I need to dichotomize four attributes in larger than typical datasets that I normally utilize, i.e., 100-200 cases (rows) but now need to analyze data of 2100 x 200 with UCINET but first want to dichotomize four continuous attributes before proceeding.? So I reduced the dataset to just 2100 x 4 attributes (only the continuous ones), converted to 1-mode, and then tried Dichotomize Interactive to look at the correlations. After 18 hr of? running with the spinning wheel in UCINET I gave up and repeated with only 2 attributes with same problem. I am running Windows 10 64bit with 64GB RAM (although UCINET 32bit seldom uses more than 4GB) with AMD Ryzen CPU, dual fast SSDs,? and a good graphics card.
?
I looked to see if Pajek if could do this but did not see that option. For now I using an a priori approach which I feel is not optimal but works, i.e., forcing the creation of only 2 clusters via different methods. I find that mixing continuous data with binary gives distorting results based on the scale of the continuous data after sums-squared reduction (even with normalizing and with better scale matching).
?
Any ideas would be appreciated otherwise I will continue with the a priori method that was covered by Steve Borgatti in the excellent technical document (().? Thanks.
