Transcript Методологические подходы к оцениванию п

Estimation of Employment for Cities,
Towns and Rural Districts
Workshop of BNU Network on Survey Statistics
Tallinn, August 25 – 28, 2014
Olha Lysa
Ptoukha Institute for Demography and Social Studies,
National Academy of Science of Ukraine
Kyiv, Ukraine
Task
To estimate the employment rate for cities, towns and
rural districts (administrative territorial units ATU) of Ukraine based on the annual LFS
dataset
Data Sources
o
o
o
o
o
Sample survey of households (LFS);
Sample survey of enterprises (BS);
Register of unemployment;
Administrative data reported by enterprises ;
Census data and demography statistics.
Survey description
S
S
S
S
S
stratified, multistage sample design with
systematic selection proportional to size;
11,1 thousand households are selected every
month, representing all country;
rotational scheme 3-9-3 (2/3 of sample was
observed in previous month);
all hh’s members of age 15-70 years old are
interviewed about their economical activity;
complex weighting procedure: design weights,
non-response adjustment, calibration to
population sex-age structure
Problems
o
o
o
o
Small sample size or 0 at ATU level;
High variance of employment rate estimates for
ATU;
All rural districts are represented in the sample
but not all cities and towns;
High variation between estimates of employment
rate in ATUs
Proposition
I.
II.
III.
Correction of direct estimates
Microlevel modelling of probability to be
employed
Multilevel composite estimator (Longford 2010)
Cities/towns representation in sample
In sample
City 1
City 2
City 3
City 4
City 5
City 6
City 7
Design weight
1
1
2
3
4
6
19
I. Direct estimation
1. Sampled cities/towns – Horvitz-Tompson estimator;
2. Cities/towns are not in the sample – synthetic estimator
(use the estimate of sampled city/town which represents
them);
3. Rural districts – composite estimator of urban and rural
populations of district
Information Using Scheme
ULFS,
current survey
ULFS,
previous surveys
ULFS,
current survey
REGIONAL
LEVEL
Auxiliary information:
demography statistics
LOCAL LEVEL
ULFS,
current survey
MICROLEVEL
II. Microlevel Model
Empirical probability to be employed
Fitted model
p = 0,327+0,056∙M+0,059∙R+0,308∙A_2+0,380∙A_3+0,390∙A_4+
+0,404∙A_5+0,392∙A_6+0,327∙A_7+0,130∙A_8-0,095∙A_9
0.626
Average probability
R2 = 0.892; F = 20.58
Model characteristics
Factors:
1) Gender:
M – male;
2) Type of area:
R – rural;
3) Age grope:
A_2 – 25–29 years old;
A_3 – 30–34 years old;
A_4 – 35–39 years old;
A_5 – 40–44 years old;
A_6 – 45–49 years old;
A_7 – 50–54 years old;
A_8 – 55–59 years old;
A_9 – 60–69 years old.
Composition between regression and
direct estimates:
~
ˆ

Pj  1     Pj    Pj
Pj
Pˆ j
~
Pj

– composite estimate;
– direct estimate;
– modelled estimate;
– weighting coefficient
III. Multilevel Composite Estimator
Pj,t  1 j  Pj,t  2 j  Pˆt  3 j  Pj,t 1
Pj,t – composite estimate for ATU;
Pj,t – composite estimate based on microlevel model for ATU;
Pˆt – direct estimate for region what includes the estimated ATU;
Pj,t 1 – composite estimate for ATU from previous survey (year);
1 j ,2 j ,3 j – weighting coefficients
Potential covariates:
Employment rate (previous), %
Level of insured workers, %
Average wage, UAH.
Hire employers level,%
Fire employers level,%
Employment rate (current)
0,940
-0,346
-0,385
0,375
0,320
Results of Simulation
Direct estimate
Multilevel composite estimate
80
70
RRMSE, %
60
50
40
30
20
10
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Administrative Territorical Units
Conclusions
Proposed approach based on multilevel composite two-stage
estimation.
Results of implementation in a simulation show improvement in
accuracy of employment rate estimates for the cities, towns and
rural districts. The RRMSE of estimates of ATUs was reduced
by 45% on average.
Using a microlevel model decreases variation between estimates
of employment rate in ATUs
We can obtain estimates for ATUs that are not in the sample.
References
1. Ghosh M., Rao J. N. K. Small Area Estimation // An Appraisal,
Statistical Science. – 1994. – Vol. 9, № 1. – P. 55–93.
2. Rao J.N.K. Small Area Estimation. – New York: Wiley, 2003. –
314 p.
3. Longford N.T. Simulation of small-area estimators of the
poverty rates in the oblasts of Ukraine. – SNTL and UPF,
Barcelona, Spain. The report prepared for the Social Assistance
System Modernization Project, Ukraine, Kyiv, 2010.
Thank You for Attention!
Olha Lysa
Ptoukha Institute for Demography and Social Studies
National Academy of Sciences of Ukraine
[email protected]