Peck R C, Arstein-Kerslake G W, Helander C J
Department of Motor Vehicles, Program and Policy Administration, Sacramento, California 95818.
J Stud Alcohol. 1994 Nov;55(6):667-78. doi: 10.15288/jsa.1994.55.667.
The primary objective of this study was to assess the extent to which drunk-driving (DUI) recidivism and DUI treatment program compliance could be predicted from psychometric, biographical, drinking history and prior-driving-record variables. These analyses were performed on data from 7,316 DUI offenders initially collected in Sacramento County, California, from September 1977 through January 1981. For most analyses, the recidivism measure was a composite of major convictions (DUI, reckless, hit-and-run), nighttime (6 PM-6 AM) and alcohol-related accidents during the 4-year interval following treatment assignment. The prediction of recidivism was highly significant for both the construct sample and the 25% cross-validation sample. The predictive accuracy was low, however, as evidenced by multiple Rs of < .30. The predicted rates of recidivism generated for each individual by the regression equation were cross tabulated by other criteria of interest, including total accidents and total injury and fatal accidents. Offenders at high risk of recidivating had substantially higher rates of accidents. The results indicate that reasonably accurate prediction of recidivism is only possible for discriminating between offenders at the extremes of the recidivism expectancy distribution. The above approach was also used to isolate factors predictive of program compliance (successfully completing treatment). In all cases, the prediction of compliance was highly statistically significant. In general, compliance was much more predictable than was subsequent DUI recidivism. Those offenders having a high probability of being noncompliant were much more likely to recidivate and have accidents than were those with favorable compliance expectancies.
本研究的主要目的是评估从心理测量、个人经历、饮酒史和既往驾驶记录变量中能够预测酒后驾车（DUI）累犯及DUI治疗项目依从性的程度。这些分析是基于1977年9月至1981年1月在加利福尼亚州萨克拉门托县最初收集的7316名DUI罪犯的数据进行的。对于大多数分析而言，累犯指标是指在治疗分配后的4年期间内，重大定罪（DUI、鲁莽驾驶、肇事逃逸）、夜间（下午6点至上午6点）及与酒精相关事故的综合情况。对于构建样本和25%的交叉验证样本，累犯预测均具有高度显著性。然而，预测准确性较低，复相关系数小于0.30就证明了这一点。通过回归方程为每个个体生成的累犯预测率，根据其他感兴趣的标准进行交叉制表，包括总事故数以及总受伤和致命事故数。累犯风险高的罪犯事故发生率显著更高。结果表明，只有在区分累犯预期分布两端的罪犯时，才有可能对累犯进行合理准确的预测。上述方法还用于找出预测项目依从性（成功完成治疗）的因素。在所有情况下，依从性预测在统计学上都具有高度显著性。一般来说，依从性比后续的DUI累犯更具可预测性。那些有高概率不依从的罪犯比那些有良好依从性预期的罪犯更有可能累犯和发生事故。
