Confirmatory Factor Analysis in SPSS

Hello,

So I haven't used SPSS in years, my boss can't seem to understand the
SAS output for confirmatory factor analysis. So I am trying to do this
in SPSS. Is there a way to do Confirmatory factor analysis in SPSS
without AMOS? I have been googling instructions on how to do it for
the last hour and it keeps pointing me to AMOS, which we don't have
and I can't install due to admin rights.

Thanks
Tan

 

Look up the FACTOR command in the help menu. I'm not positive whether
this comes with base SPSS or not though. In my version (17) if you go
to Windows -> Dimension Reduction -> Factor is the access to the GUI
menu.

 

Thanks, i have saw that. but I think that is regular factor analysis.
I don't know much about it. But the test that I need is Confirmatory,
which I didn't see an option for confirmatory

 

If you are trying to do a formal confirmatory factor analysis to see if
a sample is from a possibly distinct population, then you need AMOS or
need to check if there is an R program you can call from SPSS.

If you are trying to confirm at the level of seeing if the same scoring
key would be used, you could just limit the factoring to the same number
of factors, and check that the loading are clean and the items are
grouped together the same way.

Why does your boss want the confirmatory factor analysis?


Art Kendall
Social Research Consultants

 

Sadly we do not have AMOS. But we do have R, which I do not know how
to use. By any chance do you know of anyplace i can look for that?

I was able to do CFA in SAS but the researchers can't seem to read and
understand the output. I only recently learned about CFA via textbook.

so basically we are using CFA to validate 2 constructs that were
identified in a previous study using exploratory factor analysis. This
second study is looking at a different population. We have like 24
questions and the two constructs looks at 1) self-efficacy and
2)stress eating. based on the statisticians, CFA is the right
approach.

 

CFA is the "right" approach.

Before things like AMOS were available we did things the way I suggested.

Trying that approach first should take a very short amount time.
Knowing that your specifying retention of two factors does or does not
result in the same scoring key is a start.

You can also see if you get similar results by plotting both sets of
eigenvalues against the eigenvalues from doing parallel analysis on each
of the data sets.

When you say the "two constructs" do you mean the underlying constructs
in writing the items or do you mean that these are the interpretations
after the first factor analysis?

Are you intending to use summative scores is further work, or is the CFA
the main point of your work?

Art Kendall
Social Research Consultants

 

Is it possible to tell by plotting the eigenvalues of 1 dataset? I
only have access to 1 dataset and not the other.

the two constructs are the two scales we are hoping to form with the
24 questions (1) self-efficacy and
2)stress eating).

We will most likely run a logistic regression on the 2 scales we will
form based on that.

 

If those were the constructs which you used to write the original items,
what stopping rule did you use in the exploratory factor analysis?



Did the previous study report the eigenvalues?
The only other thing you need from that study is the number of cases and
number of variables. With that you generate the eigenvalues you would
obtain from say 10,000 random sets of data the same size. You then
compare those to the eigenvalues of the actual data.


Are the numbers of cases very different in the two studies? If the
number of cases is about the same in the two studies and you have the
same set of items you may only need to run 1 parallel analysis.

How many of the original set of items did you retain for the second study?


Art Kendall
Social Research Consultant

 

Google is your friend. Googling on <confirmatory factor analysis r> too
me here, for example:

http://www.statpower.net/Content/312/Handout/Confirmatory Factor Analysis with R.pdf

James Steiger is very well-known and trustworthy, by the way.

 

"Before things like AMOS were available we did things the way I
suggested."
Damn Art, you must be REALLY old. I was slamming LISREL code around
on big iron back in the mid 1980's (35+ years ago). Shame that SEM is
*STILL* a relatively ARCANE path ;-(

 

"the researchers can't seem to read and understand the output."
Probably wouldn't understand the output from *ANY* other SEM program
either unless they are familiar with the technique. Time to hit the
books?
--

 

I'm not ready to be old yet. I started doing scale development factor
analysis recently - 1971.

I don't really know what the OP was trying to do. I have found few
instances when all or the original items intended as measures of a
construct came out as clean measures of the construct. In fact when I
started out it was a rule of thumb to have about 150% of the number of
items that one hoped to end up with. Post hoc, it often was obvious why
an item did not load cleanly.

In my experience, finding different groupings of items than originally
intended often gave insight into important distinctions.

Art