Factor Analysis....

#1
:confused:

Hi everyone,

I have a couple of FA questions for my thesis..if anyone is able to offer some guidance it would really be appreciated.

My research is looking at the prevalence of depression in a specific group of parents, and the factors that are related to whether the parents are depressed or not. It is hypothesised that certain factors like attachment to their child, spousal support, whether their expectations of parenthood are met, and their social support (amongst others) will all impact on depression levels.

I have managed to measure depression levels via a realiable and valid scale that is well researched.

However, the attachment and parental expectations scales have previously only been used with another very different parent population, therefore I believe that I need to check the factor loadings etc when they are used on my parent population before utilising them in the study.

Here is my problem....

Because my particular parent population is quite restricted, I have only managed to gain 50 participants (with no chance of any more due to timeframes). Is it possible to factor analyse scales with such a small population? The attachment scale is exactly the same as the one that was previously used on a different parent population, but the expectations scale has had some items removed and some added in order to make it appropriate for my participant group.

When I try to run the factor analysis I am getting 10 factors on the expectations scale (I was expecting 4) and 8 on the attachment scale (was expecting 3 or 4). Could this be due to small sample size? Is there any way around this?

Any help would be greatly appreciated!
Thank you.
 

Xenu

New Member
#2
Some of the factors are most likely just a result of random noise. You should only take into account those factors that explain most of the varaiance. As a rule of thumb, don't include factors with eigenvalue less than one, as it doesn't explain more than the individual original variables. Also, you can plot the eigenvalues (scree plot) for some guidance. A suggestion is to not include the variables after the "knee" in the plot. This is just guidlines though.

Of course, you may exclude those factors that you don't have any reason to believe is related to depression. Use only those that are relevant for your study.

And yes, your sample size may give you trouble, but hopefully that won't be insurmountable (and factor analysis migh not even be absolutely necessary).
 
#3
You can also run your correlation coefficient for the set of expectation variables and look for convergent and discriminant values. In other words, check to see if those variables that should be highly correlated are, in fact, highly correlated and those that should not be correlated are not.

Jenny Kotlerman
www.statisticalconsultingnetwork.com
 
#4
If I were you, I'd "force" a 3 or 4 factor solution (whichever you have theoretical rationale for). In SPSS, click on the Extraction button in the Factor analysis window. The default is for it to show eigenvalues greater than one, but you'll see a place where you can indicate how many factors you want. Then, your stats program will show you factor loadings for however many factors you request. If the loadings are substantial (e.g., above .30 at minimum) you can go with that solution. 50 people might cause a few issues with factor analysis, but nothing too serious. Good luck!
 
#5
You should note in your text that with the relatively small sample size that the results may not generalize to other samples. But if there is a literature to back up your conclusions, you should be OK. Remember that the Kaiser criterion (eigenvalue > 1) is calculated from the data. For example, if you had ten more participants it might change the number of factors that have eigenvalues > 1. So if I were worried about sample size I would do either someone else says and force SPSS to give a certain number of factors.