Is it worth identifying service employment (sub)centres when modelling apartment prices?

The use of the attributes of the central business district and several subcentres instead of the characteristics of all the land parcels or zones can be seen as a higher level of analysis in real estate valuation. However, old technological limitations on considering smaller territorial units are being successfully overcome. The question is whether or not we still need generalisation, i.e. to identify urban centres when modelling real estate prices, or whether it is preferable to operate at a lower spatial level. The application of the traditional approach of identifying centres is compared with an 'objective' centrality index and a 'subjective' accessibility index calculated for each zone. The purpose is to find out which of the three concepts best fits a regression model of apartment prices and provides the best prediction. Both global and geographically weighted ordinary least squares regressions are used as well as spatial lag and spatial error models. We conclude that if a model is spatially weighted or the spatial effects are controlled, it is not that important which of the concepts is applied. Nevertheless, in most cases the highest predictive capacity is obtained with duocentric models.