New Sampling Design of INSEEs Labour Force Survey

1
New Sampling Design of INSEEs Labour Force
Survey
Sébastien Hallépée Vincent Loonis Manchi
Luc INSEE, Unité de méthodes statistiques
2
Outline

• Background
• Sample redesign
• Current sample design
• Options for redesign
• New sampling design and potential benefits
• Concluding remarks

3
Background

• In France, 2 series of unemployment data
• Annual unemployment rate estimated from LFS
• only source of quarterly estimates of
  unemployment rate according to ILO definition
• Monthly unemployment data derived by applying
  growth rates obtained from administrative data

4
Background (continued)

• Annual average of unemployment rate for 2006
• Labour Force Survey 9.8
• Monthly series (monthly admin data benchmarked on
  annual unemployment rate) 9.1

5
Background (continued)

• Beginning of 2007
• No publication of 2006 unemployment rate
• INSEE launches a series of studies and plans a
  publication in the fall
• September 2007
• Stop publication of monthly unemployment rates
  according to ILO definition
• Admin data will be published monthly by National
  Employment Agency
• INSEE will publish quarterly results issued from
  LFS (starting Dec. 2007)
• December 2007
• Action plan for improving precision and quality
  of the survey
• Redesign of LFS

6
Sample redesign

• Increase sample size by 50 by introducing a new
  sample beginning in January 2009.
• Control new sample phase-in with gradual end of
  current sample (exhausted in July 2010).

7
Current Sample Design

• Sampling frame based on 1999 Census of population
• Sample size 54 000 dwellings quarterly
• 1st stage with stratified, geographically defined
  PSUs
• 2554 PSUs
• 2nd stage with  Sectors  as SSUs
• 1 sector per PSU ( 120 and 240 dwellings)
• 3rd stage with  Area  as tertiary sampling
  units
• 1 area 20 dwellings
• An area remains in sample for 6 consecutive
  quarters
• Each quarter, 1/6 of area sample replaced by
  another area in the same SECTOR

8
Current Data Collection

• When an area enters the sample, the interviewer
  has to accomplish a field listing to capture new
  dwellings
• 1st and 6th interviews face-to-face
• In between by telephone

9
Limitations of Current Sample Design

• Ratio betwwen LFS estimates and admin data for
  the number of new dwellings

Urban Unit
of Paris
France
métropolitaine
2003 T3
2003 T4
2004 T1
2004 T2
2004 T3
2004 T4
2005 T1
2005 T2
2005 T3
2005 T4
2006 T1
2006 T2
2006 T3
2006 T4
2007 T1
2007 T2

• LFS underestimates the number of new constructions

10
Redesign Constraints

• Statistical constraints
• Maintain area sampling
• Improve sampling method
• Improve update of new building constructions
• Control transition between current sample and new
  sample
• Coordination with other household surveys
• Organizational constraints
• Maintain interviewers work loads
• Tight schedule

11
Options for New Sample Design

• Tax files
• Interesting statistical properties
• Careful of quality of sectors
• Homogenous auxiliary data
• New Census annual surveys
• Interesting properties for area formation
• Problems relating to different rotating schemes
  between Census and LFS
• Problems caused by different types of auxiliary
  information
• After various studies ? Tax files chosen

12
New Sampling Frame Features

• Tax files
• Contain data for all building premises subject to
  tax declaration
• Year N data available on 1st quarter of year N1
• Main features
• Unique identification code, housing
  characteristics (address, occupants name,
  cadastral data )
• Comprehensive data base
• No duplicate records
• Very recent data
• Many auxiliary data

13
Objectives of Sample Redesign

• The sample redesign aims at improving
• Sampling design
• Handling of non-response
• Data collection

14
Improving Sampling Design

• High quality of geographical information allows
  allocation of 32 000 000 dwellings to 200 000
  sectors with similar size ( 120 main residences)
• Reduce sampling design by 1 stage ? direct
  sampling of sectors
• Constant sample size of sectors ? reduce variance
• Balanced sampling can be used to select sectors
  based on large number of criteria, in
  particular
• Total income of a sector
• Number of  HLM  (low-rent) dwellings
• Number of collective dwellings

15
Improving Sampling Design (continued)

• Sampling method for new building constructions
• Record matching of 2 consecutive year tax files
  based on unique dwelling ID
• ?allows identification of new building
  constructions
• Procedure for updating sampled areas can be
  centralized, automated and carried out regularly

16
Improving Handling of Non-response

• For all sampled dwellings, information less than
  2-year old
• Category of dwellings (main, vacant, )
• Type of dwellings ( individual, collective,
  social )
• Household incomes within dwelling
• Household status (single, married, of people)
• Dwelling size
• All these auxiliary information are likely to
  improve non-response models. In current sample,
  only partial information available based on 1999
  Census.

17
Improving Data Collection

• In current sample, areas contain 20 dwellings (of
  any category). In future sample, 20 main
  residences ? stabilize interviewers work load
• Annual update of new dwelling identification
  (name and address)
• Use of Internet sites to help locating new
  dwellings
• Google map, Mappy, Geoportail,
• Cadastral website

18
(No Transcript)
19
Concluding Remarks

• Many benefits from new sampling design
• Better coverage
• Better field operations
• Improved sampling methodology
• Improved estimation (handling of non-response,
  calibration)