Kenya - Integrated Labour Force Survey 1998-1999, Second Round

The sample for the 1998/99 Integrated Labour Force Survey (ILFS) was drawn from the NASSEP III master sample frame, which was developed from the population count of the 1989 Population and Housing Census. The frame covered all the districts (excluding Turkana, Marsabit, and Samburu) that were in existence during its inception in 1989. The master sample frame, which is a two stage stratified cluster design, is multi-purpose for household-based surveys.

In the design of NASSEP III, the Enumeration Areas (EAs) of the 1989 population census were the Primary Sampling Units (PSUs). The PSUs were selected using the Probability Proportional to Size (PPS) method, and were then segmented into smaller units of about 100 households, constituting one Measure of Size (MOS). One segment from each PSU was selected randomly for the creation of a “cluster”. The frame was further categorised into urban and rural sub-strata. The urban component comprised 329 clusters (of which 209 are operational) spread over 63 urban centres, with population 10,000 and over, including all district headquarters with the exception of Marsabit, Mararal and Lodwar towns. The rural component of the frame had a total of 952 clusters (of which 930 are operational) spread over 34 districts as constituted in 1989, but excluded Turkana, Marsabit, Samburu and the North Eastern districts of Wajir, Garissa and Mandera. In creating the rural component of the frame, each of the 34 districts covered was treated as a stratum. The allocation of the PSUs to the rural districts was done proportionately to the population size. The allocation of the clusters to the districts varied between 12 and 36 clusters, with sparsely populated districts assigned fewer clusters than densely populated districts.
Sample Size Determination

The child labour phenomenon was used in determining the appropriate sample size, so as to increase the chances of capturing working children in sampled households since the child labour incidence is a rare event. First, it was estimated that children aged 5-17 years constituted 37.0 percent of the listings of 1996, and also in the December 1998 population projections. Also, the proportion of working children falling in this age interval was estimated to lie between 15 percent and 19 percent (using the results of the 1989 Population and Housing Census). Using a margin of error of 5 percent and a confidence level of 95 percent with an adjustment for the design effect of 2.0, a sample size of 54,000 persons was estimated for the survey. Working with average household size of 4.2 persons, the sample size translated into 12,814 households, which were selected by a systematic selection of every tenth household in each cluster. Where the calculated number fell below 10 households, a minimum of 10 households was taken in all such cases. The resultant sample size was observed to be sufficient to provide national and provincial estimates.


The survey, as stated earlier, covered 1,109 clusters out of the 1,139 selected clusters, giving 97.4 percent response rate. The remaining 30 clusters constituting 2.6 per cent were not covered, mainly due to inaccessibility caused by flooding and insecurity prevailing in these clusters.

At household level, 11,049 out of 12,814 selected households participated in the survey, giving 86.2 percent response rate. The household response rates varied between districts and urban/rural sub-strata as shown in Annex 1. The lowest response rate was recorded in Garissa district while the highest response rates were observed in Mandera and Embu districts. The rural component had a higher response rate of 87.5 percent, that is 9,111 respondents out of a total of 10,413 selected households. In the urban areas there was a response rate of 80.7 percent based on 1,938 households that responded from 2,401 selected households. Among the provinces, the lowest response rate was experienced in North Eastern where 74.6 percent while Eastern Province had the highest response rate of 92.7 percent.


As to the weighting procedures, weighting of the sample data was done because the selection process of the sample was not self-weighting; and in the accompanying computation process, adjustment was done for cluster and household non-response. In addition, the adjustments took into consideration both the listed populations in the clusters and population growth.