Appendix D

Completeness of Census Returns

This appendix summarizes research from the 1990 and 2000 censuses documenting the generally more complete population coverage on mail returns compared with enumerator-obtained returns. See Chapter 7 and Appendixes G and H for documentation that mail returns generally have lower missing data rates for content items on the short and long forms compared with enumerator-obtained returns.

D.1 COVERAGE COMPLETENESS: 1990

Research from the 1990 census, based on a match of P-sample and E-sample records in the 1990 Post-Enumeration Survey (PES), found that mail returns were substantially more likely than returns obtained by enumerators to cover all people in the household. Only 1.8 percent of mail returns had within-household misses, defined as cases in which a mail return in the E-sample matched a P-sample housing unit but the P-sample case included one or more people who were not present in the E-sample unit. In contrast, 11.6 percent of returns obtained by enumerators had within-household misses (Siegel, 1993; see also Keeley, 1993). These rates did not vary by type of form: within-household misses were 1.9 percent and 1.8



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



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

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

OCR for page 409
The 2000 Census: Counting Under Adversity Appendix D Completeness of Census Returns This appendix summarizes research from the 1990 and 2000 censuses documenting the generally more complete population coverage on mail returns compared with enumerator-obtained returns. See Chapter 7 and Appendixes G and H for documentation that mail returns generally have lower missing data rates for content items on the short and long forms compared with enumerator-obtained returns. D.1 COVERAGE COMPLETENESS: 1990 Research from the 1990 census, based on a match of P-sample and E-sample records in the 1990 Post-Enumeration Survey (PES), found that mail returns were substantially more likely than returns obtained by enumerators to cover all people in the household. Only 1.8 percent of mail returns had within-household misses, defined as cases in which a mail return in the E-sample matched a P-sample housing unit but the P-sample case included one or more people who were not present in the E-sample unit. In contrast, 11.6 percent of returns obtained by enumerators had within-household misses (Siegel, 1993; see also Keeley, 1993). These rates did not vary by type of form: within-household misses were 1.9 percent and 1.8

OCR for page 409
The 2000 Census: Counting Under Adversity percent for short-form and long-form mail returns and 11.7 percent and 11.3 percent for short-form and long-form enumerator-filled returns. In an analysis of the 1990 PES for 1,392 poststrata, Ericksen et al. (1991:Table 1) found that both the gross omission rate and the gross erroneous enumeration rate were inversely related to the “mailback rate” (equivalent to the mail response rate) for PES cases grouped by mailback rate category.1 The relationship was stronger for omissions than for erroneous enumerations—the omission rate was 3 percent in the highest mailback rate category and 19 percent in the lowest mailback rate category, compared with 4 percent and 10 percent, respectively, for the erroneous enumeration rate. Consequently, the net undercount rate also varied inversely with the mailback rate. D.2 COVERAGE COMPLETENESS: 2000 Using data from the original Accuracy and Coverage Evaluation (A.C.E.) P-sample and E-sample, we carried out several analyses of the relationship between mail returns and population coverage for 2000. The analyses are not as comparable as we would have liked to the 1990 analyses summarized above: not only are there differences between the PES and the A.C.E., but also it is difficult a decade later to determine exactly how the 1990 analyses were performed. Nonetheless, the work is sufficiently similar that we are confident that the findings, which largely confirm the 1990 results, are valid. D.2.a Within-Household Omissions and Erroneous Enumerations by Type of Return We linked P-sample and E-sample records in the same housing units to provide a basis for calculating rates of within-household omissions for 2000 that could be compared to the 1990 rates from Siegel (1993). We also developed other classifications of linked P-sample and E-sample households. Table D.1 shows our results: E-sample mail returns received before the cutoff for determining the nonresponse follow-up workload 1   Ericksen et al. (1991) defined 10 mailback rate categories: one for under 55 percent, eight intervals of 5 percentage points from 55–59.9 percent to 80–84.9 percent, and one for 85 percent and over.

