摘要
由于交通事故数据在记录与获取方面的问题,发展中国家学者进行事故频次建模中往往受到小样本量问题的约束。为扩大样本量,传统广义线性模型将每年事故频数作为独立变量分析,但无法考虑每年事故频数间的时间相关性,估计结果存在偏差。采用广义估计方程进行小样本交通事故数据的事故频数建模,并与传统的广义线性模型估计结果进行了对比。结果表明,传统的广义线性模型低估了说明变量系数的标准误差,论文采用的广义估计方程考虑了交通事故数据间的时间相关性,对于预测变量对于交通事故频数的影响估计更加准确,因此适合小样本交通事故数据的事故频数建模分析。
Due to incomplete records in crash reporting systems,researchers in developing counties are often confronted with small sample size in crash frequency modeling.To expand sample size,traditional generalized linear models(GLM)treat crash count in each year as a separate observation,and therefore they cannot account for the temporal correlation among crashes from different years,which could bias the model estimates.This study evaluates the application of generalized estimating equation(GEE)to crash frequency modeling based on small-size crash data.The results of GEE are compared to those of GLM.Study results show that the traditional GLM underestimate the standard errors of the coefficients for explanatory variables.The GEE with an exchangeable correlation structure successively captures the temporal correlation in the longitudinal data used in this study and therefore is considered to outperform the traditional GLMs in crash frequency modeling.
出处
《交通信息与安全》
2013年第5期123-126,130,共5页
Journal of Transport Information and Safety
关键词
交通安全
事故频数
广义线性模型
广义估计方程
时间相关性
traffic safety
crash frequency
generalized linear model
generalized estimating equation
temporal correlation
