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 30
Modernizing the U.S. Census 2 Population Coverage and Its Implications Although the census count of the U.S. population has never been complete, public concerns about the incompleteness have increased in recent decades. The census is the sole basis for apportionment of congressional seats and is relied on heavily for the distribution of federal funds. Improved statistical and demographic techniques permit the Census Bureau to estimate the incompleteness of the census with greater accuracy than in the past. Thus, concern about census incompleteness springs, ironically, from the improved professional work of Census Bureau staff and from extraordinary expectations for a "complete" census count. Some undercount of the population occurs in all censuses. This chapter reviews what is known about census coverage for recent U.S. censuses. The chapter also discusses two major implications of census incompleteness: political representation at the congressional level and the distribution of federal funds that are allocated on the basis of population. COVERAGE ESTIMATES Coverage estimates, which measure the extent to which the census counts all the people, are made by two methods. One method is to conduct a large sample survey in conjunction with the decennial census (called the Post-Enumeration Survey, PES, in 1990), match all individuals in the survey to those reported in the census, and then estimate the number of unenumerated people in the census by age, sex, and race. The second method, demographic analysis, is to develop an estimate of the population independent of the census, using birth
OCR for page 31
Modernizing the U.S. Census and death records for previous years, immigration and emigration data, and previous censuses. Demographic estimates are the primary means for comparing coverage for censuses over time for the nation as a whole. Demographic analysis has two main methodological weaknesses. One is deficiencies in the data—particularly the immigration and emigration data. The magnitude of illegal alien flows into and out of the United States in recent decades is unknown, as is the magnitude of emigration of citizens and legal resident aliens. The second shortcoming is that demographic analysis cannot provide estimates by states or other subnational areas. Demographic estimates of coverage (see Table 2.1) show that the net national undercount (the number of people omitted minus the number over-counted) was estimated at 7.0 million in 1940, 6.3 million in 1950, 5.6 million in 1960, 5.5 million in 1970, 2.8 million in 1980, and 4.7 million in 1990. The undercount rate dropped from 5.4 percent in 1940, to 4.1 percent in 1950, 3.1 percent in 1960, 2.7 percent in 1970, and 1.2 percent in 1980, then rose to 1.8 percent in 1990 (Robinson et al., 1993:13). Coverage Errors Based on the criterion of net undercount, the 1990 census was somewhat worse than the 1980 census. Comparisons of net undercount, however, fail to reveal some other kinds of deficiencies in census counting. Net undercount figures reflect three elements: People who were not counted in the census or who were omitted from the census in their proper place or residence. These people are called omissions. People who were enumerated more than once, were ineligible to be counted in the census (e.g., babies born after Census Day), or were counted at their incorrect place of residence. These people are called erroneous enumerations. People whose existence was ascertained (by the judgment of the enumerator or other evidence) but whose characteristics were missing and so had to be "borrowed" from another enumerated person. These people are called Substitutions. The combination of erroneous enumerations and substitutions (b and c) can be added together to produce the total number of counting errors, which in 1990 were estimated to include 16 million people. A substantial proportion of counting errors are people who were incorrectly placed in the wrong geographic areas, resulting in an omission in the correct location and an erroneous enumeration in the incorrect location. From the national perspective, people assigned to an incorrect location were counted in the population (and thus not part of the national
OCR for page 32
Modernizing the U.S. Census TABLE 2.1 Net Population Undercount in the Census by Demographic Analysis, 1940-1990 Undercount Rates and Numbers 1940 1950 1960 1970 1980 1990 Total Population (millions) 131.7 150.7 179.3 203.3 226.6 248.7 Undercount rate (%) 5.4 4.1 3.1 2.7 1.2 1.8 Number undercounted (millions) 7.0 6.3 5.6 5.5 2.8 4.7 Nonblacks Population 118.8 135.7 160.5 180.7 199.9 218.2 Undercount rate 5.0 3.8 2.7 2.2 0.8 1.3 Number undercounted 5.9 5.2 4.3 4.0 1.6 2.9 Blacks Population 12.9 15.0 18.9 22.6 26.7 30.5 Undercount rate 8.4 7.5 6.6 6.5 4.5 5.7 Number undercounted 1.1 1.1 1.3 1.5 1.2 1.8 Difference: black-nonblack net undercount rate 3.4 3.6 3.9 4.3 3.7 4.4 Note: Alaska and Hawaii became states in 1959. For 1950 and earlier, the population data and undercount estimates are for the 48 coterminous states. For 1960 and after, the data include Alaska and Hawaii. Source: Population data: Bureau of the Census (1993b, Table 1). Undercount rates: Robinson et al. (1993).
