Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 44
Modernizing the U.S. Census 3 Census Cost Increases and Their Causes The cost of census activities has increased sharply since 1960. In 1990 dollars, the 1960 census cost about $520 million. The 1990 census cost $2.6 billion, an increase of 400 percent after adjusting for inflation (U.S. General Accounting Office, 1992). Population and, more important, the number of housing units have increased over this period but, even after making allowances for these factors, cost escalation has been severe. The average cost of the census was less than $10 dollars per housing unit in 1960 (in 1990 constant dollars) and was still only $11 per housing unit for the 1970 census. It escalated to $20 per housing unit in 1980 and $25 in 1990, an increase of 150 percent in real terms over 30 years. GENERAL FACTORS A substantial decline in the population's response rate to the mailed census questionnaire, from 78 percent in 1970 to 65 percent in 1990, has been an important cause of the cost escalation, although far from the only one (U.S. General Accounting Office, 1992:36).1 In 1990, households that did not mail back their questionnaires were personally visited, as many as six times, in an attempt to collect census information. If mail response rates continue to decline, as they have done for the past several censuses, the national mail response rate may be less than 60 percent in 2000.2 The U.S. General Accounting Office (1992:41) estimates that the 2000 census would cost $4.8 billion, in 1992 dollars, if the 2000 census is conducted using the same methods as the 1990 census.3 If increased census costs had resulted in census improvements, particularly
OCR for page 45
Modernizing the U.S. Census in better coverage and a decreased differential undercount of minorities, it would be possible to argue the merits of costs versus coverage. However, as demonstrated in Chapter 2, both the overall percentage undercount and the differential undercount (the difference of the undercount between the black and nonblack populations) apparently worsened in the 1990 census (Robinson et al., 1993). It is of course possible that the net undercount and differential undercount might have been even worse in 1990 if less had been spent on the census. Some may claim that even greater funding was needed in 1990 to decrease the undercount. However, we cannot assess these hypothetical claims in the absence of evidence. The observable evidence is that substantially more was spent on the 1990 census, and the undercount did not improve. The fact that spending more money did not produce a more accurate census was, in large part, at the center of the criticisms levied at the 1990 decennial census.4 Costs have been affected by increased legal demands for accurate small-area data (see Appendix C for extended discussion of legal requirements for census data). The ''one-person, one-vote" rulings of the Supreme Court in the 1960s and the Voting Rights Act of 1965 as extended and amended in the past 30 years have substantially expanded the requirement for accurate population data, cross-classified by age and ethnicity at the small-area level, for legislative redistricting and related purposes. And although the statutes do not specify the geographic level of detail that is required, the census now provides it at the level of individual census blocks in order to provide flexible building blocks for subsequent reaggregation by census data users. Also, over the past three decades, the number of congressional statutes calling for the use of demographic and related data to apportion federal funds among states and localities has mushroomed. Although the statutes often do not specify the use of data based on the decennial census, in practice the use of census data for these purposes is ubiquitous in the absence of alternative sources. During this 30-year period there has also been a virtual explosion among state and local governments and private business firms in the development of computer data banks and computer models based on the use of census data at the block level for purposes of planning and operations. This has generated another set of demands for accurate small-area data. It has also raised important questions about relationships among federal, state, and local governments and private business firms with respect to the appropriate development and use of large-scale geocoding and geographic data systems (discussed in Chapter 8). These developments, the ways that the Census Bureau has responded to them, and the ways in which they are perceived by the public have interacted with each other both to raise the costs of taking the census and to generate increased perceptions of inequity in the resultant census statistics. Thus, for example, the depressed response rate is typically lowest precisely in areas with high concentrations of minority groups, leading to substantially increased costs in an effort to minimize differential undercount and produce accurate data. Despite
OCR for page 46