OCR for page 409
The 2000 Census: Counting Under Adversity included proportionally fewer cases with one or more omissions or possible omissions (2.8 percent) than did E-sample returns that were obtained by enumerators in nonresponse follow-up (7.0 percent). The difference was in the same direction as in 1990, but it was not as pronounced. Perhaps more striking, enumerator-obtained returns in 2000 included a much higher proportion with one or more erroneous or unresolved enumerations (15.5 percent) than did mail returns (5.3 percent) (comparable data are not available for 1990). Such enumerations included duplicates, geocoding errors, people lacking enough reported data for matching, and other erroneous and unresolved enumerations. If the corrections for additional duplications and other A.C.E. measurement errors, which were first reported by the Census Bureau in October 2001 (Executive Steering Committee for A.C.E. Policy, 2001b), and corrections for other measurement errors could be incorporated in the analysis, they would affect the percentages cited in the text. However, they would not likely affect the relationship in which erroneous enumerations occur more frequently in enumerator-obtained than mail returns. Indeed, the A.C.E. Revision II results showed lower estimated correct enumeration rates for enumerator return poststrata compared with mailback poststrata (U.S. Census Bureau, 2003c:Table 6). By housing tenure, both owner-occupied households and renter households showed the same patterns: mail returns included proportionally fewer cases of within-household omissions or cases with one or more erroneous or unresolved enumerations than enumerator returns. Consistently, renter households had higher proportions of these kinds of households than owner households (comparable data are not available for 1990). D.2.b Omissions and Erroneous Enumerations by Mail Return Rate Deciles In an analysis similar to Ericksen et al. (1991), Table D.2 classifies P-sample cases and E-sample cases in the 2000 mailback universe into 10 mail return rate categories, with each category defined to include 10 percent of the total. (The decile cutoffs are very similar between the two samples.) The mail return rate associated with each case is the rate for the census tract in which the P-sample or

OCR for page 409
The 2000 Census: Counting Under Adversity Table D.1 Composition of 2000 Census Households, as Measured in the Original A.C.E.E- Sample, by Enumeration Status, Mail and Enumerator Returns, and Housing Tenure (weighted)   Total Owner Renter Type of Census (E-Sample) Household Mail Return (%) Enumerator Return (%) Mail Return (%) Enumerator Return (%) Mail Return (%) Enumerator Return (%) No Valid Omissions from P- Sample 97.1 93.0 97.5 93.8 96.3 92.3 All Members Match with P-Sample 81.5 62.9 84.6 69.4 73.0 56.1 All Matches or Correct (nonmatched) Enumerations 10.3 14.6 8.7 12.7 14.7 16.6 At Least One Correct and One Erroneous or Unresolved Enumeration 2.1 4.2 1.9 3.6 2.8 4.9 All Erroneous or Unresolved Enumerations 3.2 11.3 2.3 8.1 5.8 14.7 One or More Omissions or Possible (Unresolved) Omissions 2.8 7.0 2.5 6.2 3.7 7.6 One or More Omissions 2.3 5.7 2.2 5.3 2.9 6.0 One or More Possible Omissions 0.5 1.3 0.3 0.9 0.8 1.6 NOTES: Households in all categories may contain P- sample cases classified as unresolved or removed. Mail returns are those received before the cutoff for determining the nonresponse follow-up universe (variable NRU = 1); enumerator returns are those obtained in nonresponse follow-up (variable NRU = 3). Enumeration status is before imputation for unresolved cases. SOURCE: Tabulations by panel staff from P-Sample and E-Sample Person Dual-System Estimation Output Files (U.S. Census Bureau, 2001b), provided to the panel February 16, 2001. Data files matched by household and weighted using the median value of TESFINWT within households.

