National Academies Press: OpenBook

The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition (2015)

Chapter: 5 Taking the Census I: Improving the Count

« Previous: 4 Evaluating the Decennial Census: Past Experience
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

5

Taking the Census I: Improving the Count

The charge to the Panel on Decennial Census Methodology called for investigation of methods of conducting the decennial census that could prove more cost-effective than the methodology used in 1980. The 1980 methodology, as described in Chapter 3, included numerous programs designed to improve coverage in hard-to-count areas and of hard-to-count populations and stipulated that all follow-up and coverage improvement operations be carried out as completely as possible. The panel was asked to consider possible alternative methodologies, for example, a methodology that would incorporate adjustment for coverage and content errors. Adjustment, if appropriate methods can be developed and implemented, might not only increase accuracy but also lessen costs by leading to a decision to give somewhat less emphasis to coverage improvement programs during the conduct of the census. Similarly, the panel was asked to consider the uses of sampling for the count and of administrative records as means of reducing costs compared with the 1980 methodology.

Most programs directed toward coverage improvement are expensive in absolute terms and often in terms of the cost per person or housing unit identified and added to the census. Moreover, some coverage improvement programs as well as other census procedures may have introduced some overcounts in 1980 by duplicating persons or otherwise erroneously adding persons. In general, however, the panel believes that the costs of well-designed and well-executed coverage improvement programs represent money well spent for improving the count. The panel, from the beginning of its work, identified as a key issue that of reviewing coverage improvement methods with the purpose of identifying particularly prom-

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

ising approaches that should be part of the methodology for conducting the enumeration.

This chapter begins by summarizing the literature on what is known about the characteristics of hard-to-count areas and groups in the population to provide the necessary background for evaluating the cost-effectiveness of coverage improvement programs. The section also summarizes what is known about the problem of overcounting. (Appendix 5.1 provides a more detailed review of the literature on undercounting and overcounting.)

The chapter then reviews the history of efforts directed specifically toward coverage improvement in both the 1970 and 1980 censuses and the Census Bureau’s plans for testing coverage improvement methods for 1990. Finally, the chapter presents the panel’s recommendations for priority areas for research and testing with regard to coverage improvement.

HARD-TO-COUNT GROUPS IN THE CENSUS: WHAT IS KNOWN

Experience in 1980

Evaluation studies of the completeness and accuracy achieved in the 1980 census are still in progress. Estimates published to date, based on the method of demographic analysis, show the rate of net undercount for the total population in the range of 0.5 to 1.4 percent, depending on the estimate of the number of resident undocumented aliens in the country (see Table 5.1). The highest net undercount rate estimated by demographic analysis for 1980 (1.4 percent) is about three-fifths of the rate estimated for 1970 (2.2 percent) and only two-fifths of the 1950 rate (3.3 percent). The differential rate of undercoverage between the black population and all others has narrowed somewhat for the nation as a whole, as the table shows. The differential in 1980 of 5.5 percentage points between net undercount rates for blacks and all other legal residents is about three-fourths of the 1950 differential of 7.2 percentage points. However, the 1980 differential is over 90 percent of the 1960 and 1970 differentials of about 6 percentage points, and most of the gain achieved by 1980 in narrowing the differential resulted from better coverage of black women and not black men.

Rates of gross and net undercount in 1980 varied by population group and by geographic area, with rates considerably higher for certain groups than for the population as a whole. The 1980 census experienced overcount as well, and rates of erroneous enumerations also differed to some extent among groups in the population. In addition to demographic analysis, studies that shed light on the kinds of persons who were more poorly counted in 1980 include the Post-Enumeration Program and the IRS-Census Match.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

TABLE 5.1 Net Undercount Rates by Race and Sex from Demographic Analysis, 1950 to 1980 Decennial Censuses (estimated population minus census population as a percentage of estimated population)

Population Category 1950 1960 1970 1980
Total population   3.3   2.7   2.2   0.5–1.4a
     Male   3.8   3.3   3.1 N.A.
     Female   2.8   2.2   1.4 N.A.
Legally resident population N.A. N.A. N.A.   0.5
     Male N.A. N.A. N.A.   1.5
     Female N.A. N.A. N.A. –0.4
Black population   9.7b   8.0   7.6   5.3
     Male 11.2   9.7 10.1   8.0
     Female   8.2   6.3   5.3   2.7
White and other races population   2.5c   2.1   1.5 –0.2
     Male   2.8   2.5   2.1   0.6
     Female   2.1   1.7   0.9 –0.9

NOTE: A minus sign indicates net overcount.

N.A. = not available; difference between total and legally resident population probably negligible (except for 1980).

aLower percentage assumes presence of 2 million undocumented aliens in the estimated population; upper percentage assumes presence of 4 million undocumented aliens. The census population used in calculating the total population rates is the actual count, including an estimated 2 million undocumented aliens that were counted.

bBlacks and other nonwhites.

cWhites only.

SOURCES: For 1950: Siegel (1974:Table 3). For 1960: Siegel (1974:Table 2, Set D estimates). For 1970: Passel et al. (1982:Table 1, column labeled “modified census count”). For 1980: For total population rates shown, Passel et al. (1982:Table 2, assumptions 2 and 4); for all other rates, Passel and Robinson (1984:Table 2). All 1980 rates shown, except for total population, include only legal residents in both the estimated and the census populations.

Demographic Analysis

Demographic analysis provides independent estimates of the national population by age, race, and sex that, when compared with the census counts for these categories, result in estimates of net undercount. (See Chapter 4 for a description of the methodology, which is based on birth and death records and estimates of net immigration.) For 1980, the presence of a significant but unknown number of undocumented aliens for whom immigration data do not exist complicated the analysis. Preliminary results

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

are available and work is in progress on refining these results. A report on that effort indicates that the preliminary findings will not be altered to any significant degree (Passel and Robinson, 1984).

The net undercount rates for 1980 derived to date by demographic analysis are graphed in Figure 5.1. All points shown are based on estimates of the legally resident population only. Clearly, blacks were more poorly counted than the remainder of the population and men more poorly counted than women. Black men ages 25-54 experienced the highest rates of net undercount, followed by black men ages 20-24 and ages 55-59. Black children of both sexes under age 10 also experienced high rates of net undercount. According to this set of estimates, nonblack women experienced small net overcounts in almost every age group. Black men and women ages 65-74 also showed net overcounts.

Data are not shown for undocumented aliens. Warren and Passel (1983) estimated that the 1980 census successfully counted 2.1 million undocumented aliens, and their estimates for age-race-sex groups were subtracted from the census population totals to obtain the net undercount estimates graphed in Figure 5.1. Hill, in a review of recent work on estimating the stock of undocumented aliens (Levine et al., 1985:App. B), concludes that the number of illegal aliens counted in the census can reasonably be estimated in the range from 1 to 2.5 million. He also states that (p. 243), “though no range can be soundly defended, a population of 1.5 to 3.5 million illegal aliens appears reasonably consistent with most of the studies.” No firm conclusions are possible about the net undercount among undocumented aliens, given the broad range of estimates both of the total illegal alien population and of those illegal aliens recorded in the census.

Demographic analysis does not provide coverage estimates either for population groups other than the basic age-race-sex categories or for subnational geographic areas. Moreover, demographic analysis does not permit further analysis of net coverage rates in terms of gross undercount and gross overcount. Finally, net undercount rates from demographic analysis for specific age-race-sex subgroups reflect reporting errors, such as age overstatement or understatement, as well as coverage errors per se.

The Post-Enumeration Program

The Post-Enumeration Program (PEP) matched interview records from the April and August 1980 Current Population Surveys to 1980 census records to measure underenumeration in the census and rechecked a sample of census records to detect erroneous enumerations. The PEP is a source of information about differential rates of net and gross coverage errors among groups in the population. (See Chapter 4 for a description of the PEP and of the estimates of net undercoverage produced from it.)

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

To assess the inequity among geographic areas and population subgroups resulting from differential coverage in the census, one must ultimately look at net undercoverage rates. But to identify groups in the population that, for one reason or another, are particularly hard to count and, conversely, groups that are more likely to be overcounted, it is the gross omission and gross overenuneration rates that one needs to examine. Preliminary findings on gross error rates from exploratory analysis of the PEP data at the Census Bureau are summarized below and more fully reviewed in Appendix 5.1. The results presented are largely from the PEP 3-8 series of estimates, which was the first series to be put into a computerized form suitable for this kind of analysis. Examination of gross error rates from several other PEP series of estimates generally confirms the picture shown by the 3-8 series regarding the population groups that were relatively harder to count in 1980 (see Appendix 5.1).

The PEP 3-8 series estimated an overall rate of gross omissions (i.e., persons in the Current Population Survey for whom corresponding records were not found in the census) of 5.4 percent, and an overall rate of gross overenumerations of 3.6 percent. (The reader should note that both the gross omission and gross overenumeration rates are overestimates and cannot be subtracted to give an estimate of the net undercount, which was estimated to be 0.8 percent in the 3-8 series—see Appendix 5.1 for explanation.) Given the problems the PEP encountered in implementation and the resulting uncertainty attached to the estimates, and given the exploratory nature of the analysis that was conducted of gross error rates, we assigned gross omission and gross overenumeration rates for population subgroups to broad categories prior to making comparisons.

With regard to gross omissions, the PEP results indicate the following patterns:

  1. Categorizing the population by ethnicity (race and Hispanic origin), the gross omission rates for blacks, Puerto Ricans, and “other” Hispanics (those not classified as Cuban, Mexican, or Puerto Rican) were over twice the average rate.
  2. Categorizing the population by household relationship, gross omission rates for persons not related to the head of household and for relatives other than parent, child, or spouse were over twice the average rate. In contrast, spouses had a below-average rate of gross omissions.
  3. In contrast to the findings by race and household relationship, the PEP did not estimate large differences in rates of gross omissions between men and women or among age groups. Similarly, large differences were not evident by region of the country or type of area, although central cities of large standard metropolitan statistical
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

areas (SMSAs with 3 million or more population) had a moderately high gross omission rate compared with the average. Areas enumerated using conventional techniques rather than mailout-mailback approach had a below-average rate.

  1. Cross-classifying ethnicity and type of place by the mail nonreturn rate for the district office (i.e., 100 percent minus the percentage rate at which questionnaires were mailed back from households) produced striking differences in gross omission rates. Blacks and Hispanics in district offices with mail nonreturn rates of 30 percent or higher exhibited gross omission rates more than three times the average, while the gross omission rate for blacks in district offices with mail nonreturn rates of under 15 percent was only moderately above the average rate and the gross omission rate for the corresponding group of Hispanics was close to the average. Similarly central cities of both large and small SMSAs with mail nonreturn rates of 35 percent or higher had gross omission rates more than three times the average, while those cities with mail nonreturn rates below 10 percent had below-average rates.

Mail nonreturn rate appears to be a good indicator of gross omissions. Of course, the mail nonreturn rate is a symptom and not a cause of various problems pertaining to an area that result in higher-than-average rates of omissions (including not only the unwillingness of persons to be counted but also problems related to census procedures such as difficulty in delivering mail to individual households in some multiunit structures). Nevertheless, the mail nonreturn rate appears to provide valuable information to locate geographic areas in which coverage is particularly difficult. Further research on the characteristics of areas with high mail nonreturn rates that could assist development of effective coverage improvement techniques is hampered by the small sample sizes in the PEP for these areas. Moreover, at present, information on socioeconomic characteristics of the nonmatched PEP cases—for example, income and occupation—that might be useful to examine along with demographic and geographic characteristics is not in a ready form for analysis at the Census Bureau. (Fellegi, 1980a, provides estimates by a broad range of characteristics for persons missed in the 1976 census in Canada, as estimated by the reverse record check methodology.)

As already noted, the whole story regarding coverage problems in the census does not emerge solely by looking at gross omissions. In every census, some persons and housing units are counted more than once or are otherwise erroneously included (e.g., via “curbstoning” or counting as an occupied unit one that was actually vacant on Census Day). The phenomenon of overenumeration may have more to do with census procedures, for example, quality control of the address list, than with the propensities of

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

persons to be counted; nevertheless, it is necessary to examine gross overenumerations as well as omissions to obtain a complete picture.

With regard to gross overenumerations, the PEP results indicate the following patterns:

  1. Population groups with relatively high gross omission rates also tended to have relatively high rates of gross overenumerations. However, the dispersion in gross overenumeration rates was less than the dispersion in gross omission rates.
  2. By ethnicity categories, blacks, most Hispanics, and members of other nonwhite races had gross overenumeration rates moderately above the average. By household relationship, persons not related to the household head and relatives other than parent, child, or spouse had moderately high gross overenumeration rates relative to the average.
  3. Gross overenumeration rates also varied by type of enumeration procedure. Within mailout-mailback areas, enumerations obtained through follow-up for nonresponse exhibited a rate of gross overenumerations more than twice the average rate, while enumerations resulting from mail returns exhibited a below-average rate. Enumerations obtained in conventional areas also had a below-average rate of gross overenumerations.

The IRS-Census Match

A methodological study conducted after the 1980 census, the IRS-Census Match, provides, as a by-product, information indicative of differential rates of gross omissions from the census. (The Internal Revenue Service provided a sample of tax returns to the Census Bureau for the analysis but had no access to the census data for these returns.) This study, which matched a sample of about 11,000 filers of 1979 tax returns to 1980 census records, found the following patterns of gross omission rates (see Appendix 5.1 for further details):

  1. Categorizing tax filers by sex and ethnicity, black men had a gross omission rate more than twice the average for the study, while white women had a below-average rate.
  2. Categorizing tax filers by marital status (proxied by joint versus single return) and income level, blacks filing single returns at all income levels and most Hispanics filing single returns had gross omission rates more than twice the average, as did blacks filing joint returns with low incomes (less than $8,000) and most Hispanics filing joint returns. In contrast, blacks filing joint returns
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

and whites filing single returns with higher incomes ($15,000 or more) and most whites filing joint returns had below-average gross omission rates.

