure 1, there was a significant decline in the percent of adjustments and refugees that report their occupation in 1998 and 1999; only 12% of the working-age adjustment and refugee population reports an occupation, down from 40% in the early 1990s. This decline is associated with a dramatic rise in the labor market quality of adjustments and refugees (see Figure 4). To determine the nature of the selection bias, we compare the sample of working-age adjustments and refugees who report an occupation with those who do not report an occupation in 1998 and 1999. Based on the distributions of regions of origin and entry preferences, adjustments and refugees that do not report occupations would likely have less predicted educational attainment than those included in our analysis (i.e., those who report occupations). Specifically, nonreporting adjustments and refugees have a larger percent of immigrants originating in Latin America and more (fewer) immigrants entering under family (employment) preferences than reporting adjustments. These differences would yield lower predicted educational attainment than our proxy suggests.

However, new arrival immigrants have less predicted educational attainment than refugees and adjustments over most of the time series (when selection bias is not an issue), with a smaller gap between new arrival immigrants and adjustment and refugee immigrants in recent cohorts (based on the regression results in Table 2). The oversampling of new arrival immigrants, who are likely to be less skilled, causes a drop in the overall predicted educational attainment of immigrants for the 1998 to 1999 period. (21) However, the overstatement of the skills of adjustments and refugees (as discussed previously) causes an increase in the overall predicted educational attainment of immigrants. It is probable that the latter effect dominates so that the observed rise in immigrant labor market quality for 1998 and 1999 is overstated. Even so, the dramatic improvement in the overall labor market quality observed in 1998 (in Figure 3) is not likely to be entirely explained away by selection bias and occupation recoding. However, for 1999, the effects of occupation recoding and selection bias seem large enough to offset much of the increase in predicted educational attainment (relative to pre-1998 levels), making it difficult to assess the labor market quality of immigrants in this year.

While unobservable factors are overstating immigrant skill in 1998 and 1999, changes in observable factors such as entry preferences and region of origin explain part of the increase in skill. Fewer immigrants are entering under low-skill refugee preferences and more are entering under high-skill diversity preferences, relative to previous cohorts (refer to Figures 6 and 7). In addition, all of the region coefficients (in column 4 of Table 2) are larger relative to the 1994-1997 cohort, indicating that all groups of immigrants are improving relative to the Latin American group. Thus, Latin American immigrants are once again declining in quality relative to all of the other regions of origin. Figure 9 illustrates a drop in the percentage of immigrants from Latin America (who are of lower skill), a subsequent rise in Asian immigrants, and a slight increase in Western European immigrants (who are the most skilled, according to the regression results) during the 1998-1999 period. However, changes in the distribution of source countries would be less severe if more adjustments reported their occupation (as discussed previously).

In general, the occupation recoding and sample selection bias in the 1998-1999 INS data make it difficult to compare this cohort of immigrants with previous cohorts. Given the magnitude of the observed increase in predicted educational attainment in 1998, it is likely that immigrant quality improved relative to the 1994-1997 cohort, albeit to a lesser extent than our skill proxy suggests. However, the results for 1999 seem unreliable. We confirm these findings using CPS data (in the Appendix), where we find a significant improvement in immigrant quality in 1998, followed by a dramatic decline in 1999.

6. Conclusion

In this paper, we examine the changing skill composition of U.S. immigrants using the public use files produced by the U.S. Immigration and Naturalization Service. Since the INS data do not contain information on the educational attainment or income of immigrants, we use the CPS to generate a skill proxy based on occupation. We focus on new legal immigrants (male and female) who were employed in their home country or in the United States at the time of immigration. The sample used in this analysis represents approximately six million new legal immigrants who entered the United States between 1972 and 1999.

Our results support the work of Barrett (1996) and Cohen, Zach, and Chiswick (1997) who find that the decline in immigrant skill experienced in the 1970s and early 1980s reverses in the late 1980s. We extend the literature by including a comprehensive study of recent immigrants. Between 1987 and 1993, immigrants from all major regions of the world improve in labor market quality. By 1994, however, immigrants from the primary source regions (Latin America and Asia) decline in skill, leading to another decline in overall immigrant skill in the United States. This period of decline is shorter and less severe than that of the 1980s; new legal immigrants once again improve in quality in 1998. However, the length and magnitude of the improvement in the late 1990s is unknown due to problems associated with the data set.

