Statistical inference based on judgment post-stratified samples in finite population Section 5. Example
In this section we apply the proposed estimators to estimate corn production in Ohio based on 2012 United States Department of Agriculture (USDA) census. The population consists of counties in Ohio (One of the county is excluded from the population since census data did not have any entry for it). Variable of interest is the total corn production in bushels. We use 2007 USDA census corn production as an auxiliary variable. Mean and standard deviation of corn production in 2012 are and bushels, respectively. The correlation coefficient between and is 0.963. Using this population, we performed another simulation study to estimate the corn production and constructed confidence intervals for the population mean. Samples are generated for sample and set size combinations Simulation and bootstrap replications sizes are taken to be 3,000 and 200, respectively. Rao-Blackwellized estimators are computed based on 50 replications.
Relative efficiencies of the estimators with respect to and coverage probabilities of the confidence intervals are given in Table 5.1. Table 5.1 indicates that Rao-Blackwellized design-2 estimators outperforms all the other estimators we considered. Coverage probabilities appear to be slightly smaller than the nominal level 0.95.
Relative Efficiencies, | Coverage probabilities | |||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
10 | 2 | 2.301 | 1.981 | 1.829 | 1.448 | 1.468 | 1.280 | 1.181 | 0.883 | 0.896 | 0.924 | 0.925 |
15 | 3 | 3.745 | 3.188 | 2.353 | 1.612 | 1.994 | 1.454 | 1.200 | 0.907 | 0.919 | 0.940 | 0.907 |
20 | 4 | 5.707 | 4.402 | 2.901 | 1.624 | 2.476 | 1.143 | 1.341 | 0.920 | 0.920 | 0.946 | 0.873 |
Coverage probabilities are computed from bootstrap percentile confidence interval. Coverage probabilities are computed from |
Table 5.2 presents the estimates of the standard deviation of the estimators of population mean from simulations and from analytic expression in equation (2.5), (2.6), (2.8), (3.2). It is again clear that estimates of the standard errors are reasonably close to the estimates from simulations. The standard deviation estimates of the estimators of the population total are obtained by multiplying the entries in Table 5.2 with the population size
Estimates from equations (2.5), (2.6), (2.8), (3.2) | Estimates from simulation | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
10 | 2 | 1,108,818.7 | 1,027,289.0 | 1,027,717.4 | 883,847.8 | 833,711.4 | 1,156,300.5 | 1,028,629.9 | 929,090.9 | 854,940.3 |
15 | 3 | 815,371.3 | 687,605.0 | 689,118.9 | 602,682.4 | 545,000.2 | 810,156.1 | 670,521.9 | 578,608.4 | 528,146.5 |
20 | 4 | 652,734.4 | 472,231.5 | 477,888.6 | 454,368.3 | 392,990.3 | 638,755.1 | 478,007.6 | 434,365.0 | 375,040.7 |
These variance estimates are obtained from simulation. |
- Date modified: