- Joined
- May 2, 2002
- Professional Status
- Certified General Appraiser
- State
- Arkansas
Totally agree, again I try to keep the issues to a minimum. I have RealStat and you can input numerable variables but frequently this muddies the waters. But use of those dummy sets to check for Fireplace, etc. and that is why I say what I did about fireplaces....they simply don't show up consistently as an obvious value increaser and are much more likely to reflect a quality or age issue. Quality in that certian styles of homes seem to have fireplaces and in the cold clime of the late 1970s a much higher percentage of homes had a functional fireplace - often with doors, blowers, etc. Many older homes had shallow fireplaces that were decorative and most houses with fireplaces in the last 15 years in our area are either gas log or decorative.On a somewhat related note, use of binary variables (1 = Yes, 0 = No) should be used very sparingly, overloading a model with several different binaries can tend to cause autocorrelation and heteroskedacity. A good rule of thumb if you use binaries (or any variable for that matter) run the model with a variable. Run the model again without the varaiable...do the numbers change R2 better or worse. Sometimes the siginificance of a variable is just modeling off another variable...removal of a significant variable, that results in no or little change to R2 should tell you whether or not there is a lot of auto correlation in a model with that variable or another...ie bedroom count to GLA.