探討成人暴力型與非暴力型自殺之相關因素：多層次與時空變異分析

Abstract

目的：自殺方式有其異質性，依據方法別的不同，暴力性的程度亦有相當的差異。本研究藉由空間資料分析，探討臺灣地區成年人口在1999~2007年間暴力型與非暴力型自殺率的空間群聚情形，並比較不同時期自殺率熱區在地理分布上的變異。進一步以多層次分析，探討自殺者的人口學特性變項、居住地的地區變項以及跨層次的交互作用，對其選擇暴力型自殺方式與否之影響。 方法：自殺死亡者之資料來源為行政院衛生署1999~2007年全國死因統計檔，為一橫斷式次級資料研究；地區層次變項資料取自2000年主計處戶口及住宅普查、內政部統計處戶籍人口統計年報、財政部綜合所得稅申報核定統計專冊，以及衛生署醫療機構現況及醫院醫療服務量統計等，以主成分分析萃取出高社經和家庭脆弱性兩項構念分數並進行LISA指標分析，作為後續多層次分析的地區變項。自殺率等空間資料，以全域型空間自相關(Moran’s I)和區域型空間自相關(Local Moran’s I)指標，進行群聚型態和熱區的分析。另外，本研究資料分為個體及地區兩個層級，故利用階層羅吉斯迴歸模型進行多層次分析，以瞭解鄉鎮(市區)變項對自殺者使用暴力型方式與否之脈絡效果。 結果：空間資料分析結果顯示，暴力型與非暴力型自殺率皆存在明顯的聚集型態。進一步檢視其熱區，兩者的分布不盡相同，且坐落位置隨自殺時期的不同亦有所變動，但熱區大致分布於台東縣、花蓮及屏東等地區，其中暴力型自殺率在北部亦存在有零散的熱區。多層次分析結果顯示，不同鄉鎮間使用暴力型自殺方式與否存在顯著差異，且自殺者之年齡、性別、婚姻狀況、工作有無和自殺時期，皆與其使用暴力型方式與否相關。此外，若自殺者居住之鄉鎮為(1)高社經熱區，對其選擇暴力型方式與否有顯著的正向直接效果；而在(2)高社經冷區有負向直接效果。自殺者選擇暴力型方式與否亦受到鄉鎮變項的跨層次交互作用影響。 結論：隨著年齡增加、女性、無工作以及春夏季的自殺者，較傾向選擇暴力型自殺方式；當居住地為高社經空間熱區，自殺者使用暴力型方式的傾向較高，高社經空間冷區則有相反的結果。此外，高社經與家庭脆弱性空間聚集兩地區變項，對自殺者使用暴力型自殺方式與否之關係存在跨層次的脈絡調節效果。

Objectives: The premimary purpose of this study is to investigate and compare the geographical variations of violent and nonviolent suicide rates in three different periods, by mapping across 350 townships in Taiwan. Secondary purpose, this study applies a multilevel logistic regression analysis to access the individual socioeconomic characteristics, township-level variables, and cross-level effects of suicidal deaths that use violent or nonviolent method and the people aged 20-64 during 1999 to 2007(n=27,970). Methods: Data files for suicides and undetermined deaths were provided by the Department of Health, from 1999-2007. To find out the changes in spatial patterning over three time periods, Moran’s I statistic and Local Spatial Autocorrelation (LISA) were used to test the spatial clustering/autocorrelation of violent and nonviolent suicide age-standardized rates. Multilevel logistic models including random coefficient model, intercept-as-outcome and slopes-as-outcome model were employed to access the relationship between township-level variables and the individual’s choice of violent/nonviolent suicide methods. Individual-level variables included gender, age, marital, employment status and suicidal periods. Township-level variables included high socio-economic and familial vulnerability hot/cold spots, and data were derived from the 2000 census of population and housing (Directorate-General of Budget Accounting and Statistics), Ministry of Finance (Financial Data Center), Ministry of the Interior (Department of Statistics), and Department of Health. Results: Both violent and nonviolent suicide rates demonstrated a significant spatial clustering characteristic for the periods 1999-2001, 2002-2004, and 2005-2007. High clusters (hot spots) of suicide rates were more frequently detected in part of rural Hualien, Taitung and Pingtung areas. Furthermore, the spatial clustering of high violent and nonviolent suicide rates was located at different towns during three periods. In addition, results from multilevel logistic models indicated that individual’s gender, age, marital and employment status and suicidal periods were associated to the choice of violent/nonviolent suicide methods. Moreover, after controlling for individual-level variables, suicidal person who lived in a high socio-economic cluster (hot spot) was more likely to commit suicide by violent suicide method. Additionally, person who lived in a low socio-economic cluster (cold spot) tended to use nonviolent method. Conclusions: As age increases, male, unemployed and suicides occured in the spring and summer had a significantly higher likelihood of commiting suicide by violent method. Additionally, compared to low socio-economic, suicidal person who lived in a high cluster of socio-economic also had a significantly higher risk to choose violent suicide method. Furthermore, cross-level interaction effect was observed in multilevel logistic models. The results demonstrate that, besides individual characteristics, suicide prevention strategies should also consider the socio-economic difference between urban and rural areas.