批發業動態調查抽樣設計之研究

許多的研究需要藉由調查所得的資料作為分析基礎，以提供決策參考或學術上建立和驗證理論之用，由於主客觀環境因素及資源條件限制，在進行資料蒐集工作時，往往不可能全數進行調查，因此如何結合抽樣設計以取得更具代表性的樣本，使調查統計結果更具真實性及提高估計精確度愈顯重要。現行經濟部統計處之批發業動態調查採分業截斷抽樣法，其每年抽樣的廠商大致相同，除了造成估計的偏誤外，亦將造成這些廠商長期填報資料之困擾與負擔，因此本研究之目的在於提出一較佳的分層抽樣設計，並利用蒙地卡羅模擬（Monte Carlo Simulation）研究法比較分析本研究所提之分層隨機抽樣與現行分業截斷抽樣及簡單隨機抽樣之精確度。研究步驟首先將整體批發業依行業別分成14個副母體，由於大型廠商對產業之影響甚大，先採Glaser方法尋求截略點，將廠商分成全查層與與抽樣層，抽樣層廠商採用蒙地卡羅模擬研究法，模擬比較簡單隨機抽樣及分層隨機抽樣兩種抽樣方法，分層隨機抽樣以普查年之「營業收入」為分層變數，並透過Hodges-Dalenius最適分層法釐訂層界，採用Neyman配置各層樣本家數。模擬比較分業截斷抽樣、簡單隨機抽樣及分層隨機抽樣設計在營收估計之偏誤及其估計式的變異數，結果證明本研究所提出之分層隨機抽樣較現行分業截斷抽樣及簡單隨機抽樣之精確度高。根據本研究所提出之分層抽樣設計，除大型企業維持全查外，其餘廠商進行分層抽樣，各層內之廠商或可每年進行輪替，一來可解決固定廠商長期填報之問題，二來更可提高估計結果的精確度。

關鍵字

批發業；抽樣設計；分業截斷抽樣法；分層隨機抽樣；蒙地卡羅模擬

並列摘要

Survey data are widely used for research to help policy-making or improve studies. Owing to limitation of circumstances and resources, it is hard to do census to collect data. It thus becomes very important to obtain a representative sample based upon an appropriate sampling design to increase the accuracy of the estimation. A cut-off sampling method with industry separation has been used for years by The Department of Statistics, Ministry of Economic Affairs, to conduct the activity survey for the wholesale trade. The sampled firms are almost same every year based upon the sampling design which could not only lead to estimation bias but also bother the interviewed firms for being long-term information providers. This research aims to present a better sampling design, and conduct a comparison analysis to analyze the accuracy of the stratified random sampling with the present cut-off sampling method, and the simple random sampling by using Monte Carlo Simulation. First of all, the wholesale trade is divided into 14 sub-populations in accordance with industrial classification. The large-scale firms which have a great influence on the industry are suggested to be all selected. The population thus classified into take-all sub-population and take-some sub-population by the cut-off point derived by Glaser method. Monte Carlo Simulation is used for the take-some sub-population to simulate and compare estimation accuracy among the stratified random sampling, the simple random sampling and present cut-off sampling. The stratified random sampling with Neyman allocation proposed by this thesis takes the operating revenues in the census year as a stratified variable, and uses the Dalenius-Hodges approximate optimum method to obtain the optimum stratification. The study results show that the stratified random sampling is more accurate than two other sampling methods based upon the bias and variance of the estimators. Besides, based on the stratified sampling, some other firms are sampled by stratification except large-scaled firms which are in the take-all sub-population. The sampled firms could be rotated every year, so that the problems of the long-term provision of information for some large-scaled firms will be solved and the accuracy of the estimate will also be enhanced as well.

並列關鍵字

Wholesale trade ； Sampling design ； Cut-off sampling with industry separation ； Stratified random sampling ； Monte Carlo Simulation

參考文獻

1. Cochran, W. G.., 1977. Sampling Techniques. New York:Wiley.

2. Dalenius, T. and J. L. Hodges, 1959. “Minimum variance stratification,” Journal of American Statistical Association. 54: 88-101.

3. Glasser, G. J., 1962. “On the Complete Coverage of Large Units in a Statistical Study,” Review of the International Statistical Institute. 30:28-32.

4. Hidiroglou, M. A., 1979. “On the Inclusion of Large Units in Simple Random Sampling,” Business Survey Methods Division, Statistics Canada, Canada.

5. Hidiroglou, M. A., 1986. “The Construction of a Self-Representing Stratum of Large Units in Survey Design,” The American Statistican. 40:27-31.

國際替代計量

批發業動態調查抽樣設計之研究

未授權

主題瀏覽