OCR for page 33
Modernizing the U.S. Census undercount), but they do produce counting errors in the census for subnational geographic areas.1 For people missed from the census or counted at the wrong location (called omissions), Ericksen and DeFonso (1993) estimated the number in 1990 at 20 million people; they estimate net undercount of about 4 million, which is quite close to the estimate of 4.7 million, cited earlier, on the basis of demographic analysis. Both omissions and counting errors were higher in 1990 than in 1980; omissions increased proportionately more than counting errors, so the result was a larger net undercount. The sum of both types of census error, omissions and counting errors, is called the gross coverage error. The gross coverage error was 23-25 million people for the 1980 census and 36 million for the 1990 census. There was a substantial increase in the gross coverage error between the 1980 and 1990 censuses. A point to bear in mind when evaluating estimates of omissions and counting errors (and hence the level of gross coverage error) is that, for the nation as a whole and for large geographic areas, many of these errors cancel out. Thus, a significant number of people who are included in the number of omissions because they were missed at the correct location are also included in the number of counting errors because they were counted but at the wrong location. Perhaps 40-50 percent of counting errors represent people who were also counted as omissions, for which the two classifications balance out (see Hogan, 1993: Table 7). However, for small areas, these kinds of geographic location errors may make a difference. Undercount by Subgroups According to the 1990 estimates by demographic analysis, almost three-fourths of the net national undercount were nonblacks (primarily whites). The rate of undercount, however, was over four times higher for blacks than for nonblacks, 5.7 and 1.3 percent, respectively (Robinson et al., 1993:13).2 Figure 2.1 presents undercount rates for blacks and nonblacks from 1940 to 1990, as well as the difference between the nonblack and black rates. In the 1990 census, the undercount rate for both men and women was also about 4 to 5 times higher for blacks than for nonblacks; it varied from 8.5 percent for black men to 0.6 percent for white women. Demographic analysis of the difference between black and nonblack net undercount rates shows modest changes from 1970 to 1990. In 1940, the black net undercount rate was 3.4 percentage points higher than the nonblack rate. By 1970, the difference had increased to 4.3 percentage points. The differential undercount by race was reduced somewhat in the 1980 census, to 3.7 percentage points. By 1990, even with continuing efforts to reduce the differential undercount, the difference had increased to 4.4 percentage points, slightly higher than at the beginning of the massive effort in 1970.
