0

I'm doing principal component analysis (PCA) on my nutrition data. Basically, I want to condense 43 food groups into fewer groups (aka, I want to define 2-3 patterns based on the 43 food groups). The problem is, my Kaiser-Meyer-Olkin (KMO) measure is very low (~0.35) but my Barlett's test for sphericity remains significant (<0.0001). Could I still proceed with PCA or do I have to delete a few food groups based on the Anti-image correlation matrix?

Peter Flom
  • 94,055
  • 35
  • 143
  • 276
Meo
  • 163
  • 2
  • 12
  • See related point 3 in my [answer](http://stats.stackexchange.com/questions/38709/skewed-variables-in-factor-analysis) – ttnphns Nov 30 '12 at 06:41
  • 2
    Why are you doing PCA? If, in subsequent analyses, you just want to be able to work with fewer variables that still explain most of the variance in the data, it's simply a transformation to a new set of variables that doesn't make any assumptions that need to be tested. If you're using it as something akin to factor analysis, making inferences about latent variables, then that's a different matter - I can't answer. – Scortchi - Reinstate Monica Nov 30 '12 at 11:45
  • Welcome to the site. I removed your signature line because the system adds it automatically. – Peter Flom Nov 30 '12 at 12:21
  • 1
    I'm a little confused as to what you are trying to do. You say you want to condense 43 groups into fewer groups. What are these food groups? Are the things such as "vegetables", "red meat", etc? You also say you want to "define 2-3 patterns based on the 43..." but this does not seem equivalent to the first task. What sort of patterns? – Peter Flom Nov 30 '12 at 12:24

0 Answers0