Estimates of Sampling Error
It is observed that estimates based on sample survey data are potentially subject to two types of errors, namely, sampling and non-sampling errors. The latter are not easy to control since they arise from factors external to the sample design, which include coding and data entry errors among others. Sampling errors are however controlled through the design of the sample and are measured by use of variances. In the ILFS, the stratified cluster design was used. This is a complex design and its variance estimates were based on the ultimate cluster method for variance estimation. The Cenvar program in the IMPS was used in the estimation of the variances of some selected variables. The standard error estimates in Appendix I were therefore obtained and reflect the situation of the data.
The variances of the parameters were either for the totals or ratios. (For the fomulae kindly check the report)
It is observed from the estimates of the variances that most of the estimates have consistently small standard errors, with the exception of categories with very few observations. Consequently, it was found that the coefficient of variation (CV) for most of the variable were in the neighbourhood of 10 percent or below with exception of North Eastern Province which had a very small sample in the survey, or categories of the population with rare occurrence such as people with university level education or ages exceeding 50 years in some provinces, which had very few observations.
The design effect in most cases was not high with the exception of the categories having small observations in the North Eastern Province. It is, therefore, observed that estimates based on the ILFS are fairly reliable and provide a good reflection of the employment situation and associated characteristics in the country.
|
Digg
Google Buzz
LinkedIn
Stumbleupon
Delicious