OCR for page 34
Modernizing the U.S. Census FIGURE 2.1 Net undercount rates by race and difference in the black and nonblack rates, 1940-1990. Net undercount rates estimated from the PES were also higher for Asian and Pacific Islanders, Hispanics, and American Indians and Alaskan Natives in the 1990 census. The undercount of the Asian and Hispanic groups is likely to have been influenced by the relatively large numbers who are foreign-born, people who may not have understood census questionnaires and procedures. By age, undercount numbers and rates from demographic analysis varied widely, from a negative undercount rate (indicating census overcount) for white women ages 15 to 24, to a positive undercount of 14 percent for black men ages 30 to 34. Black girls and boys under age 5 were missed at an 8.6 and 8.2 percent rate, respectively; black girls and boys ages 5 to 9 were missed at a 7.5 percent and 7.7 percent rate, respectively. The undercount rate peaked for all men at ages 25 to 29, but at about 4.5 percent for nonblack men and 12.7 percent for black men. For black women, the undercount rate ranged from minus 7.7 (indicating census overcount) to 4.9 percent for each age group above age 20. The estimates of net undercount from demographic analysis differ in reliability because of differences in the basic data used to prepare them. The result is that estimates of census error are more reliable for whites than for blacks and other race and ethnic groups. Estimates of the omitted population are selective by race, sex, and age. Men and women ages 25 to 60, together with children under age 10, made up more than 90 percent of those not counted. Even more narrowly, almost two-thirds of the estimated omitted population consists of two
OCR for page 35
Modernizing the U.S. Census groups: children under age 10 and men ages 25 to 39. These findings, along with evidence of higher underenumeration rates for minority groups, are useful for assessing the effect of using statistical methods to improve the enumerated count. Some information is known about the geographic distribution of the net undercount from the 1990 PES. Figure 2.2 displays estimates of the net undercount rates for the 1990 census by state. States are classified into five categories: very low and low net undercount, average undercount, and high and very high net undercount. States with a population of the most undercounted groups tended to have higher net undercount rates. States in the South and Southwest had net undercount rates higher than the national average. States in New England and the Midwest tended to have lower than average net undercount rates. New York, Maryland, and the District of Columbia are distinctive with higher net undercount rates than their neighboring states. The undercount is higher in cities than in other areas, and the people missed in the census are disproportionately concentrated in larger cities. The most serious problems of conducting the census seem to occur in inner cities, despite intensive efforts to count the population in those areas. There are several implications of undercount for minority groups. In political representation and funding based on population, undercounted groups get less credit for their population than they are due. Political districts drawn relative to population are "overpopulated" for undercounted areas (i.e., overpopulated means a larger actual population, including people who were unenumerated, than counted in the census). Overpopulated districts result in underrepresentation of minority areas (i.e., fewer districts) at all levels of government—federal, state, and local—that base political representation on population size. Errors in Small-Area Data The accuracy of the census population counts can have different meanings in different contexts. For example, for a government program with cutoff points for state funding, the degree of inaccuracy is critical only when a state population is close to the point at which funding decisions would be affected. For congressional redistricting or for local-area decisions that involve relatively small areas, the relative accuracy of the population count for blocks and aggregations of blocks is important. In response to the panel's request for information about the accuracy of small-area data in the 1990 census, the Census Bureau provided special analysis of census blocks from the PES. That block-level analysis revealed several types of errors in data from a stratified sample of 5,290 PES clusters in the 1990 census (Diffendal, 1994).3 More than 25 percent of the clusters had no erroneous inclusions—housing units assigned to the sampled cluster that are actually outside the search area. In the census, for clusters with erroneous enumeration, duplication occurs fairly often (in about 25 percent of the total clusters).
OCR for page 36
Modernizing the U.S. Census FIGURE 2.2 Net undercount rates for states, 1990.
OCR for page 37
Modernizing the U.S. Census The distribution of erroneous inclusions is skewed; most of them occur in a small proportion of clusters. About 1 percent of the clusters (about 50 out of 5,290 clusters) in the 1990 census account for more than one-third of erroneous inclusions. This means that relatively few areas contribute a substantial proportion of erroneous inclusion errors to the census. The most common types of census nonmatches—the situation in which the PES results are not matched to census results—are: (1) nonmatch with other persons in the household matched (occurs in more than one-half of clusters), (2) nonmatch with the entire household missed for a matched address (occurs in more than one-third of clusters), and (3) nonmatch with the entire household missed for a missed housing unit (occurs in about one-fifth of clusters). Typically, for the three common types of nonmatches, there are only 3 to 6 cases per cluster, of about 30 housing units per cluster. Census nonmatches are also concentrated in a relatively few number of clusters. The initial analysis by Diffendal does not reveal if clusters with a high proportion of erroneous enumerations are in vacation areas, wealthy suburbs, or other areas where the overcount may be biased in favor of the more mobile segment of the population and among those who have more than one residence. The analysis presented in Diffendal (1994) is a good beginning to understanding the quality of small-area census data. It would be useful to see further research on the overlap of erroneous enumerations and census nonmatches (showing possible joint occurrence in the same clusters) and on the sum of erroneous enumerations, census nonmatches, and substitutions (i.e., the level of gross error). In addition to the errors in small-area data discussed above, there are other sources of error, including the errors in the information provided by respondents and in data processing. Regarding nonsampling errors, Appendix L provides information from the 1990 census on errors stemming from nonresponse to items in the census questionnaire. IMPLICATIONS OF UNDERENUMERATION Underenumeration in the census has serious political, economic, and social implications. The decennial population count, reported in the census, affects the state apportionment of seats in the U.S. House of Representatives and the geographic boundaries for congressional districts, state legislative districts, and city council districts. Under the "equal proportions" methods for federal apportionment, a shift of relatively few people could result in the change of a state's representation. For all state and local districts, the possibility that undercoverage will have an effect on a district's boundaries depends on the size of the district; the coverage rates by age, sex, and race; the distribution of the population by age, sex, and race; and the undercoverage rates of contiguous districts.