Modernizing the U.S. Census these efforts, the results still fall short of meeting heightened public demands for fairness and evenhandedness in redistricting legislatures and apportioning public monies. In turn, the Census Bureau over recent decades has sought to deal with the combined pressures of falling response rates and increased demands for detailed, geographically fine-grained and accurate data principally through the use of highly labor-intensive enumerative techniques for follow-up and coverage improvement. SPECIFIC FACTORS AND MARGINAL COSTS The increase in census costs in recent decades has been dramatic. Expressed in dollars of 1990 purchasing power, the full cycle costs of the decennial census rose from a little over $520 million in 1960 to $2,600 million for the 1990 census. The fourfold increase in costs over a 30-year period arises from several factors, including factors not directly attributable to census operations themselves. Three key factors need to be considered, looking at costs in constant dollars: census content, the growth of population and housing units, and declines in mail response rates. Census Content Congressional concern with census cost escalation has led to a search for remedies for the 2000 census. One suggestion has been to reduce census content, perhaps by eliminating the long-from census questionnaire. Every person in the United States was asked during the 1990 census for some basic demographic information (including race, age, relationship to household head, and sex) on a short-form questionnaire. In addition, a 1-in-6 sample of people was given a more detailed long-form questionnaire with questions about income, schooling, occupation, and related social and economic items. Residents in smaller governmental areas, with populations of about 2,500 or less, were sampled at a rate of 1 in 2 in order to obtain sufficient responses for accurate estimates of smaller areas. People in areas with about 2,000 or more housing units were sampled at a 1-in-8 rate. Special rules were made for people in group quarters and those residing in areas in which special enumeration was required. Overall, the sample design for the long form produced a sample of 1 in every 6 households in the nation. But, with one caveat, we do not believe that the content of the census is driving cost increases and coverage problems (see Chapter 6 on the special issue of the cost of the long form). Some items on the long-form census questionnaire, such as those pertaining to income, may be subject to greater than average response error and require more follow-up verification. But the long-form questionnaire has remained at relatively constant length for the past four censuses, whereas costs have increased dramatically. Furthermore, the sampling rate for the long-form questionnaire
OCR for page 47
Modernizing the U.S. Census has declined, a change that should have yielded lower overall costs.5 The panel acknowledges the congressional concern about substantial cost escalations in the census; however, we do not find evidence that census content is a primary factor producing the cost increases. Although we believe it most unlikely that the addition or subtraction of a limited number of questions would significantly affect overall census costs, it is certainly possible that very large differences in the complexity and length of the census questionnaire might affect public willingness to cooperate, reduce the response rate, and through that route affect costs—although, even then, such cost increases can affect only some 17 percent of households, since all other households are already receiving only the short form. There is some evidence on this point from recent Census Bureau experiments with a very truncated form. We note the historical evidence that strongly suggests that matters of content have not been a major driving force behind the recent and threatened rise in census costs. The length of the census questionnaire has not changed appreciably during the 1960 to 1990 period. The sample of households receiving the sample long-form questionnaire has decreased slightly during the period. The sampling fraction has decreased from 1 in 4 in 1960 to 1 in 6 in 1990. During this period of four censuses, the questionnaire delivered to U.S. households included a short form with 17 questions in 1960, 18 in 1970, 15 in 1980, and 13 in 1990. The questionnaire length of the long form was 49 questions in 1960, 38 or 52 questions in 1970,6 65 questions in 1980, and 58 questions in 1990. Overall, the length of the long form has remained relatively constant, while the sampling fraction has declined. Hence, we cannot attribute any increased census costs to the content of the census. Changes in Population and Housing Some of the census cost increases can be attributed to growth in the population and the number of housing units to be counted. In Table 3.1, the first part summarizes some of the basic data relevant to analyzing costs for the four censuses beginning in 1960. The second part converts the nominal costs to dollars of constant 1990 purchasing power and shows them on a per housing unit basis. Since census costs depend primarily on the expense of delivering a mail questionnaire to a household or having an enumerator visit a housing unit, it is more realistic to relate cost growth to the rise in the number of housing units rather than to population growth. Even when housing units are vacant and contain no household, there is a cost to ascertaining that fact, so that the number of housing units rather than the number of households is the relevant unit to consider for cost analysis. Between 1960 and 1970 census costs, when adjusted for the growth in inflation and in the number of housing units, increased only modestly (see the last line in Table 3.1); thereafter costs grew rapidly. Overall, the growth
