Page 42 - Regression Guideline for AMC
P. 42
MODEL STABILITY
• We have not yet discussed two key final points: how many proper6es should be selected and how many predictors should be considered?
• The short answer to how many proper6es should be selected is that this depends on the number of predicted variables in the regression model. It is op6mum to have 40 proper6es per predictor. In our model of Orlando Florida, we analyzed 402 proper6es and had 7 predictors. This is well above the op6mum number needed, which would have been 280 proper6es. Our database of 402 proper6es would have been adequate to test up to 10 property characteris6cs.
• Why are that many proper6es necessary to model this number of characteris6cs? The answer, in a word, is shrinkage.
42