Experience from Previous Censuses

Coverage evaluation programs for previous censuses provide additional information about groups in the population that are more apt to be undercounted compared with other groups. It is important to look at data available from previous censuses both for clues as to the correlates of the undercount and also to determine if there are any patterns over time. That is, are some population groups apparently getting easier to count and others harder to count? Any time patterns that can be discerned have implications for choice of coverage improvement methods in the next census. Unfortunately, only the post-enumeration survey program for the 1950 census provides separate gross overcount as well as undercount figures, and the great differences in enumeration methods make it hard to compare the 1950 with the 1980 results.

Demographic Analysis

Previous censuses show similar patterns, though higher levels, of net undercount for broad population groups as in 1980, using the method of demographic analysis. In every census since detailed coverage analysis began in 1950, blacks were more poorly counted than others and men more poorly counted than women (see Table 5.1).

Looking at patterns of undercount for more finely stratified age, race, and sex groups reveals some intriguing differences over time. Black men of working age were the most heavily undercounted group in 1980. This has also been true in previous censuses, but the data show a shift in the age groups most affected (see Figure 5.2). In 1960, black men ages 15-39 were most heavily undercounted; in 1970, the age group experiencing the greatest undercount among black men had shifted to the range from 20 to 49; in 1980, black men with the greatest undercount rates were in the age range from 25 to 54.

This pattern does not clearly support a conclusion that undercount among black males is age-specific nor a conclusion that high rates of undercount are specific to a particular cohort of the population. Nevertheless, the data suggest that a group of black men who were ages 15-34 in 1960 is still proving particularly hard to count as the cohort grows older. The data also suggest that in every census young black men age 20 and older are much harder to count than black male teenagers. The phenomenon of black children under age 10 of both sexes being relatively hard to count appears

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

to be a new pattern evident in 1970 and 1980 but not 1960 (based on data not shown for black female children as well as the data shown for males).

Post-Enumeration Surveys

Post-enumeration surveys conducted in previous censuses provide data on relative rates of undercoverage for various population groups. Appendix 5.1 reviews the findings of these surveys in detail. Highlights of the survey results include:

  1. With respect to household relationship, the 1950 and 1960 survey results corroborate the finding from the 1980 PEP that persons not belonging to the nuclear family are harder to count than household heads, spouses, and their children.
  2. Survey data from 1950 suggest that fewer years of schooling are associated with a higher-than-average gross omission rate.
  3. Findings with regard to labor force status, occupation, and income are mixed. The 1960 survey found relationships of low income and unemployment to higher rates of gross omissions, but, in the case of income, the relationship appeared stronger for whites compared with blacks. Both the 1960 and 1950 survey results estimated high gross omission rates for persons employed as agricultural laborers, while farmers and farm managers had below-average gross omission rates.

Resident Observer Studies

The techniques of resident observation employed in ethnographic studies were used on one occasion to investigate factors affecting the coverage of household surveys. The findings from this study support and extend the findings about hard-to-count groups in the census based on traditional methods of coverage evaluation. Appendix 5.1 provides a full description.

Housing Coverage Studies

Rates of omission of housing units do not necessarily translate into comparable rates of missed persons; nevertheless, studies of completeness of coverage of housing units conducted in every census since 1950 are another source of information on relative rates of gross omissions in the population. Persons can be missed in the census because the entire household is overlooked or because one or more persons in an otherwise enumerated household are missed. In 1950, the post-enumeration survey indicated that three-quarters of all missed persons were in whole households that were

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

missed, while only one-quarter were in enumerated households (Bureau of the Census, 1960:Table C). By 1970, this distribution had changed: only half of missed persons were in missed households and the other half were in otherwise enumerated households. Among blacks, nearly three-quarters of those missed were in enumerated households (Siegel, 1975). With improvements in compilation and review of the address list used for the census, the remaining problem of coverage has, to a great extent, shifted from a problem of locating structures to one of finding everyone who is associated with a particular household. Appendix 5.1 reviews findings from the 1980 census and previous censuses on characteristics of missed housing units.

COVERAGE IMPROVEMENT PROGRAMS: PAST EXPERIENCE

In past censuses, the Census Bureau has implemented programs designed to improve coverage. Those programs have included general advertising and publicity to increase awareness of the census and encourage response, programs directed toward improving the quality of staff and operational procedures, and, finally, special programs targeted specifically to known problem areas. This section reviews the special coverage improvement programs implemented in 1970 and 1980 to address specific problem areas.

Coverage Improvement in 1970

The Census Bureau adopted specific coverage improvement procedures for the 1970 census predicated on three assumptions:

  1. The need for even greater accuracy in the population count than achieved in the past because of the use of the data for legislative redistricting under “one man, one vote” court requirements and the growing use of the data for fund allocations.
  2. The perception that it was becoming increasingly difficult to obtain a complete count in the absence of additional coverage efforts.
  3. The belief that new methods would be required to effect any coverage improvement. As a history of the coverage improvement efforts in 1970 notes (Bureau of the Census, 1974b:1):

The 1950 and 1960 programs were predicated on the assumption that undercounts were due largely to the enumerator’s failure to follow instructions. Hence, stress was placed on simplified procedures, training, and quality control. Analysis of the results of the 1960 evaluation program . . . indicated that the reasons were more complex. In particular, a substantial part of the undercount appeared to be due either to deliberate attempts by some segments of the population to be omitted from the census or to the fact that they did not fit into

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

any households by the conventional rules of residence. Even where the undercount was due to complete households being missed, the causes were frequently such that additional enumerator training, exhortation to the enumerators, and similar approaches appeared potentially capable of only marginal gains.

Programs to encourage public cooperation with the census, particularly among hard-to-count groups, were important components of the Census Bureau’s strategy to obtain complete coverage in 1970. These programs included public information efforts and community education programs, assistance centers set up in 20 cities that the public could call or visit for help in filling out census forms, and providing instruction sheets and questionnaires in Spanish and Chinese where needed. Special efforts to improve enumerator performance in the 20 largest cities were also adopted.

The Census Bureau also implemented specific coverage improvement programs designed to add housing units and persons to the count, most of which were also used in the 1980 census. These programs are identified in Table 5.2, which indicates the number of housing units and persons added by each program, total costs, and costs per housing unit and person added (all costs are in 1980 dollars). The table categorizes the programs as: (1) programs carried out prior to Census Day with the primary purpose of correcting the address list, both in terms of entire structures and units within structures, (2) programs carried out during the data collection phase and designed to locate missed units within structures or to verify the occupancy status of listed units, and (3) programs carried out during data collection and designed to add missed persons. Note that the cost estimates provided are only approximate, as are the estimates of numbers of housing units and persons added to the count.

In brief, the 1970 coverage improvement programs included:

1.1 Advance Post Office Check (APOC). The APOC involved a check of the address list carried out from February through October 1969 by the U.S. Postal Service in areas for which the Census Bureau purchased commercial mailing lists. These areas included about three-quarters of the mailout-mailback population or 45 percent of the total population.

1.2 Precanvass. The Precanvass was an additional check that Census Bureau enumerators made several weeks before Census Day of the address list in selected enumeration districts of 17 large metropolitan areas expected to prove difficult to count. The enumerators concentrated on identifying multiple units within structures.

1.3 Casing and Time of Delivery Checks. These checks involved review of the address lists by the Postal Service just prior to Census Day

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

TABLE 5.2 Additions and Costs of 1970 Census Coverage Improvement Programs

Housing Units Added Persons Added Cost (1980 dollars)
Number (1,000s) Percentage of Total Number (1,000s) Percentage of Total Total (1,000s) Per Added HU Per Added Person
Programs to improve address list prior to data collection
     Advance Post Office Check (APOC)a 1,200 1.7 3,600 1.8 8,250   6.88   2.29
     Precanvass    108 0.2    234 0.1    743   6.88 3.18b
     Casing and Time of Delivery Checks 1,800 2.6 5,400 2.7 N.A. N.A. N.A.
     Subtotal 3,108 4.5 9,234 4.5 8,993 6.88c 2.35c
Programs to improve housing unit count during data collection
     National Vacancy Check __d __d 1,069 0.5    225 __   0.21
     Post-Enumeration Post Office Check (PEPOC)    174 0.3    484 0.2 1,538   8.84 3.18b
     Report of Living Quarters Check    126 0.2    380 0.2 1,207   9.58 3.18b
     Subtotal    300 0.4 1,933 1.0 2,970 9.15e   1.54
Programs to improve person count during data Collection
     Missed Persons Campaign __ __ __ __ N.A. N.A. N.A.
     Movers Check __ __      15 __    635 __ 42.33
     Supplemental Forms Operation      40 0.1    122 0.1    388   9.70 3.18b
     Subtotal      40 0.1    137 0.1 1,023 9.70f   7.48
TOTAL 3,448 5.0 11,304 5.6 12,986 7.36g 2.20h
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

NOTES: Total 1970 housing unit and population counts, used as bases of percentages, were 68,672,000 and 203,302,000, respectively. Many programs were conducted in only some areas of the country. Hence, when evaluating the cost-effectiveness of such programs in adding to the count of persons or housing units, care should be taken to use the appropriate denominator as noted in the text. Detailed percentages may not add to subtotals due to rounding.

aBureau of the Census (1976:3-39, 3-42, 4-21 to 4-24). Housing unit additions are approximate estimates; person additions are housing unit additions times 3 persons per unit. The figure for housing units added through APOC represents net additions (4.4 million additions minus 3.2 million deletions). The APOC also corrected about 1.8 million addresses within structures.

bCosts for the Precanvass, PEPOC, Report of Living Quarters Check, and Supplemental Forms Operation were estimated by the Bureau of the Census (1974b) at about $1 to $2 per person, and hence were calculated as $1.50 times 2.118, or $3.18, for this table.

cPer housing unit cost calculated as $8,993/1,308; per person cost calculated as $8,993/3,834; i.e., denominator includes only housing unit or person additions for programs for which total costs are available.

dThe National Vacancy Check resulted in 250,000 housing units or 0.4% of the total being reclassified from vacant to occupied (Bureau of the Census, 1973c:15).

ePer housing unit cost calculated as $2,745 (cost of PEPOC and Report of Living Quarters Check)/300.

fPer housing unit cost calculated as $388/40.

gNumerator = $12,126 ($8,933 + $2,745 + $388); denominator = 1,648 (1,308 + 300 + 40).

hNumerator = $12,986; denominator = 5,904 (3,834 + 1,933 + 137).

SOURCE: Number of housing unit and person additions are from Bureau of the Census (1974b:Table A), except where otherwise noted; costs are approximate estimates from Bureau of the Census (1974b:18), except where otherwise noted. All costs are expressed in 1980 dollars (1970 cost estimates times 2.118).

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

both in mailout areas for which the Census Bureau purchased lists and in prelist areas for which Census Bureau enumerators developed the mailing list.

2.1 National Vacancy Check. In the National Vacancy Check, the Census Bureau carried out a sample survey of about 13,500 housing units originally classified as vacant to determine their occupancy status. On the basis of the findings, imputation procedures were used to reclassify 8.5 percent of all vacant units as occupied and to impute persons to these units.

2.2 Post-Enumeration Post Office Check (PEPOC). The PEPOC was administered in conventionally enumerated areas of 16 Southern states. The Postal Service checked the address lists developed by enumerators for completeness and Census Bureau staff followed up a sample of missed addresses in the field. On the basis of this effort, housing units and persons were added to the census records via imputation.

2.3 Report of Living Quarters Check. This check involved comparing respondents’ answers to Question A about number of living quarters at their address with the number recorded on the census address list. For structures listed as having fewer than 10 units, for which the respondent indicated a greater number of units than noted in the census list, enumerators made a field verification of the number of units.

3.1 Missed Persons Campaign. In this operation the Census Bureau left cards with community and other local organizations to distribute to persons in casual settings, such as carry-outs, barbershops, etc. The cards, which asked for minimal demographic information, were to be returned to the Census Bureau to match to the census records.

3.2 Movers Check. In the same metropolitan areas in which the Precanvass was conducted, the Census Bureau attempted to follow up persons reporting a change of address to the Postal Service during the census enumeration period.

3.3 Supplemental Forms Operation. The Census Bureau mounted special “Were you counted?” campaigns and enumerated persons who came forward on special forms. Residents traveling overseas were also enumerated with supplemental forms. In most cases, these forms were processed for an area and persons added only when the total number of supplemental forms represented 1 percent or more of the enumeration district population.

The programs in category 1 added about 4.5 percent to the housing unit and person count and were reasonably cost-effective (recogniz-

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

ing that cost-effectiveness of coverage evaluation programs is difficult to measure, particularly in the absence of information regarding the proportions of housing units and persons correctly added to the count, i.e., not overcounted). The APOC added 1.7 percent to the overall housing unit count—3.8 percent in the commercial mailing list areas in which the program was conducted—in addition to correcting many addresses. The Casing and Time of Delivery Checks added 2.6 percent overall and fully 4.4 percent in the mailout areas in which these checks were performed. The Precanvass added only 0.2 percent to the total housing unit count, but the program was implemented in selected areas of only 17 metropolises. In these selected areas, the Precanvass added 2.3 percent to the housing unit count.

The 1970 programs carried out during the data collection phase and aimed at checking the count of housing units and their occupancy status (category 2) proved cost-effective as well, although these programs added a much smaller percentage to the population count than the address check programs. The National Vacancy Check added 0.5 percent to the population count and reclassified 0.4 percent of total housing units from vacant to occupied. The program cost very little per added person, because it was carried out on a small sample (about 0.2 percent) of units originally classified as vacant. Of course, in determining the cost-effectiveness of a coverage improvement program based on a sample survey, one must look not only at the cost per added person but also at the reliability of the data obtained. A program with a smaller sampling fraction will cost less on a per person added basis compared with a more extensive program, but may also produce less reliable data.

Evaluation of the 1970 National Vacancy Check indicated that data quality was high, even with the error introduced by sampling (Waksberg, 1970, 1971). The program was implemented in a conservative manner in several respects. First, units in the sample of 13,500 were reclassified from vacant to occupied only if the enumerator determined that the same family had continuously occupied the house during the census enumeration period. On this basis, 11.4 percent of the sample units were reclassified. In the imputation procedure applied to the complete set of census records, instead of 11.4 percent, a total of 8.5 percent of vacant units (to attempt to account for the smaller average household size of misclassified units in the sample compared with correctly classified units) were changed to occupied and persons imputed to these units. It turned out that this procedure imputed somewhat fewer persons than expected because the imputed household size for the reclassified units on average was yet smaller than the average household size targeted for the imputation. The best estimate is that no more than another 0.1 percent should have been added to the population count (Bureau of the Census, 1974b:12-13).