OCR for page 48
Modernizing the U.S. Census TABLE 3.1 Factors Associated with Increases in Census Costs, 1960-1990 Decennial Census Cycle 1960 1970 1980 1990 Basic Data Full-cycle census costsa (in millions, nominal dollars) $120 $231 $1,136 $2,600 Population (in millions) 179.3 203.3 226.5 248.7 Housing units (in millions) 58.9 70.7 90.1 104.0 Mail response rates (in percent) —b 78 75 65 Derived Data Census costs in 1990 dollars (in millions)c $523 $744 $1,795 $2,600 Cost per housing unit (in 1990 dollars) $9 $11 $20 $25 a Total census costs are shown in Figure 2.6 of a report from the U.S. General Accounting Office (1992) and in original source documents from the Census Bureau. We should note, however, that Figure 2.6 of the General Accounting Office report is mislabeled as showing costs in constant 1990 dollars. The graph shows costs in current dollars. Total expenditures for the 1970 census, for instance, were $231 million, which are shown in that amount in the General Accounting Office graph. b The 1960 census was not conducted using primarily mailout/mailback questionnaires. c Using GDP implicit price deflator to adjust to 1990 dollars. of housing units accounts for about $350 million of the cost increases from 1970 to 1990.7 Mail Response Rates Some of the cost growth per housing unit can be attributed to the fall-off in mail response rates.8 If a mail questionnaire was not returned to the census office, a field worker visited the address in an attempt to count the household. Under the 1990 census procedures, as many as six visits could have been made. Census Bureau staff have provided estimates that under 1990 census procedures a decline of 1 percentage point in the mail response rate would increase costs by 0.67 percent.9 Between 1970 and 1990 the mail response rate fell 13 percentage points, a decline that translates into a $225 million cost increase.10 However, this estimate assumes the existence of the highly labor-intensive cost structure of 1990. If, instead, the same adjustment percentage for the decline in the response rate is applied to the lower cost base of 1970, the resulting cost increase would have been only about $95 million.11 The change in response rates has been qualitative as well as quantitative. Not only has the number of nonrespondents increased, but they are also increasingly difficult to locate and enumerate. The
OCR for page 49
Modernizing the U.S. Census cost impact based only on the decline of mail response rates probably underestimates the impact using the 1970 cost base. Thus, we estimate that somewhere between $95 and $225 million of cost increases between 1970 and 1990 can be attributed to the fall in the response rate. And the higher number is valid only if one accepts as "necessary" all of the other cost increases between 1970 and 1990. Table 3.2 summarizes the cost analysis to this point. Of the roughly $1.9 billion inflation-adjusted cost increase between 1970 and 1990, only $445 to $575 million can be explained by growth in the number of housing units and the decline in the mail response rate. Cost increases not explainable by these factors amounted, at the lower end of the range, to some $1.3 billion, a rise of 100 percent relative to the 1970 base, after adjustment for changes in housing units and response rates.12 Other Factors Having taken into account housing unit growth, mail response rate declines, and inflationary changes, we must conclude that other factors account for the TABLE 3.2 Increases in Census Costs, 1970 to 1990 (in 1990 dollars, in millions) Census Costs 1970 1990 Nominal census costs (in millions) $231 $2,600 Census costs in 1990 dollars (in millions) $744 $2,600 Housing units (in millions) 70.7 104.0 Cost per housing unit, in 1990 dollars $11 $25 Factors Behind Cost Increases 1970-1990 Total census cost increase, in 1990 dollars (in millions) $1,856 Less increase due to: • Growth in housing units, in 1990 dollars (in millions) $350 • Decline in mail response rates, in 1990 dollars (in millions) $95-225 Equals unexplained increase, in 1990 dollars (in millions) $1, 281-1, 411
OCR for page 50