OCR for page 38
Modernizing the U.S. Census Effects on Congressional Apportionment If undercoverage were eliminated, population would increase in areas with a large number of such high-undercount groups as minorities and inner-city residents. An estimate of what the 1990 census results would have been, if surveys were used to complete the count, can be obtained by using PES information to make a correction for each state's population. If used in 1990, a correction, using a set of adjustment factors taking age, sex, race, region, and urban-rural characteristics into account, would have been made for each of 7 million census blocks in the nation. The number of congressional seats is fixed at 435: an addition of a congressional seat for one state means that another state loses a seat. Congressional apportionment of seats in the House of Representatives is based on the method of equal proportions. In this method, each of the 50 states receives one congressional seat, no matter what its population. Next, a priority value is calculated for the nth seat for each state (for example, n equals 2 for a state's second seat after it has received the first seat), using the state's apportionment population multiplied by the factor: The factor decreases as the number of seats increases: the factor equals 0.71 for the second seat and 0.41 for the third seat. The state with the largest priority value receives the next congressional seat. When each state has exactly one congressional seat, the state with the largest population receives the 51st seat. But, after it does, the population apportionment factor is recalculated and compared with the priority value for all other states to decide which state receives the 52nd seat. The repeated calculation of priority values and assignment of congressional seats continue until all 435 seats are distributed. One state will receive the 435th congressional seat, and the state with the highest priority value for the 436th seat will not receive one. Adding population to a state will increase the priority values for the state, but it will not necessarily increase the likelihood of gaining an additional seat. The critical question for a state is whether a change in its priority value affects the assignment of the last several congressional seats. Corrections to the census population count would typically change the congressional delegations for those few states with priority values close to the 435th cutoff. As illustrated in Table 2.2, Washington received the 435th seat in the reapportionment based on the 1990 census. If Washington's population had been 10,200 persons fewer, or 0.21 percent of the state's total, it would not have qualified for the 435th seat. At the same time, Massachusetts failed to qualify to gain another seat by a population of 12,607, or 0.21 percent of its total population.
OCR for page 39
Modernizing the U.S. Census TABLE 2.2 Apportionment Population for States, 1990 House Seat State Number of Seats in House Delegation Apportionment Population Population Change Required Amount Percent Possible Losers 428 California 52 29,839,250 –236,002 –0.79 429 Texas 30 17,059,805 –104,241 –0.61 430 Mississippi 5 2,586,443 –15,648 –0.61 431 Wisconsin 9 4,906,745 –29,004 –0.59 432 Florida 23 13,003,362 –72,490 –0.56 433 Tennessee 9 4,896,641 –18,900 –0.39 434 Oklahoma 6 3,157,604 –9,036 –0.29 435 Washington 9 4,887,941 –10,200 –0.21 Possible Winners 436 Massachusetts 11 6,029,051 +12,607 +0.21 437 New Jersey 14 7,748,634 +22,698 +0.29 438 New York 32 18,044,505 +98,765 +0.55 439 Kentucky 7 3,698,969 +34,258 +0.93 440 California 53 29,839,250 +401,972 +1.35 441 Montana 2 803,655 +11,002 +1.37 442 Arizona 7 3,677,985 +55,242 +1.50 443 Georgia 12 6,508,419 +109,885 +1.69 Source: Passel (1991).
