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 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 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 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 Correction of direct estimates Microlevel modelling of probability to be employed Multilevel composite estimator (Longford 2010)

Cities/towns representation in sample Design weight City 1 1 City 2 City 3 2 City 4 3 City 5 4 City 6 6 City 7 19 Special research of estimates quality for labour force indicators has shown, that mid-annual, quarter and month estimates indicators of economic activity and employment are suitable for quantitative analysis both on nation-wide, and at the regional level except of month estimates for Sevastopol city, where CV ≥ 10% (for national level CV < 1,5%, for regional level CV < 6%); mid-annual estimates of unemployment rates can be used for quantitative analysis on nation-wide level, CV < 3% (table 1). As to regional level, for 7 regions from 27 indicators estimates are reliable, for other 19 regions CV doesn’t exceed 19% and only for Sevastopol city CV ≥ 25%; quarter estimates of unemployment indicators can be used for quantitative analysis on nation-wide level (CV < 5%), at the regional level it is possible to use only estimates for separate regions; month estimates of unemployment indicators can be used for quantitative analysis on nation-wide level (CV < 5%). At the regional level indicators estimates aren’t suitable, in most cases CV ≥ 15%. On the basis of reliability rate analysis of state SEAP data it is possible to draw the conclusion: for regional level direct indicator estimates are less accurate than for Ukrainian level: it is caused first of all by considerably smaller sample size for each separate region. And for indicators of population economic activity and employment accuracy of received estimates, for example for regions, is satisfactory. Low reliability of population unemployment indicators is also caused by rather small values of the indicator.

I. Direct estimation Sampled cities/towns – Horvitz-Tompson estimator; Cities/towns are not in the sample – synthetic estimator (use the estimate of sampled city/town which represents them); Rural districts – composite estimator of urban and rural populations of district State sample surveys of population economic activity in Ukraine are carried out on the basis of interview of household non-institutional set that is formed on the procedure of stratified multistage random selection. From 2004 for monthly survey 11.1 thousand households are selected that represent all regions of Ukraine. With the purpose of reliability increase of economic activity, employment and especially unemployment indicators in rural areas on the basis of sample of household agriculture activity survey 7.4 thousand households are selected additionally for interview on the program of SEAP. Inside each of 25 regions of Ukraine allocate two strata: urban settlements (cities and towns) and rural administrative districts; besides city Kyiv and city Sevastopol are surveyed. The sample size is distributed on strata proportionally number of population. All regions are represented in the sample almost equally, part of sample by regions is varied from 0,06% to 0,08%.

Information Using Scheme Special research of estimates quality for labour force indicators has shown, that mid-annual, quarter and month estimates indicators of economic activity and employment are suitable for quantitative analysis both on nation-wide, and at the regional level except of month estimates for Sevastopol city, where CV ≥ 10% (for national level CV < 1,5%, for regional level CV < 6%); mid-annual estimates of unemployment rates can be used for quantitative analysis on nation-wide level, CV < 3% (table 1). As to regional level, for 7 regions from 27 indicators estimates are reliable, for other 19 regions CV doesn’t exceed 19% and only for Sevastopol city CV ≥ 25%; quarter estimates of unemployment indicators can be used for quantitative analysis on nation-wide level (CV < 5%), at the regional level it is possible to use only estimates for separate regions; month estimates of unemployment indicators can be used for quantitative analysis on nation-wide level (CV < 5%). At the regional level indicators estimates aren’t suitable, in most cases CV ≥ 15%. On the basis of reliability rate analysis of state SEAP data it is possible to draw the conclusion: for regional level direct indicator estimates are less accurate than for Ukrainian level: it is caused first of all by considerably smaller sample size for each separate region. And for indicators of population economic activity and employment accuracy of received estimates, for example for regions, is satisfactory. Low reliability of population unemployment indicators is also caused by rather small values of the indicator.

Empirical probability to be employed 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 Average probability 0.626 Model characteristics R2 = 0.892; F = 20.58 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: Estimates of general indicators of labour force are calculated on the basis of estimator (1). They are: number of economically active, employed and unemployed population as well as rates of economic activity, employment and unemployment and so on. Data of last population census, current data of demographic statistics, and current data of social statistics as for number and placement of institutional population are used as external information. One is used by the calibration of statistical weights system for what special procedures are developed – modelled estimate; – composite estimate; – direct estimate; – weighting coefficient

III. Multilevel Composite Estimator – composite estimate for ATU; – direct estimate for region what includes the estimated ATU; – weighting coefficients – composite estimate based on microlevel model for ATU; – composite estimate for ATU from previous survey (year); Estimates of general indicators of labour force are calculated on the basis of estimator (1). They are: number of economically active, employed and unemployed population as well as rates of economic activity, employment and unemployment and so on. Data of last population census, current data of demographic statistics, and current data of social statistics as for number and placement of institutional population are used as external information. One is used by the calibration of statistical weights system for what special procedures are developed Potential covariates:   Employment rate (current) Employment rate (previous), % 0,940 Level of insured workers, % -0,346 Average wage, UAH. -0,385 Hire employers level,% 0,375 Fire employers level,% 0,320

Results of Simulation

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 Ghosh M., Rao J. N. K. Small Area Estimation // An Appraisal, Statistical Science. – 1994. – Vol. 9, № 1. – P. 55–93. Rao J.N.K. Small Area Estimation. – New York: Wiley, 2003. – 314 p. 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. State sample surveys of population economic activity in Ukraine are carried out on the basis of interview of household non-institutional set that is formed on the procedure of stratified multistage random selection. From 2004 for monthly survey 11.1 thousand households are selected that represent all regions of Ukraine. With the purpose of reliability increase of economic activity, employment and especially unemployment indicators in rural areas on the basis of sample of household agriculture activity survey 7.4 thousand households are selected additionally for interview on the program of SEAP. Inside each of 25 regions of Ukraine allocate two strata: urban settlements (cities and towns) and rural administrative districts; besides city Kyiv and city Sevastopol are surveyed. The sample size is distributed on strata proportionally number of population. All regions are represented in the sample almost equally, part of sample by regions is varied from 0,06% to 0,08%.

Thank You for Attention! Olha Lysa Ptoukha Institute for Demography and Social Studies National Academy of Sciences of Ukraine Olysa@ukr.net