Estimation in the Swedish LFS –
an example of combining survey data from independent samples
Martin Axelson & Frida VidellStatistics Sweden
e-mail: [email protected]
Outline
• The current sampling design for the Swedish LFS
• Combining data from independent samples – some theory
• Estimation in the Swedish LFS
9th Workshop on Labour Force Survey Methodology2
The current sampling design for the Swedish LFS – project 1• Carried out by Statistics Sweden in 2008
• In response to increased interest in groups outside or with a weak attachment to the labour market
• Goal: To suggest a cost-efficient way to secure better statistics for domains of particular interest
• Suggestion: Combine data from two samples
• One drawn according to the design in use
• One drawn according to a design constructed with high precision for specific parameters and domains of study in mind.
9th Workshop on Labour Force Survey Methodology3
The current sampling design for the Swedish LFS – project 2
9th Workshop on Labour Force Survey Methodology4
9th Workshop on Labour Force Survey Methodology5
• Stratification after• Region
• Sex
• Age
• Born in Sweden (Y/N)
• Not working (Y/N)
• Not working• Derived using available register information
• Important register: Longitudinal integration database for health insurance and labour market studies
• Prediction of “Not working” during the reference week at the individual level
9th Workshop on Labour Force Survey Methodology6
9th Workshop on Labour Force Survey Methodology7
Combining data from independent samples – some notation
9th Workshop on Labour Force Survey Methodology8
Combining data from independent samples – explicit weighting
9th Workshop on Labour Force Survey Methodology9
jsk kj
kkjj
ygt
,
,ˆ
ksk kj
k
sk kj
kxkj
jj
g xxxx
t 1
,,
, )()(1
Combining data from independent samples – implicit weighting
9th Workshop on Labour Force Survey Methodology10
J
j sk kj
kkjkImp
j
ybht
1 ,
,ˆ
k
J
j sk kj
kkkjJ
j sk kj
kkjxk
jj
bbh x
xxxt 1
1 ,
,
1 ,
, )()(1
Combining data from independent samples – some relevant facts
9th Workshop on Labour Force Survey Methodology11
Estimation in the Swedish LFS – project 1• The possible gains in precision presented by
project 1 were unrealistic
• Numerical results on expected gain based on (almost) optimal explicit weighting of GREG-estimators
• Such weighting amounts to using parameter specific weight systems, i.e., different weights are used for different parameters, which NSIs typically try hard to avoid
9th Workshop on Labour Force Survey Methodology12
Estimation in the Swedish LFS – project 2
9th Workshop on Labour Force Survey Methodology13
Estimation in the Swedish LFS – project 3
9th Workshop on Labour Force Survey Methodology14
Estimation in the Swedish LFS – today
9th Workshop on Labour Force Survey Methodology15
Estimation in the Swedish LFS – examples of precision gains
9th Workshop on Labour Force Survey Methodology16
.
5.0, )]ˆ(ˆ/)ˆ(ˆ[ LFSOLFS tVtV
Estimation in the Swedish LFS – examples of precision gains
9th Workshop on Labour Force Survey Methodology17
Table 1: Population 20-64 by age and labor status, January 2013
Age 5.0
, )]ˆ(ˆ/)ˆ(ˆ[ LFSOLFS tVtV
Unemployed Not in the labor force
20-24 0.86 0.85 25-34 0.83 0.82 35-44 0.79 0.81 45-54 0.82 0.77 55-64 0.76 0.88
.
Estimation in the Swedish LFS – examples of precision gains
9th Workshop on Labour Force Survey Methodology18
Table 2: Population 20-64 not in the labor force by gender, January 2013
Sex 5.0, )]ˆ(ˆ/)ˆ(ˆ[ LFSOLFS tVtV
Full-time students
Working at home
Jobseekers, not available
Retired Long-term ill
Others
Male 0.83 0.72 0.79 0.93 0.80 0.86 Female 0.84 0.85 0.78 0.91 0.82 0.90
Estimation in the Swedish LFS – concluding remarks
9th Workshop on Labour Force Survey Methodology19
Finding solutions to methodological challenges – reflections from a methodologist
•Typically requires using (challenging) theory and methodology
•Theory and practice must go hand in hand – let implementation be part of the challenge
•Should be seen as an investment, as the short-term cost is expected to be followed by long-term savings
•Sampling design and estimation are best treated together, given the strong dependency between them
9th Workshop on Labour Force Survey Methodology20