The challenge of assessing the well-being of the foreign born has increased the demand for statistical information about this population. Until recently, the decennial census was the sole source of foreign-born population data. In 1994, the Census Bureau added nativity and citizenship items to the Current Population Survey (CPS). Currently, we are compiling CPS data into an expanding time series of key indicators that can be used to monitor the welfare of the foreign born. In this paper, we update our review and evaluation of these data. First, we highlight aspects of the CPS data development process. Second, we discuss the new nativity items, with a special focus on the citizenship data. We conclude our paper by addressing recent CPS data development issues affecting the series.

 The authors wish to thank Gregory D. Weyland, Current Population Surveys, Demographic Surveys Division for sharing his knowledge of the Current Population Survey; Dennis Schwanz, Demographic Statistical Methods Division for statistical review; Frederick W. Hollmann, Population Projections Branch, Population Division, for providing information about population estimates and projections used in the second stage weighting, and peer reviewing this paper; Kristin Hansen, Migration and Journey to Work, Population Division for her pioneering contribution to the CPS nativity series, including development of the data edit specifications; and the hundreds of Census Bureau interviewers and programmers who collect and process the CPS nativity data monthly.

We also wish to thank the Immigration and Naturalization Service in the Department of Justice, and the National Center for Child Health and Human Development in the Department of Health and Human Services for their ongoing financial support of the CPS nativity series.

This paper was presented in the session: "Measurement and Modeling Issues in Immigration and Emigration" at the annual meeting of the Population Association of America, Chicago, Illinois, April 1-3, 1998. The views expressed are attributable to the authors and do not necessarily reflect those of the Census Bureau.

If you would like a paper copy of this paper, please call the Population Division Statistical Information office at 301-457-2422 or send an e-mail message to pop@census.gov. Include the name of this report and the authors in the body of the message.

Before adding the new nativity items to the CPS, the Census Bureau conducted cognitive research designed to increase the validity and reliability of the information collected. The research showed: the nativity questions could become core questions asked each month; these questions should follow the labor force questions, and precede the family income-related question; skip patterns alleviated respondent burden and improved accuracy; and certain question phrasing facilitated understanding by respondents with poor English language skills or those unfamiliar with terms such as 'born abroad,' 'American parent' or 'citizen by naturalization.'¹⁵

All foreign-born cases are asked at least one of the citizenship questions: a) Are you a citizen of the United States? b) Were you born a citizen of the United States? c) Did you become a citizen of the United States through Naturalization? A "no" to the first question becomes "not a citizen" in the editing process. A "yes" to this question is asked question two. A "yes" to question two skips question three. A"no" to question two is asked question three (see Appendix A for detailed questions).

Prior to editing, the citizenship data include several "citizen" subgroups: a) native, born in one of the U.S. states; b) native born in a U.S. outlying areas; c) foreign born who will be classified as citizens born abroad of American parents because at least one parent was born in the U.S. or U.S. outlying areas; d) foreign born who told us they were not citizens; e) foreign born who told us they were "born citizens", but their parents are foreign born; f) foreign born who told us they are naturalized citizens; g) non-respondents.

When we edit the citizenship item, we use a multi-stage approach. First we assign U.S. citizenship to the cases: a) 'born in the United States'; b) 'born in Puerto Rico or an outlying area'; and c) 'born abroad of American parents'. Next, when we can, we assign cases with a blank value an imputed value based on the characteristics of other household members. If this is not possible, we use an allocation matrix to match a case with a missing value to a case with a citizenship value. Cases with responses to one or more of the citizenship questions are edited and assigned 'naturalized citizen' d), or 'not a citizen' e).

Passel and Clark provide evidence of a discrepancy between the number of persons naturalized in a specified period, as reported by the Immigration and Naturalization Service (INS), and the number of CPS cases reporting naturalized citizenship status who entered the United States to live within five years preceding the survey collection date.¹⁹ Several mitigating factors may underlay the observed discrepancy. First, the inherent nature of each of these data sets make them not strictly comparable. INS data reflect the incidence of an administrative process taking place over time, while the CPS data are a snapshot of current conditions. For example, some naturalized citizens emigrate from the United States, and return 'to stay' or 'to live' at a later date.²⁰ A respondent naturalized in 1989, who left the United States in 1993 and returned in 1995, might truthfully answer' "yes" to the naturalized citizenship question, and provide a year of entry under five years. Second, although immigrants are generally not eligible for naturalization until they have fulfilled a five-year waiting period, there are exceptions not reflected in the INS naturalization data. Some spouses and adopted children acquire citizenship in less than five years.

