<
<        I read what followed but you were terribly non-commital about the
<main point I made. Do you agree that each of the concepts in a real 6
<concept model (sparse and the  step-3 of the four-step) need not account for
<as much covariance as each of the corresponding six factors in a factor
<analysis with the same items? Do you agree that, the fact that the sixth
<factor in factor anlaysis explains little covariance does NOT imply that any
<one of the six factors in the sparse and theoretically postulated six
<concept model MUST explain correspondingly little covariance?

The issue in factor analytically driven studies is dimensionality and
what dimensionality contains the covariation. What you do may not fit that
paradigm, because you often work with so few indicators of a latent
variable that it would be hard to overdetermine them for a factor analysis.
This is not at this point a criticism, just an observation.  I do think I
have learned something in contemplating the situation of 5 factors raised
by you, that it would be quite possible (in fact it must be common) for
a smaller number of factors to appear to be sufficient to account for
the common variance in a set of manifest variables, when in fact, the
more appropriate model would require more. This would be case where the
common factors have high correlations with other common factors.  So
the manifest variables dependent on them will seem to reside mostly in
a smaller dimensional space. Thus if you were merely willing to settle for
a good approximation to the data, you might miss out on requiring more
factors.  This will certainly happen to those who believe most factors
should be orthogonal to one another. We certainly can't work with that
idea in SEM. (I leave the implication of that to the student to work out).

Anyway, I have other flame wars to attend to....

Stan Mulaik