Kohonen self-organizing feature maps
Correct usage of the maximum likelihood method criteria described above for both the traditional and alternative confirmatory factor analysis to identify the values of free model parameters and estimate the model goodness-of-fit measure needs testing multivariate normalcy of distributions of either observed variables or residual vector components. This procedure is laborious and frequently impossible because of deficiency in observed data.
To overcome this problem a new technique that uses the capabilities of self-organizing feature maps (SOFM), or Kohonen networks, is proposed.
Advantages of the proposed technique in estimating goodness-of-fit measures are the following:
no need to test multivariate normalcy of distributions of either observed variables or residual vector components;
simple procedure of estimating type 2 statistical errors is available;
it is possible to reveal the most probable percentage component-wise structure of statistically significant deviations for the pseudosolution residual vector;
higher reliability of obtained goodness-of-fit measures because of unrestrictedness of generated random samples of variances and covariances ingressed in the pseudosolution and the following unlimited goodness-of-fit estimation accuracy.