【摘要】 本文将EM方法应用于含有欠测数据的因子分析的最大似然估计问题,利用EM法统一处理了为保证估计唯一性的种种条件下的最大似然估计问题,得到了相当统一的迭代算法。这种处理方法可以简单地将根据欠测样本估计正态分布的期望向量与协方差阵的问题及回归问题作为其特例。从而使建立具有泛用性的计算程序成为可能。更多还原

【Abstract】 Most of the current procedures for multiple factor analysis are concerned with complete observations, and to obtain MLE by the ordinary Flettcher-Powell method or Newton-Raphson method. From the practical point of view, however, usual observations contain missing values and occasionally restriction exist among parameters in multiple factor model. The present paper is to propose a systematic procedure to obtain the maximum likelihood estimation (MLE) under the general necessity described above. The procedure is fundamentally based on Dempster-Laird-Rubin’s EM algorithm, and gives an excellent property to obtain iteratively and surely the MLE with monotonous approach from initial values.更多还原