11-27-2013 03:09 PM
Hello. Does anyone happen to know if it is possible to look at a general regression model (multiple regression), and to determine from a full set of 100 independent variables, what are the 10 most highly important variables in the regression equation, and their associated relationships.
For example, I would like to be able to determine which variables are the most highly important, and also determine what type of relationship they have to the independent variable, whether it is linear, exponential, etc..
It has always been my general understanding that this basically can't be done in statistics and most models are based upon some "assumed" relationship between variables, meaning that the entire process is extremely manual. Unfortunately a request has fallen on me to automate this process, (I honestly am not sure if this is possible, I am kinda leaning towards it not being so), so I thought I'd come here for general ideas since this forum has always been so helpful in the past.
If what I am asking does not make sense please let me know!
11-27-2013 04:22 PM
Principal components can help to determine the variables that are the most important and I have seen someone who developed a full automated model building solution that I think factored in some of the above considerations.
If you drop me a line I'll provide his info to you, but for obvious reasons won't do it on here