5
Demographic Analysis

We begin our assessment of population coverage in the 2000 census by reviewing demographic analysis (DA), which has been used extensively by the Census Bureau for coverage evaluation of past population censuses. Demographic analysis was also advanced by the Bureau as a benchmark against which not only the census counts, but also the survey-based Accuracy and Coverage Evaluation (A.C.E.) estimates could be compared. Given the role of DA in contributing to the Census Bureau’s March 2001 recommendation not to adjust the census data for redistricting, we believe it important to assess the strengths and weaknesses of this tool for census coverage evaluation. In subsequent chapters, we examine the A.C.E. methods and results and the effects on population coverage of people requiring imputation and people reinstated in the census too late to be included in the A.C.E. processing.

METHODOLOGY: OVERVIEW

The methodology of demographic analysis uses aggregate data from administrative records, supplemented by census and survey results, to estimate the total population by age, sex, and race (two categories—black and all other). Since 1970, estimates have been constructed separately for people under age 65 and aged 65 and older (see Robinson, 2001a:App.A; National Research Council, 1985:133–139, 148–151).

Estimates of the population under age 65 are constructed for single-year birth cohorts by sex and race. The procedure uses the number of births as the starting point (e.g., estimates of people aged 60 in 2000 begin with the number of births in 1940), subtracts deaths, and adds estimates of net immigration in each year to the estimation year.1 Birth and death data are from vital statistics records collected by the states in a common format and forwarded to the federal government. (All states and the District of Columbia have been part of the vital statistics registration system since about 1933.) Corrections are made for

1  

In practice, the DA estimate for a census year (2000) usually begins with the DA estimate for the previous census year (1990), updated with estimates of births, deaths, and net immigration in the decade between the two years.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 77
The 2000 Census: Interim Assessment 5 Demographic Analysis We begin our assessment of population coverage in the 2000 census by reviewing demographic analysis (DA), which has been used extensively by the Census Bureau for coverage evaluation of past population censuses. Demographic analysis was also advanced by the Bureau as a benchmark against which not only the census counts, but also the survey-based Accuracy and Coverage Evaluation (A.C.E.) estimates could be compared. Given the role of DA in contributing to the Census Bureau’s March 2001 recommendation not to adjust the census data for redistricting, we believe it important to assess the strengths and weaknesses of this tool for census coverage evaluation. In subsequent chapters, we examine the A.C.E. methods and results and the effects on population coverage of people requiring imputation and people reinstated in the census too late to be included in the A.C.E. processing. METHODOLOGY: OVERVIEW The methodology of demographic analysis uses aggregate data from administrative records, supplemented by census and survey results, to estimate the total population by age, sex, and race (two categories—black and all other). Since 1970, estimates have been constructed separately for people under age 65 and aged 65 and older (see Robinson, 2001a:App.A; National Research Council, 1985:133–139, 148–151). Estimates of the population under age 65 are constructed for single-year birth cohorts by sex and race. The procedure uses the number of births as the starting point (e.g., estimates of people aged 60 in 2000 begin with the number of births in 1940), subtracts deaths, and adds estimates of net immigration in each year to the estimation year.1 Birth and death data are from vital statistics records collected by the states in a common format and forwarded to the federal government. (All states and the District of Columbia have been part of the vital statistics registration system since about 1933.) Corrections are made for 1   In practice, the DA estimate for a census year (2000) usually begins with the DA estimate for the previous census year (1990), updated with estimates of births, deaths, and net immigration in the decade between the two years.

