Sampling Procedure
The sampling algorithm behind the NETS data was designed to ensure sucient variation in employment status. Given that the KIDS panel was not meant to reveal anything about the racial dimensions of living standards, all Indian households were deleted from this sample which led to a target sample of 677 individuals. Clusters (or magisterial districts) known to have been fabricated by fieldworkers in the first wave of the panel (cluster numbers 217 and 218) were then deleted (see Carter et al (2003) for more details). Owing to the wide geographic dispersion of the areas contained in the data and the cost implications of this, all clusters with fewer than 5 households were deleted from the frame. Thus clusters 74, 76, 79, 200, 202, 208, 210, 212, 215, 219, 226, 230, 231, and 239 were not sampled, in addition to all clusters where no individuals experienced a transition out of unemployment and into employment. This left a total of 358 households located in 45 separate clusters spread throughout the province of KwaZulu-Natal covering 1749 individuals aged 19-64 in 2002. This target sample contained the majority of the original 677 individuals, save for those no longer in the sample owing to deletion of the clusters mentioned above. These individuals were denoted as "core" members of the target sample. Demographic details of these individuals and other spatial data such as hand-drawn maps and aerial photographs were used to locate the geographical position of the 358 households that these individuals were observed as residents of in 1998. The tracking process began by plotting the rough geographic location of each sample cluster of households. In addition to those household members denoted as core, the questionnaire also allowed for new economically active individuals joining the household to be captured. The names of those individuals identied as core persons were pre-listed on each household questionnaire. Since individuals aged 15-60 were interviewed in 1998, the target sample thus ultimately comprised the 19-64 age cohort (though allowing for new household members effectively increased the range of ages beyond the 64 year cut-off). In addition, a further 206 new household members were interviewed, increasing the potential sample to 1955. As far as was practical, individual members of each household were interviewed directly. This meant that in certain cases, more than one visit to the household was required in order to complete the interview. Individuals identied as core respondents who were no longer resident within the household had to be tracked and interviewed with a separate survey instrument. The tracking rule we applied was as follows: if the person had moved to within a five-kilometre radius of the original household, then a face-to-face interview was completed, if the individual could be located. If the individual had moved further away, they were contacted telephonically, if this was possible.