OCR for page 409
The 2000 Census: Counting Under Adversity Table D.2 Rates of P-Sample Omissions, E-Sample Erroneous Enumerations, and P-Sample and E-Sample Unresolved Cases in the Original 2000 A.C.E., by Mail Return Rate Decile of Census Tract (weighted)   P-Sample Rates (%) E-Sample Rates (%) Census Tract Decile (Return Rate Range)a Omissionsb Unresolved Casesc Erroneous Enumerationsd Unresolved Casese 10th (82.8−100.0) 3.8 1.2 2.5 1.3 9th (79.7−82.7) 4.8 1.7 2.7 1.7 8th (77.3−79.6) 5.2 1.9 3.2 1.9 7th (74.9−77.2) 5.8 2.0 3.6 1.9 6th (72.6−74.8) 6.8 2.2 3.9 2.1 5th (69.9−72.5) 7.6 2.3 4.6 2.8 4th (66.8–69.8) 8.3 2.4 4.6 2.6 3rd (63.2−66.7) 9.4 2.5 5.0 3.5 2nd (57.7−63.1) 11.6 3.0 5.7 4.3 1st (19.9−57.6) 14.8 3.6 7.2 4.5 a The return rate ranges shown are for the P-sample; the ranges for the E-sample are almost identical. b The omission rate is omissions divided by the sum of omissions plus matches (excluding unresolved cases, cases removed from the P-sample as not appropriately in the sample of Census Day household residents, and inmovers who were not sent through the matching process). c The unresolved rate for the P-sample is unresolved cases divided by the sum of omissions, matches, and unresolved cases. d The erroneous enumeration rate is erroneous enumerations divided by the sum of erroneous enumerations, matches, and other correct enumerations. e The unresolved rate for the E-sample is unresolved cases divided by the sum of unresolved cases, erroneous enumerations, matches, and other correct enumerations. SOURCE: Tabulations by panel staff from P-Sample and E-Sample Dual-System Estimation Output Files (U.S. Census Bureau, 2001b), provided to the panel, February 16, 2001; weighted using TESFINWT. E-sample person resided. Within each P-sample decile, the omission rate is calculated as the ratio of valid P-sample cases that did not match an E-sample person to the total of nonmatches plus matches. Within each E-sample decile, the erroneous enumeration rate is calculated as the ratio of erroneous enumerations (duplicates, fictitious persons, etc.) to the total of erroneous enumerations plus correct enumerations. The omission rate ranges from 3.8 percent in the highest mail return rate decile to 14.8 percent in the lowest decile. The erroneous enumeration rate ranges from 2.5 percent in the highest return rate decile to 7.2 percent in the lowest mail return rate decile. The differ-

OCR for page 409
The 2000 Census: Counting Under Adversity ences are in the same direction as those estimated for 1990, although they are not as pronounced.2 The rate of unresolved cases in the P-sample and E-sample (cases whose match or enumeration status could not be resolved even after field follow-up) also shows a relationship to mail return rate deciles (see Table D.2). The unresolved rate (unresolved cases as a percentage of unresolved plus matches and nonmatches for the P-sample, and as a percentage of unresolved plus correct and erroneous enumerations for the E-sample) ranges from 1.2 percent in the highest P-sample mail return rate decile to 3.6 percent in the lowest decile and from 1.3 percent in the lowest E-sample mail return rate decile to 4.5 percent in the highest decile. These results indicate that it was easier to determine match or enumeration status in areas with higher mail return rates. D.2.c Erroneous Enumerations by Domain and Tenure In an analysis of 2000 data—for which we have not seen comparable findings for 1990—we examined erroneous enumeration rates for E-sample people (including unresolved cases) by whether their household mailed back a return or was enumerated in the field, looking separately at cases in the mailout/mailback and update/leave/mailback universes. This analysis finds considerable overall differences and for domain (race/ethnicity; see Appendix E, Table E.3) and tenure groups. The analysis differs from that reported above for mail return rate deciles for 1990 and 2000 in that it uses the mail return status of the individual household to classify E-sample people and not the return rate of either the poststratum (as in 1990) or the census tract (as in 2000). As shown in Table D.3, the rate of erroneous and unresolved enumerations for people on mail returns in the mailout/mailback universe is 4.3 percent, compared with 12.6 percent for people on returns obtained by a nonresponse follow-up enumerator in that universe. For most race/ethnicity groups, people on mail returns have lower erroneous and unresolved enumeration rates than do 2   We cannot determine the comparability of the 1990 and 2000 analysis due to missing details on how mail return, omission, and erroneous enumeration rates were calculated for 1990 by Ericksen et al. (1991). Hence, the comparison between the 1990 and 2000 results should be limited to order of magnitude.