Modernizing the U.S. Census remaining cost increases. These other factors account for a sizable proportion of overall census cost increases. We are unable to account for $1.3 billion, or almost three-fourths of the total 1970 to 1990 cost increases. This unexplained portion of census cost increases is not contributed by inflation, by population or housing growth, or by declines in the mail response rates. What other factors account for the increases? The persistent disparity in the undercount of different population groups, with undercount rates greater for minorities than for others, coupled with the consequences of greater demands for accurate small-area data, has generated pressures on the Census Bureau to expand substantially its efforts to decrease the differential undercount and to produce more detailed accurate data. The results, which still did not satisfy public demands in the 1990 census, have led to a highly labor-intensive and expensive effort to enumerate the U.S. population. Many questions have been raised about the educational qualifications of census enumerators. It seems likely that there have been declines in their education levels and job skills. For earlier censuses, a more abundant supply of part-time, motivated, better-educated housewives formed a large portion of the temporary census staff. With declines in married-couple households and increases in the proportion of women working full time, there are relatively fewer women with these qualifications available for part-time census employment. The panel attempted to examine this issue but has not been able to acquire data on either the educational qualifications for the temporary census labor force or the relationship between educational qualifications and productivity. It is therefore difficult to assess the impact of this factor on census staff productivity and on overall census costs. Census Cost Changes The panel reviewed some of the specific procedures that have been associated with cost changes in the 1980 and 1990 censuses. Our review raised questions about the justification for these cost increases. All costs for the following five areas are in 1993 dollars. The cost estimates and census cost headings are taken from Neece and Pentercs (1994). Census Planning Activities. The cost of planning direction and review increased from about $30 million for the 1970 census to $105 million for the 1990 census. Most of this threefold increase seems to be associated with headquarters management, since it explicitly does not include geography, address list development, content planning, or statistical research and evaluation. What is not known adequately is what has been bought for the extra expense and was it worth the price. Data Collection. The biggest single increase reported in Census Bureau cost figures is related to what is called data collection: from $378 million in
OCR for page 51
Modernizing the U.S. Census 1970 to $1.4 million in 1990. The unit cost of data collection (per housing unit) was $5.51 in 1970 and $13.86 in 1990. More district offices were created and kept open longer, the supervisory ratios were reduced considerably, the method of payment was changed from piece rate (with exceptions for specially difficult areas and for the final cleanup work) to hourly rates. Paying hourly rates seems to be particularly open to abuse by very short-term employees working in an almost entirely decentralized (and necessarily only loosely supervised) mode. What is more striking about these cost increases is that they occurred despite heavy investments in automation. This whole activity, more than any other, needs a fresh examination in the new environment for taking future censuses. Data Processing. The two census data processing components, data processing and data processing systems, had combined costs that increased from $122 million in 1970 to $433 million in 1990. Given the fact that computer power per unit of expenditure increased at an exponential rate during the past 20 years, this is one area where the panel would have anticipated at most a modest level of growth in overall expenditures, particularly since there appear to be no offsetting savings elsewhere in the census system. The explanations by the Census Bureau staff concern an increase in automation, not whether this improved the process or its results by a factor commensurate with the observed cost increases. Data Dissemination. The costs of data tabulation and publication increased fivefold: from $22 million in 1970 to $114 million in 1990. This increase, too, seems to raise questions about whether improvements were commensurate with cost increases. How much detail should be pretabulated and how much produced on demand? How should the product line change from print to electronic media? Are the quite modest gains in timeliness worth the substantial cost increases? Census Testing. The costs for test censuses and dress rehearsals increased from $10 million in 1970 to $78 million in 1990. Yet the shift in methodology from the 1960 to 1970 census (to a mailout/mailback census for most of the country) was arguably more fundamental than methodological changes implemented since. Also, very few changes were made in terms of content, and these tended to be deletions. The panel supports research and testing for new methods and new content, but with an emphasis on a critical and strategic approach to testing: the Census Bureau should concentrate on testing a selected few high-priority items for inclusion in the next census. Census Staff Productivity The panel studied changes in labor productivity by examining the number of housing units and the total population enumerated by census staff in the 1960, 1970, 1980, and 1990 censuses. Based on census data assembled by Beresford