A third underlying factor related to the discrepancy Passel and Clark suggest may be the editing process described in Section 2.5 above. All cases are assigned citizenship edit flags based on responses provided to the citizenship questions during the survey interview. The edit flags indicate the action to be taken during the editing process. Cases with a "0" edit flag value leave the edit process after citizenship is assigned. Tables A1. and A2. show all the non-zero edit flag cases from the March 1997 CPS where the respondent indicated a year of entry five years or less before the survey collection date. Table A.1 contains a complete listing of all these cases, and Table A.2 contains a listing of only the edit flag Code '10' cases from Table A2.

Cases with a Code 10 edit flag provided a "yes" to the question "Were you born a citizen?" (question two in Section 2.4 above), as well as indicating they had been born abroad and that both parents had been born abroad. Although Code 10 cases and both parents are born abroad, some of these cases may be "born abroad of foreign born naturalized citizens." We do not ask the citizenship status of parents who are not household members, so we do not know whether the parents of Code 10 cases are citizens who have been naturalized or were born abroad of American citizens themselves. The March 1997 CPS editing process assigned the 31 Code 10 cases with an "entry" less than five years before the survey collection date the citizenship status 'Naturalized citizen' (Citizenship Code 4).²¹

In sum, respondents classified as naturalized citizens with a year of entry less than five years before the survey date may meet one of the criteria described above. Some may be naturalized citizens returning from abroad. Some may be adopted children. Some may be spouses who have acquired citizenship in under five years. Some may not be naturalized citizens, but U.S. citizens born abroad of foreign-born U.S. citizens.

Figure 1. Naturalized Citizens: CPS and Naturalizations FY93 - FY96: INS provides an illustration of the relationship in recent years between estimates of the number of naturalized citizens based on the March CPS and naturalization activity at the INS. The CPS data are lagged on the INS data, because 50% of the CPS respondents joined the survey one year earlier and information about citizenship status is obtained when a respondent joins the survey. The CPS data are from the March 1994 to March 1997 surveys. The naturalization data covers the INS fiscal years (October to September) preceding each March survey. The CPS data reflect a net effect of many years of naturalization including the death of some naturalized citizens, while the INS data show only recent activity. However spurious, these two data series are related and the similar direction of their trend lines is encouraging.

Following every decennial census, the Census Bureau re-calibrates the CPS sampling frame. Beginning in April 1994 and ending in June of 1995, the Census Bureau phased in a new CPS sampling frame based on the 1990 census. Prior to April 1994, the frame had been based on the 1980 census. Nativity data collected before July 1995, are not exactly comparable with data collected after this date.²²

The CPS is sample based and therefore subject to sampling error. For example, the estimate of the size of the foreign-born population March 1997 is 25,779,000 and the standard error for this estimate is 438,000. A calculation of 1.645 x 438,000 produces a 90% confidence interval of 25,058,000 to 26,498,000 for the estimate. In other words, there is a 90% chance that the true value lies somewhere between 25.1 and 26.5 million. The estimated size of the foreign born population for March 1997 is based on 13,835 sample cases. Estimates for smaller numbers will have relatively higher standard errors and confidence intervals.

In Table A.3, for example, the 1996 CPS estimate for persons born in Jamaica is 506,000. In comparison, the 1997 figure is 397,000. Did the population born in Jamaica and living in the U.S. decline by 109,000 persons between these two dates? The 90 percent confidence interval for 1996 is 506,000 ± 75,000 x 1.645 persons. Theoretically, this means the results from 90 of 100 surveys in 1996 of a similar size, would have produced a number between 383,000 and 629,000--a relatively large range. And the other 10 theoretical surveys, would have produced a number outside this range. Comparable figures for Jamaica for 1997 are 397,000 with a range of 288,000 to 505,000. Although these ranges overlap, Table A.3 shows that the ratio of the absolute difference of the population estimates for the two estimates, to the standard error of the difference of the estimates is below the 1.645 required for significance at the 90 percent confidence level. Even when a factor for the fifty percent sample overlap is added into the equation, the ratio remains below 1.645 at 1.303. The apparent decline is not statistically significant. In fact, the CPS data cannot tell us whether the foreign born population of Jamaica changed at all.

