Page 29 - Regression Guideline for AMC

Normal distribu6on of data
• Given these changes in the predictor values, we can righyully ask, which model is “right”? In a sense, both models are right for the data on which they are based. The first model includes several very expensive proper6es and tries to fit the data to explain the values of these proper6es as well as the others. In so doing, the model does not fit the more in-­‐range proper6es as well as the second model, which is only fiRng the data to a more homogeneous but s6ll varied group of proper6es.
• We can evaluate the overall fit of the models by looking at a sta6s6c that we will explain shortly, R-­‐square. We can also look at the 95% confidence interval around the example property to see if the new model more accurately predicts its last known sale price.
• The 95% confidence interval around the point es6mate is $394,702 to $329,611. The last sale price of the subject property is within this range. From the perspec6ve of this property, the model is predic6ng much be.er following removal of the sales price outliers.
• Note that we have achieved this level of accuracy by considering only a limited set of predictors. In actual modeling, we could consider up to 14 possible predictors, adjust for 6me of sales, property condi6on, etc. These addi6onal adjustments could poten6ally further improve model accuracy.
29