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

The Post-Enumeration Post Office Check added 0.3 percent to the housing unit count overall and 0.2 percent to the population count. The program added 1.3 percent in the conventionally enumerated areas of the South in which it was carried out. The recheck of units in which the respondent reported more living quarters than there were addresses for the structure on the mailing list added at a minimum about 0.2 percent to the population count overall. The Census Bureau was only able to estimate the effects of this program for questionnaires returned by mail. For the latter universe, the added persons shown in Table 5.2 represent about 0.3 percent of the total. The Census Bureau estimated that the Report of Living Quarters Check was erroneously omitted in one of three cases; if the check had been made for all applicable addresses, at least 0.3 percent would have been added to the total population count (Bureau of the Census, 1974b:4).

The programs directed toward finding missed persons (category 3) were least effective in terms of additions to the count. The Supplemental Forms Operation added less than 0.1 percent to the total population count, although, as previously noted, these forms were generally processed only where they represented 1 percent or more of the initially enumerated population. The Movers Check added a negligible number of persons overall and 0.6 percent to the population of the areas in which it was performed (the same 17 large metropolitan areas in which the Precanvass was implemented). The Census Bureau estimated that the Movers Check would have added another 0.6 percent to the population of these 17 areas if the program had been carried out completely according to specifications (Bureau of the Census, 1974b:8).

As noted in a previous section, the 1970 census missed proportionately more persons in otherwise enumerated households than in missed households compared with the 1950 experience. This result is probably due, at least in part, to the relative effort and success achieved by the programs aimed toward housing unit coverage (the first and second categories in Table 5.2) versus the programs aimed at identifying missed persons. Another point to emphasize regarding the 1970 coverage improvement strategy is that many programs were carried out on a selective basis in areas in which it was felt they would be particularly effective or for which the effort was believed to be justified in terms of cost. Two programs, the National Vacancy Check and the PEPOC, were carried out on a sample basis and the results used to impute persons to the census. Finally, there was some effort evident in the 1970 program, specifically in the National Vacancy Check, to guard against overcounting as well as undercounting.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

Coverage Improvement in 1980

The 1980 coverage improvement strategy exhibited three differences from 1970:

  1. The resources put into coverage improvement in 1980 exceeded the resources spent in 1970 (expressed in 1980 dollars) by several orders of magnitude, reflecting the belief that every effort was necessary to obtain accurate coverage to satisfy needs for fund allocation, redistricting, equal employment opportunity actions, and other important public policy uses of census data. Programs aimed at increasing public cooperation, particularly among hard-to-count groups, such as special publicity efforts, assistance centers, and foreign-language questionnaires, were greatly expanded, as were the number and extent of programs designed specifically to add housing units and persons to the count.
  2. The Census Bureau made a deliberate decision to conduct most specific coverage improvement programs on a nationwide basis and to avoid the use of sampling and imputation. However, some programs were implemented selectively in areas specifically designated for the purpose.
  3. Several new programs were adopted to tackle the problem of within-household undercoverage, although most programs, as in 1970, were directed toward improvement of the address list either before or after Census Day.

Table 5.3 provides statistics from Census Bureau evaluations (see Thompson, 1984; updated in Bureau of the Census, 1985c) regarding coverage improvement efforts in 1980 for programs implemented prior to Census Day directed at improving the address list and programs implemented during data collection. Again, estimates of cost and added housing units and persons are approximate. Overall, the 1980 census coverage improvement effort, including all the programs listed in Table 5.3, accounted for almost 9 percent of the total costs of the census and added about 8.4 percent to the total population count. In 1970, the coverage improvement programs listed in Table 5.2 accounted for about 3 percent of the total census costs and added about 5.6 percent to the total population. As in 1970, the cost-effectiveness of specific 1980 census coverage improvement programs varied greatly.

The 1980 census address list improvement programs carried out prior to Census Day (Table 5.3, Panel A) proved extremely cost-effective. The Advance Post Office Check performed by U.S. Postal Service staff in summer 1979 and the Precanvass carried out by Census Bureau staff in early

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

TABLE 5.3 Additions and Costs of 1980 Census Coverage Improvement Programs

Program Housing Units Added Persons Added Cost (1980 dollars)
Number (1,000s) Percentage of Total Number (1,000s) Percentage of Total Total (1,000s) Per Added HU Per Added Person
Panel A: Programs to improve address list prior to data collection
     Advance Post Office Check (APOC) 2,000a 2.3 5,120 2.3 6,970 3.49 1.36
     Precanvass 2,360b 2.7 6,030 2.7 11,800 5.00 1.96
     Casing and Time of Delivery Checks 2,060 2.3 5,280 2.3 9,290 4.51 1.76
     Subtotal 6,420 7.3 16,430 7.3 28,060 4.37 1.71
Panel B: Programs to improve housing unit count during data collection
     Local Review 53c 0.1 76c __ 4,310 44.74e 31.20e
     Post-Enumeration Post Office Check (PEPOC) 50 0.1 130 0.1 990 19.80 7.62
     Prelist Recanvass 120 0.1 220 0.1 10,290 85.75 46.77
     Vacant/Delete Check 409d 0.5 1,720 0.8 36,320 36.41f 21.12
     Subtotal 632 0.7 2,146 0.9 51,910 45.16g 23.29h
Panel C: Programs to improve person count during data collection
     Casual Count __ __ 13 __ 250 __ 19.23
     Coverage Questions and Dependent Roster Checki 93 0.1 240 0.1 7,500 80.65 31.25
     Nonhousehold Sources Program __ __ 130 0.1 9,820 __ 75.54
     Were You Counted 17 __ 71 __ 270 15.88 3.80
     Subtotal 110 0.1 454 0.2 17,840 70.64j 39.30
TOTAL 7,162 8.1 19,030 8.4 97,810 8.99k 5.04l
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

NOTE: Total 1980 housing unit and population counts, used as base of percentages, were 88,207,000 and 226,546,000, respectively. Many programs were conducted only in some areas of the country. Hence when evaluating the cost-effectiveness of such programs in adding to the count of persons or housing units, care should be taken to use the appropriate denominator, as noted in the text. Detailed percentages may not add to subtotals due to rounding.

aAlso corrected 2.9 million addresses.

bAlso transferred 570,000 units from one geographic area to another.

cAlso transferred 48,000 housing units and 56,000 persons from one geographic area to another.

dAlso reclassified 590,000 units from vacant to occupied, in addition to the 409,000 units reclassified from “delete” to housing unit additions. Also deleted 507,000 vacant units from the housing inventory.

ePer housing unit and per person cost calculated as 55% of total costs (share attributable to additions as opposed to transfers) = $2,371/53 and 76, respectively.

fPer housing unit cost calculated as 41% of total costs (share attributable to additions as opposed to reclassification) = $14,891/409.

gNumerator = $28,542 ($2,371 + $990 + $10,290 + $14,891).

hNumerator = $49,971 ($2,371 + $990 + $10,290 + $36,320).

iHousing unit and person additions and costs based on evaluation of report of living quarters question (H4) edit only.

jNumerator = $7,770 (costs of Coverage Questions and Dependent Roster Check plus Were You Counted).

kNumerator = $64,372 ($28,060 + $28,542 + $7,770).

lNumerator = $95,871 ($28,060 + $49,971 + $17,840).

SOURCE: Calculated from Bureau of the Census (1985c:2-3).

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

1980 each added well over 2 percent to the U.S. total housing unit count. Both of these programs were limited to the tape address register (TAR) areas (i.e., city delivery areas for which the Census Bureau had developed computerized geographic coding files and purchased commercial mailing lists), and, in those areas, they added between 4 and 5 percent each to the housing unit count for a cost of about $4 per added housing unit. Comparable figures for the 1970 programs are 3.9 percent of housing units added by APOC in the TAR areas and 2.3 percent of units added by the Precanvass in 17 metropolitan areas, for a cost of about $7 (in 1980 dollars) per added unit. The 1980 Casing and Time of Delivery checks—implemented by Postal Service staff just prior to Census Day in the entire mail census area (including 95 percent of the population in TAR plus prelist areas)—also added over 2 percent to the U.S. total housing unit count for about the same cost as the other two programs. In 1970, these checks added over 4 percent of the housing units in the mailout areas.

The programs carried out during data collection that were primarily directed at checking the address list or at determining whether units were correctly classified as occupied or vacant (Table 5.3, Panel B) proved much more expensive than the pre-Census Day programs. These programs included:

  • Local Review. The Census Bureau provided preliminary housing unit and also population counts to local officials after completion of the first stage of follow-up. Officials reviewed the counts and indicated problem areas for checking.
  • Post-Enumeration Post Office Check. In contrast with 1970, PEPOC was carried out in all conventionally enumerated areas of the country on a 100 percent basis as part of the second stage of follow-up.
  • Prelist Recanvass. In Prelist areas, the address list was rechecked during the second stage of follow-up. In some areas, only selected enumeration districts were recanvassed.
  • Vacant/Delete Check. In contrast with 1970, the 1980 Vacant/Delete Check was implemented on a 100 percent basis during the second stage of follow-up. Each of 8.4 million housing units originally classified as vacant or as “delete” because they were not residential was rechecked in the field.

These four programs added about 1 percent to the total population count for an average cost of $23 per person added. Over 80 percent of this improvement was due to the Vacant/Delete Check. The Prelist Recanvass had the highest unit costs, and there is evidence that it experienced severe operational problems that diminished its effectiveness. The Local Review program also had high unit costs and added less than 0.1 percent to the

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

population count. Local Review was very unevenly implemented across the country; many areas did not participate. The effectiveness of the PEPOC in terms of adding persons is understated in Table 5.3 because it was carried out in conventional areas representing only 5 percent of the total U.S. population. In these areas, PEPOC added 1.2 percent to the population count, about the same as the performance in 1970, although the cost to add a person in 1980 was almost two and a half times the 1970 cost, reflecting the difference between a 100 percent and a sample operation. The Vacant/Delete Check, as discussed further below, probably introduced a measure of overcounting as well as reducing the undercount. The 1980 program added 0.8 percent to the population count compared with 0.5 percent for the 1970 effort. The cost to add a person from the 1980 Vacant/Delete Check was fully 100 times the 1970 cost, reflecting the great increase in the number of units that were rechecked in the field.

There are data on the characteristics of persons added to the census in 1980 from some of these programs. Evidence suggests that the Prelist Recanvass replicated the race distribution in the general population and hence did not help reduce differential undercount (Thompson, 1984:12). This further lowers the panel’s assessment of its relative cost-effectiveness. The Vacant/Delete Check, by contrast, made a measurable impact on differential coverage rates. Based on available data, it appears that this program may have reduced the black versus white differential undercoverage by 0.5 percentage points (estimated from Thompson, 1984:23).

The programs carried out to improve the person count during data collection (see Table 5.3, Panel C) proved least cost-effective. These programs included:

  • Casual Count. This operation was similar to the 1970 Missed Persons Campaign, except that, instead of relying on community organizations, the Census Bureau sent special enumerators about 6 weeks after Census Day to places frequented by transients who might otherwise be missed. The operation was limited to centralized (city) district offices.
  • Coverage Questions and Dependent Roster Check. This program was directed not only toward adding housing units and persons within households but also toward reducing erroneous inclusions. Responses to questions on number of units in the building and the roster of household members were edited and followed up as appropriate.
  • Nonhousehold Sources Program. This operation—an innovation in 1980—involved matching several administrative lists to census records for selected census tracts in urban district offices. The lists used were driver’s license records, immigration records, and public
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

assistance records in New York City. About 6.8 million persons were checked against census records.

  • Were You Counted. This program was similar to the 1970 Supplemental Forms Operation.

The above four programs added only 0.2 percent to the total 1980 population count for a cost of over $39 per added person. The component of the Coverage Questions Check that involved rechecking buildings in which the respondent reported more living quarters than there were addresses on the mailing list appeared less effective in adding persons and more costly than the comparable Report of Living Quarters Check in 1970. (The Census Bureau was not able to evaluate the effectiveness or cost of the other coverage questions in 1980.) The Casual Count and Were You Counted programs had negligible impact in both the 1970 and 1980 censuses. The one major innovation for 1980, the Nonhousehold Sources Program, which had appeared promising in pretests, added only 130,000 persons (less than 0.1 percent of the total population and less than 2 percent of the total number of administrative list entries checked against census records), for a cost of over $75 per person added. If the Nonhousehold Sources Program had been more effective in terms of persons added, the program could have had a pronounced effect on differential coverage rates. Among the small group of persons identified through the Nonhousehold Sources list matching operation, about one-third each were white, black, and Hispanic, compared with the breakdown in the general population of over 80 percent white, 12 percent black, and 6 percent Hispanic (Thompson, 1984:18-19).

In addition to programs designed to add persons during the data collection stage, the 1980 census effort included a program called Whole Household Usual Home Elsewhere, which was designed to increase the accuracy of the count by area. In this effort, about 1 million persons were transferred from one enumeration district to another in accordance with the Census Bureau’s rules of usual place of residence. For example, persons residing in a vacation home on Census Day had their data transferred to the location of their usual home. Other programs, such as Local Review and the Precanvass, also produced transfers as well as net additions.

Evaluation of Coverage Improvement Experience in 1980

Looking at the 1980 coverage improvement programs, it appears evident that programs carried out prior to Census Day to check the address list were important in improving the count and low in cost in terms of dollars per housing unit added to the list. Moreover, because these programs were implemented before the enumeration, any additions that were in fact duplications could be corrected subsequently.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

The costs per person added by the programs administered during data collection were quite high. The Nonhousehold Sources Program stands out in this regard, as does the Prelist Recanvass. The Vacant/Delete Check, although not the most costly on a per person added basis, was the most expensive program in total costs—but it significantly reduced the differential undercount, which is of key importance.

