冤案的成因探究:以羅吉斯回歸為基礎

Abstract

從近年來的冤案研究可知，單一個案的成因探究，以及單一、多項因子的歸因已無法了解冤案的發生全貌，因此本文借鏡文獻上以對照組案件作為控制組的研究設計，並輔以「法實證研究方法」、「Firth 羅吉斯回歸」以及「雙變項相關分析」之方式，以了解哪些是釀成既成冤案的獨特性因素，以及造成判決結果歧異的原因為何?冤案因子間是否存在一定程度之交互作用，以及其對判決結果之影響為何?本文共篩選出 25 件既成冤案樣本、70 件 near miss 案件樣本，以及 70 件正確判決樣本。

首先，本文藉由文獻回顧及細究判決理由之方式，共擇定 7 項冤案因子: 「虛偽自白」、「錯誤證詞」、「科學或科技證據的瑕疵」、「指認錯誤」、 「證據不具同一性」、「錯誤的論理法則」及「錯誤的經驗法則」，並透過共線性分析及 Cramér’s V 相關性檢定，篩選出最後欲放入回歸模型的冤案因子。從模型結果得知，「錯誤證詞」為貫穿整個刑事司法程序的冤案因子，「虛偽自白」 及「科學或科技證據瑕疵」則為區辨案件從正確判決走向冤案，以及案件從 near miss 案件走向冤案的關鍵因素。接著，本文並透過雙變項相關分析得出，「虛偽自白」與「科學或科技證據瑕疵」，以及「指認錯誤」與「科學或科技證據瑕疵」這兩組冤案因子間存有中度相關性，並在既成冤案中呈現一定程度之交互作用。

依此，本文認為，案件走向冤案歧路係源於「補強法則難以發揮作用」、「法院對科學或科技證據的獨斷」以及「對供述證據信用性判斷之恣意」，並提出應從「強健補強法則之內容」、「揭示科學證據之論證過程及潛藏風險」、「重建供述證據信用性之判斷」以及「建立證據隔離措施」，作為未來防免冤案之方向。

Research on the causes of miscarriages of justice in recent years has revealed that exploring individual cases or attributing errors to single or multiple isolated factors is insufficient to grasp the full complexity of how wrongful convictions occur. Accordingly, this study adopts a research design that utilizes a control group of "near miss" cases for comparative analysis. Supplemented by Empirical Legal Research (ELR) methods, Firth Logistic Regression, and Bivariate Correlation Analysis, this thesis aims to identify the unique factors that lead to established wrongful convictions and the reasons behind disparate judicial outcomes. Furthermore, it explores the extent of interactions between various "innocence factors" and their impact on case dispositions. The sample size for this study consists of 25 established wrongful convictions, 70 near miss cases, and 70 correct conviction cases.

Initially, through a literature review and a detailed examination of judicial reasoning, seven innocence factors were identified: false confessions, erroneous testimony, flaws in forensic science or technology evidence, eyewitness misidentification, lack of chain of custody (identity of evidence), erroneous logical reasoning, and erroneous rules of experience. Following collinearity analysis and Cramér’s V correlation testing, the final variables were selected for the regression model. The results indicate that "erroneous testimony" is a pervasive factor throughout the entire criminal justice process. Meanwhile, "false confessions" and "flaws in forensic science or technology evidence" are the critical determinants that distinguish correct convictions from wrongful ones, as well as near miss cases from actual miscarriages of justice. Bivariate correlation analysis further demonstrates a moderate correlation between "false confessions" and "flawed forensic evidence," as well as between "eyewitness misidentification" and "flawed forensic evidence," showing a significant degree of interaction in established wrongful convictions.

In conclusion, this study argues that the divergence toward a miscarriage of justice stems from the ineffectiveness of the corroboration rule, judicial arbitrariness in forensic interpretation, and discretionary abuse in judging the credibility of statements. To prevent future wrongful convictions, this thesis proposes several reformative directions: strengthening the substance of the corroboration rule, disclosing the deductive processes and inherent risks of forensic evidence, reconstructing the criteria for assessing statement credibility, and establishing evidence segregation measures.