Confirmatory v Exploratory Factor Analysis (Principle Component Analysis)

[Orlando Villella]

Could you help shed some light on a situation please?

We have an instrument that was used in an organizational setting and has
since been adapted for use with a different population. The confusion seems
to be that since a CFA was done in the first studies with this instrument
in an organizational setting, that CFA is appropriate with the adapted
insturment for use with families. Or, would Expoloratory Factor Analysis be
more appropriate since we don't know at all if this instrument would be
useful with a dramatically different population?

Any help you could offer would be most welcome.
Orlando

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Maguin, Eugene]

Orlando,

Perhaps there are different opinions on this but it seems that you'd like to
show that the original instrument works well in a new population (families).
My opinion is that you should do a CFA. Actually, a series of tests using
CFA. You might find a reference such as Rex Kline (Principles and practice
of structural equation modeling) or Barbara Byrne (she has done several
books, tailored to specific SEM packages) useful. Without being too
specific, you'd expect a high degree of item relevance and appropriateness
in the new population and the same items loading on the same factors in the
new population. If the results of the original CFA were at hand, you might
test whether 1) the factors loadings and 2) the item error variances were
plausibly the same between the two samples. If there are multiple factors,
you could then look at equality of factor variances and covariances. And,
lastly, factor means.

Gene Maguin

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Maguin, Eugene]

Orlando,

I'd like others to see the conversation because I know there are other, very
experienced folks who would have a view on this question.

That said, I think that doing a principle components analysis (PCA) is not
the thing to do. I think that PCA assumes a different underlying model than
the common factor model, is normally done with 1's in the diagonal and
assumes orthogonal factors. I'd use a common factor solution (PAF) with
iterated communalities and an oblique rotation to simple structure. The
oblique rotation is particularly important to use unless you believe that
the factors are truly uncorrelated with each other in the population.

I can't comment on the parallels between an organization in crisis and a
family in crisis, except to say that since people are involved in each there
might be some similarities. I think the item relevance and appropriateness
test is a critical, nonstatistical test. If you judge that the items
adequately represent the family crisis process, I would go straight to CFA
and would forget the factor analysis.

Gene Maguin


>>Gene -
I thought that CFA might have been appropriate at first, but then it made so
sense to me to theoretically conclude that an organization (in crisis) would
react the same way as a family (in crisis). They seem to be two completely
different situations, so an instrument appropriate for an organization might
not work so well for a family in crisis. After running the PCA, there was
nothing to suggest that the instrument in its current form would work well.
By working well, I mean to say that the appropriate questions loaded on the
appropriate subscales. The questions themselves seemed to be loading on many
different factors rather than the three subscales comprising the original
instrument. What do you think? Is further analysis with CFA needed? What
might that tell us?
Thanks for your input.
Orlando

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Swank, Paul R]

I agree with Gene. First of all, there is a known structure and it would
be helpful to know if it fits in this other population and EFA won't
tell you that. If it doesn't fit, then you can begin exploring, either
with the CFA or going to EFA. I would also agree with Gene that
principal components is not the way to go for the exploratory analysis.
If you are interested in the factor structure rather than the
components, then I would say go with principal axes methods, maybe even
maximum likelihood EFA. I myself am particularly fond of Mplus for this
work, but any major package, including SPSS, would also do it.

Paul R. Swank, Ph.D
Professor and Director of Research
Children's Learning Institute
University of Texas Health Science Center
Houston, TX 77038


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Gene Maguin
Sent: Monday, January 26, 2009 2:21 PM
To: [hidden email]
Subject: Re: Confirmatory v Exploratory Factor Analysis (Principle
Component Analysis)

Orlando,

I'd like others to see the conversation because I know there are other,
very
experienced folks who would have a view on this question.

That said, I think that doing a principle components analysis (PCA) is
not
the thing to do. I think that PCA assumes a different underlying model
than
the common factor model, is normally done with 1's in the diagonal and
assumes orthogonal factors. I'd use a common factor solution (PAF) with
iterated communalities and an oblique rotation to simple structure. The
oblique rotation is particularly important to use unless you believe
that
the factors are truly uncorrelated with each other in the population.

I can't comment on the parallels between an organization in crisis and a
family in crisis, except to say that since people are involved in each
there
might be some similarities. I think the item relevance and
appropriateness
test is a critical, nonstatistical test. If you judge that the items
adequately represent the family crisis process, I would go straight to
CFA
and would forget the factor analysis.

Gene Maguin


>>Gene -
I thought that CFA might have been appropriate at first, but then it
made so
sense to me to theoretically conclude that an organization (in crisis)
would
react the same way as a family (in crisis). They seem to be two
completely
different situations, so an instrument appropriate for an organization
might
not work so well for a family in crisis. After running the PCA, there
was
nothing to suggest that the instrument in its current form would work
well.
By working well, I mean to say that the appropriate questions loaded on
the
appropriate subscales. The questions themselves seemed to be loading on
many
different factors rather than the three subscales comprising the
original
instrument. What do you think? Is further analysis with CFA needed? What
might that tell us?
Thanks for your input.
Orlando

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[John Fiedler]

I think I may be even more worried about Windows updates than SPSS updates.
I removed ver 16 of SPSS and ran with 15 until the patch of 17 was
available. I spend a day getting rid of Vista 64 and going back to XP.

Has anyone run SPSS under the W7 beta with any success?

Thanks as always,

JOHN

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD