提高社區面訪調查回應率之可行策略與成效分析

目標：探討依據戶籍抽樣之社區面訪調查，若針對未居住於原抽戶籍地址樣本加以追蹤，能否提昇調查回應率及改善完訪樣本代表性，並估算所需投入行政成本。方法：資料來源為2008年「台灣第十次家庭與生育力調查」，該調查以設籍台灣20至49歲婦女為訪查對象，以內政部戶籍資料為抽樣母體清冊。運用描述性統計方法，計算不同追蹤程度回應率，再以類別變項比例檢定，比較原抽戶籍地址完訪與追蹤完訪樣本之背景特性差異。結果：若僅就居住在原抽戶籍地址樣本進行訪查，整體回應率僅4成5左右，就未居住於原抽戶籍地址樣本追蹤訪查，整體回應率可提高至7成以上，投入之行政成本只相對增加6.3%。原址完訪與追蹤完訪樣本在年齡、居住地區以及經濟狀況等，均具統計顯著差異，驗證樣本居住於原抽戶籍地址與否具選擇性。結論：針對未居住在原抽戶籍地址樣本追蹤訪查，為提高社區面訪調查回應率與改善完訪資料代表性之可行且具成效策略。

關鍵字

面訪調查；回應率；選擇性偏誤

並列摘要

Objectives: This study had two specific aims. The first was to determine if the tracing of household interview survey samples could improve response rates and data representativeness. The second was to estimate the proportion of increased administrative cost arising from sample tracing as compared with the proportion of increased response rate. Methods: The data used for analysis were from the ”Women's Health and Fertility Survey” conducted in 2008 by the Bureau of Health Promotion. Household registration data managed by the Ministry of the Interior were used as the sampling frame and all females aged 20 to 49 were eligible. We used a descriptive statistical method to calculate response rates by level of administrative input and used the Chi- Square test to detect significant differences in background characteristics between those who were residing at the registered address and those who were not. Results: Results showed that only a 45% response rate could be achieved without tracing those who did not reside at the registered address. By tracing those who were not residing at the registered address, the response rate increased, remarkably, to greater than 70%, whereas only a 6.3% increase in administrative cost was required. There were significant differences between those who were residing at the registered address and those who were not in terms of age distribution, urbanization of residence, and income. Conclusions: We concluded that survey results were biased if those who did not reside at the registered address were not traced. To improve the response rate and data representativeness, household interview surveys that use household registration data as their sampling frame need to trace samples of those who do not reside at the registered address.