I would grateful for some advice. I have inherited a project in which the
dataset consists of 52 variables and 780 results. The variables are
clinical indicators, and are a mixture of binary and ordinal responses.
There are no missing values. The dataset covers 7 tumour types. The object
is to select those variables that "best" characterise each type and then
use these to develop a "scale" to predict tumour type. I have considered
cluster or factor analysis and even discriminant analysis based on random
sampling from the dataset. Unfortunatley my stats degree did not include a
real world problem such as this. Can anybody recommend any particular
technique or suitable references? I have been using Everitt as my prime
reference.
Many thanks,
Edith
|