在抽樣設計中有許多方法，舉凡簡單隨機抽樣、分層抽樣、群集抽樣…等，皆是為了降低成本、提高效率，根據資料特性的差異而發展出不同抽樣設計，樣本是否能有效表達出母體特性則受抽樣方法的影響，因此樣本的選擇是抽樣設計中極重要的一環。雖然在過去所提出的最佳化抽樣策略可以使均方預測誤差最小化，但卻帶來大量且繁瑣的運算，且需要完整的母體假設，增加在實際運用上的難度。

本文中將調適型抽樣策略結合多變量工具提出一種於二階段抽樣策略來選取樣本之方法，在給定第一階段樣本下更新母體的共變異矩陣，利用第一階段所觀察到的樣本來選取第二階段樣本，並進行最後調整，使得選擇到的單元能盡可能解釋母體變異，藉由模擬研究及實際資料，證明所提出的抽樣策略優於簡單隨機抽樣的預測結果並將方法加以運用。所提出的抽樣策略不需完整的母體模型假設且與母體機率分佈及預測量無關，僅需要母體間的相關矩陣，可去除複雜的計算過程，使其在運用上有相對優勢。

There are several different types of sampling methods, such as simple random sampling, stratified sampling, cluster sampling, and so on. All of them are developed to reduce costs and collect the data more quickly. With the difference between data, various sampling designs would be constructed. The representative sample is affected by sampling methods. As a result, the selection of samples is crucial in sampling design. Although the optimal sampling strategies proposed previously can minimize the mean-squared prediction error, but the exact population model is required. It takes intensive computational load, contributing to great
difficulty in practice.

In this research, a two-phase sampling strategy is proposed based on adaptive sampling strategy and some multivariate analysis techniques. The observed values of the first-phase sample are utilized to select the second-phase sample according to the updated population covariance matrix given the first-phase sample units. Furthermore, the selected sample is adjusted in order to get the units that can account for as much population variability as possible. It shows that the proposed sampling strategy conduces to better prediction than simple random sampling without replacement by means of simulation results. A real data on the utilization is also presented. These proposed sampling methods require only a population covariance matrix.
They are independent from the population probability distribution model and the predictor that are used. Reducing the complicated calculation makes it advantageous in application.

1.Bolfarine, H. and Zacks, S. (2012). Prediction theory for finite populations. Springer Science & Business Media.
2.Chao, C. T. (2004). Selection of sampling units under a correlated population based on the eigensystem of the population covariance matrix. Environmetrics: The official journal of the International Environmetrics Society, 15(8):757–775.
3.Chao, C. T. and Thompson, S. K. (2001). Optimal adaptive selection of sampling sites. Environmetrics: The official journal of the International Environmetrics Society, 12(6):517–538.
4.Cressie, N. (1993). Statistics for spatial data. Wiley series in probability and mathematical statistics: Applied probability and statistics. J. Wiley.
5.Johnson, R. A. and Wichern, D. W., editors (1988). Applied Multivariate Statistical Analysis. Prentice-Hall, Inc., Upper Saddle River, NJ, USA.
6.Lin, Y. H. (2018). A model-based sampling selection method based on classical multivariate analysis methods.
7.Mukhopadhyay, P. (2008). Multivariate statistical analysis. World Scientific Publishing Company.
8.Ribeiro Jr, P. and Diggle, P. J. (2001). geor: a package for geostatistical analysis, r news 1/2:15–18. Find this article online.
9.Thompson, S. and Seber, G. (1996). Adaptive sampling. Wiley Series in Probability and Statistics. Wiley.
10.Zacks, S. (1969). Bayes sequential designs of fixed size samples from finite populations. Journal of the American Statistical Association, 64(328):1342–1349.