Multiple regression explores and quantifies the relationship between two or more components of known and available data (sale prices and property characteristics) to generate a market value. In essence, this method uses aspects of both the paired sales and cost approach methods by determining which property characteristics are the primary contributors (cost approach) and the amount they contribute (paired sales).
Regression does not require strict similarity between property sales because it estimates the value contribution (coefficient) for each attribute using a "goodness of fit," or error-minimizing technique. This produces statistics about the quality of the attribute contribution that the other methods cannot provide. These statistics help evaluate the predictive accuracy of the regression equation, or essentially, the ability to predict sales price.
Multiple Regression Statistics
Several of the confidence measures that are most useful during the analysis include the coefficient of determination, or R2, standard error of the estimate and t-value. The adjusted R square provides a total score, or measure, for all the property characteristics used in the model. Depending on the data type, R2 ranging between the mid-.80s and .90s indicate the model can explain a high percentage of the variables. The standard error of the estimate indicates the level of variation in the "goodness of fit." The t-value is an individual score for each property characteristic and its significance in explaining sale price.