We find that fluctuations in overall immigrant quality are driven by three primary sources: (i) changes in the quality and quantity of immigrants obtaining an adjustment, (ii) immigration policy that alters the distribution of immigrant entry class type, and (iii) the distribution of source regions.

In addition, we find several other interesting features concerning the trends of new legal U.S. immigrants. First, immigrant skills, as measured by our skill proxy, differ significantly over the time series and are more volatile than native skills, validating our study of annual fluctuations in immigrant quality. Second, the predicted educational attainment of Latin American immigrants between 1972 and 1999 exhibits little fluctuation while immigrants from all other source regions improve in quality. Third, immigrants who obtain an adjustment to their immigrant status compared with those who enter directly from their home country exhibit high levels of variability in their labor market quality over the time series; this has a significant impact on overall trends. Lastly, changes in immigration policy that control the types of immigrants admitted to the United States, including the 1980 Refugee Act and especially the Immigration Act of 1990, have a direct effect on the overall skill quality of new immigrants.

Appendix

To determine the effectiveness of our skill proxy to accurately describe trends in immigrant skill, we compare our results with immigrant trends in recent CPS data. It is likely that our skill proxy overstates immigrant education levels since native education levels are used to generate the proxy. However, Jasso, Rosenzweig, and Smith (2000b) state that "... it would be closer to the troth to use the CPS native-born education to characterize legal immigrants than to use the CPS foreign-born population" (p. 132). Even so, our aim is to analyze immigrant trends. As long as the errors in the proxy are constant over time, our methods are credible. To get a sense of how much the proxy misstates actual immigrant education levels and to see if the proxy is constant over time, we compare actual education levels of new CPS immigrants with predicted levels of education for new CPS and new INS immigrants. Recall, however, that immigrants in the CPS and INS samples are quite different (i.e., the CPS includes illegal immigrants).

According to Figure A1, predicted education levels of CPS immigrants are significantly higher than the actual CPS values (0.4 years of education on average), but they move together over time. (22) CPS data (both actual and predicted) suggest a significant improvement in immigrants who enter the United States in 1998, consistent with the INS data. Different than INS immigrants, CPS immigrants who enter the United States in 1999 decline in skill relative to the 1998 cohort. Both the CPS and INS samples indicate that immigrant skill was fairly constant between 1994 and 1997. Somewhat surprising is that CPS immigrants have slightly more predicted education than INS immigrants between 1996 and 1998. However, our null hypothesis is that CPS- and INS-predicted educational attainments are the same; we cannot reject this hypothesis (F = 0.03). This provides at least partial support for our use of a skill proxy to analyze immigrant trends.

Our results are consistent with preliminary evidence from the New Immigrant Survey. Jasso, Rosenzweig, and Smith (2003) report that new legal immigrants in the 1996 cohort of the New Immigrant Survey had 12.7 years of schooling; this is precisely the predicted level of educational attainment of new legal immigrants in our INS sample, providing additional support to our proxy.

The standard deviation of actual educational attainment of CPS immigrants is quite large between 1994 and 1999, ranging from 4.4 to 4.9 years. This contrasts with the relatively low and constant standard errors of the predicted educational attainments of CPS and INS immigrants (1.9 to 2.2 years). In addition, the standard error of the predicted educational attainment for INS immigrants is relatively constant between 1972 and 1999.

An additional way to build credibility for our skill proxy is to compare the educational attainment of INS and CPS immigrants by occupation. Figure A2 plots actual and predicted education levels for CPS immigrants as well as predicted education levels of INS immigrants in most occupations. (23) One should expect some level of overstatement of education levels for low skill occupations, like fanners (FFF) and laborers (LAB), where the gap between the proxy (predicted education levels) and actual education levels is significantly large and positive. However, there are several occupations, most of which are health related (DOC, HLD, HLT, TNH), where the proxy understates immigrant education levels. Closer examination of these differences indicate that in all health-related occupation besides doctors (DOC), even considering our small sample size, we cannot reject the hypothesis that actual and predicted education levels are significantly different (using z-tests). (24) Thus, in general, the proxy does a good job in estimating education levels for many occupations (i.e., LAW, NUR, and SLS); it overstates education levels for a few common low skill immigrant occupations (FFF and LAB), but only significantly understates education levels for doctors. Since a majority (60%) of CPS immigrants is in low skill occupations that are overstated by the proxy, the overall predicted education of CPS immigrants is overstated (in Figure A1).

