A Comparison of Strategies for Estimating Population Totals

In survey sampling, estimating the total for a finite population is a common objective. In power systems load research, the totals of several finite populations must be estimated from the same sample. While the sampling plan may be optimal for estimating one parameter, it is likely not to be optimal for estimating all parameters of interest. Thus, a common strategy in such cases is to choose estimators for the other parameters of interest having minimum mean square error (MSE) for the selected sampling plan. The MSEs of selected sampling plan and estimator combinations (sampling strategies) are compared in this model. The estimators studied are the mean per unit, ratio, Beale, Tin and jackknife modified ratio, and regression. The comparison is made by Monte Carlo simulation, using several electric utility data sets, based on 2 models: smoothed empirical resampling and a probabilistic model called gamma on gamma.