The implementation of new methods for population censuses

Reengineering of the population census
the french experience and its lessons for
implementing new censuses
Jean-Michel Durr
Ministry of Education, France
Former director of the project for redesign the french census – INSEE
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
The goals of a census
• Provide official population (enumeration) : state, regions,
municipalities… :
– to share public funds on the population basis
– Seats at parliament
• Describe characteristics of population and dwellings :
–
–
–
–
–
–
Sex, age
Employment
Migrations
Place of work, commuters
Households
….
– >>> at any geographical level
• Sample base for household surveys
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
The traditional census
• Main features :
– Exhaustiveness
– Simultaneity
– Every ten years or about
• Advantages :
– Coherence of the data : snapshot of the population and dwellings
– Data for any geographical level available
– Mobilisation of the population and the teams
• Drawbacks :
– Sample for some variables
– Refreshment only every 10 years
– Problem of quality due to the importance of the operation and the lack
of capitalized experience
– Fixed dates : may not fit to the evolution
– Risk for boycott
– Cost of te operation (inc. intercensal headcounts)
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Freshness
• Refreshment only every 10 years : sufficient in the past, but no
more :
• Increasing demand for more frequent local data :
– changes in population are going faster :
• urbanization,
• migration,
• social behaviour...
– More and more utilisation at a local scale : decentralization, local area
policies
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Quality
•
•
•
•
•
Huge operation, difficult to manage and to ensure quality
unsufficiently skilled enumerators
problem to capitalize experience
allways changes in methodology or definition
problem to compare censuses over time :
– the example of migration :
• Populationcensus N+10 - Populationcensus N is supposed to measure the
evolution of population :
= Births + Deaths + balance of migration
• but affected by differences of quality ot the two censuses :
if the balance of migration is 50 000
and census N overestimates population of 25 000
and census N+10 underestimates of 25 000,
the balance of migration will be estimated to zero.
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Fixed date
• The dates of the censuses may structure the analysis of the
evolution and may hide the real process :
% people living in periurban
areas
1980
1985
1990
1995
2000
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
2005
The situation in the world
• Mainly decennial census,
– Registers in certain countries, but often problem of quality
– Sometimes intercensal headcounts (very costly)
• Increasing demand :
– Rapid changes in population and urbanization
– At a local scale : decentralisation, more power to local authorities
– Users ask for more frequently refreshed information
• Concerns about quality and efficiency of the operation
A need for finding new ways
• A lot of projects or studies for new methods of census :
–
–
–
–
–
–
Spain, Germany, Israel : register based census
Netherlands : virtual census based on surveys linked to the register
France, USA, Peru : use of rolling census
Brazil : studies for redesign the census
UK, Mexico : studies (surveys + register)
...
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
The situation in the world - ECE
Method
2000
2010
Traditional census
35
22
Based on registers
3
7
Registers + headcount
5
4
Registers + surveys
1
8
Headcount + rolling census
-
1
Rolling census
-
1
44
43
Total
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
New methods ?
• Get free of the dogmas of :
– Exhaustiveness :
• Use sample
– Simultaneity :
• Collect information at different time and combine it
• And many combinations are available :
– Sample/exhaustive surveys
– Register/administrative sources
– Multi annual combination
– …
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
The situation in the 90’s in France
• General censuses since 1801, every 5 years until 2nd world war
• Increasing periods between censuses : 7, 8, 9 years due to budget
consideration
• Insufficient legal basis :
• periodicity
• Insee and the municipalities’ function
• Increasing demand for information :
• At a local scale : décentralization, more responsibility for municipalities, regions
• Fresher data
• No population register
• Progress in :
– Sampling techniques (well-balanced)
– Geographic information systems (GIS)
• >> Reconsider traditional general census using rolling
census techniques
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Rolling census
• Principles :
– Ideas of Kish (1980, 1990…)
– Combination of data collected at different periods
– Cumulative survey covering a whole country over a
period ot time rather than a particular day :
• Produces annual and multi-year estimates
• Produces information for all areas
• updated every year
– Exhaustive enumeration or sample
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Another rolling census experience
• American Community survey :
– Implement a national sample of 3 million addresses
– Profiles every year for communities of 65,000 or more, beginning in
2004 and each year thereafter
– 3 to 5 year cumulations for communities
of less than 65,000 population
– but insufficient sample to replace the decennial headcount
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Methodology
• constraint :
– Same budget as for a general census : every year, 1/7 th of the general
census budget
• 5 years cycle :
– Produce 3 years old data
 sampling : survey 5/7 (70%) of population over 5 years
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Communes under 10 000 inhab.
France
numbers almost
36 000 communes
under 10 000
inhabitants (almost
30 millions
inhabitants).
These communes
Have been spread
into 5 groups
WHEN ?
Each year, in january and february,
to avoid problems of seasonal effect
WHO ?
One group enumerated each year,
More than 7 000 communes
HOW ?
Survey on every inhabitant of the
communes of the group.
In 2004, every inhabitants in the communes of first group,
In 2005, every inhabitants in the communes of second group,
etc.
HOW MUCH ?
