標題: 分層隨機抽樣下母體平均數推論之研究
On the Inference about the Mean under Stratified Random Sampling
作者: 李琇玉
丁承
經營管理研究所
關鍵字: 蒙地卡羅模擬;分層隨機抽樣;t 統計量;Monte Carlo simulation;stratified random sampling;t statistic
公開日期: 2005
摘要: 分層隨機抽樣法在抽樣調查實務應用上很普遍,但對該調查資料母體平均數做進一步相關之估計或檢定,則甚少見到較嚴謹的作法。究其原因,分層抽樣母體平均數可視為單變量變異數分析線性比較(linear contrast)之特例,雖可採用正確 t(exact t)檢定法 (以texact 表之),惟該法須滿足各層變異數一致的要求,此一條件十分嚴苛,導致實務應用相當受限。Satterthwaite 雖已提出一近似 t (approximate t)檢定法 (以tappr 表之),但公式較繁複,故實證上較少採用。實務上,或有為求操作便利,勉強採用簡單隨機抽樣所對應的 t 檢定法為之,惟此係錯誤之作法。 為使分層抽樣的統計推論更為簡化,更利於應用,本研究透過模擬研究,探討在分層隨機抽樣架構下,是否可放寬各層變異數一致之限制,以及放寬此一限制後仍可採用 texact 的條件。所採方法係以蒙地卡羅模擬法,就各種不同的分層數、各層所占母體比例、樣本數以及層內變異數之設計組合,每一組合均進行 1000 次之模擬後,再計算並比較 texact 與 tappr 之模擬檢定力大小。模擬結果顯示放寬上述限制條件確屬可行;亦即以 texact 檢定法進行統計推論時,各分層之層內變異數均相等的假設並非必要。經歸納研究結果,texact 與 tappr 二者適用之條件如下:(1) 在分層抽樣設計下,當各層所占母體比例相等時,可直接用 texact 進行推論。(2) 當樣本數大於50時,可直接使用 texact 進行推論。(3) 樣本數小於50,則使用 tappr 進行推論。 欲以 texact 進行統計推論,必須先滿足各分層均符合常態性且所占母體比例已知之條件。在實證方面,本研究分別以2001年 5 月香港來台旅客及2001年第 1、2季來台觀光旅客消費調查之樣本資料進行實例分析。本研究所提出之 texact 檢定建議,可適用於任何期間、任何型態的觀光、休閒消費支出之統計推論。此一成果,亦可再擴大其應用層面,提供旅遊者、觀光政策的主管當局以及旅遊產業經營者的參考資訊,協助制定有效之經營管理策略。
It is of interest to make inference about the mean for a population consisting of different strata. Assuming that prior probabilities of strata and normality of the measure are given, traditional methods require that the equality of within-stratum variances be first examined. If the variances are equal, then use the exact t statistic; otherwise use the approximate t statistic. In this study, we have investigated the necessity of the equal-variance assumption for the exact t statistic by comparing the power of the two t tests. Based on extensive Monte Carlo simulation, it has been found that the equal-variance assumption can be ignored if prior probabilities are equal. In addition, the sample size of 50 is large enough to make the exact t test effective for the problem of inference about the population mean, regardless of the equality of within-stratum variances. A simple procedure has been recommended to facilitate statistical inference. The procedure, illustrated with Taiwan’s inbound tourism expenditure data, is particularly useful for the studies where stratified random sampling is conducted and inference for the mean daily expenditure is desirable. The procedure applies for any type of tourism expenditure during any time period. The resulting information is useful for tourism policy makers, the tourism industry, and tourists.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT008531810
http://hdl.handle.net/11536/77124
Appears in Collections:Thesis


Files in This Item:

  1. 181001.pdf
  2. 181002.pdf
  3. 181003.pdf

If it is a zip file, please download the file and unzip it, then open index.html in a browser to view the full text content.