OCR for page 77
The 2000 Census: Interim Assessment TABLE 5-1 Total Population for 2000   Difference from Census Count Source Population (in millions) No. (in millions) Percent Census count 281.4   DA base estimate 279.6 –1.8 –0.65 DA alternate estimate 282.3 0.9 0.32 A.C.E. estimate 284.7 3.3 1.15 NOTE: See text for discussion. SOURCE: Robinson (2001a:Table 3). underregistration of births, using the results of birth registration studies adjusted by interpolation and extrapolation (the latest study covered 1964–1968). Administrative data from the Immigration and Naturalization Service are used to estimate legal immigration, but they must be adjusted to fill in gaps for undocumented (illegal) immigration and emigration. For the population aged 65 and over, estimates are constructed from Medicare enrollment statistics. The Medicare data are adjusted for those not enrolled, who were estimated to be 3.7 percent of people aged 65 and over in 2000. THE 2000 ESTIMATES The Census Bureau’s concerns about differences in the census, DA, and A.C.E. estimates of the total population are illustrated by the several different figures available for 2000 shown in Table 5-1. The census count is that reported for April 2000, including household and group quarters residents in the United States. The DA base estimate is that developed by demographic analysis as described above, which, at the time of the census, was the Census Bureau’s initial and best estimate of the “true” population to be used as a measure for evaluating overall census coverage. The DA alternate estimate is that developed by the Census Bureau in early 2001, which incorporated a higher allowance for net undocumented immigration than that included in the DA base estimate. The A.C.E. estimate is that developed by dual-systems estimation for the household population from the results of matching census enumerations and an independent postenumeration survey in a sample of blocks. For comparability with DA and the census count, the A.C.E. estimate is augmented by approximately 7.5 million people to cover residents of group quarters and people enumerated in other special operations not included in A.C.E. (e.g., the remote Alaska enumeration).2 2   There is also a postcensal estimate for 2000, which is the 1990 census count (not adjusted for net undercount) carried forward with data on births, deaths, and net immigration. We do not discuss the Census Bureau’s postcensal estimates program in this report.

OCR for page 77
The 2000 Census: Interim Assessment Thus, we can summarize the data in Table 5-1. The DA base estimate for 2000 of 279.6 million is 1.8 million lower than the census count of 281.4 million, implying a net overcount in the census of 0.7 percent. However, the A.C.E. estimate is 284.7 million, which implies a net undercount in the census of 3.3 million (1.2%). There is a need then to reconcile a difference of 5.1 million between the two estimates—base DA and A.C.E. —to come to closure in choosing the “best” estimate for evaluating the census count of the U.S. population in 2000.3 The alternative DA estimate developed by the Census Bureau reduces the difference from the A.C.E. estimate somewhat, but not entirely (the difference decreases from 5.1 million to 2.4 million). By age and sex, coverage patterns estimated by DA are broadly similar between 1990 and 2000 in that women were better counted then men, and older people were better counted than younger people. The 1990 PES and 2000 A.C.E. also show these patterns. However, the 2000 DA net undercount estimates are much lower than the 1990 DA estimates, even when the 2000 DA population estimates are adjusted to allow for a greater number of undocumented immigrants than originally estimated; see Table 5-2. UNCERTAINTY IN IMMIGRATION ESTIMATES The main area of uncertainty in the DA estimate of the total population lies with the immigration component, especially the number of undocumented immigrants. Other components, such as emigrants and some categories of legal immigrants, also add to the margin of error. The DA base estimate assumes that there are about 6.0 million undocumented immigrants living in the United States under age 65–3.3 million from the 1990 DA estimate plus a net increase during the 1990s of 2.7 million.4 The estimated net increase during the decade essentially represents an extrapolation of net undocumented immigration derived from estimates that mainly reflect changes between 1992 and 1996. Considering a number of factors, Census Bureau researchers believe that 6 million is a reasonable lower-bound estimate of the number of undocumented immigrants at the time of the 2000 census (Robinson, 2001a). For purposes of comparative analysis, the Census Bureau simply assumed a doubling of net undocumented immigration over the decade, to 8.7 million (bringing the total population to 282.3 million. This alternate estimate (see Table 5-1) implies a census undercount of 0.9 million, or 0.3 percent, still far below the 3.3 million (1.2%) indicated by the A.C.E. 3   In 1990, in contrast, the DA estimated a higher net undercount than that estimated by the Post-Enumeration Survey (PES). The DA estimate of the net undercount was 1.9 percent; that of the PES was 1.6 percent. 4   For people age 65 and over the adjustment for Medicare underregistration presumably includes undocumented immigrants. See Robinson (1991) for a discussion of the Medicare adjustments for the 1990 DA estimates.