There is evidence that the Vacant/Delete Check contributed to overcount as well as importantly reducing undercount (see Bureau of the Census, 1985c:Ch. 8). The 1980 program (in contrast to the 1970 National Vacancy Check) was designed not only to verify the status of units originally classified as vacant or delete, but also to identify and enumerate persons who were missed in the census because they were moving from an old to a new residence. Enumerators were instructed to ask residents of units originally classified as vacant whether they had moved in since Census Day, and, if so, whether they had been counted at their previous residence. Movers who stated that they had not been counted were enumerated at the new address. However, people were often enumerated without being aware of the fact (e.g., because some other household member filled out the form), and, hence, movers located in the Vacant/Delete Check were at risk of being counted twice.

Other 1980 census coverage improvement programs, such as Whole Household Usual Home Elsewhere, also probably contributed to overcount. The fact that all the coverage improvement programs were implemented clerically, with no use made of automation, undoubtedly served to increase cost and reduce effectiveness. This was particularly true for programs that were carried out in the final stages of follow-up, when there was great pressure on the district offices to close out their operations.

Overall, the three pre-data collection coverage improvement programs, together with the Vacant/Delete Check, accounted for over 95 percent of persons added but only 66 percent of the coverage improvement budget—casting doubt on the cost-effectiveness of the other approaches. These comparisons would be even more favorable to these specific programs if the Vacant/Delete Check had been carried out on a sample basis, as in 1970.

CENSUS BUREAU PLAN FOR TESTING COVERAGE IMPROVEMENT PROGRAMS FOR 1990

The Census Bureau’s testing program for the 1990 census began in spring 1984 with tests in several urban and rural localities of improved methods of address list compilation—a key element in achieving completeness of coverage (Bureau of the Census, 1984b). Included in plans for 1986 pretests are many tests related to coverage improvement (see Johnson, 1984; updated in Bureau of the Census, 1985b). Almost all the programs

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

implemented in 1980 are scheduled for further testing in 1986, along with some new programs. Current plans call for testing improved techniques and procedures for the following programs that were used in 1980:

  • Advance Post Office Check. As a high priority pretest objective, the Census Bureau proposes to test the use of mailout-mailback procedures in rural areas that were conventionally enumerated in 1980. One procedure to be tested would be to have Census Bureau staff prelist the area, followed by an APOC, with the Postal Service delivering the questionnaires. There is a proposal in urban areas to test enhancing the APOC by adding identification of problem addresses (e.g., addresses where there is a mail drop for an entire building).
  • Precanvass. The Census Bureau proposes testing an enhancement of the Precanvass that includes correcting addresses within all multiunit structures, even where the count in the structure from the Precanvass agrees with the count on the address register, and also to extend both the APOC and the Precanvass operations to prelist as well as tape address register areas.
  • Casing and Time of Delivery Checks and Local Review. Various improvements to these operations are proposed for testing.
  • Vacant/Delete Check. The Census Bureau proposes to test ways of improving the effectiveness of this program, not including, however, consideration of conducting the program on a sample basis.
  • Casual Count. Tests of automating the process of searching for persons identified in the Casual Count operation and of adding them to the census are proposed.
  • Coverage Questions and Dependent Roster Check. The Census Bureau proposes to examine the combination of questions used in 1980 to check within household coverage to determine if rewording, new instructions, or other changes will increase their effectiveness, and also to test adding questions about multiple residences that could help minimize overcounting. The Census Bureau also proposes to test improvements in the Whole Household Usual Home Elsewhere program.
  • Nonhousehold Sources Program. Various possible improvements are proposed for testing in this program, such as the use of new sources of administrative lists and the use of automated matching and searching techniques.

The only programs used in 1980 that are not proposed for testing in 1986 are the Prelist Recanvass and the Post-Enumeration Post Office Check. (Conventional area enumeration methods, which include PEPOC,

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

will not be tested in the 1986 round of pretests and may not be used at all in 1990.) A new program being considered for testing is the use of an update list/leave procedure in prelist areas, in which Census Bureau enumerators instead of the Postal Service would deliver questionnaires and at the same time update the address list. Update list/leave is also proposed for testing (although perhaps not until 1987) in multiunit structures in urban areas that pose special problems for mail delivery.

The Census Bureau has outlined an ambitious testing program related to coverage improvement. The panel stated its belief in Chapter 3 that the Census Bureau must choose among the ideas proposed for testing. In the following sections, we offer recommendations regarding priorities for testing and research in the area of improving the count. The discussion first addresses needed research on hard-to-count groups and on problems of overcount.

NEEDED RESEARCH ON UNDERCOUNT AND OVERCOUNT

The panel supports further work at the Census Bureau to analyze the characteristics of population groups and areas more subject to census undercount and also those more likely to be overcounted. The panel also supports further analysis, to the extent available data permit, of the effectiveness of coverage improvement programs in reducing differential undercount. This research can contribute importantly to the planning of special coverage improvement efforts for the next census and also to the planning of evaluation programs to determine the completeness of coverage that was achieved. At the present time, the undercount research staff at the Census Bureau is continuing investigation of gross undercount and overcount with the data from the Post-Enumeration Program including analyzing enumeration districts that contain nonmatched cases (i.e., gross omission cases in the Current Population Survey) on characteristics such as percentage not speaking English and percentage low income.

Recommendation 5.1. We recommend that the Census Bureau assign a high priority to the completion of studies of undercount and overcount in the 1980 census.

Research on the characteristics of hard-to-count groups and of groups and areas prone to overcount will be more useful for planning coverage improvement and evaluation programs for the next census to the extent that the research is completed expeditiously. To be designing pretests for 1990 without having completed research on undercount and overcount diminishes the value of the research results and can result in less well-designed tests.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

Recommendation 5.2. We recommend that the Census Bureau set up a timetable and assign staff to permit completion of the analysis of 1990 coverage evaluation results in time to be used in planning the first pretest of the 2000 census.

ISSUES IN COVERAGE IMPROVEMENT: QUESIONNAIRE CONTENT

Next the panel discusses priorities for research and testing of coverage improvement programs, beginning with consideration of items on the questionnaire that relate to coverage. These items include the questions on race and Hispanic origin as well as questions designed specifically to help coverage, such as number of living quarters or addresses in the respondent’s building. The population counts for race and Hispanic groups are affected by the accuracy of reporting race and ethnicity as well as by coverage errors, and it is important to understand what responses to these questions mean if appropriate estimates of coverage rates are to be developed (e.g., from demographic analysis).

Race and Hispanic Origin Questions

Information about race and ethnicity, including particularly Hispanic origin, is required for the implementation of a number of federal and state laws pertaining to political representation, civil rights, and assistance to disadvantaged groups. Even if it were not for these specific legal requirements, such information would be needed as a basis for understanding the political and economic status of various racial and ethnic groups. The legal uses of racial and ethnic categories reflect basic political and economic concerns of U.S. society today. These concerns are evident in the importance attached to completeness of coverage in the census for race and ethnic groups. Differential rates of net undercoverage—for example, the net undercount rate of greater than 5 percent estimated for blacks in 1980 compared with a rate probably considerably less than 1.5 percent for all others—have excited more attention than the undercount rate for the entire population.

Information about race has been collected in each census since 1790. A specific separate question on Hispanic origin was introduced for the first time in 1970, when it was asked on a sample basis. In 1980, a question on Hispanic origin was included on the short form.

For 1990, issues related to the panel’s work include:

  1. Whether question design can be improved to yield more accurate and/or more useful information, including whether the design
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

should explicitly strive for comparability with other sources of race and ethnicity information, such as vital statistics.

  1. Whether, for considerations of coverage improvement, minimizing respondent burden, or other reasons, part of the race and ethnicity information could more appropriately be collected on a sample basis.

Race and Ethnicity Questions in Earlier Censuses

Changing information needs and societal attitudes about race and ethnicity have been reflected in changes in the design, content, and enumerator instructions for the race and ethnicity question(s) from one census to the next. The frequent changes severely limit data comparability across succeeding censuses.

In 1920, persons of mixed white and Negro blood were classified as Mulatto. Anyone who was not classified as White, Black, Mulatto, Chinese, Japanese, or Indian was classified as “Other.” In 1930, the Mulatto designation was dropped. Enumerators were instructed to list persons with any Negro blood, no matter how small the percentage, as Negro. Persons of Mexican birth or parentage were to be listed as “Mexican” unless definitely Negro, Indian, Chinese, or Japanese. In 1940, Mexicans were listed as white unless definitely Indian or some other race.

There were apparently no further major definitional changes in 1950 or 1960. In 1960, racial designations, and, in 1970, ethnic designations, were placed on a self-identification basis, although, where data were collected by an enumerator, the enumerator was allowed to fill in blanks by observation when possible. In 1980, however, enumerators were no longer allowed to enter race by observation. In every modern census, missing responses have been filled in via editing and imputation routines.

The Directive to Standardize Federal Race and Ethnicity Information

Increased legal and program uses of racial and ethnic designations in the 1960s and 1970s produced a proliferation of race and ethnic data collections by various agencies, using a variety of concepts and definitions. To improve data comparability, the Office of Management and Budget’s (OMB) Statistical Policy Division in 1977 established standard categories to be used by all federal agencies collecting data on race. The prescribed racial categories are: white, black, American Indian or Alaskan Native, and Asian or Pacific Islander. The ethnicity categories are: Hispanic origin, not of Hispanic origin. Alternatively, Statistical Policy Directive 15 allows agencies to use a combined race and ethnicity categorization: white (not

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

Hispanic), black (not Hispanic), Hispanic, American Indian or Alaskan Native, Asian or Pacific Islander.

The 1980 Census

The race question on the 1980 census was designed with the aim of obtaining accurate information that could be aggregated into the OMB prescribed groupings with minimum need for hand tabulation. Since there was evidence that many respondents might be unaware that their racial background was one that the federal government includes in the “Asian and Pacifíc Islander” group, nine separate race or ethnic groups for aggregation into this category were listed. Also listed were white, black, Indian (Amer.), Eskimo, Aleut, and “other,” for a total of 15 categories (see Figure 5.3 for question format). A question on Spanish/Hispanic origin appeared separately (and with two other questions intervening) on the 1980 census. It requested information for four separate Spanish/Hispanic categories (see Figure 5.3). Thus, two of a total of seven population questions on the 1980 census short form were about race or ethnicity. Together these two questions took up about 30 percent of the space on the population part of the short form.

Almost 6 million individuals identifying themselves as Hispanic on the Hispanic origin question (about 40 percent of total Hispanics) marked “other” on the race question. In contrast to 1970, when similar responses were classified as “white” during tabulation, these responses were kept in the “other” category in 1980. This change reflected a joint OMB-Census Bureau decision that the great majority of the Hispanics who responded in this way understood the race question and did not consider themselves white. Some data users were critical of this decision, which they argued impairs the comparability of the 1980 data with the data from the 1940 through 1970 censuses. The 1980 data have been tabulated and published in such a way as to permit users to reclassify this group if they wish, however. Even with such reclassification, data are not fully comparable from one census to the next, due to a variety of other changes in question design, enumerator instructions, and editing rules.

Considerations for 1990

Collection of information on race and ethnicity in a large, diverse country such as the United States is inherently difficult. With the introduction of the concept of self-identification, the racial and ethnic categories moved away from their former precise, or pseudo-precise, anthropological definitions and toward definitions stemming from commonly perceived cultural categories. This shift was appropriate. Certainly, questions requiring

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

images

FIGURE 5.3 Race and Hispanic origin questions on the 1980 census short form.

information about percentages of Negro and Indian blood (used, at least in theory, through 1950) would be generally regarded as offensive today.

The quest for accurate self-identification by respondents and the feasibility of computer tabulation produced, in 1980, a “race” question that was in fact a mix of racial, ethnic, and geographic categories. This was not inappropriate, but it does raise the question of whether the questions on race and Hispanic origin could be combined.

A related question is the possible need for information on additional ethnic or geographic categories in 1990. Since 1980 there has been substantial entry into the United States of refugee populations from Cambodia, Haiti, El Salvador, and elsewhere. These groups remain tiny relative to the total size of the U.S. population, but it may be that, as groups of particular

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

policy concern, detailed information on their geographic location will be sought. This situation suggests the desirability, in the interest of keeping the short-form question of manageable length, of moving some of the detail on Asian and Pacific Islander categories to the long-form sample. Arguing against this is the probable difficulty of obtaining accurate short-form responses without listing all the detailed categories.

There is no clear evidence that inclusion of detailed race and Hispanic origin questions on the short form in 1980 was a barrier to a complete count. The group that logically might find these questions most irritating and irrelevant is the white, non-Hispanic majority of the population. Undercoverage among this group is believed to have been minimal. Other population groups seemed, in general, willing to supply race and ethnic information and, in many cases, insistent, on doing so.

Design of the race and ethnicity question(s) is complicated by the limitations and ambiguities of common English-language usage. It is difficult to find brief, unambiguous, readily understood phraseology for distinguishing Indians (from India) from Indians (native U.S. tribes). The 1980 phraseology “Indian (Amer.)” is ambiguous. Does it apply only to tribes native to the United States or does it encompass all Indians of North and South America? Presumably, the former was intended, but should those of Mexican, South, and Central American origin not also have the opportunity to conveniently identify their origin—and would this not be useful information?

The matrix of information that may need to be collected on the short form is illustrated in Figure 5.4, but this illustration is not intended as format or phraseology for use in an actual census question.

images

FIGURE 5.4 Race and Hispanic origin information that may be required in 1990.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

Recommendation 5.3. We recommend that the Census Bureau test a variety of question designs for the race and ethnicity information to be collected in the 1990 census, including some that combine the collection of information on Hispanic origin with the other race and ethnicity information.

Developing Race and Ethnicity Questions

The Census Bureau does not have many opportunities to test important questionnaire changes, such as changes in the race and Hispanic origin questions prior to a census. Moreover, it is expensive to mount full-scale questionnaire wording tests, as was done prior to 1970 and 1980 and is planned for 1990 in a national content test, currently scheduled for 1986.

The focus group technique has been successfully employed to design survey questions. This approach, originally developed in market research, involves in-depth discussions with small, usually homogeneous groups (Higginbotham and Cox, 1979; Slavson, 1979). Focus groups offer the advantage of being able to probe for underlying meanings and hidden associations evoked by different question wording that may affect responses in unforeseen ways. This feature may be particularly useful for the testing of questions on race and ethnicity. While focus group findings cannot be directly generalized, focus groups can help narrow the range of question alternatives that warrant testing with larger—and more costly—samples selected scientifically.

