● 中文摘要：

　　本文以矯正機關現行「毒品再犯危險性評估表」中所列10個因子為基礎，結合文獻中所發現對毒品再犯有影響的18個因子，利用再犯預測模型，對女性毒品受刑人出獄後的再犯情況予以預測，希能提升機關現有再犯預測效率。 本研究以97至100年間高雄女子監獄出監毒品受刑人1,029人為樣本，追蹤至100年底再犯罪情形，考慮兩組危險因子，第一組為僅利用10個因子的「評估表因子」，第二組為包含全部28個因子的「整體因子」，分別運用Logistic迴歸分析及存活分析，除篩選影響再犯及再犯間隔時間危險因子外，更進一步預測其出監後可能再犯機率及再犯間隔時間。 實證結果發現，以評估表因子運用Logistic迴歸進行預測，再犯篩檢率為58.4%；若以整體因子預測，再犯篩檢率提升至73.3%。以存活分析四種常用機率分配預測再犯間隔時間，中位數預估結果遠優於平均數，依評估表因子預測，四種機率分配預測再犯時點與實際再犯時點誤差日數小於60天及小於180天個數，占實際再犯個案比例最高為2.5%、9.6%；若依整體因子預測，誤差日數小於60天及小於180天個數，占實際再犯個案比例提升至10.2%、27.3%。

● English Abstract:

     The paper constructs recidivism prediction models for women drug prisoners, using the 10 factors evaluated in "drug recidivism risk assessment form" by correctional institutions and the 18 factors studied in the literature. With the new recidivism prediction model, I hope to help improving the prediction accuracy of women drug prisoners’ recidivism. The sample in the paper includes 1,029 drug prisoners released from Kaohsiung Women''s Prison between 2008 and 2011. All criminal records are traced until the end of 2011. Two sets of potential risk factors of recidivism are considered in the paper. The first set only contains the factors in the evaluation form, and the second set includes all relevant factors. Using Logistic Regression Analysis and Survival Analysis, the effects of potential risk factors on recidivism are examined. I also predict the probability and the time interval of recidivism. Using the Logistic regression model with the risk factors only in the evaluation form, 58.4% of recidivism can be correctly predicted. While extending the set of potential risk factors, the screening rate of recidivism can be enhanced to 73.3%. The median forecast results are far superior to the average forecast in Survival Analysis. With the potential risk factors in the evaluation form, the difference of predicted recidivism date and the actual date is less than 60 days and less than 180 days in 2.5% and 9.6% of sample respectively. With all relevant risk factors, prediction, the share of sample whose difference of predicted recidivism date and the actual date is less than 60 days and less than 180 days are significantly improved to 10.2% and 27.3% respectively.

● 文章連結：

http://handle.ncl.edu.tw/11296/ndltd/18379617716587886935

● 資料來源：

臺灣博碩士知識加值系統