Key points are not available for this paper at this time.
The statistical model for three‐mode factor analysis developed by Bentler & Lee (1979) represents a model for the covariance structure of random variables. In this model, arbitrary rescalings of the variables will destroy the covariance structure. A slightly modified model is proposed in this paper to handle the problem. A statistical development of the model is discussed based on the generalized least squares approach. Estimates, standard error estimates and a goodness‐of‐fit statistic are obtained via the Gauss‐Newton algorithm. It is shown that either the sample correlation matrix or the sample covariance matrix can be used for estimation. An example is presented to illustrate the theory.
Lee et al. (Tue,) studied this question.