As a case in point, prior to the 1980 census the Census Bureau conducted numerous tests of different wording of the question on Hispanic origin. The various pretests and dress rehearsals tried out variations of this question, as did the 1976 National Content Test, which had a sample size of 28,000 housing units. A number of serious response problems were encountered. For example, in almost every case in which a question had a category with the term “American” such as “Central or South American” or “Central or South Amer. (Spanish)” there was evidence that some non-Hispanic Americans checked these responses (Fernandez and McKenney, 1980). Holding a number of focus group sessions at an early stage in the questionnaire content planning would probably have provided timely evidence, for a relatively low cost, of this behavior and other response problems. In a similar situation, the Social Security Administration successfully used focus group interviews to identify problems and ambiguities with the race and ethnicity items on a proposed revised application form and designed operational tests of using alternative versions based on the focus group findings (Scherr, 1980; Scherr and Nelson, 1980).

Focus groups cannot and should not replace other methods of questionnaire development, including sample surveys with alternative questionnaires

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

and controlled laboratory or classroom experiments. (The Census Bureau conducted a number of classroom experiments prior to the 1980 census that provided useful findings regarding placement of instructions, the position of particular items on the questionnaire, requiring respondents to make machine-readable entries for date of birth, and the use of graphics; see Rothwell, 1983.) However, we believe that the use of focus groups for questionnaire development of sensitive and ambiguous items such as race and ethnicity would be very useful. We initially recommended the focus group technique in our interim report (National Research Council, 1984), and we note that the Census Bureau used focus groups in the 1985 pretest in Tampa to elicit reactions to the questionnaire format.

Recommendation 5.4. We recommend that the Census Bureau, in addition to other methods that it has traditionally employed, use the technique of focus group discussions as one means to develop questions on particularly sensitive items such as race and ethnicity.

Comparability Considerations

Although changes in question wording and categories for the race and ethnicity items may be necessary to improve the information, it is vitally important to strive for historical comparability of race and ethnicity data from one census to the next to the extent possible. Historical comparability is important to permit reliable analysis of changes in the status of various groups. Cross-temporal comparability is also important for evaluation of completeness of coverage, for example, using demographic analysis or reverse record check methodologies.

Recommendation 5.5. We recommend that, in 1990, as it did in 1980, the Census Bureau collect, tabulate, and release data on race and ethnicity in such a way that the data can be aggregated as necessary to obtain maximum feasible comparability with 1980 and 1970.

Comparability of race and ethnicity data from the census with race and ethnicity information collected in vital statistics records is also important for at least two reasons. First, vital statistics on births and deaths are large components of the total population estimates by race that are compared with the census counts to estimate net undercoverage using the technique of demographic analysis. Second, vital rates, such as birth, death, marriage, and divorce rates, which have vital statistics data in the numerator and census counts in the denominator, are important social indicators that are commonly analyzed by race. For both of these purposes, it is desirable that the data on race from both vital statistics and the census be as comparable as possible.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

There will probably always be differences between the concepts of race and ethnicity as collected in vital statistics and in the census, if only because the methods of data collection vary: self-enumeration in the census versus identification by others in vital statistics (parents or medical staff for newborns and relatives or medical staff for decedents). Nevertheless, discrepancies due to differences in categories and editing rules could be minimized.

Currently, definitions of race and ethnicity differ in vital statistics from those used in the decennial census in several important ways. First, vital statistics records include Mexicans, Cubans, and Puerto Ricans in the white race category. Second, not all states determine Hispanic origin, although the 22 states that do are estimated to account for 90 percent of Hispanic births. Third, there are some differences in editing rules when race is mixed or unclear. For example, in vital statistics birth records, newborns of mixed parentage are assigned the race of the father, unless the father is white or the mother is Hawaiian, in which case the child is classified according to the mother’s race (see National Center for Health Statistics, 1982a, 1982b). In the 1980 census, by contrast, persons of mixed parentage who could not specify a single category were coded according to the race of the mother (in 1970, the rule was to use the race of the father—see Bureau of the Census, 1983c.)

At present, the National Center for Health Statistics is reevaluating the standard certificates for vital events. Specifically, the center is requesting comments on whether the birth and death certificates should include a question on ethnic origin or descent separate from the race item and whether the question should ask simply for Hispanic origin or ask for origin in every case, such as Italian, English, Cuban, etc.1

Recommendation 5.6. We recommend that the Census Bureau, the National Center for Health Statistics, and other relevant federal agencies work closely together to design questions and response editing rules on race and ethnicity that minimize conceptual differences between census and vital statistics records to the extent feasible. The Office of Management and Budget should act as necessary to facilitate such coordination.

Coverage Questions

The 1970 and 1980 censuses included several questions on the short form designed to aid in achieving a complete and accurate count. In 1970, question H-A asked, “How many living quarters, occupied and vacant, are at this address?” with categories provided from 1 to 10 or more. Answers

__________________

1Personal communication from John E. Patterson to Miron L. Straf, October 26, 1984.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

to this question were checked against the address list for structures with under 10 units to identify missed households. In 1980, the same question was asked as Question H-4 and edited as in 1970. In addition, the 1980 questionnaire included as question 1 on the first page a space to list the name of each person living there on Tuesday, April 1, 1980, or who was visiting and had no other home (see Figure 5.5). An edit was performed to check that the number of names listed in this household roster agreed with

images

FIGURE 5.5 Coverage questions in the 1980 census.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

the number appearing on the inside of the questionnaire; field follow-up took place if there were more names on the roster than inside. Finally, the 1980 questionnaire included 3 questions (H-1, H-2, H-3) that probed for persons whom the respondent either failed to list in Question 1 or improperly included (see Figure 5.5).

As discussed above, evaluation indicated that the H-4 edit in 1980 was less successful in adding housing units and persons than the comparable edit in 1970. Neither effort added more than a fraction of 1 percent to the population count. Review of questionnaires in 1980 that failed the H-4 edit indicated that the census office staff had a difficult time in conducting the edit and also that some respondents may not have correctly interpreted the question (Thompson, 1984:15).

The Census Bureau was unable to evaluate the effectiveness of the household roster (Question 1) edit, because of the absence of appropriate records, but looking at Figure 5.5 suggests that respondents may well have had problems with the instructions indicating which persons to list in Question 1 and which to omit. Similarly, the instructions do not seem at all clear for households that on Census Day were at a vacation residence but had a usual residence elsewhere.

The panel believes it is important that the questions and instructions regarding composition of the household be clearly communicated to respondents and that responses to such questions be given special attention by the field offices. This extra care is needed to minimize the possibilities for incorrect enumeration, whether it be undercount, overcount, or misallocation of persons and/or housing units among geographic areas.

Americans have always been highly mobile—one-sixth of the population changes residence every year, and some of those persons are in the process of moving at the time of the census (Hansen, 1984:Table A). Movers complicate both completion of an accurate count and evaluation of the count. Households with second (vacation) homes also complicate accurate enumeration. The 1970 census found that about 5 percent of households had a second home (Bureau of the Census, 1982c:751), and the percentage is growing. Finally, recent trends in living arrangements, retirement, and the workplace have resulted in populations with two or more “usual” residences that present special problems for accurate census-taking. Some examples include:

  • Retired persons who have two “usual” homes, one for the winter months in a warm climate and the other for the summer months in a cool climate;
  • Children of divorced families in which the parents have joint custody and the children spend a substantial part of the year, month, or week with each parent; and
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
  • Two-career couples with jobs and residences in two different locations.

It is debatable how each of these kinds of persons should be counted. Leaving aside the matter of assignment to a specific household and geographic area, populations such as these appear more than usually at risk of undercount as well as overcount.

The Census Bureau is proposing as part of its 1986 pretest program to consider alternative coverage and household roster questions and to test adding questions about multiple residences that could help minimize miscounting. The Census Bureau also plans to test improvements in the program that attempts to assign households found at their second (vacation) home to their regular residence (the Whole Household Usual Home Elsewhere program).

A range of questionnaire design techniques, including focus groups, would appear useful to employ for these questions to determine wording and formats that are clear to the respondent and also easy for census office staff to process. Research on trends in mobility, second homes, and multiple residences should also assist in decennial census planning. Identifying geographic areas particularly affected by these phenomena might suggest special efforts targeted to particular populations in these areas. It is particularly important in this regard to assess future trends. If, as appears probable, a growing part of the population is likely to have two or more usual places of residence, to own a second hone, or to be moving between an old and a new residence during the census enumeration, then planning for a complete and accurate count should give high priority to dealing with these groups.

Recommendation 5.7. We recommend that the Census Bureau give high priority in its planning for 1990 to research and testing of questions and enumeration procedures that address problems of accurately counting persons in the process of moving, households with second (vacation) homes, and persons with more than one usual place of residence.

A Specific Suggestion for a Coverage Improvement Question

In the 1977 pretest in Oakland, California, the Census Bureau tested the concept of “network” or “multiplicity” response rules for coverage evaluation (Sirken et al., 1978). Such rules include asking parents to provide names and addresses of children and vice versa. Published results suggested that the address information furnished was not of sufficient quality to warrant further investigation of this method as part of a coverage evaluation program that included matching samples of persons to census records.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

However, the panel believes that the concept of generating lists of individuals in an area from the census operation itself to use as a procedure to improve coverage is worth investigating, at least for hard-to-enumerate areas, and comparing with other procedures in terms of costs and effectiveness. The procedure would be to ask respondents in the census for lists of specific types of relatives not living in the household. Information needed for nonresident relatives to facilitate locating them and determining if they had been included in the census would include address and also basic demographic characteristics, such as age and sex. The panel believes that relatives may be at least as good a source of up-to-date address information to use for a matching operation for coverage improvement as other lists that are commonly suggested, such as driver’s licenses or welfare records.

The Oakland results suggested that address information supplied by parents was somewhat more accurate than information supplied by most other categories of relatives. Moreover, parents would probably be the most reliable source of information on a critical match item: birth date. Hence, asking parents to provide basic demographic information and addresses for children not living in the household could improve coverage, particularly of hard-to-count groups such as young adult black and Hispanic males and young black and Hispanic children in central cities.

Recommendation 5.8. We recommend, as one procedure to consider for improving coverage of hard-to-count groups, that the Census Bureau pretest a question asking parents for names and addresses of children who are not part of the household. This question should be included in the 1986 pretests.

Specifically, we propose that a question similar to the following be added to the census form:

Does anyone living in this household have a son or daughter living somewhere else?   Yes     No

If yes, please list sons and daughters below.

Name________________________________________________________

(Last                  First                Middle)

Sex _____           Age _____            Birth Date ________________

(Month-Day-Year)

Address _____________________________________________________

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

The object is to improve coverage in hard-to-count areas, particularly of young children, and hence it would not be cost-effective or even feasible to follow up all children reported as not living in the household. Instead, the goal would be to examine census returns from areas identified as hard to enumerate and to follow up those children reported by their parents as living in the same area. The question suggested above is phrased to ask parents for the addresses of all children not living in the household, so that there is no opportunity for misinterpretation of which children should be listed, but the follow-up could be restricted to children in target age-race-sex groups.

The answers to this question would provide a list of individuals that can be matched against the census. Presumably the list could be constructed and follow-ups (perhaps on a sample basis) of nonmatches done during the census operation. Operational questions for a test include the accuracy of birth date and address obtained from parents, the method of identifying addresses that are from hard-to-enumerate areas and should be followed up, the method of locating addresses, the use of different procedures in urban and rural areas, and the method of sharing information in cities with multiple offices. The effects on response rates of asking this question also need to be examined.

The panel recognizes that there are problems in adding a question to the census short form, particularly a question that requires a lot of space and that may be viewed as invasive of privacy. Indeed, given its intended follow-up on a sample basis, the question should perhaps be included on the long form only.

Research and testing of the suggested multiplicity coverage question and of other such questions, including ones on multiple residences, should be closely coordinated. The wording and format of all such questions must be carefully considered to ensure that the entire package is communicated clearly. If there is concern over the increased respondent burden, particularly for recipients of the long form, the Census Bureau should consider deleting other questions. Chapter 6 suggests some long-form housing questions that could be deleted and collected instead from other sources. The Census Bureau should also consider the possibility of different questionnaire formats in different areas. For example, it might be possible to include the multiplicity coverage question only on questionnaires for enumeration districts with certain expected characteristics, such as a high poverty rate and, conversely, to include questions on multiple residences only on questionnaires administered in other kinds of areas.

Because the multiplicity question appears promising for coverage improvement and also relates to other coverage questions that the Census Bureau is proposing to test in 1986, it is important that the multiplicity question be tested in 1986 as well. The panel, in fact, recommended in its interim report (National Research Council, 1984) that the multiplicity

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

question be tested in the first 1990 census pretest, that is, in 1985. The Census Bureau has proposed delaying a test until 1987. For the reasons outlined above, the panel believes that high priority should be given to testing coverage questions in 1986 and that these tests should include a multiplicity question.

ISSUES IN COVERAGE IMPROVEMENT: SPECIÀL ENUMERATION PROCEDURES

The panel does not propose to comment in detail on each of the various coverage improvement procedures used in 1970 and 1980 and proposed for testing in 1986. We believe we can be most useful to the Census Bureau by recommending general strategies for deciding the priorities to assign in its 1990 census research and testing program. The Census Bureau staff exercised some selection in the process of drawing up the proposed package of 1986 pretests of coverage improvement programs, but the package still seems much too ambitious for the likely available staff and budget resources and for the time available to design, execute, and evaluate the results from this and subsequent pretests prior to 1990. Pretest results that cannot be assimilated in time to affect the next pretest or the dress rehearsals represent largely wasted effort.

The panel believes that the goal of a complete enumeration is very important and, as discussed more fully in Chapter 7, that adjustment procedures should not be viewed as an alternative to obtaining as complete a count as possible through cost-effective means. The panel also believes that special coverage improvement programs can make important contributions to improving the count. However, the panel does not subscribe to the view that every coverage improvement idea that is suggested should be pursued or that programs used in past censuses should automatically be included in the plans for the next census. The panel believes that evaluation results for coverage improvement procedures used in prior censuses should be carefully reviewed and that further research and testing should be conducted only for programs that meet certain criteria of cost-effectiveness, particularly in reducing differential undercounts. Similarly, proposed ideas for new kinds of procedures should be assessed against several criteria to determine the extent to which they appear promising and feasible.