Over 5 years, every inhabitant of the
communes under 10 000 inhabitants
is enumerated
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
A year for every commune
(exemple of Bas-Rhin)
"communes 2004«
(less than 10 000 inhabitants)
Haguenau
"communes 2005"
(less than 10 000 inhabitants)
Saverne
"communes 2006"
(less than 10 000 inhabitants)
Strasbourg
"communes 2007"
(less than 10 000 inhabitants)
"communes 2008"
(less than 10 000 inhabitants)
Communes concernées chaque année
(10 000 inhabitants or more)
Sélestat
In 2009,
we go on …
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Communes over 10 000 inhab.
WHEN ?
France numbers
almost 900
communes over 10 000
inhabitants :
near
30 millions
inhabitants
WHO ?
HOW ?
HOW MUCH ?
Each year, in january and february,
to avoid problems of seasonal effect
Every commune every year
Survey on a 8 % of the dwellings sample
each year
Over 5 years, 40 % of the population is
surveyed
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
A group of buildings sampled every year
In 2009,
we go on …
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Plano amostral : resumo
Estratos equilibrados
< 10.000 habitantes (36.000 comunas/30 milhões hab.)
 10.000 habitantes (900 comunas/30 milhões hab.)
•Comunas < 10.000 hab.
– Seleção de 20% das comunas por ano
– Enumeração exaustiva nas comunas selecionadas
– Todos os endereços visitados em 5 anos
•Comunas >= 10.000 hab.
– Todas as comunas visitadas anualmente
– Visita a 8% dos endereços de cada comuna/ano
– Serão visitados 40% dos endereços em 5 anos
 Estratos adicionais : domicílios coletivos (internatos, instituições
etc)
- 20% visitados a cada ano
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
The new census will produce every year :
• De jure population for every commune and administrative district :
enables to take more frequently in account changes in population
– > more equity
• Detailed statistical results :
– At every geographical level
(micro data file for tabulation)
– Combining data collected during the 5 last years
– Relevant for the medium year (Y-2)
• Overall estimations at national and regional level : based on data
collected the last year
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Cumulating information
Year of collection
2004
2005
2006
2007
2008
2009
2010
2011
1
2
3
4
5
Group 1 Group 2 Group 3 Group 4 Group 5 Group 1* Group 2* Group 3*
* Including updates
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
How to estimate population of
communes < 10 000 inhab. in Y-2 ?
Y-5
I
Y-4
Y-3
RIIA3
A2
RIII
IV
~
RIVA2
RVA5
Y
~
RIIA2
III
V
Y-1
~
RIA2
RIA4
II
Y-2
RIVA1
~
RVA2
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
RVA
Communes > 10 000 inh.
• 8% sample every year
• Mobile sum of the data collected 5 latest years
• Weighted according to the Y-2 RIL (number of dwellings)
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Why Y-2 ?
• Medium year of the set of data
• At most 2 years between observation and estimation
• Mobile average in large communes
• …
• But it is possible to estimate Y with the same set of data collected
over Y-4 <-> Y
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Agenda
• 1995 : first report
• 1998 : beginning of the project
• 2004 : beginning of the new census
• 2005 : 1st overall estimations (France-regions) : some criteria
• 2006-2007 : Overall estimations (France- regions) : all criteria
• End of 2008 : first de jure populations
• 2009 : « cruising speed » dissemination
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Utilisation of the new rolling census
data
• General census : a snapshot every 10 years
• New census : not a movie, rather a slideshow
– communes < 10 000 inhab. : more frequent full enumeration
– communes > 10 000 inhab. : sample survey, but annual update :
• Suburban districts evolution : for example, changes in the social
characteristics
• local decisions
• no more dependent on the date of the census
• Any level : annual update
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Rolling samples : two approaches
– Things are not changing too much during the period : as if you
take a picture with a long exposure time. A certain fuzziness is
acceptable
• If the period is long, for example 5 years, you may consider you
estimate the middle year to consider a shorter period (2 years)
between observation and estimation
– Things may change over the period :
• you may collect relevant information for each period and
combination gives a stronger estimate for an average mobile mean.
• Otherwise, possible use of small area method (with complementary
sources or not)
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Flow data : example of migration
• For example, estimate migration over 5 years :
– At country, state, region level, the annual sample is relevant
enough :
• compare population between the present year and 5 years before to
calculate the balance of migration. Less problems of quality
differences to compare
• use the question « Where did you live 5 years ago » to build the
matrix of flows (region x region)
– For small communes, enumerated every 5 years :
• compare population of the year to population 5 years ago to
calculate the balance of migration
• matrix of flows over 5 years :
– the former place of residence of immigrants has been collected at the
last survey
– the new place of residence for internal emigrants has been collected
over the last 5 years in the other communes
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Example : commuters
Urban area of
Montpellier :
- 380 000 inhab.
- 45 000 surveyed
every year
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
How to take benefit of that experience
for implementing new censuses
Not a unique model :
• When get free of the two dogmas of exhaustivity and simultaneity, a
lot of frameworks are possible :
–
–
–
–
Sample/not (e.g. : spread the whole population over 10 years)
Level of sampling : commune, building, dwelling…
Use of administrative data/not
Data collection over year/one moment in the year
• Example : french and american experiences are very different but
based on the same principles
• So : necessary to find the best way according to the national
context
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro
Thank you for attention
Any questions ?
2° CONFEST - 21-25 de agosto de 2006 - Rio de Janeiro