协方差结构模型不收敛原因及解决方法

Analysis on Cause and Solution of Non-convergence for Covariance Structure Model

协方差结构模型实际上是一般线性模型的扩展,它主要利用一定的统计手段对复杂的理论模式加以处理,并根据各种拟合指数对模型作出检验与评价,从而达到证实或证伪事先假设的理论模型的目的。笔者在利用协方差结构模型对北京市居民住房消费行为和意愿进行量化研究时发现协方差结构模型存在不收敛问题,文章提出,导致模型不收敛的原因,一是缺失数据处理方法不当,可采用期望最大化算法(EM算法)和马尔科夫链蒙特卡罗法(MCMC算法)处理数据缺失;二是变量间存在多重共线性,可去掉设置不合理的潜变量以避免共线性问题;三是模型过于复杂,收敛条件苛刻,可调整模型使之简单化,并重新设定收敛条件,促使模型收敛。

The author aims to develop covariance structure model to research purchasing behaviors and intentions on housing consumption. The causes mentioned in this paper about non-convergence for model are unreasonable methods to process missing data which can be solved by EM and MCMC algorithm, multicollinearity among variable and too strict convergence conditions for model solved by re-adjustment of them.