A Quantitative and Empirical Study of Sentencing Disparities in Aggravated Fraud Cases through Statistical and Machine Learning Approaches

• In Event: Judicial Discretion in the Sentencing Process 2

Abstract

In response to long-standing concerns about inconsistencies in Taiwan’s sentencing practices, this study focused on aggravated fraud, aiming to elucidate how various legal parameters shape the final prison terms. To quantitatively investigate this important subject, bridge this gap, we compile a comprehensive dataset of 1,617 verdicts in 2019, coding each case for the presence or absence of statutory aggravating, mitigating provisions factors. We then apply both inferential statistical methods as well as machine learning approaches to investigate the significance of these factors through a sentence prediction model. We found that aggravating provisions lead to an average increase of roughly one month in prison terms compared to cases without them. By contrast, mitigating provisions led to dramatic reductions, cutting sentences by nearly 50% for eligible defendants. To further evaluate the consistency of sentencing practices, we constructed a sentencing prediction model designed to capture how these statutory components influence judicial decisions. Achieving a Mean Absolute Error (MAE) of 1.84 months, the model displayed a close alignment between predicted and observed actual sentences. However, cases without either aggravating or mitigating circumstances saw a slightly higher MAE of 1.99 months, implying that judges might rely on additional subjective considerations without explicit statutory cues. Particularly noteworthy was the category of cases containing both aggravating and mitigating factors — 72.73% are classified as outliers, reflecting considerable unpredictability when judges must weigh conflicting legal considerations.

Authors

• Yun Tiao Lin, National Tsing Hua University

• Meng-Chi Lien, National Tsing Hua University

• Daw Wei Wang, National Tsing Hua University