A more appropriate method might be to perform a factor analysis and based on its results, select the variables that fall in the most significant category.
I'm struggling about a statistics problem regarding combining several individual dependent variables into one.
I'm doing Repeated ANCOVA for behavioural data (numerical). However, since some variables may be working together, I want to combine them and create a new dependent variable. The problem is that those variables are not necessarily significantly correlated and taking the mean seems not a very good solution.
So my question is, in this case, what is the best way to combine those variables? Is taking the interaction term better? Do I need them to be significantly correlated?
Thanks very much!
Hi, many thanks for your reply. I was also thinking about this. But I don't know if it is because the sample size is small or if it is because there are not much difference between the individual variables. There are finally two components and it is mostly due to the different body sections (eyes movs are clustered together, hands movs together, etc). Thus I don't know if it is still the best option to do so. Also I wonder if I shall take the weighted mean of the individual variables on the basis of the PCA. Any further suggestions? Or if you know where can I find more information, please let me know. Thanks a lot!
If number of variables and cases are too many then perhaps you should keep all variables in an analysis like a mixed model and then by eliminating variables one by one try to maximize the -2log-likelihood ratios...this results in the best fit to your data.
Take a look at:
-2log restricted likelihood