While the size of the population is a major factor in a statistically significant estimate, change over time may be significant for one large group but not for another. For example, although the population born in Mexico is larger than the population born in China/Taiwan, the change in the size of Chinese foreign-born population between 1996 and 1997 was statistically significant at the 90 percent confidence level, while the change in the size of the foreign born population from Mexico was not. Between 1996 and 1997, the foreign-born population of Chinese extraction increased by an estimated 273,000 persons (Diff/SE = 1.969), but the estimate of change for Mexico is not statistically significant (Diff/SE = 1.187).

Statistical confidence levels can also be calculated for percentages. For example, the 1997 CPS data in Table 1. indicate 9.7 percent of the population was foreign-born compared to 9.3 percent in 1996. In 90 percent of all possible samples taken in 1996, the percent foreign-born would fall into a range from 9.0 to 9.6 while a similar range for 1997 is 9.3 to 10.0. It appears the percentage of foreign-born persons increased between the two March dates. When we calculate the ratio of the difference to the standard error of the difference for these two samples we derive a factor of 1.918 which tells us the two estimates are statistically different at the 90 percent confidence level.

In summary, use caution when interpreting CPS data. Estimates of the current size or proportion of the foreign born population, as well as estimates of year-to-year change (difference) should include calculations of a confidence interval. In general, the larger the sample (e.g., born in Mexico, or regional groupings of individual country data), the more reliable will be the estimate and the interpretations based on the estimate. Estimates of the size of the large country-specific foreign-born groups are subject to less sampling variability than those of smaller countries. Year-to-year intra-country change may not be statistically significant at the 90 percent confidence level, if the residual is not very different from zero.

The CPS sample was lowered in January 1996, from 56,000 to 50,000 eligible households. Table 2 shows the effect of this reduction on the number of foreign-born sample cases. The reduction affected New York City, Los Angeles county, and selected states. All these areas continue to have representative samples for the national labor force series, which shows data for the population by race and sex. The sample reduction may have affected the reliability of estimates for small population groups, however.

In New York City and Los Angeles, the CPS primary sampling units (PSUs) generally cover several contiguous neighborhoods and many apartment buildings. The last step of the CPS sample selection process within a PSU is the 'cluster' stage where several housing units in the same block or building are selected for interview. Year-after-year as the survey rotates through a PSU, different neighborhoods and buildings will be included in the cluster stage. Characteristically, newly arrived immigrants live in ethnic clusters in neighborhoods where 'like' ethnic enclaves exist side-by-side. In New York City for example, Jamaicans, Dominicans and other migrants from the Caribbean live in close proximity.²³ In successive years, as the CPS sample falls in different neighborhoods and buildings in the PSU, the size of the sample of different small groups fluctuates. This underlying process may explain the not statistically different swings in size of the weighted estimates of the foreign born from Jamaica and the Dominican Republic seen in Table A.4.

Concurrent with the introduction of the new sampling frame, the Census Bureau phased in the new 1993 Metropolitan Area definitions. Sub-national geography changes were introduced, that may affect tabulations for the foreign born in metropolitan areas.

After we evaluated the nativity data collected in 1994, we determined that the forty country designations available in the survey instrument were insufficient. Tables A.4 to A.8 show the original list of foreign countries we identified in 1994. We could not identify about 3.6 million of the 23 million foreign born, who are listed in Table A.4 as 'Other country'. Beginning in 1995, we increased the number of country designation choices available on the CPS computer assisted screens from about 40 to over 100.²⁴

Following the introduction of the new 'place of birth' screens, the 1994 - 1996 data for country of birth appeared inconsistent for countries beginning with the letters T through Z. (Table A.4). Although sampling variation is always a factor, these differences seemed to fall outside sampling error. For example, the decrease between 1994 and 1996 of those reporting Vietnam as a place of birth (496,000 to 47,000) seemed excessive.

Further review of the data revealed that a processing error, now corrected, had caused these inconsistencies. Specifically, after the introduction of the new nativity screens to the survey instrument in 1995, the Census Bureau did not immediately revamp the processing procedures and 'place of birth' data were coded incorrectly. Responses for 'place of birth' for countries beginning with letters at the end of the alphabet were dropped and 'blanked'. Next, these 'blank' cases were assigned new values based on the nativity editing specifications. The 'hot deck' procedure provided values from donor cards with a high probability of selection owing to their sample size, i.e. persons born in China, India, or the Philippines. The 'blanked' place of birth values were mostly changed to these countries. The result was an overstatement of some of the already large groups and an understatement of the groups from the dropped screen.

During CPS processing, country of birth values are assigned when the country was not reported. These edits introduce an inconsistency with 1990 census data, because country was not allocated for about 4 percent of the foreign-born population reported in the census. Table A.9 illustrates the numerical effect on country groupings if--to be consistent with the CPS editing--unknown country is assigned in 1990. As an example, the number of Mexican-born persons would increase by about five percent from 4.3 million to 4.5 million. The percent increase by country grouping does not vary much.

A second change instituted in the CPS, beginning in January 1994, was the modification of the CPS population "controls" used for second stage weighting.²⁵ In Section 1.5, we discussed the factors designed to ameliorate census and survey under- coverage built into the population estimates used in second stage weighting. As shown in col. 5 of Table A.9, an illustrative (not official) under-coverage adjustment for the 1990 data would increase the figures for Mexico (5.1 percent) and other Western Hemisphere countries by a greater magnitude than Europe and Asian countries.

The last column in Table A.9 illustrates the combined effect of the two changes included in the CPS foreign-born estimates but not in the published decennial results. The impact is greatest again for the Western Hemisphere countries--for example, the overall Mexican-born estimate for 1990 is increased by 0.5 million (10.9 percent) from 4.3 to 4.8 million.²⁶

In Table A.10 March 1997 CPS estimates of the foreign-born for country groupings are compared to the 1990 census estimates for the same country groupings. Columns 1 to 3 contrast the 1997 CPS totals with 1990 data to make rough inferences about consistency and growth since 1990. The 27 percent increase in the foreign born population from 20.4 million in 1990 to 25.8 million in 1997 (column 7) is consistent with growth trends, where sustained immigration has more than offset mortality and emigration. As Table A.10 shows, this change varies by region. For example, the increase of the Mexican-born population since 1990 (2.3 million or 47 percent), contrasts markedly with the 15 percent decline in the European-born population. The growth of the Mexican-born reflects high immigration levels and the low mortality of a younger group, while the reverse is true for the European born.

The "pre-1990" CPS estimates in columns 4 to 6 narrow the consistency comparison in terms of effects such as mortality and emigration. In the absence of coverage differences or reporting errors, the 1997 CPS estimates for the pre-1990 entry cohorts should be lower than the corresponding 1990 census counts--because immigration since 1990 is not included to offset the deaths and emigration that occurred between 1990 and 1997. The relative larger decline in the European-born population that entered before 1990 (-28 percent) can be attributed to the combined forces of mortality and emigration, especially mortality. The anomalous increase in the Mexican- born population represents some residual inconsistencies in the data (such as misreporting of period of entry, different coverage levels in the CPS and Census, or other sources).

To become a naturalized citizen, most immigrants must reside in the United States for five years.²⁷ Currently, the citizenship question is asked only when a new person is first included in the survey. CPS does not ask respondents about their immigration status (not all foreign born are immigrants and therefore eligible for citizenship), or if they become citizens after they enter the survey. Currently, the Census Bureau is exploring the feasibility of re-asking the citizenship questions. In spite of the implied understatement of naturalized citizenship status, Tables 3 and 4 show the number and percent of the naturalized citizen foreign-born population increased significantly between 1996 and 1997. March 1994 and 1995 data are confounded by the irregular one- time exposure of the entire March 1994 sample to the nativity questions, and the 16 month interval before the initial respondents were phased out of the survey.