OCR for page 52
Modernizing the U.S. Census TABLE 3.3 Comparison of Population Enumerated in the Census and Number of Census Bureau Personnel Taking the Census, 1960 to 1990 Panel A 1960 1970 1980 1990 Total Population (millions) 179.3 203.3 226.5 248.7 Total Housing Units (millions) 58.9 70.7 90.1 104.0 Undercount Rate (%) Total 3.1% 2.7% 1.2% 1.8% Blacks 6.6% 6.5% 4.5% 5.7% Nonblacks 2.7% 2.2% 0.8% 1.3% Difference in black and nonblack rates 3.9% 4.3% 3.7% 4.4% Number of District Offices 399 393 412 449 Time Period Open (months) 3-6 4-6 6-9 9-12 Staff Headquarters 1,846 2,186 4,081 6,763 Field 187,500 223,038 458,523 510,000a Processing 2,000 2,600 5,400 11,000 Total 191,346 227,824 468,004 527,763 Average Hourly Pay Manager $3.35 $5.88 $9.41 $15.25 Field supervisor 2.75 4.68 7.37 12.00 Crew leader 1.95 2.96 4.66 8.50 Enumerators 1.60 2.56 4.03 7.50 Clerks 1.70 2.32 3.66 5.75 Panel B: Productivity ratio = housing units/number of staff 1960 1970 1980 1990 Headquarters 31,907 32,342 22,078 15,378 Field 314 317 197 204 Processing 29,450 27,192 16,685 9,455 Panel C: Percent change in productivity ratio by decade 1960-1970 1970-1980 1980-1990 Headquarters +1.4% -31.7% -30.3% Field +1.0% -37.8% +3.6% Processing -7.7% -38.6% -43.3%
OCR for page 53
Modernizing the U.S. Census Panel D: Percent change in productivity ratio since 1960 1970 1980 1990 Headquarters +1.4% –30.8% –51.8% Field +1.0% –37.3% –35.0% Processing –7.7% –43.3% –67.9% a The 1990 census staff number has been adjusted to make it comparable with earlier censuses. The Census Bureau reports that there were 650,000 field staff in 1990 (Beresford, 1993). This number includes an unknown number of persons who took training, received training pay, and then voluntarily quit. Beresford (1994) provides preliminary estimates that the average field staff worked 22 days, with 2 days for training. If there were 150,000 quitters, the field staff estimate would be: 650,000 – 150,000 + (2/22 × 150,000) = 513,635, which we round to 510,000. Note: Beresford relies on Census Bureau estimates for the staff numbers in Panel A. The Census Bureau provides different sources of information for each census, but there are no comparable records for prior censuses. Source: Beresford, 1993, 1994. (1993), Table 3.3 summarizes trends in staff productivity. During the 1960 to 1990 period, as the Census Bureau increased efforts to reduce the undercount, the number of enumerated households (or persons) per census employee decreased. The organization of census staff changed over the period. The number of district offices increased, from 399 in 1960 to 449 (plus 9 in Puerto Rico) in 1990. Also, the organization of district offices evolved during this period, with the addition of 13 regional offices in 1990 that provided oversight to the data collection efforts of district offices. District offices also stayed open longer, increasing their operational period from 3 to 6 months in 1960 to 9 to 12 months in 1990.13 The length of operation for local census offices is comparatively long for the U.S. census, compared with the national censuses of other countries. The numbers shown in the table have limitations. A major limitation to the staff figures, which are from the Census Bureau, is that no full-time-equivalent data are available for the field staff. Full-time-equivalent numbers for field staff were developed as a necessary part of the 1990 census administrative budget process; however, no figures are available for actual employees. Data for headquarters and processing are for full-time staff, but (as discussed below) it is not known how long the temporary field staff worked on the census. It is thought by census staff that a substantial proportion of the field staff worked for only a short period in 1990. Also, we do not have data on the actual hours worked. Lack of hourly information is a problem particularly for interpreting trends in productivity of temporary staff because they generally work for a limited period of time, ranging from a few days to several months. If, as appears to be the case for 1990, turnover in the temporary staff is particularly high, then the observed staff
OCR for page 54
Modernizing the U.S. Census numbers are higher than the unknown full-time-equivalent figures. Such higher temporary staff numbers, with turnover, will also evidence lower productivity. Panel A of the table shows that there was a substantial increase in staff for headquarters, field operations, and data processing. Compared with 1960, the 1990 census staff was almost 3 times larger to count 1.4 times more people and 1.8 times more households. The 1990 census employed many people as census enumerators who received training but did not actually carry out census work. The panel has no estimates from the Census Bureau of the numbers of enumerators who were trained and did not work, although Table 3.3 includes assumptions that there was a substantial proportion of the total number of enumerators—150,000 of the 650,000 hired enumerators—who received training and voluntarily departed without actually working on the census. Panel B reports the