被忽略的相關性

The Forgotten Correlation

2024年總統大選前，「藍白合」的呼聲高唱入雲。然而，藍白雙方對如何以民調高低決定正副總統候選人，各持己見、互不相讓。本文在不違背藍白「六點共識」的前提下，對虛無假設：「侯柯配」的支持率（大於）等於「柯侯配」的支持率，與對立假設：「侯柯配」的支持率小於「柯侯配」的支持率，建構出一個顯著水準約0.05的檢定規則。我們的研究顯示，必須考慮民調中兩者支持率之間的相關性才不至造成錯誤的推論。我們也討論了如何適當地估計上述相關性，並在此基礎上推導出符合統計學理的檢定規則。

Before the 2024 presidential election, calls for a Kuomintang—Taiwan People’s Party (KMT-TPP) coalition reached unprecedented levels. However, both sides of the alliance maintained firm positions on how to select the presidential and vice-presidential candidates based on polling data. In this study, while adhering to the KMT-TPP“Six-Point Consensus,”we propose a testing rule with a significance level of approximately 0.05. Specifically, the rule tests the null hypothesis that the“Hou-Ko pairing”support rate is at least as high as that for the“Ko-Hou pairing”against the alternative hypothesis that it is lower. Our findings indicate that accounting for the correlation between the poll support rates is essential to avoid erroneous inferences. Additionally, we propose a method for reasonably estimating this correlation and, based on that, derive a testing rule that conforms to sound statistical principles. The proposed approach ensures a fair and objective candidate selection process, with potential applications in broader political decision-making scenarios where survey-based comparisons play a critical role.