OCR for page 77
The 2000 Census: Interim Assessment TABLE 5-2 Net Census Undercount, by Sex and Age, as Measured zby Demographic Analysis and Post-Enumeration Surveys, 1990 and 2000 (in percent)   1990 2000   Demographic Analysisa   Category Demographic Analysis PES Base Alternate A.C.E. Male   Total 2.79 1.93 –0.13 0.91 1.51 0–17 years 2.16 3.17 –0.51 0.27 1.53 18–29 years 2.15 3.16 –2.57 0.34 3.45 30–49 years 3.83 1.85 1.28 2.26 1.81 50 years and over 2.72 –0.57 0.15 0.29 –0.24 Female   Total 0.94 1.25 –1.16 –0.25 0.79 0–17 years 2.43 3.20 0.06 0.87 1.54 18–29 years 0.64 2.81 –3.07 –0.66 2.11 30–49 years 0.50 0.88 –0.91 0.04 0.95 50 years and over 0.24 –1.20 –1.43 –1.28 –0.76 Total 1.85 1.58 –0.65 0.32 1.15 NOTES: Minus sign (–) indicates a net overcount of the population. Net undercount is the difference between the estimate (A.C.E., PES, or demographic analysis) and the census divided by the estimate. Total population includes household and group quarters populations; the census count of group quarters is added to the A.C.E. for comparability with DA. aBase is the originally produced DA estimate which includes an allowance for 6 million undocumented immigrants; alternate is a DA estimate that arbitrarily doubles the flow of undocumented immigrants between 1990 and 2000, allowing for 8.7 million undocumented immigrants total. SOURCE: Robinson (2001a:Tables A, 5). There are no direct, comparative measures for evaluating the net immigration component, especially the undocumented component, of DA. At present, a “residual” process is used to estimate the number of undocumented immigrants: that is, an estimate of the expected number of foreign-born people legally residing in the country is derived from reported data on legal immigration, and this figure is compared with the number of foreign-born people reported in the census long-form sample or, more recently, in the Current Population Survey (CPS).5 The difference between the two represents the number of undocumented immigrants included in the census (or CPS). The computations are carried out in some detail by country (or region) of birth and year of entry, which is believed to add to the validity to the estimates. (Data on country of birth and year of immigration are now included regularly in the CPS so that the computations can be carried out more frequently, perhaps adding some stability in the estimates over time.) 5   The census long-form sample included about 17–18 million households in 1990 and 2000; the CPS includes about 50,000 households each month.

