Representing ordinal predictors in real estate valuation with multiple regression

Abstract

There are several different valuation methods used in real estate valuation. Statistical methods such as multiple regression and machine learning techniques such as ANN and Ensemble Methods are among the key ones implemented in recent studies. In all these methods the types of predictor variables for estimation of the price/value range from nominal, ordinal to continuous. Although the representation of nominal and continuous variables are agreed in prediction models, the representation of ordinal variables can change in every model. This paper provides a comparison of the different representations of the same ordinal variable in models. The findings indicate that representation of ordinal variables as continuous can be beneficial when the ordinal variable has many levels.