Computing Estimates and Their Sampling Errors from Complex Samples
Jean Dumais, J. Heward Gough · 2012
Illustrates how sampling variance (sampling error) is estimated in most assessment surveys and the importance of correctly incorporating the sample design into that estimation, and explains how estimates of sampling error can be obtained using replication. Sampling error, not attributable to human factors, measures the extent to which an estimate from various possible samples of the same size and design, using the same estimator, differ from one another. When the sample is large enough and the number of strata is moderate, alternative jackknifing strategies are available. Jackknifing is quite efficient at estimating variances for totals and continuous functions of totals (for example, ratios, proportions, or correlation coefficients), but is not as good with respect to discontinuous nonlinear or order statistics (for example, Gini coefficients or medians). A number of statistical software products claim to specialize in survey processing, but they give inaccurate results if they fail to take into account that a survey was based on a complex sample design.