Author : Dadang Sukandar

Summary : In this study, sampling on two occasions refers to sampling from the same conceptual population at two different and successive points of time. Sampling designs considered in this study are one-stage and two-stage stratified sampling were simple random sampling without replacement is used in both occasions. The study evaluated the effect of change in stratification classes on properties of estimates and derived estimators of means and the difference of means, the variances of the estimators, and estimates of the variances. There are four cases of change in stratification considered. Case one, which may be considered the "ideal" situation, refers to the case when the strata in both occasions do not change. Case two refers to the situation where some strata in the first occasion are split into two or more new strata in the second occasion. The third case occurs when some strata in the first occasion are merged into some new strata in the second occasion. The fourth case is the combination of case two and case three. In this, the Best Linear Unbiased Estimator (BLUE) of means in the first and second occasions were derived. Simulation was used to determine properties of estimates with regards to unbiasedness of the derived estimators of the estimates, and their variance estimates. Revenue of establishment for some five-digit PSIC industries in the Philippines for the years 1988 and 1989, were used in the simulation of one-stage stratified sampling on two occasions. Results showed that the estimators in all cases are unbiased while the estimates are biased. Numerically as obtained from test data, the relative bias of estimate of change is higher than estimate of means. The lowest relative bias of estimate of change is equal to about 8 percent, while the highest is equal to 28 percent. While the relative bias of estimates for the means is equal to about 0.4 percent while the highest is equal to 7 percent. The simulation also shows that average of estimates for variances of means and difference ranges.