Download Full Text (315 KB)
The scatterplot is one of the most useful graphical displays of bivariate data. It allows one to see general trends and atypical points simultaneously, as well as other aspects of the data. A second feature of survey data is that some of it may be imputed to account for item non-response. A third feature is that the sample sizes can be large. A fourth feature is that the observations may have intraclass correlation due to cluster sampling. Survey designs typically specify that individuals are to be sampled with unequal probabilities of selection. The sample weight associated with an individual is the inverse of that individual's probability of being included in the sample, adjusted, if necessary, for non-response. For some applications, it may be useful to sample points for a sampled scatterplot not just proportionally to the sample weights. More direct approaches to estimating smooth conditional percentile or mean curves are possible using the original ungrouped data.
Statistics and Probability