basic productivity ratio by dividing the number of housing units by the number of staff. We assume that the major cost of the census is related to the number of housing units because census questionnaires must be prepared, delivered, and processed for each housing unit.14 But the number of housing units increased more rapidly than the population, increasing the number of units requiring census questionnaires and follow-up for nonresponse. Overall census costs are affected more by the number of housing units than by population numbers. In 1960, about 31,907 housing units were processed by each headquarters staff person. This productivity ratio increased slightly, by 1.4 percent (see Panel C), between 1960 and 1970; decreased substantially, by 31.7 percent, between 1970 and 1980; and declined a further 30.3 percent between 1980 and 1990. Overall (see Panel D), productivity for headquarters staff dropped 52 percent over the four censuses from 1960 to 1990. The field staff includes the bulk of employees working on the decennial census. Most of the field staff are temporary workers and receive salaries considerably lower than those of the permanent staff in headquarters or those handling the processing of the census. For this examination, numbers of full-time-equivalent employees are preferable, but such numbers were not reported for the 1990 census. The numbers recorded for the census field staff include all employees who worked on the census, including some persons who worked only a few days. Our understanding is that turnover in temporary field staff was higher in the 1990 census than in prior censuses. The best estimate from those familiar with the 1990 census is that about 150,000 of the 650,000 field staff quit after receiving two days of training. Adjusting for the quitters, we estimate an equivalent of about 510,000 for the number of field staff who worked throughout the enumeration period on the census.15 From 1960 to 1990, there was a decrease in the number of housing units per temporary field staff, from about 314 in 1960 to 204 in 1990, a decrease of about 35 percent over the period. A factor in the decrease of the total number of housing units per field staff in 1990 is that mail response rate was lower than
OCR for page 55
Modernizing the U.S. Census 1980, increasing the proportion of housing units requiring personal follow-up.16 The major declines in field staff productivity occurred between 1970 and 1980. Finally, the processing staff increased from about 2,000 in 1960 to 11,000 in 1990, a 4.5-fold increase. (Processing staff receive average salaries that are lower than headquarters employees.) There were sharp decreases in productivity ratios during the period, from 29,000 housing units enumerated per processing staff in 1960 to 27,000 in 1970, a 7.7 percent decline, followed by a 39 percent decline in 1980 and a further 43 percent decline in 1990—a cumulative decrease of 68 percent over the four censuses. SUMMARY The costs of taking the census have escalated sharply, especially since 1970. Even after adjusting 1970 wages and prices to 1990 levels and allowing for the growth of the number of housing units and the decline of the mail response rates, an unexplained increase in census costs of approximately $1.3 billion between 1970 and 1990 remains. There are important caveats to cost analysis of the census over the past 20 years. The work done by census staff (both field and headquarters staff) changed drastically during the 20-year period as the Census Bureau tried increasingly complex and labor-intensive approaches. Although the figures in this chapter emphasize productivity data, we note that the cost increases are primarily associated with issues of census design and not management. The rise in costs did not produce a ''better" census, at least as measured by the accuracy with which the total population or its major components were counted. We believe that a large part of the cost increase was driven by the way in which the Census Bureau responded to several important outside pressures: there was an increased demand for accurate population counts at very detailed geographic subdivisions and in hard-to-enumerate areas for purposes of congressional and legislative redistricting and otherwise carrying out the Voting Rights Act and its amendments. At the same time, public cooperation with the census process, as measured by the mail response rate, declined and was lowest precisely in the areas in which the pressures for an accurate count were greatest. The Census Bureau responded by pouring on resources in highly labor-intensive enumeration efforts to count every last person. For example, census district offices remained open for longer durations and the supervisory ratios increased during the period. It is widely believed among those associated with the census process over the last several decades that the quality of the temporary personnel employed during the decennial census deteriorated over the period, as evidenced by higher turnover and lower productivity. Although we have not been able to assess the quantitative importance of this phenomenon, the view is so widespread as to lend it some credence.
