Page 47 - Regression Guideline for AMC

MODEL STABILITY
• Using a small number of cases on which to run our analyses also affects sta:s:cal power. This concept refers to the ability to detect significant effects for the property characteris6cs analyzed. With a smaller number of proper6es in the analysis, some predictors that are actually important to the market value of proper6es in a given area might be assessed by the regression model as being non-­‐significant. If so, we might not adjust the comparables by these characteris6cs, thereby also contribu6ng to a less accurate es6mate of the value of the subject property.
• To illustrate these points, we randomly selected samples of 50 and 100 cases from the 402 Orlando proper6es and ran the same regression model on these subsets of proper6es in the same market.
• The following graphs show that as sample size decreases, the proper6es are more spread out around their predicted values as indicated by the regression lines. This means the regression model produced larger residuals for the es6mated property values and less accurate predic6ons.
47