Whilst we're on the subject and just to clarify Jeremy, is the sign still arbitrary if some loadings are negative are some are positive within the same factor? Or is this usually indicative of a problem such as lack of reverse scoring of certain items?
Kathryn
>>> Jeremy Miles <[log in to unmask]> 12/08/09 3:08 PM >>>
Kathryn is correct - the signs are arbitrary. If you want to 'flip'
every sign relating to a factor, you can. So take one factor, and
take each loading on it - change every positive to negative, and every
negative to positive. Then take each of the correlations of that
factor and the other factors, and flip those signs.
You can repeat for another factor too - you might find that some
correlations flip twice - that's OK. Sometimes minor changes in your
analysis or rotation cause every item to flip. You can also suggest
which parameters are positive or negative to start off with (although
that's difficult in SPSS) to try to make it end up with signs where
you want them. (You might want to do that if you want to compare
factor analysis solutions from different groups - like the same scale
in different languages).
Jeremy
2009/8/12 Kathryn Jane Gardner <[log in to unmask]>:
> So long as you have correctly reverse scored your items, as far as I understand the sign of the factor loading in factor space is somewhat arbitrary and I think different factor methods can in fact produce different signs. If you intend to use the resulting factors as scale scores, I think people usually deal with this by creating summated scale scores rather than the saving factor scores method(i.e., after reflecting those conceptually negative items, simply sum the items on each factor to create subscales) and the items will be in the correct direction here.
>
> Hope this helps.
> Kathryn
>
> Dr. Kathryn Gardner
> School of Psychology
> Darwin Building
> University of Central Lancashire
> Preston, UK
> PR1 2HE
> Tel: +44(0)1772 89 4463
> [log in to unmask]
> www.onlinepsychresearch.co.uk
>
>
> UCLAN Psychology: The best psychology graduate employment prospects in the North West (National Student Survey, 2008) Committed to world-changing teaching and research
>
> See www.uclan.ac.uk/scitech/psychology/ for more information
>
>>>> Caroline Wilson <[log in to unmask]> 12/08/09 8:47 AM >>>
> Hi,
> I've just done an exploratory factor analysis on some data (using principal axis factoring and oblique rotation) and all is well except that three of the factors are made up of items which are ALL showing loadings with negative values where there should be positive values. Apart from the negative sign the loadings are pretty much as I would have expected.
> I'm scouring my codebook to see if I've done something silly like re-code positive items instead of negative (so that they're all now negative) but am otherwise at a loss for why the negative sign.
> Has anyone come up against another reason for negative loadings?
> Many thanks
>
>
> Caroline Wilson
> PhD Research Student
> Institute of Energy and Sustainable Development
> De Montfort University
>
--
Jeremy Miles
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
|