With regard to assigning priorities for research and testing of coverage improvement programs with which the Census Bureau has prior experience, the panel recommends the following strategy:

Recommendation 5.9. We recommend that the Census Bureau review coverage improvement programs used in past censuses and proceed with research and testing directed toward use in 1990 of those pro-

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

grams that: (1) exhibited a high yield in terms of numbers of missed persons correctly added to the count and/or contributed significantly to reducing differential undercoverage, (2) exhibited low-to-moderate costs per person correctly added, and (3) did not add many persons incorrectly. Programs that do not satisfy these criteria should be dropped from consideration unless: (1) the program exhibited low total dollar costs and had demonstrable public relations or goodwill value in previous censuses or (2) there is some particular reason to believe a revised program will yield greatly improved results.

The above recommendation does not quantify terms such as “high yield” or “low cost.” Obviously, the previous performance of specific coverage improvement programs should be carefully and appropriately measured and a decision whether to include a program in the 1990 pretest plans carefully arrived at. For example, the proportion of housing unit or person additions to the count should be measured using the appropriate denominator. In the case of a program administered in specified areas, the denominator should be the total count only for those areas. Admittedly, the available evaluation data are subject to margins of error that may be wide for some programs. Nonetheless, it seems possible to assign priorities through a hard look at the information in hand.

Based on the data in Tables 5.2 and 5.3, it appears that the various address checking programs carried out in advance of Census Day easily qualify for further consideration both in terms of proportion of additions to the count and in terms of cost per addition. Some other programs, such as the Were You Counted program, yielded very little but were quite inexpensive and appear to have goodwill value in conducting the census. Still other programs are more problematical. Although the coverage questions and roster checks were low yield and costly to administer in 1980, it appears essential, as discussed above, that research be carried out to develop optimal formats and processing procedures for these questions to minimize problems of undercount, overcount, and misallocation among geographic areas. The Vacant/Delete Check met a minimum standard of 0.5 percent additions to the count, but it was costly in 1980. A high priority for further research on this program would involve investigation of ways to reduce costs, for example, by returning to the use of sampling, as in 1970 (see further discussion in Chapter 6, where the panel recommends that the Census Bureau conduct research on the use of sampling for the Vacant/Delete Check and possibly other coverage improvement programs). The Nonhousehold Sources Program appears to fail on tests both of additions to the count (even when the denominator is the number of addresses that were selected for matching) and cost. It is possible that the use of automation could improve the cost-effectiveness of this program as might selection

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

of other kinds of lists, but, given that choices must be made, it appears that the program should be given low priority.

For ideas with which the Census Bureau has little or no experience, the panel suggests that questions such as the following be asked:

  1. To what extent is the proposed coverage improvement program directed toward known problem areas? For example, the Census Bureau is considering tests of a number of means of handling multiunit structures for which mail delivery is often problematical, including an update list/leave procedure that was tried experimentally in a few district offices in 1980. As another example, the proposed multiplicity coverage question is directed toward hard-to-count groups, specifically young minority children.
  2. To what extent does any available evidence suggest that the proposed procedure night prove effective? For example, although the update list/leave procedure in 1980 resulted in significantly higher initial mail return rates in the experimental offices compared with the controls (81 versus 71 percent—see Mikkelson and McKelvey, 1983), no significant differences in coverage have been determined (see Bailey and Ferrari, 1984; Mikkelson, 1984).
  3. Do rough paper-and-pencil estimates of cost and yield suggest that the proposed program is likely to be cost-effective?
  4. Can the program be implemented in targeted areas as a means of improving cost-effectiveness, or can cost savings be effected through judicious use of sampling?

For coverage improvement procedures that the Census Bureau decides to retain in its research and testing program, the panel believes it is important to further categorize them into programs that need early field testing versus those that can be researched with other, less expensive, and less staff-intensive methods. For example, it may not be necessary to include the Casual Count program in any early full-scale pretest. Other procedures, such as trying out various address checks in prelist and conventional areas, probably need early testing, particularly to work on integrating these operations with the various automation efforts that are being given high testing priority. A strategy that does not attempt costly field tests of every program should help make the Census Bureau’s budget and staff resources stretch farther and help reduce the problem of a proliferation of tests producing results that cannot be assimilated in time for 1990. Finally, one cost-effective means of gaining useful information for improving coverage programs would be to conduct focus groups that include members of hard-to-count populations. Such groups could consider reasons for failure to be counted and consider as well the likely impact of particular programs.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

Recommendation 5.10. We recommend that the Census Bureau conduct full-scale pretests in 1986 only of those coverage improvement programs that require such testing. Furthermore, we recommend that the Census Bureau use focus groups that include members of hard-to-count populations as one means to explore coverage improvement techniques and to narrow the range of options to be field-tested.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

APPENDIX 5.1
GROSS OMISSIONS AND GROSS OVERENUMERATIONS IN THE CENSUS

This appendix presents results of coverage evaluation programs that identify groups in the population that appear to have been less well counted—through omission and/or erroneous enumeration—in the 1980 and previous decennial censuses. Most of the results represent estimates of gross omissions from the census. Results of the demographic analysis method of coverage evaluation are discussed in the text but not in this appendix, because demographic analysis provides estimates of net undercoverage but not of the gross omission or gross overenumeration components. Results of studies of the completeness of census coverage of housing units are also briefly reviewed.

GROSS OMISSIONS OF PEOPLE

Findings from 1980 Census Coverage Evaluation Programs

The Post-Enumeration Program

The PEP developed estimates of gross omissions in the 1980 census through matching interview records from the April and August Current Population Surveys (the P sample) to census records in the same small geographic areas (enumeration districts). The PEP design resulted in gross omission rates that represent overestimates because, among other reasons, the rates include persons who were enumerated in the census but at a location so far removed from their address in the CPS that it was outside the area of search for a match in the census records. The PEP also encountered many problems in implementation. To date, there are 12 separate sets of estimates of undercount developed from PEP based on different treatment of problems such as nonresponse to the CPS (see Chapter 4 for a description and evaluation of PEP). Hence, the discussion that follows seeks to determine the order of magnitude of differences in gross omission rates among population subgroups, but it cannot provide precise estimates. The tables shown in this section express findings from the PEP in terms of ratios of the gross omission rate for a population group to the average rate experienced for the total population. Population groups are placed into one of five categories of relative gross omission rates:

  1. Very high: greater than or equal to 3 times the average rate;
  2. High: greater than or equal to 2 times and less than 3 times the average rate;
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
  1. Moderately high: greater than or equal to 1.25 times and less than 2 times the average rate;
  2. Average: greater than 0.75 times and less than l.25 times the average rate;
  3. Below average: less than or equal to 0.75 times the average rate.

Several tables show relative gross omission rates from the PEP for population groups categorized by household relationship, race and Hispanic origin (ethnicity), type of place, and by ethnicity and type of place crossed with rates of nonreturn of the mail questionnaires in the district offices. The data represent the results of preliminary exploratory analysis conducted by the Census Bureau of the PEP 3-8 series. This series was based on matching the April CPS to the census and estimated an average gross omission rate for the total population in 1980 of 5.4 percent.

The 3-8 series happened to be the first series to be put into a computerized form suitable for this kind of analysis at the Census Bureau. Further work is necessary to confirm that the 3-8 findings are reliable and representative of the results shown by other series of estimates. More limited tabulations of several other PEP series of estimates were recently prepared by the Census Bureau, and they generally confirm the picture shown by the 3-8 series regarding the population groups that were relatively harder to count in 1980 (see discussion at the end of this section).

Findings from the PEP 3-8 Series

The preliminary findings from the 3-8 series with regard to which population groups proved relatively harder to count are not surprising, but the dispersion among the five categories of relative gross omission rates is not always as great as one might expect. On the dimension of household relationship, members of the nuclear family—head, spouse, and son or daughter—were relatively easy to find compared with other household members (see Table 5.4). Persons not related to the household head and relatives other than parents had high rates of gross omissions compared with the average for the total population. On the dimension of ethnicity (see Table 5.5), blacks were among the hardest-to-find groups, with a high relative gross omission rate. Persons classified as Hispanic had a moderately high rate overall but showed dispersion when further categorized by place of origin. Puerto Ricans and “other” Hispanics had high relative gross omission rates; the rate for Mexicans was moderately high; while the rate for Cubans was within the average category. Finally, American Indians and Asian Americans had moderately high gross omission rates, while the rates for non-Hispanic whites and other races were average.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

TABLE 5.4 Relative Gross Omission Rate Category by Household Relationship for a Sample of Persons, Post-Enumeration Program-Census Match (1980, PEP Series 3-8)

Relative Gross Omission Ratea Household Relationship
Very high
High Nonrelative
  Other relative
  Brother or sister
Moderately high Mother or father
Average Head
  Son or daughter
Below average Spouse

NOTE: Average gross omission rate for the 1980 PEP was 5.4%.

aCategories of relative gross omission rates are as follows:

(1) Very high: greater than or equal to 3 times the average rate.

(2) High: greater than or equal to 2 times and less than 3 times the average rate.

(3) Moderately high: greater than or equal to 1.25 and less than 2 times the average rate.

(4) Average: greater than 0.75 and less than 1.25 times the average rate.

(5) Below average: less than or equal to 0.75 times the average rate.

SOURCE: Hogan (1983b:3).

Distributions are not given for other demographic variables such as age and sex. Males and females both had average rates of gross omissions, as did most age groups. (The PEP findings of very little difference in coverage rates between men and women contrast with the results of demographic analysis, which showed worse coverage for men, particularly among blacks.) From unpublished PEP 3-8 series tabulations, young adults ages 15-24 had a moderately high relative rate of gross omissions, while persons age 45 and older had below-average rates. The PEP does not provide separate estimates of coverage of undocumented aliens, although the PEP sample probably included some representation of this group.

Among the variables displayed, the dimension of type of place (see Table 5.6) shows the least dispersion. Central cities of large standard metropolitan statistical areas (SMSAs with 3 million or more population) had a moderately high relative gross omission rate, while all other place types had average rates. Note that areas enumerated using conventional techniques rather than a mailout-mailback approach had a below-average relative gross omission rate. Distributions are not given for urban versus rural areas or region of the country (Northeast, Midwest, Southwest), as

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

TABLE 5.5 Relative Gross Omission Rate Category by Ethnicity and by Mail Nonreturn Rate and Ethnicity for a Sample of Persons, Post-Enumeration Program-Census Match (1980, PEP Series 3-8)

Relative Gross Omission Rate Categorya Ethnicity (detailed categorization) Mail Nonreturn Rate (mail areas only)b and Ethnicityc
Very high 30% or higher: Black
Hispanic
High Black (non-Hispanic) 30% or higher: Total
Hispanic: Puerto Rican 15-29%: Black
Other
Moderately high Hispanic: Total 30% or higher: White
Mexican 15-29%: Hispanic
American Indian Less than 15%: Black
Asian
Average Hispanic: Cuban 15-29%: Total
White (non-Hispanic) White
Other race (non-Hispanic)
Below average Less than 15%: Hispanic
Less than 15%: Total
White

NOTE: The average gross omission rate for the 1980 PEP was 5.4%.

aCategories of relative gross omission rates are as follows:

(1) Very high: greater than or equal to 3 times the average rate.

(2) High: greater than or equal to 2 times and less than 3 times the average rate.

(3) Moderately high: greater than or equal to 1.25 and less than 2 times the average rate.

(4) Average: greater than 0.75 and less than 1.25 times the average rate.

(5) Below average: less than or equal to 0.75 times the average rate.

bThe mail nonreturn rate is the percentage of occupied households that did not mail their questionnaires to census offices.

cThe three ethnicity categories shown are exhaustive: black non-Hispanic, Hispanic of all races, and white and other race non-Hispanic.

SOURCE: Hogan (1983b:2, I983a:attached graphs).

all of these categories had average gross omission rates. This is not to say, however, that more in-depth analysis would not reveal interactions between region or urban versus rural and other variables.

When either ethnicity or type of place is crossed with the mail nonreturn rate for the district office (i.e., 100 percent minus the percentage rate at which questionnaires were mailed back from households), the dispersion in relative rates of gross omissions increases dramatically. While blacks on average had a high relative gross omission rate, those blacks in district offices with mail nonreturn rates of 30 percent or more had a very

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

TABLE 5.6 Relative Gross Omission Rate Category by Type of Place and by Mail Nonreturn Rate and Type of Place for a Sample of Persons, Post-Enumeration Program-Census Match (1980, PEP Series 3-8)

Relative Gross Omission Rate Category Type of Place Mail Nonreturn Rate (mail areas only)d and Type of Place
Very high 35% or higher: Central city, large SMSA
Central city, small SMSA
High 30% or higher: Other, SMSA
Outside SMSA
25-34%: Central city, large SMSA
Moderately high Central city, large SMSAb 25-34%: Central city, small SMSA
10-24%: Central city, large SMSA
Average Central city, small SMSAc 15-29%: Other, SMSA
Other, SMSA Outside SMSA
Outside SMSA 5-24%: Central city, small SMSA
All mailout-mailback areas
Below average Conventional areas 0-14%: Other, SMSA
Outside SMSA
0-9%: Central city, large SMSA
0-4%: Central city, small SMSA

NOTE: The average gross omission rate for the 1980 PEP was 5.4%.

aCategories of relative gross omission rates are as follows:

(1) Very high: greater than or equal to 3 times the average rate.

(2) High: greater than or equal to 2 times and less than 3 times the average rate.

(3) Moderately high: greater than or equal to 1.25 and less than 2 times the average rate.

(4) Average: greater than 0.75 and less than 1.25 times the average rate.

(5) Below average: less than or equal to 0.75 times the average rate.

bLarge standard metropolitan statistical area (SMSA) is defined as an area with over 3 million population.

cSmall SMSA is defined as an area with 3 million or less population.

dThe mail nonreturn rate is the percentage of occupied households that did not return their questionnaires to census offices.