OCR for page 40
Modernizing the U.S. Census What would have happened to the reapportionment process if the state populations had been corrected using the PES, as was debated in 1990? It is difficult to determine exactly how the application of the PES results to the 1990 census would have affected actual congressional reapportionment, because the adjustment would have been done for small geographic areas; the calculation here uses average PES values to estimate the state's population data (Passel, 1991, presents the calculation cited here). Applying the average PES values to the 1990 state population counts, three states would have gained a congressional seat: Georgia (a 12th seat), Montana (a 2nd seat), and California (a 53rd seat). The losing states would have been Oklahoma, Pennsylvania, and Wisconsin. It should be noted that very small population differences affect the assignment of congressional seats close to the cutoff—regardless of whether the decisions are based on population figures including or excluding corrections for undercoverage. Because of California's large population size, relatively small percentage changes in its population could add or subtract a congressional seat; California shows up as both a possible winner and possible loser in Table 2.2. In this instance, the correction would have resulted in a distribution of congressional seats that is not substantially different from the allocation of seats based on the uncorrected counts. Congressional redistricting would be affected to a greater extent than apportionment because virtually all congressional districts, except for those in single–state districts, would have their boundaries changed by adjusted census block data. Moreover, a census that is corrected for undercoverage in the physical enumeration would affect the redistricting for state legislatures and city councils, which rely on decennial census data. Effects on Distribution of Federal Funds The undercount also affects the distribution of federal and state funds, which are allocated on the basis of population. Funds for education, health, transportation, housing, community services, and job training are all allocated to geographic areas on the basis of population size and social and economic factors. In 1990 the federal government disbursed about $125 billion to state and local governments, and nearly half of this amount was distributed using formulas involving census population data. Several studies have examined the effect of adjusting for census undercount on the distribution of funds to state and local governments. All studies of the 1970 and 1980 censuses concluded that the impact of census population adjustment on grant allocations would be small.4 More recently, Murray (1992) has reported results for the 1990 census. The total federal allocation for grants involving census population counts was $58.7 billion in 1989, or about $236 per capita for eligible population jurisdictions. However, adjusting the allocation for the undercount would not simply result in an additional $236 per unit of net undercount for several reasons. First, population
OCR for page 41
Modernizing the U.S. Census is only one of several factors in many federal formula grant allocations; in such programs, an increase of population results in only a partial increase in funding. Second, although many grant allocations increase with population gains, there are some programs (such as the Community Development Block Grant Program in the Department of Housing and Urban Development) in which funding is reduced with population growth. Finally, and most important, federal grant formulas are largely fixed in their total amount. In reality, as the total U.S. population is increased by correcting for the undercount, smaller amounts of funds per capita would be available for allocation. If, for example, a fixed sum was apportioned among geographic areas on the basis of population size alone and the population of every geographic area doubled, there would be no change in funds allocated to any area. Table 2.3 shows the effects of correcting for the census undercount on 108 federal programs that are affected by population counts. Overall, the 108 programs had obligations of $58.7 billion in 1989, with five major programs accounting for $51.8 billion of the total. Of the major programs, Medicaid is the largest, and the Highway Planning and Construction Program is the second largest. As shown in the third column of Table 2.3, the overall obligation per capita for all federal programs was $236 in 1989. When each state and relevant local jurisdiction is adjusted for undercount, some states and local areas lose and some gain.5 The overall amount per net undercounted person among gaining areas is about $56, considerably less than the average per capita obligation. Only 34 to 42 percent of areas would gain because of using corrected population counts. Many governments with only modest population undercounts would not, in fact, actually gain additional federal grant monies. The effect on