The addition of the new 'place of birth' screens improved our reporting accuracy for this item, but about 20 cases per month have no value provided by the respondent because the interviewer does not understand the response, perhaps due to the respondent's inability to speak English. We are evaluating a change to the instrument that would allow the interviewer to write-in a response for 'place of birth' and then subject matter experts would code the information before processing.

The new CASES software will allow us to collect multiple data items on a single screen or grid format. For example, "Where were you born?", "Where was your mother born?" and "Where was your father born?" could be combined and then the interviewer could enter a "same" option to report the identical country, instead of typing it over and over.

Currently, there is a range check for the 'year of entry' item. This check indicates when an inconsistency occurs, for example, an age too young for the entry year given. With CASES, the interviewer will be able to see the cause of the inconsistency, rather than a message that says "try again". This will allow the interviewer to understand and convey information about the nature of the problem to the respondent.

This paper poses the question, "How Well Does the Current Population Survey Measure the Foreign-Born Population in the United States?" The answer is, it depends. All data can be used or abused, so the goal of the analyst is to determine when the data effectively measure a condition, and when the range of error is so great as to make assessment impossible.

The CPS nativity data provide a reliable basis for tracking change in the size of the total foreign-born population at the national level. Although comparability problems remain with 1994 and 1995, the data from the 1996 and 1997 March CPS files are relatively problem free. The 1996 and 1997 samples, about 13,000 cases each year, produce internally consistent and statistically significant measures at the 90 percent level of confidence. The 50% overlap in the year-over-year samples strengthens these statistics. Reliable statistics for the size of the naturalized citizen subgroup are also available from the 1996 and 1997 data. Precision in derived measures, such as change in the proportion foreign born or naturalized, is affected by simultaneous movement in the denominator as well as the magnitude of the difference and may not always be statistically significant.

Inter-country static comparisons (e.g. size of Mexico compared to the Philippines) are generally significantly different from zero and therefore more likely to produce statistically significant results. Measures of intra-country trends over time are generally not significantly different from zero, and therefore less likely to be statistically significant. For example, the CPS sample for Mexican foreign born produces a very reliable estimate of about seven million. The ratio of the difference of the 1996 and 1997 estimates over the standard error of the difference for those two years is not statistically significant. Perhaps there was no change. Or, perhaps, the rate of change was too small to be detected from the CPS sample. In contrast, the sample and weighted estimate for the foreign-born population from China is statistically significant and the change in the population between 1996 and 1997 is large enough to produce a significant result.

The nativity data series will be used to monitor change over time. The potential for tracking key social and economic indicators associated with the foreign-born population in the United States increases as the series expands. The accumulating data series will allow analysts to derive two- year and three-year moving averages for tracking educational attainment, income, poverty, program usage, and other measures of well-being. Changes in moving averages can be assessed with time series analysis techniques that mitigate some of the limitations of un-averaged year-to-year data comparisons. Another methodological approach that can overcome some of the data limitations involves the use of annual averages, based on 12 months of data and comparable to the annual averages found in Employment and Earnings.

The new Current Population Survey nativity questions provide a rich and useful body of information about a growing segment of the population of the United States of America. With the data from the CPS, policy makers will have access to information that tracks the process by which these new entrants become integrated into the U.S. economy and society. Analysts will no longer be limited to the once-every-ten-years snapshot provided by the decennial census.

