Sampling design, survey implementation, weighting and nonresponse

The target population for the Labour Force Survey consists of persons aged 15 to 74 permanently resident in Finland. The sample was drawn with systematic sampling from the Central Population Register arranged on the basis of domicile codes, by which the sample was divided regionally in the same way as the target population. The sample was first post-stratified by region, gender and 10-year age group into 252 strata. After that the weights were calibrated according to the number of unemployed in the jobseeker register, region, gender and 5-year age groups to correspond to the register data. Calibration produces the correct population distributions for all the auxiliary variables used in the weighting, provided that the calibrated weights differ as little as possible from the original ones. Use of additional information increases the accuracy of the results and corrects the bias caused by nonresponse.

In the survey period, that is, during the last quarter of 1999, the size of the LFS sample was 37,010, of whom 31,715 took part in the interview. Of them, 15,807 were employees aged 15 to 64 (the target population of the in-service training statistics). Estimation of in-service training data was made similarly as in the Labour Force Survey.

The sampling design also involves a rotating panel, in which the same person is included at an interval of three months. In that case, there is no overlapping in the sample during the three months and each person is included in the in-service training survey data only once.

In the last quarter of 1999, nonresponse for the LFS was 14 per cent. Error possibly caused by nonresponse was corrected by weighting the response data to the level of the target population using the calibration method.

If a person to be interviewed was not be reached, the information could be inquired of a proxy, who was usually some other person belonging to the household of the target person. As of the 1998 data collection, questions relating to in-service training could also be addressed to proxies. Previously questions were not asked of proxies, but the data were produced by means of imputing.

To the start

Updated 1.1.2001