Page 33 - Regression Guideline for AMC
P. 33
R-‐Square
• While there is no threshold above which we can defini6vely say the R-‐square value indicates an acceptable or very good predic6ve model and below which we would reject the model as being too inaccurate, we do want R-‐square to be as high as possible. All else being equal, higher values of R-‐square indicate be.er model fit, a stronger associa6on between the predictors and the dependent variable, and more accurate predic6ons.
• Consider the next two graphs which depict the regression model predic6ons and actual sales prices for proper6es in Orange County, California where the regression R-‐square was . 26 (26% of the sales price accounted for by the model) and the graph we saw previously for the proper6es in Orlando, Florida where the R-‐square value was .47 (47% variance in sales price accounted for by the model).
33