OCR for page 409
The 2000 Census: Counting Under Adversity people on enumerator-obtained returns, and owners have lower rates than renters. There are two exceptions: American Indian and Alaska Native owners living off reservations and Native Hawaiian and other Pacific Islander owners, for whom the rates of erroneous and unresolved enumerations are similar between mail and enumerator-obtained returns; and American Indian and Alaska Native renters living on reservations, for whom the rates of erroneous and unresolved enumerations are higher for mail returns than for enumerator-obtained returns. For the update/leave cases, the patterns are similar to mailout/ mailback cases: the rate of erroneous and unresolved enumerations for people in update/leave areas is 3.1 percent, compared with 9.2 percent for people on enumerator-obtained returns, with rates for owners lower than those for renters (see Table D.3). The exceptions are American Indians and Alaska Natives living on reservations, for whom people on mail returns have higher (not lower) rates of erroneous enumerations than do people on enumerator-obtained returns, and for whom owners have higher rates than renters. There were very few list/enumerate cases, but they have a high rate of erroneous and unresolved enumerations—18.3 percent (data not shown). In contrast, the rates are relatively low for rural update/enumeration (5.2 percent) and urban update/enumeration (2.9 percent). For all other enumerator-obtained returns (e.g., those obtained from new construction), the rates of erroneous and unresolved enumerations is high—14.3 percent.

OCR for page 409
The 2000 Census: Counting Under Adversity Table D.3 Rates of E-Sample Erroneous Enumerations and Unresolved Cases, in Mailout/Mailback and Update/Leave Types of Enumeration Area (TEA), by Mail or Enumerator Return, Race/Ethnicity Domain, and Housing Tenure, Original 2000 A.C.E. (weighted)   Percent Erroneous and Unresolved Cases of Total E-Sample Enumerations   Mailout/Mailback TEA Update/Leave TEA Race/Ethnicity Domain and Tenure Category Mail Return Enumerator Return Mail Return Enumerator Return American Indian and Alaska Native on Reservation   Owner 0.0 6.5 13.5 9.0 Renter 5.1 1.9 10.2 3.9 American Indian and Alaska Native off Reservation   Owner 7.2 7.7 2.2 8.7 Renter 9.1 12.8 5.5 11.3 Hispanic   Owner 3.4 8.2 4.0 7.3 Renter 7.0 15.2 7.1 15.1 Black   Owner 4.6 11.7 3.7 7.3 Renter 8.4 16.8 5.6 10.6 Native Hawaiian and Pacific Islander   Owner 6.6 6.9 7.4 6.7 Renter 5.3 13.8 10.8 13.5 Asian Owner 4.0 8.8 3.5 8.5 Renter 8.4 17.5 8.4 23.2 White and Other Non-Hispanic   Owner 3.0 8.5 2.6 8.0 Renter 7.2 16.6 5.2 12.9 Total 4.3 12.6 3.1 9.2 NOTES: See Table E.3 in Appendix E for definitions of race/ethnicity domains. Mail returns are those received before the cutoff for determining the nonresponse follow-up universe (variable NRU = 1). Enumerator returns are those obtained in nonresponse follow-up (variable NRU = 3). SOURCE: Tabulations by panel staff from E-Sample Person Dual-System Estimation Output Files (U.S. Census Bureau, 2001b), provided to the panel February 16, 2001; weighted using TESFINWT.