1 A Study of Sources for the Error Structure in Estimates of Census Coverage Error Components Mary H. Mulry U.S. Census Bureau 2009 International Total Survey Error Workshop June 16, 2008
2 Census Coverage Error Definitions Net census coverage error = omissions – erroneous enumerations Components of coverage error Erroneous enumerations Omissions Estimated net error in Census 2000 was small, but evidence indicated component errors were larger
3 Net census coverage error DSE used to estimate net coverage error Case-by-case matching of enumeration(E) & independent population(P) samples Processing employs balancing of errors that improves net error estimates Net error estimate is unbiased if no model error: net error = DSE – census However, balancing of errors causes upward bias in weighted nonmatches and weighted erroneous enumerations Not suitable for component errors
4 Components of coverage errors omissions & erroneous enumerations Component error estimation needs processing without balancing of errors needed for net error Collect more data from respondents More processing of DSE data Different estimators Estimators: EEs = weighted erroneous enumerations Omissions = net error + EEs
5 Error structure in component errors Recent studies (Mulry 2008, Spencer 2008) Error structure in estimate of erroneous enumerations yields understanding of error structure in estimate of omissions Some offsetting of errors in estimates of omissions Errors present in estimate of EEs for net error offset in estimate of EEs for components
6 Definition of Components of Census Coverage Error Erroneous enumerations Duplicate enumerations People born after Census Day People who died before Census Day Enumerations for people not residents of a HU in the U.S. Omissions People who should have been enumerated in the Census but were not
7 Definition of Correct Location for Enumeration For net error Persons must be enumerated in a HU within the search area of their ‘usual residence’ For component errors Persons must be enumerated in a HU once anywhere in the U.S.
8 Varying amounts of data reported for Census enumerations E1E1 E0E0
9 Data-defined Enumerations E 1 has sufficient info for net error CE 1 = correct enumerations EE 1 = erroneous enumerations WL 1 = enumerations in wrong location, but only enumeration for person E 0 has insufficient info for net error CE 0 = correct enumerations EE 0 = erroneous enumerations WL 0 = enumerations in wrong location, but only enumeration for person
10 Estimates of Erroneous Enumerations
11 Notation for errors in status in enumeration sample True status coded status
12 True status vs coded status for enumeration sample Subscript is coded status True values are sums of columns Estimates are sums of rows
13 Net error terms are important for component error estimates
14 Types of errors in data Identification of duplicate enumerations Membership in housing unit population Usual residence Geocoding housing unit containing the enumeration
15 How Errors Occur Failure to detect False detection Types of errors Duplication Population member Usual residence Geocoding
16 Correct Enum coded Erroneous False duplicate Undetected HU pop member Undetected usual residence Has duplicate that is misclassified as usual residence Erroneous Enum coded Correct Undetected duplicate Falsely HU pop member False usual residence Has duplicate that is usual residence
17 Correct Enum coded Wrong Location Undetected usual residence Another HU misclassified as usual residence & not enumerated there False geocoding error & only enumeration Wrong Location coded Correct Enum False usual residence Another HU is usual residence & not enumerated there Undetected geocoding error & only enumeration
18 Erroneous Enum coded Wrong Location Undetected duplicate Misclassified as only residence, but also enumerated at usual residence Falsely HU pop member Misclassified as in HU pop at wrong location Wrong Location coded Erroneous Enum False duplicate Usual residence outside search area & not enumerated there Undetected HU pop member at wrong location
19 Sources of errors Processing errors 2 studies evaluate 2010 CCM Data collection errors 4 studies evaluate for 2010 CCM
20 Info on processing error Matching Error Study All types of errors Administrative Records Study Types of error: Duplication, HU pop
21 Info on data collection error Respondent debriefings Types of error: usual residence, HU pop Study of Missed Housing Units Type of error: geocoding
22 Info on data collection error Recall bias study Type of error: usual residence Comparison of census operations with CCM results Type of error: geocoding
23 Summary of error sources Synthesis of info from CCM evaluations Designing simulation study to aid analysis of error structure Develop better understanding of error structure
24 U.S. Census Bureau