Factor analysis question

Factor analysis question

Post by Mousi » Tue, 22 Jul 2003 12:40:25



This may be a stupid question, but I can't seem to find the answer anywhere.
When reporting the results of a factor analysis, specifically the percentage
of varience the factor solution accounts for, do you report the extraction
sums of squares % of varience or the rotated % of varience?  Intuitively, I
think it should be the rotated % of varience, but I would be most grateful
for some input here.

Also, with factor analysis when looking at the loadings on each factor, some
of the items load on more than one item.  What questions should I be asking
if this happens?

Thanks very much

 
 
 

Factor analysis question

Post by Rich Ulric » Wed, 23 Jul 2003 06:30:42


On Mon, 21 Jul 2003 03:40:25 GMT, "Mousie"


> This may be a stupid question, but I can't seem to find the answer anywhere.
> When reporting the results of a factor analysis, specifically the percentage
> of varience the factor solution accounts for, do you report the extraction
> sums of squares % of varience or the rotated % of varience?  Intuitively, I
> think it should be the rotated % of varience, but I would be most grateful
> for some input here.

The percent initially extracted is the one that says something
about the quality of the factoring.  For a common-factor
interpretation, the total fraction should be about the same as
the reliability.  You look at the cumulative percent to see that
it is 'reasonable,' and you look at the 'scree'  at the  to see
where the contribution drops stops.

Quote:

> Also, with factor analysis when looking at the loadings on each factor, some
> of the items load on more than one item.  What questions should I be asking
> if this happens?

"Are these items more ambiguous than they need to be?"
"Are these cross-loadings sensible?"

You do not say what it is that you are factoring.  If they are
questions that you have composed, then you have a chance
to refine them.  

If there are a lot of cross-loadings that I did not expect,
then  I figure that the sample N  is probably too small,
given the amount of structure ('correlations') that existed;  
and this lack of delineation could be a consequence.
(Then I would drop a bunch of variables to see if the
other loadings become cleaner.)

--

http://www.pitt.edu/~wpilib/index.html
"Taxes are the price we pay for civilization."  Justice Holmes.

 
 
 

1. Factor analysis question

Hello

We have a data set which includes, for each of 96 cities, 4 estimates
of the number of drug injectors in that city.  Each of these estimates
contains error.

We ran a factor analysis to determine their commonality.

We now wish to use the results of this analysis to come up with a
better estimate of the number of drug injectors in each city.

It seems intuitively reasonable to multiply the standardized scoring
coefficients by the estimates, add these together, and then divide by
the sum of the standardized scoring coefficients.  I've even seen this
done.  But I haven't seen a good proof that this is correct (it may not
be correct!)

Any advice on how to proceed will be appreciated.

Thanks in advance

Peter L. Flom, PhD
Assistant Director, Statistics and Data Analysis Core
Center for Drug Use and HIV Research
National Development and Research Institutes
71 W. 23rd St
New York, NY 10010
(212) 845-4485 (voice)
(917) 438-0894 (fax)

2. Driver for Mita DP-570 for Win2k

3. Factor Analysis Question

4. Xmh questions

5. Factor Analysis Questions

6. Transferring files

7. Please ignore/excuse exploratory factor analysis question

8. Sprite modes, don'tcha just love 'em!

9. factor analysis question

10. Factor Analysis Question

11. factor analysis: calculate factor scores?

12. Calculating factor scores for new dataset after factor analysis

13. Three questions about factor analysis (scree, Bartlett, cumulative over 1.0)