S=? ? '+D=? 00'? '+D=(? 0)(? 0)'+D=? *? *'+D.

Thus, regardless of the value of ? used, it is always possible to transform ? by an orthogonal matrix 0 to get a new ? * which gives the same representation for ? . Furthermore, since S is symmetric, there are p(p+1)/2 distinct elements in S, and in the factor representation model there is generally a greater number, p(m+1), of distinct parameters. So in general a unique estimate of ? is not possible and there remains the problem of identification in the factor analysis model. We refer to Anderson and Rubin (1956) for a detailed treatment of this topic.