Lastly, we include the relevant sample means in Table A1. Immigrants in the CPS sample are less likely to be married, but not significantly so, and they are younger than INS immigrants. These additional levels of heterogeneity provide some evidence that the INS and CPS immigrants are capturing, at least in part, different types of immigrants.

Table A1. Sample Characteristics: INS vs. CPS, 1994-1999

 Significantly
 Different
 INS (999) CPS (5% Level)?

Percent female 40.8% 34.6% Yes
Average age 36.6 30.0 Yes
Average education 12.8 * 12.3 Yes
Percent married 45.8% 47.3% No
Sample size 1,377,485 1280

* Predicted educational attainment.

References

Barrett, Alan. 1996. Did the decline continue? Comparing the labor-market quality of United States immigrants from the late 1970s and late 1980s. Journal of Population Economics 9:55-63.

Barrett, Alan. 1998. The effect of immigrant admission criteria on immigrant labour-market characteristics. Population Research and Policy Review 17:439-56.

Betts, Julian R., and Magnus Lofstrom. 2000. The educational attainment of immigrants: Trends and implications. In Issues in the economics of immigration, edited by George J. Borjas. Chicago and London: The University of Chicago Press, pp. 51-115.

Borjas, George J. 1995. Assimilation and changes in cohort quality revisited: What happened to immigrant earnings in the 1980s? Journal of Labor Economics 13:201-45.

Borjas, George J. 1999. Heaven's door: Immigration policy and the American economy. Princeton, N J: Princeton University Press.

Chiswick, Barry R. 1986. Is the new immigration less skilled than the old? Journal of Labor Economics 4:168-92.

Cohen, Yinon, Tzippi Zach, and Barry R. Chiswick. 1997. The educational attainment of immigrants: Changes over time. Quarterly Review of Economics and Finance 37:229-43.

Duleep, Harriet O., and Mark C. Regets. 1998. Projecting immigrant earnings: The significance of country of origin. Social Security Bulletin 61:32-44.

Duleep, Harriet O., and Seth Sanders. 1993. The decision to work by married immigrant women. Industrial and Labor Relations Review 46:677-90.

Funkhouser, Edward, and Stephen J. Trejo. 1995. The labor market skills of recent male immigrants: Evidence from the Current Population Survey. Industrial and Labor Relations Review 48:792-811.

Greenwood, Michael J., and John M. McDowell. 1999. Legal U.S. immigration: Influences on gender, age, and skill composition. Kalamazoo, MI: W.E. Upjohn Institute fur Employment Research.

Jasso, Guillermina, Mark R. Rosenzweig, and James P. Smith. 2000a. The changing skill of new immigrants to the United States: Recent trends and their determinants, ha Issues in the economics of immigration, edited by George J. Borjas. Chicago and London: The University of Chicago Press, pp. 185-225.

Jasso, Guillermina, Mark R. Rosenzweig, and James P. Smith. 2000b. The new immigrant survey pilot (NIS-P): Overview and new findings about U.S. legal immigrants at admission. Demography 37:127-38.

Jasso, Guillermina, Mark R. Rosenzweig, and James P. Smith. 2003. The new immigrant survey in the U.S.: The experience over time. Washington, DC: Migration Policy Institute.

LaLonde, Robert J., and Robert H. Topel. 1991. Immigrants in the American labor market: Quality, assimilation, and distributional effects. American Economic Review 81:297-302.

Massey, Douglas S., and Nolan J. Malone. 2002. Pathways to legal immigration. Population Research and Policy Review 21:473-504.

Schultz, T. Paul. 1998. Immigrant quality and assimilation: A review of the U.S. literature. Journal of Population Economics 11:239 52.

