04-23-2013 11:42 PM
In excel, I have used a second-order polynomial forecast which is fit to intercept the Y axis at 10,000. Excel reports an R^2 of 0.25 for this relationship.
I'm trying to replicate this in SAS in order to observe significance levels/standard errors with the following model:
proc reg data=data2;
model revenue=xr xr2 /white spec;
This model outputs parameter intercepts identical to Excel's. Unfortunately, the R^2 is insane at 0.9963.
SAS results report: "Restrictions on intercept. R-Square is redefined". How should I be interpreting the R^2 then? Also, is there any way to output a graphical output which exhibits the confidence bands off the base of the data?
04-24-2013 03:34 AM
I would like to hear what is the mean of revenue. My thoughts are that if it is far from 10000 than it can make sense.
Basically the R^2 is a measure which tells you how much more of variance of your variable REVENUE can be explained by your model compared to the model with intercept only.
I would say the R^2 in excel will be estimated using the model revenue=intercept + resid for comparison (which is estimated by average of revenue),
on the other hand the SAS will use model revenue=10000+resid for comparison and since this one could be pretty weak than the valu 99% additional fit can make sense.
04-24-2013 08:45 AM
If it were me, I would subtract 10,000 from my Y variable and use the NOINT option on the model statement to consider lines that pass through the origin. (But then I'd ask myself, "why am I considering only no-intercept models.)