- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
I have a data set of 3784 participants (no missing data) who answered 16 General Human papillomavirus (a disease) knowledge items. As an example (HPV is rare) These items were asked as true, false and I don't know. I want to examine how many factors should be retained. I conducted an exploratory factor analysis (see below).
My results indicate 1 factor (by looking at eigenvalues >1, and the scree plot). This is great, as we were hoping for unidimensionality of this scale
I am wondering if I should try a rotation ? I believe a rotation is not appropriate since I have 1 factor
Thanks,
proc
factor data=WORK.A method=principal nfactor=16 heywood scree;
var Q28A Q28B Q28C Q28D Q28E Q28F Q28G Q28H Q28I Q28J Q28K Q28L Q28M Q28N Q28O Q28P;
run
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Hi Statsgirl, you don't need to rotate if you have one factor but if you do you might find the results to be only slightly different than your unrotated solution.
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Since all of your values fall into one of 3 categories, PROC FACTOR may not be your best choice for analysis. Take a look at PROC CORRESP for correspondence analysis, and in particular, you might find Example 34.1 Simple and Multipple Correspondence Analysis of Automobiles and Their Owners as a good analogy to your data.
Correspondence analysis avoids the assumption that the data are continuous (or at least ordinal) and thus might be a better fit.
Steve Denham