workshop lfs, april 15 2010 estimating the non-response bias using exogenous data on employment...
TRANSCRIPT
Workshop LFS, April 15 2010
Estimating the non-response bias using exogenous data on employment
Etienne Debauche
Corinne Prost
Workshop LFS, April 15 2010
Response rate in the French LFS
›large variability in space and in time
›special event in 2007
80
82
84
86
88
T031
T032
T033
T034
T041
T042
T043
T044
T051
T052
T053
T054
T061
T062
T063
T064
T071
T072
T073
T074
T081
T082
T083
T084
T091
T092
T093
T094
Response rate (seasonally adjusted)
Response rate (raw)
Workshop LFS, April 15 2010
Response rate in the French LFS
›non-response survey:
-« light » PAPI survey, sent to the non-respondents
-around 20% of the non-respondents answer to this survey
- they are more often employed than the respondents of the main survey
›question: is the correction of non-response in the weighting scheme efficient enough to correct for non-response bias?
Workshop LFS, April 15 2010
Response rate in the French LFS
›one study done in 2007: very small effect of the response rate on the measure of the unemployment rate
›geographical analysis
›problem: the response rate is endogenous: employment rate and response rate are both impacted by the business cycle
Workshop LFS, April 15 2010
Employment in the French LFS
›Employment: differences between the LFS estimate and the official employment estimate, coming from administrative database
-4,0
-3,0
-2,0
-1,0
0,0
1,0
2,0
3,0
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
2007
T3
2007
T4
2008
T1
2008
T2
2008
T3
2008
T4
2009
T1
2009
T2
2009
T3
LFS employment estimateAdministrative employment estimate
Workshop LFS, April 15 2010
Employment in the French LFS
›a lot of reasons why both estimates are different:
-concepts: ILO employment / job statistics; differences for sick leaves, maternity or paternity leaves
-scope: total employment / private employment; frontier workers; collective households
-quarterly average / average of the number of jobs at the end of the quarters
›And errors in the LFS: sampling errors; non-response bias; bias due to the proxy answerings
Workshop LFS, April 15 2010
Modeling the non-response rate
›behavior of the interviewers: they start with easier interviewees (for instance non-active ones). They stop when the cost of search exceeds the benefit.
wenC
enCnw
insee
insee
),,('
)),,(max(
E
E
C
n
Workshop LFS, April 15 2010
Using the administrative data employment estimates to assess the non-response bias
Response rate:
Administrative data employment:
LFS employment:
t
vE
uat
LFS E
E
ELFSE
Workshop LFS, April 15 2010
Using the administrative data employment estimates to assess the non-response bias
taE
uv
vu
LFS E)(
),cov(vE
uat
LFS Ε
E
Workshop LFS, April 15 2010
Using the administrative data employment estimates to assess the non-response bias›estimation at the region level (NUTS2 and NUTS3), with region and time fixed effects: the effect of the response rate is identified within the regions, independently of the national business cycle
›estimation on the employment rates
›weight: population of the region
trtrtrtrtLFSr taE ,,,, 11)( E
Workshop LFS, April 15 2010
Effect of the response rate
Variable (1) (2) (3)Administrative employment rate 0,40 0,36 0,02
(0,13) (0,09) (0,01)LFS Response rate 0,31 0,09 0,06
(0,04) (0,03) (0,02)
Level France NUTS2 NUTS3Period 2003q1-2009q3 2003q1-2009q3 2003q1-2009q3
TABLE 1DEPENDENT VARIABLE: LFS EMPLOYMENT RATE
ESTIMATION WITH FIXED-EFFECTS
Workshop LFS, April 15 2010
Effect of the response rate
›robustness check thanks to the natural experiment of 2007: the increase of the response rate on 2007 is due to the controversy. It should be exogenous.
Variable (1) (2) (3) (4)Administrative employment rate 0,40 0,36 0,02 0,02
(0,13) (0,09) (0,01) (0,02)LFS Response rate 0,31 0,09 0,06 0,08
(0,04) (0,03) (0,02) (0,03)
Level France NUTS2 NUTS3 NUTS3Period 2003q1-2009q3 2003q1-2009q3 2003q1-2009q3 2006q1-2007q4
TABLE 2DEPENDENT VARIABLE: LFS EMPLOYMENT RATE
ESTIMATION WITH FIXED-EFFECTS
Workshop LFS, April 15 2010
Effect of the response rate
-0,04
-0,03
-0,02
-0,01
0
0,01
0,02
0,03
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
2007
T3
2007
T4
2008
T1
2008
T2
2008
T3
2008
T4
2009
T1
2009
T2
2009
T3
LFS employment simulated LFS employmentadministrative employment
Workshop LFS, April 15 2010
Conclusion
›There is a significant effect; can the correction for non-response be improved?
›The effect is not large. In particular, it only explains a small part of the discrepancies between LFS employment and administrative data employment. Can the other sources of discrepancies be quantified?
Workshop LFS, April 15 2010
Discussion
›similar results in other countries?
›do some countries use external administrative data employment in the weighting scheme (grossing factor)?