SOURCE: Hogan (1983b:4, 1983c; and unpublished worksheets).

high relative gross omission rate (3 or more times the average rate), and conversely, those blacks in district offices with mail nonreturn rates of less than 15 percent had only a moderately high relative gross omission rate. A similar spread is evident for Hispanics and for non-Hispanic whites (see Table 5.5). The dispersion for type-of-place categories is even more extreme when the factor of mail nonreturn rates is introduced. Central cities of large SMSÀs, which on average exhibited a moderately high relative rate of omissions, had a very high rate in those areas in which the district office mail

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

nonreturn rate was 35 percent or greater, and conversely, a below-average rate in those areas of the central city in which the mail nonreturn rate was under 10 percent (see Table 5.6).

Findings from Other PEP Series

Unpublished tabulations of gross omission rates from two other PEP series, the 5-8 series based on matching August CPS records to the census, and the 14-20 series based like the 3-8 series on matching April CPS records to the census but with a different treatment of incomplete cases, generally support the findings reported above from the 3-8 series. The 5-8 series estimated an overall rate of gross omissions of 5.25 percent and the 14-20 series a rate of 3.45 percent compared with the 5.4 percent rate estimated by the 3-8 series. In relative terms, all three series found that blacks had a high relative gross omission rate, Hispanics a moderately high rate, men and women average rates, young adults ages 15-24 a moderately high rate, persons age 45 and older below-average rates, and other age groups average rates. In each case, the determination of the gross omission rate category for a population group was made relative to the average rate for the particular series. Data from the match of August CPS records to the census also generally support the 3-8 series of findings regarding the relationship of high mail nonreturn rates to high relative rates of omissions (Hogan, 1983a).

The IRS-Census Match

A methodological study conducted after the 1980 census, the IRS-Census Match, although not designed as a coverage evaluation study, provides some evidence on differential rates of gross omissions from the census. The purpose of the IRS-Census Match was to examine tracing and matching problems with pre-enumeration surveys and reverse record checks. The study attempted to match a sample of about 11,000 filers of 1979 tax returns to 1980 census records. Black and Hispanic filers were oversampled (Childers and Hogan, 1984a).

The average gross nonmatch rate for the total sample was 12.6 percent. There are many reasons for the high rate, including the facts that addresses supplied by taxpayers on IRS forms were not always the same as the residence address and that the matching study was carried out several years after the census and not intended to produce coverage estimates. Nonetheless, some insights can perhaps be gained when the IRS sample is categorized along several dimensions and gross omission rates for subgroups are compared with the average for the entire sample.

In the IRS-Census Match study, blacks and Hispanics exhibited moderately high relative gross omission rates (category 3), while the rate for

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

non-Hispanic whites fell into the average category (category 4). Adding the dimension of sex increases the dispersion, with black males having a high relative gross omission rate (category 2) and white females a below-average rate (category 5). These findings are consistent with those from the PEP and demographic analysis.

The IRS-Census Match provides data on gross omission rates for two important dimensions: marital status (proxied by type of return—single or joint) and income (adjusted gross income reported to the IRS). These two dimensions help identify hard-to-count groups (see Table 5.7). Persons filing single returns had moderately high gross omission rates, while persons filing joint returns had below-average rates. Cross-tabulating type of return with ethnicity gives the result that, while black single return filers had a high relative gross omission rate, black joint return filers fell into the average category. Similarly, white single return filers had a moderately high relative gross omission rate, while white joint return filers were below average. Type of return did not discriminate to any important extent among the Hispanic group.

Adding the dimension of income refines the picture of hard-to-count groups. Black single return filers had high relative gross omission rates regardless of income level; however, income discriminated among black joint return filers, with those reporting less than $8,000 income showing a high relative rate of gross omissions but those reporting $15,000 or more income a below-average rate. Among whites, those filing single returns with reported income under $15,000 and those filing joint returns with income under $8,000 fell into the category of moderately high relative gross omission rates, while the remainder fell into the below-average category. For Hispanics filing joint returns, the cutting point between high relative gross omission rates and average rates was an income level of $15,000; however, income level did not discriminate among Hispanics filing single returns to any great degree.

Findings from Previous Census Coverage Evaluation Programs

Coverage evaluation programs for previous censuses provide additional information about groups in the population that are more apt to be undercounted compared with other groups. This appendix reviews the findings of post-enumeration surveys and resident observation but does not discuss demographic analysis. The chapter text reviews the estimates of net undercoverage provided by demographic analysis for the 1950, 1960, and 1970 censuses.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

TABLE 5.7 Relative Gross Omission Rate Category by Ethnicity and Type of Return and by Income, Type of Return, and Ethnicity for a Sample of Income Tax Filers Ages 18 to 64, IRS-Census Match (1980)

Relative Gross Omission Rate Categorya Ethnicity and Type of Returna Income in 1979, Type of Return, and Ethnicitya
Very high
High Black: Single return Under $8,000, single return: Black
Hispanic: Single return Hispanic
Under $8,000, joint return: Black
Hispanic
$8,000-14,999, single return: Black
$8,000-14,999, joint return: Hispanic
$15,000 or more, single return : Black
Hispanic
Moderately high Black: Total filers Under $8,000, single return: White
Hispanic: Joint return Under $8,000, joint return: White
Total filers $8,000-14,999, single return: Hispanic
White: Single return White
$8,000-14,999, joint return: Black
Average Black: Joint return $15,000 or more, joint return Hispanic
White: Total filers
Below average White: Joint return $8,000-14,999, joint return: White
$15,000 or more, single return: White
$15,000 or more, joint return: Black
White

NOTE: The average gross omission rate for the 1980 IRS-Census Match was 12.6%.

aCategories of relative gross omission rates are as follows:

(1) Very high: greater than or equal to 3 times the average rate.

(2) High: greater than or equal to 2 times and less than 3 times the average rate.

(3) Moderately high: greater than or equal to 1.25 and less than 2 times the average rate.

(4) Average: greater than 0.75 and less than 1.25 times the average rate.

(5) Below average: less than or equal to 0.75 times the average rate.

bThe three ethnicity categories shown are exhaustive: black non-Hispanic, Hispanic of all races, and white and other race non-Hispanic.

SOURCE: Childers and Hogan (1984a:Tables 1 and 2).

Post-Enumeration Surveys

Post-enumeration surveys conducted in previous censuses provide data on relative rates of undercoverage for various population groups. Tables 5.8 through 5.11 show relative gross omission rates for the population categorized along several dimensions from the results of the match of the April 1960 Current Population Survey to 1960 census records and of the match of

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

TABLE 5.8 Relative Gross Omission Rate Category by Household Relationship for a Sample of Persons, CPS-Census Match (1960)

Relative Gross Omission Rate Categorya Household Relationship
Very high Brother- or sister-in-law
  Group quarters resident
High Son- or daughter-in-law
  Other relative
  Nonrelative
  Grandson or granddaughter
Moderately high Relationship not reported
  Mother or father
  Mother- or father-in-law
  Brother or sister
Average Son or daughter
  Head
Below average Wife

NOTE: The average gross omission rate for the 1960 CPS-Census Match was 6.5%.

aCategories of relative gross omission rates are as follows:

(1) Very high: greater than or equal to 3 times the average rate.

(2) High: greater than or equal to 2 times and less than 3 times the average rate.

(3) Moderately high: greater than or equal to 1.25 and less than 2 times the average rate.

(4) Average: greater than 0.75 and less than 1.25 times the average rate.

(5) Below average: less than or equal to 0.75 times the average rate.

SOURCE: Bureau of the Census (1964a:Table 19).

the Post-Enumeration Survey conducted in summer 1950 to 1950 census records. (The dimensions shown were chosen to try to present estimates based on large enough sample sizes for reliability.) The relative gross omission rate experienced for the entire population was 6.5 percent in the 1960 CPS-Census Match and 2.2 percent in the 1950 Post-Enumeration Survey. (The lower rate for 1950 attests to the deficiencies of a “pure” post-enumeration survey in which enumerators are sent out to recanvass an area.)

On the dimension of household relationship, both 1960 and 1950 data support the findings from the 1980 PEP, namely, that persons not belonging to the nuclear family were harder to find than household heads, spouses, and their children. Nonrelatives including residents of group quarters were particularly difficult to count (see Tables 5.8 and 5.10).

Looking at relative gross omission rates by extent of education, data from 1950 indicate that persons with education not reported exhibited a very high relative gross omission rate, while persons with 6 or fewer years of schooling exhibited a moderately high rate. The gross omission rate for

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

TABLE 5.9 Relative Gross Omission Rate Category by Sex and Employment Status and by Race and Income of Males for a Sample of Persons, CPS-Census Match (1960)

Relative Gross Omission Rate Categorya Sex and Employment Status (persons 14 years and over)b Race and Income in 1959 (males 14 years and over with income)c
Very high
High Female: Unemployed Nonwhite: Income under $7,500
Male: Agricultural (Ag.) wage worker Total male 14 years and over with income
Moderately high Male: Not in labor force Nonwhite: Income $7,500 or more
Unemployed White: Income under $5,000
Average Female: Total White: Income $5,000-9,999
Nonag. wage worker Total male 14 years and over with income
Nonag. self-employed
Not in labor force
Male: Total
Nonag. wage worker
Below average Male: Ag. self-employed White: Income $10,000 or more
Nonag. self-employed

NOTE: The average gross omission rate for the 1960 CPS-Census Match was 6.5%.

aCategories of relative gross omission rates are as follows:

(1) Very high: greater than or equal to 3 times the average rate.

(2) High: greater than or equal to 2 times and less than 3 times the average rate.

(3) Moderately high: greater than or equal to 1.25 and less than 2 times the average rate.

(4) Average: greater than 0.75 and less than 1.25 times the average rate.

(5) Below average: less than or equal to 0.75 times the average rate.

bGroups not shown because of small sample size are male agricultural and nonagricultural unpaid worker; female agricultural wage worker, self-employed, and unpaid worker; and female nonagricultural unpaid worker.

cAll income amounts are in 1979 dollars (1959 figures times 2.5).

SOURCE: Bureau of the Census (1964a:Tables 28 and 43).

persons with more than 6 years of schooling fell into the average category (see Table 5.10). Comparable data are not available from 1960.

Both the 1960 and 1950 census coverage evaluation programs furnish data on differential rates of undercount among the population classified by labor force status, occupation, and income levels (see Tables 5.9 and 5.11). In 1960, unemployed women had a high relative gross omission rate, and men who were unemployed or not in the labor force had moderately high rates. Looking at employed persons, male agricultural paid laborers had a high relative gross omission rate, while the rate was below average for self-employed men both in farming and other lines of business.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

TABLE 5.10 Relative Gross Omission Rate Category by Household Relationship and by Education Level for a Sample of Persons, Post-Enumeration Survey-Census Match (1950)

Relative Gross Omission Rate Categorya Household Relationship Education Level (persons 25 years and over)
Very high Nonrelative Education not reported
High
Moderately high Other relative 6 or fewer years of school completed
Average Head
Wife
More than 6 years of school completed
  Son or daughter  
Below average

NOTE: The average gross omission rate for the 1950 Post-Enumeration Survey-Census Match was 2.2%.

aCategories of relative gross omission rates are as follows:

(1) Very high: greater than or equal to 3 times the average rate.

(2) High: greater than or equal to 2 times and less than 3 times the average rate.

(3) Moderately high: greater than or equal to 1.25 and less than 2 times the average rate.

(4) Average: greater than 0.75 and less than 1.25 times the average rate.

(5) Below average: less than or equal to 0.75 times the average rate.

SOURCE: Bureau of the Census (1960:Tables E and 4).

On the dimension of income, looking only at males, there is a clearer picture for whites compared with nonwhites in 1960. In the case of white males, those with low income had a moderately high relative rate of gross omissions, while those with high income had a below-average rate. In contrast, income did not discriminate to any important extent among nonwhite males.

The 1950 data, which do not include race breakdowns, support the general patterns evident in 1960 on occupation and income. Persons with very high relative rates of gross omissions include farm laborers of both sexes and female unpaid farm workers. In contrast, male farmers and also female sales workers had below-average rates (female farmers were excluded because of very small sample size). Male nonagricultural laborers and female private household workers fell into the next category of high relative gross omission rates. On the dimension of income, low income is associated with a moderately high relative rate of gross omissions, while high income is associated with a below-average rate. Note that persons

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

TABLE 5.11 Relative Gross Omission Rate Category by Sex and Occupation and by Income of Males for a Sample of Persons, Post-Enumeration Survey-Census Match (1950)

Relative Gross Omission Rate Categorya Sex and Occupation (persons 14 years and over)b Income in 1949 (males 14 years and over)c
Very high Female: Farm laborer and unpaid worker
Male: Farm laborer
High Female: Private household worker Income not reported
Male: Nonfarm laborer
Moderately high Female: All other occupationsd Income under $3,000
Male: Farm unpaid worker
Not employed
Average Female: Not employed Income $3,000-10,499
Male: All other occupationse Total male 14 and over
Below average Female: Sales worker Income $10,500 and over
Male: Farmer and farm manager

NOTE: The average gross omission rate for the 1950 Post-Enumeration Survey-Census Match was 2.2%.

aCategories of relative gross omission rates are as follows:

(1) Very high: greater than or equal to 3 times the average rate.

(2) High: greater than or equal to 2 times and less than 3 times the average rate.

(3) Moderately high: greater than or equal to 1.25 and less than 2 times the average rate.

(4) Average: greater than 0.75 and less than 1.25 times the average rate.

(5) Below average: less than or equal to 0.75 times the average rate.

bExcludes male private household worker, female farmer, and female nonfarm laborer categories because of small sample size.

cAll income amounts are in 1979 dollars (1949 figures times 3).

dlncludes professional and technical, nonfarm manager, clerical, crafts, operative, and service worker categories.

eIncludes categories listed in note d plus sales worker category.

SOURCE: Bureau of the Census (1960:Tables 6A, 6B, and 9B).

with various characteristics, such as income, not reported in the 1950 study tended to have very high or high relative gross omission rates, indicating that these persons represented a generally hard-to-count group.