OCR for page 56
Modernizing the U.S. Census Our deliberations have led us to two general observations about the problems and issues outlined above. First, there are no conceivable changes in the collection of census data that will simultaneously meet all of the following objectives: (1) continue a highly intensive census effort, relying principally on physical enumeration and labor-intensive follow-up techniques, to overcome the consequences of a declining mail response rate; (2) provide detailed and reliable block-level data for redistricting and the Voting Rights Act; (3) provide the other housing and demographic data widely demanded for cross-tabulation at the level of small geographic areas; (4) reduce the differential undercount; and (5) keep costs from growing rapidly. There is, in short, no magic bullet. Second, from an inspection of recent trends in census costs, the panel concludes that efforts to increase differential coverage, especially through highly labor-intensive enumeration techniques, are a key factor driving up costs. Moreover, efforts to improve differential coverage (i.e., the percentage difference of population counted for whites and minorities) have had increasingly diminishing returns (coverage was better in 1980 than 1970, but there were no such gains in 1990). Differential coverage improvement efforts have been carried to the point at which additional effort and expense may yield little or no improvement in coverage or in decreasing subgroup differences in net undercount. Expensive efforts to improve census coverage are understandable given such forces as the impetus of the Voting Rights Act to provide detailed data on race and ethnicity at the block level. Nonetheless, it is appropriate to ask if this continued effort to improve differential coverage, which has been so far unsuccessful and has increased census costs, is necessary for future censuses. NOTES 1 The denominator for census mail response rates includes both unoccupied and occupied housing units. 2 Overall mail response rates were 78 percent in 1970, 75 percent in 1980, and 65 percent in 1990 (U.S. General Accounting Office, 1992). If mail response rates dropped at the rate of change of 1970 to 1990, then the projected response rate in 2000 would be 55 percent. 3 The GAO calculation assumes continued population and household growth during the decade and a decrease in the mail response rate from 65 to a range of 55 to 59 percent, involving about 50 million nonrespondent households. 4 Examining census costs and the national net undercount rate over the 1960 to 1990 period, the general trend has been moderate improvements in overall coverage accompanied by substantial cost increases. The 1980 census seems atypical in the overall trend, achieving a much lower overall net undercount rate but doubling real costs per household. After the achievements of the 1980 census, the 1990 census was not able to continue the general trend of coverage improvements.