OCR for page 77
The 2000 Census: Interim Assessment There has been much speculation about the adequacy of the immigration component used in the DA estimates. Passel (2001) argues that data now available support a significantly higher estimate of undocumented immigrants in the population in 2000 than that used by the Bureau. Specifically, looking at the March 2000 CPS foreign-born estimates, Passel concludes that one can easily support an estimate of undocumented immigrants of 7 million, which is 1 million higher than the DA base estimate. Furthermore, using 2000 census and A.C.E. data to adjust the March 2000 CPS would support even higher estimates of undocumented immigrants—perhaps in the 8–9 million range. Some recent work by Warren (2001), released by the Immigration and Naturalization Service, supports an estimate of 6.5 to 7.5 million undocumented immigrants in the United States at the time of the census—higher than the DA base estimate but lower than Passel’s high estimates. Recently released data from the Census 2000 Supplementary Survey supports Passel’s higher estimates of 8–9 million undocumented immigrants in the United States in 2000.6 Passel’s research also suggests that the Census Bureau understated two components of legal immigration: foreign-born people living in the country legally with temporary visas (e.g., foreign-born students and guest high-tech workers) and Mexicans living in the United States as legal residents. According to Passel, both of these groups increased in number during the 1990s more than indicated by data on legal admissions. Reasonable estimates for these two groups would add about another 750,000 people to the estimates. Furthermore, there are indications that the Bureau’s allowances for net emigration may have overstated the number of immigrants who left the country during the 1990s.7 In response, the Census Bureau has suggested that these higher estimates of legal and illegal immigrants imply that the proportions of foreign-born people in the U.S. population—particularly Hispanics—are much higher than are reasonable based on past history. Bureau researchers are currently looking at detailed information from the 2000 census long form and the Census 2000 Supplementary Survey on country of birth, year of immigration, and other characteristics to try to resolve competing estimates of legal and illegal immigrants. This discussion is not intended to answer the question of the quality of the DA estimate, but rather to point out the problem of fine-tuning and interpreting the DA estimate in light of the uncertainty associated with estimates of the immigration (legal and illegal) components. As already noted, there are no precise tools for evaluating the accuracy of DA. Rather, one uses analyses of such 6   The Census 2000 Supplementary Survey included 700,000 households, of which about 58,000 households were surveyed each month in 2000 by mail with telephone and personal follow-up using a questionnaire similar to the long form. It is intended to provide the basis for a transition from the census long form to the planned American Community Survey, which will sample 250,000 households each month beginning in 2003. 7   Administrative data on emigration represent only a very small fraction of emigrants and are, themselves, of uncertain quality. The method for estimating emigration is essentially an extrapolation of previous trends.

OCR for page 77
The 2000 Census: Interim Assessment TABLE 5-3 Net Census Undercount by Race, as Measured by Demographic Analysis and A.C.E., 2000 (in percent)   Demographic Analysis     Base Estimatea Alternate Estimateb   Category Model 1c Model 2d Model 1c Model 2d A.C.Ee Total Population –0.65 –0.65 0.32 0.32 1.15 Black 4.67 0.93 5.36 1.65 2.06 Male 6.94 3.26 7.64 3.98 2.37 Female 2.52 –1.27 3.20 –0.56 1.78 Nonblack –1.48 –0.90 –0.45 0.12 1.02 Male –1.21 –0.65 –0.11 0.44 1.40 Female –1.74 –1.14 –0.79 –0.20 0.64 Black-nonblack difference 6.15 1.83 5.81 1.53 1.04 NOTES: Minus sign (–) indicates a net overcount of the population. Net undercount is the difference between the estimate (A.C.E. or demographic analysis) and the census divided by the estimate. Total population includes household and group quarters populations; the census count of group quarters is added to the A.C.E. for comparability with DA. aBase is the originally produced demographic analysis (DA) estimate that includes an allowance for 6 million undocumented immigrants. bAlternate is a DA estimate that arbitrarily doubles the flow of undocumented immigrants between 1990 and 2000, allowing for 8.7 million undocumented immigrants total. cModel 1 compares the 2000 DA estimates for blacks with 2000 census tabulations for people who only reported black race. dModel 2 compares the 2000 DA estimates for blacks with census 2000 tabulations for people who reported black whether or not they reported any other race. eThe A.C.E. estimates are the average of Model 1 and Model 2, which differ by no more than 0.3 percent. SOURCE: Robinson (2001a:Table 6). data as country of origin of the foreign born, year of entry, and citizenship—to name the most important variables—to arrive at some estimate. The question then is whether the overall DA estimates can be used as a gold standard for measuring census coverage and the adequacy of surveys such as the A.C.E., given the continuing high level of uncertainty of a major component. It should be noted that a similar problem arose in 1980 when the DA estimate was initially about the same as the 1980 census count. Further analysis suggested that the DA estimate failed to allow for significant numbers of undocumented immigrants in the country (then estimated at about 3 million). The DA estimates were revised to incorporate the estimate of 3 million undocumented immigrants, which then became part of the base DA for later periods. In summary, the DA estimate of total population is subject to a high degree of uncertainty and its strength for comparative analysis is not clear. What seem to be needed are data and measuring instruments to evaluate DA, preferably before census or other survey results become available. At the moment, it seems that the DA estimate, which is supposed to help analyze and measure

OCR for page 77
The 2000 Census: Interim Assessment overall census and survey coverage, is increasingly dependent on the census and survey results to establish its own validity and provide the basis for adjustments to the estimates. ESTIMATES BY RACE The total population estimate is only one aspect, although a major one, of DA analyses. There are other relevant, informative estimates, such as the estimates by race. As stated above, DA estimates are prepared for blacks and nonblacks separately.8 Race in DA largely reflects the race assigned in the particular administrative record at the time of the event (birth, death, etc.). Thus, the comparison of the DA estimate with the census—the net undercount—will be affected if people who are classified as a particular race in DA—such as black—report a different race in the census. In 2000 the measure of the net black undercount is affected and made more complicated by the multiple race classification instruction (“mark one or more”) used in 2000. As a result, the 2000 tabulations do not include a “black” race category that is comparable with either previous censuses or the DA estimate. Rather, tabulations for the black population for 2000 can be made in one of two ways: as the number of people reporting only black as their race or as that number plus the number of people reporting black in combination with one or more other races. To deal with this problem in measuring black net undercount as reflected in DA, the Census Bureau developed two estimates, representing the range of the two tabulations indicated above. Using the DA base estimate, the black net undercount ranges from 4.7 percent based on those reporting black only in the census to 0.9 percent when black in combination with other races is included in the count of blacks—a fairly wide range with the truth somewhere in between. Note that using the Bureau’s alternate DA estimate (i.e., an estimate with 8.7 million instead of 6.0 million undocumented immigrants) serves only to slightly raise the overall level of the estimated black net undercount; the wide range remains unchanged; see Table 5-3. Another inconsistency in the race comparison arises from the millions of people—mainly Hispanic—who report their race as “other race-not specified” in the census. To make the census counts consistent with the historical race categories used in DA estimation, these people need to be redistributed to the standard race categories. In 1990 there were about 10 million people in the “other race-not specified” category who were reassigned to a specific race category. A similar modification was made in 2000; approximately 15 million people were so reclassified. 8   Administrative records do not yet provide a basis for developing demographic estimates for Hispanics or other race groups.

OCR for page 77
The 2000 Census: Interim Assessment TABLE 5-4 Sex Ratios (Men per 100 Women) from the Census, Demographic Analysis, A.C.E., and PES, 1990 and 2000   1990 2000   Demographic Analysis PES Census Demographic Analysisa A.C.E.b Censusb Black   Total 95.16 90.44 89.59 94.90 91.05 90.59 0–17 years 102.42 102.37 102.42 102.73 103.30 103.31 18–29 years 99.27 92.13 93.99 100.22 94.10 93.99 30–49 years 95.92 89.00 86.17 96.47 89.66 88.53 50 or more years 78.33 72.08 71.49 76.94 73.51 73.47 Nonblack   Total 97.19 96.54 95.89 97.66 97.88 97.15 0–17 years 105.23 105.51 105.51 104.95 105.50 105.53 18–29 years 104.94 104.57 103.78 104.81 106.89 105.27 30–49 years 102.00 100.34 99.59 101.94 101.36 100.59 50 or more years 80.79 79.86 79.38 84.17 83.54 83.10 aThe 2000 demographic analysis sex ratios are those for the base estimate, which is the originally produced demographic analysis estimate for 2000 that includes an allowance for 6 million undocumented immigrants (the base estimate sex ratios differ by less than 0.4 percent in every instance and less than 0.1 percent in most instances from sex ratios computed for the alternate estimate, which arbitrarily doubles the flow of undocumented immigrants between 1990 and 2000, to 8.7 million undocumented immigrants total). bSex ratios for the A.C.E. and the 2000 census are based on Model 1, which categorizes blacks as people who only reported black race. (Sex ratios computed under “Model 2,” which categorizes blacks as people who reported black whether or not they reported any other race, differ by less than 1.0 in every instance and less than 0.1 in most instances from the Model 1 sex ratios.) SOURCE: Robinson (2001a:Table 8). Traditionally, the DA breakdown of black and nonblack populations has been intended to provide a contrast between net undercount rates in prevailing majority and minority populations. A question about the usefulness of DA estimates is whether the black-nonblack dichotomy is still relevant to capture this contrast. The black population represents a smaller proportion of the minority population than in the past (about 40% in 2000 compared with over 50% in 1970). Correspondingly, the nonblack population is currently estimated at about 20 percent minority, and this percentage will probably increase with the passage of time. The black population is a growing fraction of the national population, and DA estimates of net undercount in the black population remain a vitally important product. However, research on a more refined indicator of summary demographic undercount than black-nonblack is needed. OTHER ESTIMATES Another component of DA analysis worth noting is the DA estimates of sex ratios, that is, the number of males per 100 females. For blacks, DA shows

