時間序列民調資料的分析：Samplemiser的運用與評估

Time Series Analyses on Opinion Survey Data: The Method and Evaluations of Samplemiser

崔曉倩(Hsiao-Chien Tsui) ； 吳重禮(Chung-Li Wu)

《台灣政治學刊》 10卷1期 (2006/06) Pp. 39-67

https://doi.org/10.6683/TPSR.200606.10(1).39-67

摘要

運用「時間序列」(time series)資料進行分析時，經常面臨的難題在於，不同時間的數據所呈現的「波動性」(volatility)，可能來自於抽樣誤差。鑑此，本文旨在引介美國政治學界所發展出的Samplemiser統計方法，藉以區分抽樣誤差與實際波動。首先，本文援引初步範例，說明Samplemiser的適用性與重要性。其次，Samplemiser運用Kalman filtering和smoothing的估計方式，對於時間序列資料進行分析與預測，本文針對其統計邏輯進行說明。再者，本文依據TVBS民調資料，以2004年總統大選陳水扁與連戰的支持度進行實證分析。經過Samplemiser處理後的民調資料顯示，兩組候選人支持度的平均變動，皆比處理前改善許多；這表示資料中因抽樣誤差所導致的變化已被排除，更能反應出實質的民意趨勢。資料亦證實，「自我迴歸參數」(autoregressive parameter)-表示前後兩期實際支持率之關係－的設定，會影響分析結果。儘管Samplemiser仍有若干限制，由於Samplemiser 4.0線上網路統計軟體已經相當便利，操作甚為簡易，值得引薦。

關鍵字

時間序列分析 ； 民意調查 ； 抽樣誤差 ； Samplemiser

並列摘要

When researchers analyze time series survey data, it is of importance to distinguish random sampling error from volatility of time varying. This study aims at introducing a statistical method of Samplemiser developed by U.S. politics scientists for distinguishing genuine movements in public opinion from random movements produced by sampling error. In this work, we first explain the applicability and importance of Samplemiser by a preliminary example. We then present the methodology of Samplemiser, the Kalman filtering and smoothing algorithm, and empirically apply this approach to the TVBS time series opinion survey data of Chen Shui Bian and Lien Chan in the 2004 Taiwan presidential election. The findings indicate that the smoothing algorithm reduces random sampling error in survey data, which implies that the data throughout the smoothing algorithm accurately gauge public opinion trends. The results also show that the estimates for autoregressive parameter (by which last period's opinion affecting current opinion) influence the accuracy with which public opinion may be forecasted. We conclude that Samplemiser with the wed-based statistical software is an available and beneficial approach although it still has some limitations.

並列關鍵字

time series analysis ； survey research ； sampling error ； Samplemiser

參考文獻

黃紀(2005)。投票穩定與變遷之分析方法：定群類別資料之馬可夫鍊模型。選舉研究。12(1)，1-37。

吳重禮(2002)。民意調查應用於提名制度的爭議：以1998年第四屆立法委員選舉民主進步黨初選民調為例。選舉研究。9(1)，81-111。

Abramson, Paul R.,Charles W. Ostrom.(1991).Macropartisanship: An Empirical Reassessment.American Political Science Review.85(1),181-92.

Abramson, Paul R.,Charles W. Ostrom.(1992).Question Response.Wording and Macropartisanship: American Political Science Review.86(2),481-86.

Abramson, Paul R.,Charles W. Ostrom.(1994).Question Form and Context Effects in the Measurement of Macropartisanship: Response.American Political Science Review.88(4),955-58.