Finally, tables are not shown by type of area or region of the country, as these dimensions did not discriminate significantly in either 1960 or 1950 on relative rates of gross omissions. In 1960 the population of urban areas, rural nonfarm, and rural farm areas all exhibited average relative gross omission rates (Bureau of the Census, 1964a:Table 8). In 1950 persons living in urban areas had an average rate of gross omissions, persons

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

classified as rural nonfarm had a moderately high relative rate, and persons classified as rural farm had a below-average rate. By region of the country in 1950, the Northeast, North Central, and West regions had average rates, while only the South had a moderately high relative rate of gross omissions. Within the South, rural nonfarm areas were hardest to count, with a high relative rate of gross omissions (Bureau of the Census, 1960:Table F).

Comparable data were not published for 1970, but unpublished data from a match of the April Current Population Survey to the census records indicate the following patterns. (These findings should be viewed as suggestive only, however, because of high variances associated with the estimates.) First, employed whites had a below-average gross omission rate, but employed blacks had a moderately high relative rate. Similarly, higher-income whites had a below-average gross omission rate, but this was not true for higher-income blacks. Unemployed and low-income persons of both races had average rates of gross omissions. Finally, rates of gross omissions were higher in rural than in other types of areas (Siegel, 1975:8-9).

Resident Observer Studies

The techniques of resident observation employed in ethnographic studies were used on one occasion to study factors affecting the coverage of household surveys. Charles A. Valentine and Betty Lou Valentine, who were trained anthropologists, conducted a resident ethnographic study of a predominantly black inner-city community in 1968-1970, partially under the sponsorship of the Center for Research in Measurement Methods of the Census Bureau (Valentine and Valentine, 1971). Interviewers for the Current Population Survey, the Health Interview Survey, and the Quarterly Housing Survey conducted interviews in the area at the time when the Valentines had been in residence for approximately 1 year. The interviewers were unaware that the Valentines were studying the area. The Valentines independently identified the residents of a number of households in the area and ultimately compared their independently derived data on household composition for a total of 25 dwelling units with the corresponding data as reported by the survey interviewers. About three-fourths of the households were black and one-fourth Hispanic. The families lived in substandard housing, some lacking basic facilities. All families were judged to have very low incomes relative to the cost of living for the area. The Valentines described the community as “a typical polyethnic inner-city slum.”

For the 25 households, the surveys reported a total of 127 individuals, whereas the Valentines identified 153 individuals as being associated with the dwelling units. Therefore, the survey procedures produced a 17 percent undercount relative to the count obtained by the resident observers. The most striking result was that the ethnographic evidence suggested that

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

61 percent (17 of 28) of the males age 19 and older were not counted by the survey procedures. The Valentines described the missed men as regularly residing in the households. The men contributed to financial support, took part in domestic activities, and shared in childrearing. In nine of the households, the men were legally married and living with their spouses. The remaining common-law unions were relatively permanent and most were intact 2 years after the study. The Valentine estimates provided a much more realistic sex ratio than did the interview results.

The Valentines described a number of reasons that led them to believe that one cannot expect traditional interview or self-enumeration procedures to identify individuals of the type missed in the study area. The Valentines felt that the respondents understood the questions. They concluded that the men were not reported because the identification of resident males in the households could be detrimental to the economic welfare of the household and that the respondents behaved in a consistent manner in failing to report these men.

GROSS OVERENUMERATIONS OF PEOPLE

Findings from the 1980 PEP Program

The whole story regarding coverage problems in the census does not emerge solely by looking at gross omissions. It is necessary to examine gross overenumerations as well as omissions to obtain a complete picture. The PEP developed estimates of gross overenumerations in the 1980 census through rechecking a sample of census records (the E sample) to identify problems such as duplicate records, persons enumerated who were not alive on Census Day, and so on. As is true for the PEP estimate of gross omissions, the estimate of gross overenumerations is overstated. Also, the rates cannot be subtracted to give an estimate of the net undercount, because they have different denominators (see Cowan and Fay, 1984, for further explanation).

Data from the PEP 3-8 series (which is the only series for which tabulations for population groups are currently available) indicate an overall gross overenumeration rate of 3.6 percent in 1980 versus an overall gross omission rate of 5.4 percent. It is the case that population groups with relatively high gross omission rates also tended to have relatively high relative rates of gross overenumeratíons. However, the dispersion in gross overenumeration rates is less than the dispersion in gross omission rates.

Table 5.12, as an illustration, shows relative rates of gross overenuneration for ethnicity and household relationship in 1980. Blacks and Hispanics on average had moderately high relative rates of gross overenumerations, as did members of other races. American Indians, Asians, white non-Hispanics,

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

TABLE 5.12 Relative Gross Omission Rate Category by Ethnicity and by Household Relationship for a Sample of Persons, Post-Enumeration Program-Census Match (1980, PEP Series 3-8)

Relative Gross Overenumeration Rate Category Ethnicity (detailed categorization) Household Relationship
Very high
High
Moderately high Black (non-Hispanic)

Hispanic: Total
Cuban
Puerto
Rican

Nonrelative Other relative Brother or sister
  Other race (non-Hispanic)  
Average American Indian Asian

Hispanic: Mexican
Other

Head Spouse Son or daughter Mother or father
  White (non-Hispanic)  
Below average

NOTE: The average gross overenumeration rate for the 1980 PEP was 3.6%.

aCategories of relative gross overenumeration rates are as follows:

(1) Very high: greater than or equal to 3 times the average rate.

(2) High: greater than or equal to 2 times and less than 3 times the average rate.

(3) Moderately high: greater than or equal to 1.25 and less than 2 times the average rate.

(4) Average: greater than 0.75 and less than 1.25 times the average rate.

(5) Below average: less than or equal to 0.75 times the average rate.

SOURCE: Hogan (1983b:2-3).

and some categories of Hispanics, in contrast, were in the average category. This pattern is similar to the pattern evidenced in Table 5.5 for relative gross omission rates, but the dispersion is less for gross overenumerations. Similar findings emerge for categories of household relationship: household members outside the nuclear family had higher rates of both gross overenumerations and gross omissions compared with nuclear family members but were not as badly overenumerated relative to the average as they were underenumerated (see Table 5.4).

The rate of gross overenumerations also varied by type of enumeration procedure. Enumerations obtained in mail areas by follow-up because the questionnaire was not mailed back showed a high relative rate of gross overenumerations, while enumerations resulting from mail returns or ob-

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

tained in conventional areas had below-average rates (Cowan and Fay, 1984:6). It is not clear how much of the gross overenumeration was due to actual double counting and other kinds of erroneous enumerations in the census as opposed to problems with unresolved cases in the E sample. These problems are known to have been worse for groups exhibiting above-average gross overenumeration rates.

Findings from Earlier Censuses

The 1970 census relied on demographic analysis as the primary method for estimating net undercoverage; the CPS-Census Match provided estimates of gross omissions of persons but not of gross overenumerations. The 1960 Post-Enumeration Survey determined gross overenumerations of persons and housing units as well as gross omissions, but only net undercoverage rates were reported, while the 1960 CPS-Census Match determined only gross omissions (see Chapter 4 for further discussion).

Only the 1950 Post-Enumeration Survey, of pre-1980 census coverage evaluation programs, reported the components of net population coverage error. As was true for the 1980 PEP, the 1950 gross overenumeration estimate of 0.9 percent is overstated, as is the gross omission estimate of 2.2 percent, because the estimates included persons counted in the wrong geographic location (see Bureau of the Census, 1960). Findings with regard to gross overenumerations in 1950 are less clear-cut than the findings for 1980. In general, most population groups fell into the same categories of relative gross overenumeration and gross omission rates. Some groups appeared to have been less often overenumerated than underenumerated relative to the average rates (as was the general pattern in 1980), while a few groups appeared to have been more often overenumerated than underenumerated. The very different enumeration procedures used in the 1950 and 1980 censuses make it difficult to compare overenumeration experiences.

HOUSING COVERAGE STUDIES

Another source of information on relative rates of gross omissions in the census is provided by studies of completeness of coverage of housing units conducted in every census since 1950. Of course, rates of omission of housing units do not necessarily translate into comparable rates of missed persons; nevertheless, data on the characteristics of missed units add to the picture of hard-to-count elements in the population. Housing coverage evaluation studies also provide information on gross overenumerations, although the estimates from the 1980 census evaluation are not comparable with estimates from previous censuses because of the use of different methods.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

Looking at missed units, the 1950 census estimated an overall gross omission rate for occupied housing units of 3 percent; the estimated rate was 2.1 percent in 1960, 1.4 percent in 1970, and 1.5 percent in 1980.

Data from 1950 on gross omissions of occupied housing units show that rented units as a group exhibited a moderately high relative gross omission rate, while owned units had a below-average rate. Within the rental category, units for which rent was not reported and with very low monthly gross rent had very high relative gross omission rates; in contrast, moderate-to-expensive units fell into the average category. The smallest units (only one room) and also units with number of rooms not reported exhibited very high relative rates of gross omissions, while the largest units (five or more rooms) had a below-average rate. Finally, close to 30 percent of missed occupied units were in buildings that were otherwise enumerated, while 70 percent were in buildings that were missed entirely (Bureau of the Census, 1960:Tables 1, 11, and 15).

By region of the country or type of place (urban versus rural or metro versus nonmetro), there were no important differences in relative rates of gross omissions for occupied housing units in 1950 (Bureau of the Census, 1960:Table K). This was also true for 1960, 1970, and 1980. In 1960, the South had a moderately high relative rate of gross omissions, as did areas outside SMSAs. In 1970, the South’s rate of gross omissions fell into the average category, while nonmetropolitan areas remained in the moderately high category. However, without special coverage efforts in the South in 1970, specifically a post-enumeration post office check of the address list, the south would have had a moderately high relative rate of gross omissions (Bureau of the Census, 1973c:Tables F and G). In 1980, rural areas had a moderately high relative rate of gross omissions and the West region and areas enumerated with conventional methods had below-average rates (Bureau of the Census, 1985a:Table 2).

In 1960 about 40 percent and in 1970 about 30 percent of missed occupied units were in buildings that were otherwise enumerated, with the remainder in structures that were missed entirely. Table 5.13 shows the percentage distribution of missed units by the enumeration status of the structure for type of place in 1960 and 1970 (comparable data are not available for 1980). A clear shift is evident from 1960 to 1970 in the distributions by area type, presumably due to the introduction of new procedures for developing address lists for using mailout-mailback enumeration procedures in 1970. The shift is toward a higher percentage of missed units within otherwise enumerated structures in central city areas in 1970 compared with 1960 and lower percentages for other metropolitan areas and areas outside SMSAs. Data not shown indicate that in 1970 one-half of the units missed within structures were in structures classified as single-unit addresses on the mailing list, with another one-third in structures classified as having two to

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×

TABLE 5.13 Percentage of Gross Omissions by Enumeration Status of the Building and Type of Area, for Samples of Occupied Units, CPS-Census Match (1960 and 1970)

Type of Area 1960 Percentage of Occupied Units Missed in Buildings 1970 Percentage of Occupied Units Missed in Buildings
Enumerated Missed Enumerated Missed
Totala 38.1 61.9 28.6 71.4
Inside SMSAb 47.4 52.6 46.2 53.8
Central city 54.5 45.5 66.7 33.3
Other 33.3 66.7 25.0 75.0
Outside SMSAc 27.6 72.4 10.0 90.0

a1970 percentages are calculated from Table G, Part A, “After Processing,” based on the 1970 CPS-Census Match and assuming that processing changes reduced the miss rate in missed buildings but not in enumerated buildings.

b1970 percentages are calculated from Table F, Part B, based on the 1970 Coverage Evaluation in Mail Areas.

c1970 percentages are calculated from Table F, Part A, “After Processing,” based on the 1970 CPS-Census Match and assuming that processing changes reduced the miss rate in missed buildings but not in enumerated buildings.

SOURCE: Bureau of the Census (1973c:calculated from Tables F and G).

four units. Three-fourths of occupied units missed in 1970 were in structures built before 1939 (Bureau of the Census, 1973c:17). Data from a study of housing units in the E sample of the 1980 census Post-Enumeration Program that contained at least one duplicated person provide an estimate that 0.9 percent of units were duplicated. This estimate, although based on methodology that the Census Bureau believes to be superior to the methodology used in previous censuses, is an underestimate because it excludes various other kinds of housing unit overenumerations. Looking at relative duplication rates, the South had a moderately high rate of housing unit duplications in 1980, as did mail areas where the address list was developed by Census Bureau staff (prelist areas), rural areas, and nonmetropolitan areas. Conventionally enumerated areas and the Midwest had below-average housing unit duplication rates (Bureau of the Census, 1985a:Table 4). The 1980 study estimated (Bureau of the Census, 1985a:30) that in 88 percent of the duplicated units the entire household was duplicated, while in the remaining duplicated units only some household members were duplicated.

Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 153
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 154
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 155
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 156
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 157
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 158
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 159
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 160
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 161
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 162
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 163
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 164
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 165
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 166
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 167
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 168
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 169
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 170
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 171
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 172
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 173
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 174
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 175
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 176
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 177
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 178
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 179
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 180
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 181
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 182
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 183
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 184
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 185
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 186
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 187
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 188
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 189
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 190
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 191
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 192
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 193
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 194
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 195
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 196
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 197
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 198
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 199
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 200
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 201
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 202
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 203
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 204
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 205
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 206
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 207
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 208
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 209
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 210
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 211
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 212
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 213
Suggested Citation:"5 Taking the Census I: Improving the Count." National Academies of Sciences, Engineering, and Medicine. 2015. The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition. Washington, DC: The National Academies Press. doi: 10.17226/21728.
×
Page 214
Next: 6 Taking the Census II: The Uses of Sampling and Administrative Records »
The Bicentennial Census: New Directions for Methodology in 1990: 30th Anniversary Edition Get This Book
×
Buy Paperback | $67.00 Buy Ebook | $54.99
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

In 1982 the Census Bureau requested the Committee on National Statistics to establish a panel to suggest research and experiments, to recommend improved methods, and to guide the Census Bureau on technical problems in appraising contending methods with regard to the conduct of the decennial census. In response, the panel produced an interim report that focused on recommendations for improvements in census methodology that warranted early investigation and testing. This report updates and expands the ideas and conclusions about decennial census methodology.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    Switch between the Original Pages, where you can read the report as it appeared in print, and Text Pages for the web version, where you can highlight and search the text.

    « Back Next »
  6. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  7. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  8. ×

    View our suggested citation for this chapter.

    « Back Next »
  9. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!