U.S. Department of Justice, Immigration and Naturalization Service. (INS) 1972-2000. Immigrants admitted into the United States. (ICPSR 8952 8956, 8958-8966, 7999, 9268-9269, 6161,6164-6165, 6449, 6456, 2267, 2534, 2955-2956, 3485-3486).

U.S. Department of Justice, Immigration and Naturalization Service (INS). 1999. Statistical yearbook of the Immigration and Naturalization Service. Washington, DC: U.S. Government Printing Office.

Warren, Robert. 1999. Unauthorized immigrants residing in the United States: Estimating the population, components of change and trends, by broad area of origin, 1987 to 1997. Unpublished paper, U.S. Immigration and Naturalization Service.

(1) Nonimmigrants include international students, tourist-overstays, and temporary workers.

(2) Greenwood and McDowell (1999) and Jasso, Rosenzweig, and Smith (2000a, b) call attention to some of these problems.

(3) Of course, the CPS and Census capture some immigrants who have recently moved into the United States since each data set records year of entry. However, the CPS sample size of new immigrants is very small because it captures relatively few immigrants in total. Only in 2000 did the Census begin recording exact year of entry. Prior to 2000, entry periods were collapsed into broad categories that lumped many years together.

(4) Both estimates are for 1997. Illegal immigrant flow data are from the Statistical Yearbook of the Immigration and Naturalization Service, 1999.

(5) The differences in reporting rates by immigrant type will be important when discussing the sources of trends in immigrant quality, in section 5.

(6) The 1980-1983 period has very low reporting rates because of observations with several missing variables, as reported in Greenwood and McDowell (1999).

(7) Details concerning the occupation mapping are available upon request. The overall results of the paper do not change with small adjustments to this mapping.

(8) We classify new immigrants in the CPS as those who lived in a different country last year. There are significant differences in the number of observations in the two samples: The CPS sample consists of 1100 observations while the INS sample includes 1.38 million observations.

(9) For example, the INS file "Immigrants Admitted into the U.S., 1997" includes immigrants who received a green card between October 1996 and September 1997.

(10) We refer the reader to their work for more details.

(11) It is likely that many of these women are entering the United States under family reunification. Still, our sample suggests that more and more immigrant women are entering the United States with recent employment history or with plans to work in the United States.

(12) Based on the discussion in Jasso, Rosenzweig, and Smith (2000a), it seems as if they did not use age to adjust their predicted earnings by occupation. They state that "earnings were assigned to respondents according to which of the 25 occupations they reported." This is problematic since the age distribution of natives and immigrants has changed over time.

(13) The INS defines children as being less than 16 years of age so we use the same age criteria. Since our sample includes only those who report an occupation, less than 0.7% of our sample is less than 18 years of age: this share remains relatively constant over time.

(14) We are assuming that the relative rankings of occupations by education levels are constant across countries. For example, in the United States, doctors are mole educated than lawyers. This may not be true in every country.

(15) While not actual education levels for natives, they are very close to actual since native data were used to produce our skill proxy.

(16) There are three breaks in trend, 1987, 1993, and 1997. The F-statistics are 33.7, 9.68, and 9.28, respectively.

(17) The list of countries in each region and immigrant entry classes are available from the authors.

(18) The Immigration Reform and Control Act (IRCA) was passed in 1986. However. immigrants admitted under IRCA preferences do not show up in our sample until 1991.

(19) The distribution of countries in the Other region changes in the 1990s. Prior to the early 1990s, the largest share of immigrants in the Other region came from islands in North America, followed by Canada, Africa, and Oceania. Since then, immigrants from Africa constitute the largest share of this category.

(20) Several studies find a strong relationship between the country of origin of immigrants and their skills at the time of entry. Examples include LaLonde and Topel 11991), Barrett (1996), and Duleep and Regets (1998).

(21) It is plausible that the decline in labor market quality of adjustment and refugee immigrants continued in the late 1990s so that they are similarly skilled, or even less so, than new arrival immigrants.

(22) In fact, we find that predicted educational attainment in the CPS data is significantly different from actual educational attainment in the CPS. The F-statistic was 5.04, slightly above the critical value of 4.96.

