Skip to content

Categorize subjects using PCA

2 messages · Dr.Vinay Pitchika, Michael Dewey

#
Hello everyone,

I have been analyzing dietary pattern data obtained via FFQ. We performed
Principal Component Analysis and decided to extract 3 dimensions (healthy,
unhealthy and mixed diet) and to adjust this in linear regression model
along with various other variables. However, our statistician was not happy
with the approach because the 3 dimensions we adjust do not have a measure.

So, I would like to classify the patients in our dataset using the PCA
results. Is it possible to classify all the patients into one of the 3
groups mentioned above?

Thank you
Vinay
#
If you want to discover groupings in your data-set I would have thought 
latent class analysis was a more principled solution. There are several 
packages for doing this.

If you want more detail about this you might be better asking on 
stats.stackexchange.com for statistical help. When you do be sure to 
define what you mean by 'do not have a measure' as it is not, to me at 
least, clear what s/he means.
On 26/01/2017 13:20, Dr.Vinay Pitchika wrote: