What adjustments should I make if I am analysing only a subset of the full dataset or my variables have missing values?
No additional adjustments are required for the survey weights. Each observation in the data file has its own weight value, stored in the weight variable (e.g. ESTWTNR). Correct weighting of one observation is therefore unaffected by the inclusion or exclusion of other observations. Nevertheless, this does not obviate the need to consider the impact of missing values on the representativeness of the sample. Analysts must still take the normal precautions that they would take with any sample to ensure that missing values do not bias the results by removing a non-random section of the population from the analysis. Many textbooks include discussions on the treatment of missing values (e.g. W.H. Greene (1993) Econometric Analysis, 2nd ed, Oxford: Macmillan, p.273-277). A separate issue can arise during variance estimation in Stata when focusing on a subset of cases (e.g. private sector workplaces), if the user encounters a situation in which any stratum is represented by just one workplace
