Interpret SAS "fit diagnostics for the response" graphs in regression results

[Hank]

Hi,

I hope there is anyone to answer this one..

When a regression is done, you get lots of free diagrams and histograms that come along with the results. I have attached two examples below. I went to a SAS course in statistics/anova/regression, but I never figured out how to interpret the lowest middle graph in the fit diagnostics table (attached in first file).

Is there anyone who can explain how to interpret these results? I.e. what do "fit mean" and "residual" show, in proportion to "proportion less"? what should the desired shape/form for the curve/line be? How should I interpret the results in the attached fil, as there seems to be a nice nonlinear line?

Most importantly, what remedies and actions should be taken to act on an eventual outcome that is not desired?

These results comes from a regression where I have hoped to rule out anything unwanted in an OLS regression (i.e multicollinearity, heteroscedasticity etc. The response has been box-cox transformed to get linear residuals- which normality test show). Therefore I am a bit worried about the curvature on the graphs.

Best regards,

Hank 

[Rick_SAS · Accepted solution]

There is a description of the Residual-Fit Spread Plot in the PROC REG doc:SAS/STAT(R) 9.3 User's Guide (search for "Residual-Fit")

Basically, you look at the heights of the two plots. If the left side (Fit) is taller than the right side (Residuals), then it means that the variables in the model explain a lot of the variation in the response variable.  If the right side (residuals) are taller, then it means that there is still a lot of unexplained variation.

[Hank]

Thanks a lot