(23) There were no counselors (COU) and only one architect (ARC) in the CPS immigrant data.

(24) This occurs because, unlike most occupations in our sample, the standard deviation for doctor's educational attainment is small: Most doctors have the same amount of education.

Linnea Polgreen * and Nicole B. Simpson ([dagger])

* University of Iowa, Department of Economics, W210 PBAB, Iowa City, IA 52242, USA; E-mail linnea-polgreen@ uiowa.edu.

([dagger]) Colgate University, Department of Economics, 13 Oak Drive, Hamilton, NY 13346, USA; E-mail nsimpson@mail. colgate.edu; corresponding author.

We are grateful for the excellent comments we received from two anonymous referees. We appreciate suggestions from Beth Ingram, Evelyn Jenkyns, and B. Ravikumar. We also thank Guillermina Jasso and her coauthors for providing the mapping of the Census and INS occupational codes. All errors are our own.

Disclaimer: The data utilized in this paper were made available by the Inter-university Consortium for Political and Social Research. The data for Immigrants Admitted to the United States, 1972-2000, were originally collected by the U.S. Department of Justice, Immigration and Naturalization Service. Neither the collectors of the original data nor the Consortium bears any responsibility for the analyses or interpretations presented here.

Received December 2004; accepted August 2005.

Table 1. INS Occupational Codes

ARC Architects
ART Writers, artists, entertainers and athletes
ASP Administrative support occupations, including clerical
COU Counselors, educational and vocational
DOC Physicians
ENG Engineers
EXC Executive, administrative, managerial occupations
FFF Farming, forestry, and fishing occupations
HLD Other health diagnosing occupations
HLT Other health assessment and treating occupations
LAB Operators, fabricators, and laborers
LAW Lawyers and judges
LIB Librarians, archivists, and curators
MCS Mathematical and computer scientists
NSC Natural scientists
NUR Nurses
PCR Precision production, craft, and repair occupations
SER Service occupations
SLS Sales occupations
SSC Social scientists and urban planners
SWK Social, recreation, and religious workers
TCO Teachers, except postsecondary
TCU Teachers, postsecondary
TNH Health technologists and technicians
TNO Technologists and technicians, except health

Table 2. Regression Results across Cohorts (Standard Errors in
Parentheses)

Variable 72-87 Cohort 88-93 Cohort

Intercept 11.201 (0.003) 11.042 (0.005)
New immigrant -0.440 (0.003) -0.531 (0.004)
Western Europe 0.949 (0.003) 1.227 (0.006)
Eastern Europe 1.115 (0.006) 1.121 (0.006)
Asia 1.344 (0.003) 1.033 (0.004)
Other regions 0.970 (0.003) 0.917 (0.005)
Family preferences 0.998 (0.004) 1.253 (0.006)
Employment preferences 3.238 (0.139) 2.617 (0.007)
Diversity preferences NA 1.337 (0.010)
IRCA preferences NA 0.974 (0.014)
Other preferences 1.587 (0.004) 1.558 (0.006)
Observations 3,062,124 1,586,231
Adjusted [R.sup.2] 0.1457 0.168
Predicted mean 12.67 12.70

Variable 94-97 Cohort 98-99 Cohort

Intercept 10.878 (0.006) 11.108 (0.025)
New immigrant -0.130 (0.004) -0.146 (0.014)
Western Europe 1.316 (0.007) 1.414 (0.016)
Eastern Europe 1.176 (0.007) 1.327 (0.015)
Asia 1.071 (0.004) 1.114 (0.010)
Other regions 1.041 (0.006) 1.045 (0.012)
Family preferences 1.007 (0.006) 0.914 (0.028)
Employment preferences 2.359 (0.007) 2.347 (0.027)
Diversity preferences 1.506 (0.009) 1.706 (0.030)
IRCA preferences 0.561 (0.023) NA
Other preferences 0.485 (0.018) 0.139 (0.025)
Observations 1,586,231 274,344
Adjusted [R.sup.2] 0.2028 0.1717
Predicted mean 12.72 13.04

All variables are significant at the 1% level.