1. "The Foreign-Born Population in the United States: March 1997". Current Population Reports P20-507. Issued March 1998. See also, The President's Comprehensive Report on Immigration, 1996 U.S. Immigration and Naturalization Service in 1998) for an overview of the contribution of immigration to the growth of the population of the United States in recent years (forthcoming).
2. In 1996, the U.S. government enacted the Personal Responsibility and Work Opportunity Reconciliation Act of 1996 and the Illegal Immigration Reform and Immigration Responsibility Act of 1996. Both of these new laws contain provisions requiring monitoring of the effects of the new legislation on the welfare of newly arrived foreign born, and both native and foreign born children of the foreign born.
3. The Immigration and Naturalization Service collects information about legally admitted immigrants and non-immigrants.
4. This paper updates and supercedes a paper presented to the annual meeting of the Population Association of America in 1996, entitled, "Comparison of Nativity Data From the Current Population Survey with 1990 Census and INS Data" by J. Gregory Robinson and Kristin A. Hansen.
5. See "The Current Population Survey: Design and Methodology," in Technical Paper 40; U.S. Bureau of the Census; 1978; and Cohany S., Polivka A., and Rothgeb J., "Revisions in the Current Population Survey Effective in January 1994," Employment and Earnings, February 1994.
6. Excerpt from, Bulletin No.2: Further Labor Force and Income Statistics for the Foreign Born Using the March, 1994 Current Population Survey, prepared by Scott M. Came under contract with the U.S. Department of Labor, Division of Immigration Policy and Research, October 1995.
7. In 1990 about 1.5 percent of the foreign-born population was institutionalized.
8. Employment and Earnings; January 1997; U.S. Department of Labor
9. For more information about the two-stage weighting procedure the Bureau of the Census uses to develop national population numbers from CPS data, see "Derivation of Independent Population Controls"; Appendix D. Current Population Survey Technical Documentation, 1997; and Technical Paper No.63.
10. About four million persons residing in the United States on census day were not enumerated in 1990. This under-enumeration is known as the 'estimated net census undercount'.
11. MICROCATI software was developed by Census Bureau staff. CASES is the trademark name of the new software purchased from an outside vendor.
12. Internal memorandum dated November 6, 1997, from Chester Bowie, Chief, Demographic Surveys Division to John F. Long, Chief, Population Division, and Daniel H. Weinberg, Chief, Housing and Household Economic Statistics Division.
13. The census of 1870 asked only if parents were foreign born.
14. The United States of America consists of fifty states and the District of Columbia. The Bureau of the Census refers to the fifty states and DC as "the states". Persons who move to the U.S. from outlying areas such as Puerto Rico are asked, "When did you come to the United States to Live", but they are citizens at birth and therefore not foreign born.
15. Wellens, Tracy. (1993) Report on the Cognitive Evaluation of the Nativity Questions for the Current Population Survey. Center for Survey Methods Research. U.S. Bureau of the Census, Washington DC 20233
16. New respondents are added through the introduction of a new rotation panel or when an old housing unit is replaced.
17. All states receive the same code in the CPS. In the decennial censuses each state has a unique code.
18. We do ask ALL persons born outside the states 'When did [the person] come to the United States to live?' We want to know about the movement of the foreign born and citizens from abroad.
19. Passel, J.S. and R.L.Clark. "How Many Naturalized Citizens Are There: An Assessment of Data Quality in the Decennial Census and the Current Population Survey". Paper presented at the 1997 Annual Meeting of the Population Association of America, Washington, D.C. March 27, 1997.
20. For example, many Italian-born American citizens returned to the United States during the 1930's in response to conditions in Italy. Personal communication with INS staff.
21. Weighted, these cases would equal roughly 60,000 persons.
22. For example, shifts in the foreign born population may not mirror shifts in the total population distribution, although this is probably not the case as many of the high growth areas are also areas of increased foreign-born population growth.
23. See "The Newest New Yorkers", American Demographics, July 1997 for a brief overview of the settlement patterns of immigrants in New York City.
24. For consistency we show the countries extant in the 1994 CPS although much more information is available.
25. The 1990 Modified Age/Race/Sex and Hispanic Origin (MARS)file discussed in 3.2.3, is not a factor in the current discussion, because the 1994 data shown in Tables A.9 and A.10 have not been revised to reflect four races. The official published 1990 census data reflect five races, but were never adjusted for under-coverage.
26. Published census data reflect an "other" category of race. The modified age-race-sex 1990 census base (MARS) for the population estimates series does not. See, Byerly, E. and D. Schmidley; (1994); Population Estimates for Counties and Metropolitan Areas: July 1, 1991; P25-1108; for a description of estimates methodology and the MARS file; and 1990 CPH-L-74--"Age, Sex, Race and Hispanic Origin Information from the 1990 Census: A Comparison of Census Results with Results Where Age and Race Have Been Modified."
27. Spouses have a three year wait. Natural and adopted child immigrants must meet other criteria.