redistribution of federal funds would be modest: only 0.32 percent, or about $190 million, of the total federal obligations would be altered by correcting the population count. By definition, the distribution of money under these programs would change if there were a differential change in the population count. The effect of the undercount on each state's share of a fixed total of funds distribution depends on state characteristics. Moreover, the amount of money gained and lost is obviously related not only to the estimated undercount rate, but also to population size. In federal funding allocation programs, social and economic factors as well as population counts are used. The use of these other factors points to the importance of enumeration and accurate data concerning the people counted for optimal program planning and equitable distribution of funds. The reduction of population undercount and the improvement of accuracy of collected data are both important for the Census Bureau, which needs to provide accurate data to ensure the fair funding of federal programs. Because in the 1990 census blacks and other minority groups had a larger undercount than whites—as in prior censuses—minorities and the communities in which they live have been disadvantaged in federal and other programs in
OCR for page 42
Modernizing the U.S. Census TABLE 2.3 Effects of One Method of Adjustment for Net National Undercount on Annual Federal Obligations to State and Local Governments, 1989 Obligations with 1990 Population Census Data Program Obligation Levela (billions) Obligation Per Capita in Eligible Population Per Capita Amount Per Miscount Among Gainers Percentage of Funds Redistributed Percentage of Jurisdictions Gaining Social Service Block Grants $2.7 $10.79 $3.78 0.53 42 Highway Planning and Construction 13.4 45.58 13.17 0.42 42 Rehabilitation Services 1.4 5.78 5.15 1.35 42 Medicaid 34.0 136.75 38.67 0.55 34 Community Development Block Grants 2.2 15.13 13.47 0.97 39 103 Other Programs 6.9 26.67 8.50 0.46 42 Total—108 Programs $58.7 $236.00 $56.00 0.32 34-42 a The obligation levels for the first five programs include total assignments of the program, totaling $53.7 billion. The actual amount distributed to the 50 states and the District of Columbia was $51.8 billion. Source: Murray (1992).
OCR for page 43
Modernizing the U.S. Census which population is an important factor for fund allocation. Representatives of the black, Asian and Pacific Islander, Hispanic, and American Indian and Alaskan Native populations have stressed their concerns about improvements in population coverage for their groups. The analysis in this chapter reports reapportionment rates and funding allocation levels for state and local governments. However, much of the effect of population coverage is at the substate level, although there are few data to document those effects. Of those variables currently measured, population coverage varies primarily by housing characteristics (tenure and types), race, sex, and age. The possible shifts in legislative representation and possible shifts in state and local monies would be greater at substate levels than demonstrated above for federal reapportionment and funding allocation, because of the greater heterogeneity in population characteristics across small areas than across larger areas. NOTES 1 The reasons why people who were counted at the wrong location are both an omission and an erroneous enumeration and why substitutions are treated as coverage errors has to do with the way in which coverage and the net undercount are estimated by a postenumeration survey (see Chapter 5). Because the postenumeration survey, by design, selects people in a sample of areas, it detects errors in geographic residence as well as omissions and erroneous enumerations at the national level. 2 Alternate estimates for the undercount from the PES showed rates that were more than 6 times higher for blacks than for whites, 4.6 and 0.7 percent, respectively (Hogan and Robinson, 1993:18). Undercount rates were 3 time higher for Asians, 7 times higher for Hispanics, and 17 times higher for American Indians than for whites. 3 PES clusters are small areas of the country that are sampled for the survey, prior to the selection of individual housing units. PES clusters were either a census block or collection of blocks in 1990, chosen as the primary sampling unit for the survey. The purpose of selecting clusters is to get the most reliable results per unit of costs. 4 Because formula grant allocations depend on factors other than population, the disbursements can be affected by errors in other data. Siegel (1975) argued that the underreporting of income in the 1970 census affects grant allocations more than population undercoverage. 5 Not all local jurisdictions are eligible for all federal allocation formula grants. The Community Development Block Grant program, for example, is restricted to cities and towns with 50,000 or more population and to urban counties.
Representative terms from entire chapter: