發刊日期/Published Date |
2010年10月
|
---|---|
中英文篇名/Title | 電話調查之抽樣設計:一項實驗性的研究 Sampling Designs in Telephone Survey: An Experimental Study |
論文屬性/Type | 研究論文 Article |
作者/Author | |
頁碼/Pagination | 7-42 |
摘要/Abstract | 本文旨在探討不同的抽樣設計對電話調查資料的影響,研究的議題主要有二:一、若採用同一種抽樣方法,但使用不同電話號碼資料庫作為抽樣底冊(sampling frame),是否會在訪問結果、樣本代表性及訪問成本上產生明顯的差異?二、若採用不同抽樣法,但使用相同的電話號碼資料庫,對於前述三頂是否也會造成影響?本研究所使用的抽樣底冊分別為是電腦輔助電話訪問(CATI)系統中內建的90年中華電信公司電話號碼資料庫,以及中華電信公司93-94年度全國住宅電話簿。在進行實地調查訪問時,是以同一問卷為基礎,分成三個專案執行:使用CATI系統之電話號碼資料庫之「舊電話簿隨機抽樣專案」(OED)、採用中華電信公司93-94年度全國住宅電話簿之「新電話簿隨機抽樣專案」(NED)及「局碼隨機抽樣專案」(APD)。 就抽樣方法來說,三個專案均是採用分層兩階段等距抽樣方式(stratified two-stage systematic sampling),第一階段都是以全國二十五個縣市作二十五個分層,並根據行政院內政部提供之民國九十三年度人口統計資料,計算台灣二十五個縣市人口數,以等比例原則計算出各縣市所需抽取之人數。惟三個專案在第一個階段的抽樣方式不同:OED及NED是分別利用各自的電話號碼資料庫以等距抽樣原則抽取各縣市所需之住宅電話號碼數,且為了降低因涵蓋率因素而造成部分電話號碼的用戶無法被抽中之問題,遂再利用隨機原則進行電話號碼尾數未四碼的選取,即結合電話號碼簿及RDD之綜合抽樣法(list-assisted RDD)。至於APD則是統計住宅電話簿中各縣市電話局碼的登記比例,以該比例作為排序的依據,運用等距抽樣法來抽取各縣市所需之電話局碼數,並搭配末四碼隨機而得。 在控制了訪員因素、訪問時間及訪問內容等可能影響訪問結果的因素後,本研究發現,以抽樣底冊的比較結果而言,NED具有比較好的執行效率;雖然合格受訪者比較少,完訪率卻比較高。其次,在比較不同的抽樣方法上,若同樣採用93-94年度全國住宅電話簿為基礎,而使用不同抽樣方法來比較執行效益來看,APD的表現則比結合電話號碼簿及RDD之綜合抽樣法要來得好,因為在整體的訪問結果、調查成本及樣本代表性都非居於最差的情況下,APD較其他二者具有更近乎均等的機率抽樣,可說是較優的抽樣設計。 This research aims to compare the results, costs, and the level of representation of computer-assisted telephone interviewing (CATI) techniques under the following conditions: (1) the same sampling approach utilizing different frames; and (2) different sampling approaches utilizing the same frame. An experimental design was devised specifically for the study. Two sampling frames were employed and both included residential numbers listed nationwide with RDD design. One of the frames was an old electronic directory (OED) accompanied by the CATI system. The other frame was a new, updated electronic directory (NED). The area code-prefix combinations in the NED frame with RDD (APD) were used as an alternative sampling approach. Six questions regarding quality of life were applied to examine the possible differences between sampling designs and between sampling frames, in addition to questions concerning the sociodemographic characteristics of the respondents. Results indicated that the efficiency of the NED design was the greatest, followed by APD. The OED design had the lowest efficiency though only slightly behind the other two. In practice, however, APD is superior to the other two because it provides a probability sample. Regional representation was achieved by all of the three sampling designs, although the results of quality-of life indicators were found to differ among them. |
關鍵字/Keyword | |
學科分類/Subject | |
主題分類/Theme | |
DOI | |
檔案下載/Download |