Surveys have been a popular research tool and have been used extensively in many fields including education. In practice, most of surveys are conducted with some part of the population, samples. As more surveys are conducted, the range of survey participants becomes wider than ever before. Groups of people, who did not attract enough educational researchers' attention because they were rare in the general population, are now considered populations of interest. However, they are hard to sample using conventional sample designs. Such situation motivated the development of a new sample design and Reckase, Kim, and Ju (2016) developed stratified inverse cluster sampling with updating process (SICSUP) in order to obtain a representative sample from such rare populations.The objective of this study is to evaluate the performance of SICSUP with respect to statistical and economic aspects. The statistical aspects are: (1) accuracy in parameter estimation, (2) required sample size to achieve desired precision that results of surveys should have, and (3) accuracy in group differentiation were examined. The economic aspect is the number of contacted schools in order to reach the predetermined sample size of elements in SICSUP as compared to that in stratified cluster sampling (SC) was investigated.The results suggest that SICSUP works as well as SC and can be a useful sample design for rare populations. Also, the results provide guidelines for the application of SICSUP in educational surveys. In terms of precision in mean, standard deviation, and standard error estimation, in general, SICSUP performs as well as SC except with small sample size (n = 50). The four replication-based standard error estimators, including the jackknife, bootstrap, BRR, and BRR with Fay's adjustment, do not make a substantial difference in standard error estimation.In terms of determination of sample size, on average, SICSUP needs a slightly larger sample than SC although the difference in sample size between the two sample designs is not sizable. With sampling weight, SICSUP and SC require a sample size about 2.30 and 2.21 times, respectively, larger than that in simple random sampling (SRS) in order to produce estimates as accurate as those in SRS.In terms of providing country rankings that are identical with those based on the population means, SICSUP works as well as or, depending on the condition, slightly better than SC. However, the results imply that rankings should be interpreted with caution.With respect to economic aspect, SICSUP needs to contact fewer schools than SC in order to reach a predetermined sample size of elements and thus, is more economical than SC. However, SICSUP might not have advantages for rare populations with large clusters or small number of strata.
Read
