臺灣地區都市化因子對竊盜犯罪影響及其區域變異：全域性與區域性迴歸分析

Exploring Influence and Spatial Heterogeneity of Urbanization Factors toward Thefts in Taiwan: Global and Local Regression Analysis

本研究嘗試運用屬於全域性（global）之多元線性迴歸與區域性（ local） 之地理加權迴歸分析方法， 針對由社會解組理論與日常活動理論相關實證文獻中所抽離之包括「人口密度」；「土地利用」；「居民流動」；「居民所得」；「貧富差距」等與都市化相關之巨觀因子模型，對竊盜犯罪率可能造成的影響進行分析，瞭解相關因子於我國竊盜犯罪之可解釋性， 並探討各因子間於警察分局（ 類似於鄉；鎮；市；區）層級之區域變異現象。本研究主要發現如下：
一；在空間探索分析部分，各變項之數值分布呈現區域差異，變項數值高者大致多集中於北；中；南三大都會區周邊區域，顯示臺灣地區竊盜犯罪率與都市化過程中產生之易犯罪結構因子多集中於都會區。
二；區域性迴歸較全域性方法具有較佳之模型解釋力，並能有效衡量資料於空間之變異，對於臺灣地區竊盜犯罪之解釋力則有南高於北之趨勢。
三；都市化因子模型中，「人口密度」與「總遷移率」等變項對於臺灣地區竊盜犯罪之解釋上具有顯著性，且均為正向關係，對於臺灣地區竊盜犯罪率之影響具有相當之區域性差異。
四；運用區域型迴歸分析方法對於犯罪因子分析結果較全域性分析更為精緻與實用，具有解釋影響因子區域性差異之優勢，故可提供相關單位制定因地制宜與權責明確化之犯罪預防策略。

Basing on official data of thefts and socioeconomics, the study first cited “urbanization model” generated from empirical researches related Social Disorganization Theory and Routine Activity Theory which formed by macro factors of population density, population of service industry rate, total population mobility rate, per capita comprehensive income, and variation coefficient of income as major concepts. The study then tried to explore influences and spatial heterogeneity of urbanization model toward thefts, and verify explanation of Social Disorganization Theory in police precinct district level area of Taiwan by using global (Multiple Linear Regression) and local (Geographical Weighted Regression) regression model. The major findings of the study are listed as below:
1. In the aspect of explanatory spatial data analysis, the values of most variables are not evenly spatial distributed in every district.Generally speaking, the higher value areas tend to concentrate in main metropolitan areas of north, middle, south, and east divisions of Taiwan, one the other hands, the lower value areas tend to distribute in mountainous, insular, or agricultural areas. These findings are supported by the literature primarily, which suggests that the higher crime risks of structural factors are concentrated in metro areas.
2. In the aspect of model analysis, the values of Local R2 are not evenly spatial distributed in every district like 0.363 in global model.Further, the higher Local R2 value districts tend to concentrate in the middle and south parts of Taiwan.
3. In the aspect of significant factors analysis, The global relationship between factors of the population density and total population mobility rate with risk of thefts are both significantly positive, suggesting that the more population density and mobility of the area, the more the relative risk of thefts. However, the local analysis results show that the contribution of the two factors has changed over the study regions, and opposite influence between south and north part of Taiwan.
4. Further, the study tried to explore the initial reasons of spatial heterogeneity distribution of local model R2 values and significant factors. We then propose that since local model can provide a useful method for describing the influence and spatial heterogeneity of the relationships between crime risks and neighborhood contextual characteristics of urbanization factors; aside of the sophistication of spatial statistics, the analysis methods also provide us with not only more realistic but also refined evidence-based results for local adaptation and accountability crime prevention policies making.