Census Coverage Error
Download
Report
Transcript Census Coverage Error
The Structure of Error Components
in 2010 Census Coverage Error
Estimation: P-sample estimates
Mary H. Mulry
& Bruce D. Spencer
U.S. Census Bureau
Northwestern Univ.
2010 International Total Survey Error Workshop
June 15, 2010
1
Census Coverage Error
• Net error = omissions – erroneous enumerations
= True Pop – census
• DSE used to estimate True Pop
• Enumeration sample (E-sample)
• Population sample (P-sample)
• Components of census coverage:
Erroneous enumerations
Omissions = net error + erroneous enumerations
(weighted P-sample not suitable for component error)
2
Error structure
for component errors
• Recent studies (Mulry 2008 2009, Spencer 2008
2009)
• Have general design for simulation to study
synthesis of sampling and nonsampling errors
• Have decompositions of E-sample & P-sample
nonsampling errors as functions of field &
processing errors for net error & component errors
• Have considered the data available from evaluations
• E-sample models presented at 2009 ITSEW
3
Dual System Estimator
DSE
=
Enumerations Correct enumeration rate
__________________________________________
not imputed
Match rate
from P-sample
4
Focus on P-sample
• Match rate estimates census enumeration
rate for DSE
• Sample selected independent of census
• 170,000 HUs in US; 7,500 HUs in Puerto Rico
• Interviews conducted Aug – Oct 2010
• Census Day was April 1, 2010
• Followup conducted early 2011 after initial
matching to collect more info on unresolveds
• Final matching after Followup
5
P-sample seeks
to answer to 2 questions
• Is person in the P-sample?
• Has to live in HU on Census Day
• Complication for usual residence: Some
move between Census Day & Interview Day
• Does the person have a matching census
enumeration at usual residence on
Census Day?
6
Interview asks
(paraphrased)
• Who lives here now?
• Did each person live here on Census Day?
• If not, where did the person live?
• includes probes to aid memory
• Did anyone else live here on Census
Day?
• If yes, where did they move?
7
People in P-sample
P-sample out-mover
moves to Group Quarters
or out of US after
Census Day
Stable Resident
lives in sample block
In-mover
on Census Day &
moves into sample
Interview Day
block after Census
Day
Sample block
8
P-sample status
Census Day
Interview Day
HU in HU out not HU HU in HU out not HU
Stable resident X
X
In-mover
X
X
P out-mover
X
X
NP out-mover
X
X
Interloper
X
X
Out-of-scope
X
Yellow = in the P-sample
9
Types of errors in data
affecting P-sample status
• Membership in housing unit population
• Usual residence on Interview Day
• Usual residences on both Census Day
& Interview Day
• Geocoding housing unit interviewed
10
How errors in P-sample
status occur
Failure to detect
False detection
Types of errors
•Population member
•IntD usual residence
•CenD & IntD usual res.
•Geocoding
11
Truth: Stable Resident in P-sample
Error
Coded out of sample
Int Day in HU
out of sample
NP out-mover
Not in HU Pop
(GQ or out of US
on Cen Day)
Out-of-scope
Int Day & Cen Day in
HUs out of sample
Interloper
12
Errors in data causing
errors in match status
• Usual residence on Census Day
• Causes search for match to look in wrong
place
• Identification of census enumeration
• Geocoding housing unit
13
Truth: Stable Resident has match
Error
Coded
Cen Day in HU
out of sample
In-mover
nonmatch
Enumeration not
found since items
missing
Stable resident
nonmatch
HU geocoded
in wrong block
Stable resident
nonmatch
14
Missing items can lead to problems
in finding a match
Census
Joe Anders
P-sample
40
Joe Anders
40
Sue Anders 38
Sue Anders 38
B. Anders
Bob Smith
17
Jim Smith
15
16
?
“B.” may be Bob, or Jim, or a 3rd son
15
Geocoding error causes listing HU in the wrong block. Census
enumerates in correct block. Match to census enumeration outside
the search area is not found. P-sample person is coded nonmatch
P-sample
x
(geocoding error)
search area
Census
x
(true)
16
Sources of errors
• Processing errors
• 2 studies evaluate 2010 CCM
• Data collection errors
• 4 studies evaluate for 2010 CCM
17
Info on processing error
• Matching Error Study
• All types of errors
• Administrative Records Study
• Types of error: ID enumeration, HU pop
18
Info on data collection error
• Respondent debriefings
• Types of error: HU pop,
CenDay & IntDay usual residence
• Study of Missed Housing Units
• Type of error: geocoding
19
Info on data collection error
• Recall bias study
• Type of error: CenDay usual residence
• Comparison of census operations
with CCM results
• Type of error: geocoding
20
Next steps
• Design estimators of E-sample &
P-sample nonsampling errors
• For use in simulation study to
synthesize errors
• Avoid double counting errors
• Continue to develop better
understanding of error structure
21
[email protected]
U.S. Census Bureau
22