Improved calculation and dissemination of coefficients of variation in the Norwegian LFS
This report documents the implementation of improved calculation and dissemination of coefficients of variation in the Norwegian Labour Force Survey. It also summarizes studies related to non-sampling errors in the Norwegian LFS.
The estimation procedure in the Norwegian Labour Force Survey (NLFS) is more complicated than simple calibration, but an empirical variance estimator was derived for the NLFS by Hagesæther and Zhang (Notater 2007/22) based on linearization.
This report documents the implementation of the improved calculation of sampling error for the main variables and groups in the regular production system of the NLFS. The empirical variance estimator is also extended with covariance elements to cover figures of change and annual averages. The new calculations of coefficient of variation (CV) and standard error (SE) are published in connection with the regular publication of quarterly and annually NLFS figures in the StatBank on our webpage. The quarterly CV and SE figures are published in the StatBank table http://www.ssb.no/en/table/09937 and annual figures in http://www.ssb.no/en/table/09938.
Some of the results are presented and discussed in this report. Average of quarterly standard error in 2011 and 2012 for total unemployment and total employment are 5.1 and 0.33 per cent of the estimated values respectively.
Eurostat has proposed new precision requirements. One of the proposed requirements is about the standard error for the estimated annual average of the proportion of unemployed at NUTS 2 level (region). Our calculations show that we will fulfil this proposed requirement for all the 7 Norwegian regions in 2010, 2011 and 2012.
The other proposed precision requirement, is about the standard error for the difference in estimated proportion of unemployed at national level between two successive quarters. In spite of the high overlap between samples in adjacent quarters, the NLFS would not fulfil this new proposed precision requirement for change estimates of unemployed persons. One reason for this is the low autocorrelation for unemployment, so the high overlap of samples is of little help for making good estimates of change in the unemployment. However, the high overlap of sample makes the change estimates for employment better. Also, the NLFS estimation procedure does not include any good register predictors for LFS-unemployment, but is in stead optimized for making good quarterly county-divided employment figures.
Also other sources of survey errors in the NLFS are described in this report.