Transcript 2000 Census PES

Directorate of Surveys
Post Enumeration Survey
Measuring Census Coverage
Post Enumeration Survey
The 2000 PES and 2010 project
Paraguay, 3-7 August, 2009
República Federativa do Brasil
Acronym: BR
Federal Capital: Brasília
Population in 2009: 191,480,630
Men: 93,962,767
Women: 97,517,863
Area: 8,514,876 km2
Density of population: 22.48
inhab/km2
Why do we evaluate Census?
•Census does not count every household or person
•In Brazil in 2000, we estimated that around 9 million persons
(5,52%) were not counted
•Over/Undercount causes bias in estimates and measuring them
helps users to better use the data.
•Not have any measure of coverage quality may affect the credibility
of Census
Why do we do PES?
•To improve the necessary sources for indirect methods is more
difficult and costly
•The PES allows to identify kinds and level of coverage errors and
helps to know where to improve more.
•Training?
•Cartography ?
•Supervision?
IBGE does not adjust Census results
based on the PES
What we calculated through the PES
– Estimate undercount of dwellings and persons
– Estimate overcount of dwellings and persons
Calculate coverage rates – gross and net
Brazil, Region, State, Urban/Rural, Age(range)
2000 Census PES
Independence of PES from the Census
–
Independent sample frame (from the EA file)
–
Different team for data collection
–
Data collection started after finishing Census
–
Confidentiality on PES sample areas
2000 Census PES
Scope
–
Private dwellings and persons living in private dwellings
PES does not cover collective dwellings
–
National coverage
•
•
Rural and urban areas (except rural of the North region)
All kind of enumeration areas including favelas
2000 Census PES
Sampling
– Size
1,356 Enumeration Areas
301,230 dwellings
1,168,494 persons (0,7%)
68% confidence level for 0,2 error
– Two-stage sampling
1st stage - selection of enumeration areas in the “States”
26 states and 1 DF (Federal District)
2nd stage - 10% of the dwellings in each EA
2000 Census PES
Questions
– Dwelling (all EA)
Full address (street, type, number, neighbourhood)
Responsible person name
Total of persons living in
– Persons (10% sample)
Name, sex, age, relationship in the household, if lived in at
census date
2000 Census PES
Data collection
Starting October 2000 and finishing February 2001
1,356 enumerators, 319 supervisor and 27 Coordinators
Matching e Reconciliation
Clerical matching (2 stages: dwelling and person)
10% in average was sent to reconciliation
Both done by the same technicians
The two operations took around 3 months
2000 Census PES
Estimation
Dual System (DSE)
PES
Counted
Not counted
Total
Counted
a
c
a+c
Census
Not counted
b
d
b+d
a found in the Census and PES
b found in the PES
c found in the Census
d NOT found neither in the Census nor in the PES
t assumed to be population
Total
a+b
c+d
t
2000 Census PES
Some results
Net undercount rate for dwellings in 2000 Census, Brazil and
Regions
14
12
10
8
%
Total
Urban
Rural
6
4
2
0
Brasil**
Norte*
Nordeste
* Does not include rural area
** Does not include rural area of North region
Sudeste
Sul
Centro-Oeste
2000 Census PES
Some results
Net undercount rate for persons in 2000 Census, Brazil and
Regions
10
9
8
7
6
%
Total
5
Urban
Rural
4
3
2
1
0
Brasil**
Norte*
Nordeste
* Does not include rural area
** Does not include rural area of North region
Sudeste
Sul
Centro-Oeste
What’s new for 2010
– Bigger sample
Around 4,000 EA
Confidence level of 68%
Error 0,10
- Include rural area of North region
– Sampling PPS
instead of random
– Automatic matching
Instead of clerical only
Where we are on the automatic
matching
– Methodology
Study of methodologies and related computational methods
– System
Conducted studies on different possibilities - Data Quality
(SAS), Quality Stage (IBM), among others
Testing FEBRL
Thanks