OCR for page 57
Modernizing the U.S. Census 5 Costs associated with attempts to improve coverage are related, to a great extent, to missing households. The problem for much of the undercoverage is not that households were delivered a questionnaire and did not respond, but that households were not located and sent a questionnaire in the first place. There are two different kinds of coverage costs associated with the long form: (1) the extra costs of getting a long-form questionnaire delivered to each sample housing unit and (2) the additional cost of getting information from mail nonrespondents. 6 The 1970 census long form used a matrix sampling approach with separate samples of 5 percent and 15 percent of households selected. The 5 percent sample had a long form with 52 questions, and the 15 percent sample had 38 questions. 7 This assumes that census costs rise in proportion to housing. Actually, costs probably rise a little less than proportionately. If so, the unexplainable part of the cost increase is a little larger. 8 We refer to both mail response and mail return rates in this report. Mail return rates are the proportion of forms mailed back by occupied housing units. Mail response rates are calculated on the basis of all housing units, including vacant units. Because about 10 percent of housing units are vacant, mail response rates tend to be slightly lower than mail return rates. 9 The actual census cost changes associated with the mail response rate show a complicated picture. Initial declines in response rates produce increases of about $14 to $15 million for each percentage point change. For greater magnitudes of declines in the mail response rate, decreases in 4 or more percentage points, there is an increased cost of about $17 to $18 million for each percent point change. For each percentage point increase in mail response rates, there is a decline of about $16 million in total census costs. 10 Although mail response rates declined for the census from 1970 to 1990, especially in the 1980 to 1990 period, similar declines did not occur in major federal household surveys. Federal surveys, however, are typically face-to-face or a combination of mail with telephone or interviewer follow-up. The census is primarily a mail questionnaire, with personal visits for nonresponse follow-up. A recent review of nonresponse in 47 major federal demographic and establishment surveys revealed that there was little evidence of declining response rates during the 1982 to 1991 period (Gonzalez et al., 1994). There were some increases, however, in refusal rates in demographic surveys. It is unclear why the census mail response rate declined substantially more than the secular trend for major surveys, unless there were improvements made in household surveys that were not duplicated in the census. 11 We derive the cost estimate using the 1970 census cost base by multiplying 13 percentage points (the 1970 to 1990 mail response decline) times the .0067 (the assumed census costs per mail response point decrease) times the estimated census costs. The estimated census costs, in constant dollars, would be
OCR for page 58
Modernizing the U.S. Census the cost per household of the 1970 census ($10.52 in 1990 constant dollars) times the number of households in 1990 (104 million), or $1.09 billion. 12 We have used the change in the gross domestic product (GDP) deflator as the measure of inflation, which rose 222 percent from 1970 to 1990. But the central conclusions of this section are unchanged if other general measures of inflation are substituted. The chain weight GDP price index rose by 207 percent. We also attempted to construct a special weighted index reflecting the increase in the wages, fringes, and costs of goods and services purchased for the census. The full set of data was not available to construct such an index, but we estimate that it might show an increase in the range of 237 to 245 percent. Even with the upper end of this range as a measure of inflation, "unexplained" cost increases over the 1970 to 1990 period would still amount to over $1.2 billion (compared with the $1.3 billion cited in the text). 13 About 110 district offices, called master district offices, were open as long as 18 to 20 months during the 1989 and 1990 fiscal years. Master district offices managed address list development activities in the year prior to the 1990 census. 14 During the period from 1960 to 1990, the number of persons per housing unit decreased. Such a decrease would reduce the length of the personal interview for nonresponse follow-up, but the impact on enumerator productivity and census costs would probably be slight. The relationship between censuses and household size is also affected by other factors. First, there is generally a positive relationship between household size and census mail response rates. Smaller households typically have lower mail response rates, which contribute to increased costs. Second, declines in household size have been associated with increases in nontraditional households (households different from married couples with or without children present). Nontraditional households, for various reasons, increase the difficulty and cost of census enumeration. Finally, households with fewer people increase the difficulty of making personal contact during nonresponse follow-up, the most expensive phase of census operations. 15 The Census Bureau intentionally "overtrained" the number of enumerators because past experience showed that many enumerators resign before census operations end. The goal was to have a sufficient number of people trained so that there would be enough workers available during the daily peak activity. 16 Census Bureau analysis (Keller, 1994) suggests that actual production rates during the nonresponse follow-up operation in the 1990 census (interviews completed by enumerators per hour) increased from 1980 to 1990. Census staff believe that these production gains can be attributed to improved training, better management reports, and incentive payments.
Representative terms from entire chapter: