Hedonic Price Indexes: A Comparison of Imputation, Time Dummy and Other Approaches

The main approaches to measuring hedonic indexes in the academic literature are the imputation approach and the time dummy method. This paper compares the two approaches, discusses some alternative methods, and comments on an interesting recent contribution by Diewert, Heravi and Silver (2007). The aim is to explain the differences between the various hedonic approaches as well as their similarities, and to point to the implications for the use of hedonic regression by statistical agencies. Hedonic indexes can be weighted or unweighted. The paper addresses the issue of choice of regression weights, including quantity weights. For unweighted indexes it is shown that, by using ordinary least squares regression, both the 'full' hedonic imputation approach and the time dummy approach leave the observable matched part of the resulting price indexes implicitly unaffected, just like quality adjustment methods such as 'single' and 'double' hedonic imputation do explicitly.