Comparison of estimators of the variance of systematic sampling

A new estimator is given of the sampling variance for single sample systematic sampling from a finite population. It approximates the conditional expectation of this variance, given a systematic sample, with respect to a Gaussian Markov stationary serial correlation model for the population values. Monte Carlo simulation is used to compare this estimator and several competitors also based on a systematic sample. An estimate of the unconditional expected variance of systematic sampling is found to be nearly as efficient in estimating the variance of systematic sampling as the new estimator. Further, for large samples, a systematic sample confidence interval for the finite population mean can be based on the estimated expected variance of systematic sampling under this model