05-05-2012 12:11 PM
I am running a regression where I get weak results, I know there should be relationship there but the OLS is not picking it up in my data.
Unfortunately I have some negative observations so I cant use natural logarithm to strengthen these results...
Does anyone else know what I could do to my observations to strengthen the r squared?
05-05-2012 08:03 PM
Have you looked at the distributions of your various variables? Is the relationship linear? Are the variables normally distributed (log is NOT the only transformation)? What is your N?
05-05-2012 09:06 PM
As a first step, you should look at your data, make some graphs. This should help you answer Art's questions. Then, keep with your hypothesis "I know there should be a relationship there" and look for outliers. Redo your regression with only the variables that form the basis of your hypothesis but use the ROBUSTREG procedure. It should detect any obvious outliers.
Another possibility you should consider is that the relationship that you suspect might be present only in a subset of the data or that it takes a different form (intercept, slope) in different subsets.
05-07-2012 01:26 AM
In addition, Check the COOK's distance to see which observation is highly influencing your model.
And check the independant variables whether there are som multi-colinearity .