OCR for page 77
The 2000 Census: Interim Assessment TABLE 5-5 Comparisons of Population Estimates for Children   Census Count/DA Estimate Age Group 2000 1990 Under 5 years 0.972 0.963 5–9 years 0.997 0.965 10–14 years 1.025 0.988 Total under 10 years 0.985 0.964 Total under 15 years 0.998 0.972 SOURCE: Data from U.S. Census Bureau. consistently significantly higher sex ratios at ages 18–29 and 30–49 than shown in the 2000 census or the A.C.E., for which sex ratios for blacks at these ages are similar and below what one would expect. In other words, both the 2000 census and the A.C.E. show net undercounts of black males relative to females at these ages, continuing a pattern exhibited in earlier censuses; see Table 5-4. This pattern is believed to be due to the lower propensities for black men to be counted in any census or survey than black women (“correlation bias”). Sex ratios for nonblacks ages 18–29, on the other hand, are somewhat higher in the A.C.E. than in DA, which was not the case in 1990. Yet another component that can shed some light on the comparison of DA with the census are the estimates for people under age 15. The DA estimates at these ages are less affected by the uncertainties associated with the immigration component, especially undocumented immigrants. Table 5-5 shows the ratios of census to DA estimates for 1990 and 2000 (base estimate) for age groups under age 15. If one assumes that DA is fairly reliable at these younger ages, either the 2000 census had higher coverage at these ages than in 1990 or there was more overcounting (duplication) than in the 1990 census. CONCLUSIONS Demographic analysis has strengths as well as weaknesses in its processes and underlying assumptions for developing estimates of the expected total population for groups. For 2000, it would appear that the demographic estimates are weakest, or at least most uncertain, for two elements of major importance: total population for use in measuring overall net undercount and differential undercount for blacks and nonblacks. The DA estimates of sex ratios by race and of younger ages are informative, but limited in usefulness for measuring the comparative coverage of the three data sets—census, A.C.E., and DA.

OCR for page 77
The 2000 Census: Interim Assessment We conclude that there are enough uncertainties about the estimates of net immigration, compounded by the difficulties of classifying people by race, so that demographic analysis should not be used as a standard for evaluating the census or the A.C.E. at this time. At present, the Census Bureau has under way extensive reanalysis and evaluation of the components of DA to help inform its upcoming decision about whether to adjust census population estimates that are used for fund allocation and other important purposes. Looking to the future, we urge the Bureau to expand its resources that are regularly devoted to estimating components of population change, particularly immigration (including the illegal component) and emigration. Resources are also needed for estimating uncertainty in the demographic population estimates. Finally, we urge the Census Bureau to lead a research effort by appropriate federal agencies and outside experts to develop improved methods and sources of data for estimating legal and illegal immigrants in surveys and administrative records as input to demographic analysis and for other uses.