An Exact Assessment of the Two-Stage EPI Sampling Method

Abstract

The Expanded Program on Immunization Sampling Method (known simply as EPI sampling) is a two-stage sampling procedure originally intended for quick estimation of disease prevalence in large geographical regions. The method was developed in the 1970s and all the subsequent assessments of its performance have been conducted by simulation. In her master's thesis, Reyes (2016) studied in detail the second-stage sampling of the method by developing formulas for the exact calculation of the household inclusion probabilities when sectors are used to identify the initial household to generate the EPI samples. The inclusion probabilities were used in turn to obtain exact mean, bias, variance and mean square error of any estimator of disease prevalence in the population. Thus, no extensive simulations are required and the results are exact rather than just estimates.

This thesis is an extension of Reyes' (2016) work. The extension is two-fold; (a) employing strips rather than sectors because they narrow the geographic area for field workers and to use strips to select the first household for the EPI sample at the secondary stage, and (b) carrying out an analysis on simulated population and sampling plans, using both stages of the EPI method. Analyzing the simulated populations showed that equal weight estimator that samples primary units with replacement with probability proportional to size (EW1) should be used when the target characteristic is thought to be spread randomly throughout the population, and the Horvitz-Thompson estimator that samples primary units systematically with replacement (HTSYS) should be used when the disease is believed to spread from a central location or through pocketing. Comparing the strip and sector sampling methods at the secondary stage using their effective areas leads to a comparative basis in which the inclusion probabilities are identical for both methods.