We have a new update on The Future of the Penny Arcade Forums.

Statistics - Multicollinearity

SeptusSeptus Registered User regular
edited March 2010 in Help / Advice Forum
I'm trying to do regression analysis on a large number of variables and use those results to cull most of those and leave myself with just 15-20. Unfortunately, when using my program, Stattools, the regression is halted because some of the variables have exact multicollinearity. The program won't tell me which ones. Should I not be able to identify them by checking for correlation between the variables? When I ran a correlation table, all I saw was the standard diagonal of 1.000 correlations between variables and themselves, and no 1.000 correlations with any others. If that doesn't work(it didn't when I tested a variable's correlation with its log transformation), are there any other tests I could run to identify these?

PSN: Kurahoshi1
Septus on

Posts

  • Zombie HeroZombie Hero Registered User regular
    edited March 2010
    Look into Principle Component Analysis.

    Zombie Hero on
    Steam
    Nintendo ID: Pastalonius
    Smite\LoL:Gremlidin \ WoW & Overwatch & Hots: Gremlidin#1734
    3ds: 3282-2248-0453
Sign In or Register to comment.