International Choice Modelling Conference, International Choice Modelling Conference 2017

Font Size: 
Dwelling and Residential Location Choice Models on Complete Market Data
Brett Smith, Doina Olaru, Stephane Hess, Matt Beck, Charisma Choudhury

Last modified: 28 March 2017


Buying a new house or apartment is one of the most difficult decisions most people face in their life, and one they generally only make a few times. A residential location decision considers the dwelling type, the attributes of the location and the tenure type. It is consequently no surprise that the modelling of these decisions has received growing attention by choice modellers over the last few years. A key issue however remains the availability of the right type of data. Efforts based on stated preference data are prone to hypothetical bias and an oversimplification of the real complexity of the choice people make. A number of authors have looked at using household survey data that incorporates a residential location dimension, but such data generally limits an analyst to study the choice of a specific residential area rather than a dwelling level context. Furthermore, a key issue arises in defining the set of available alternatives. An analyst will not generally know what other properties would have been available to the decision maker at the time he or she made their choice.

Analysts have to a certain extent ignored this issue and focussed instead on choice set generation, i.e. recognising that it is not likely that each household will consider all alternatives and will form their own limited choice set. One way analysts have inferred the choice set is by way of ruling out alternatives that vary greatly in terms of price, dwelling structure or location from the observed sale. The resulting set of alternatives is known as the evoked set (Rashidi, et al., 2012), but the process does sill not reveal the choice set. Further refinement of the assumed choice set for modelling may employ the two stage probabilistic model developed by Swait and Ben-Akvia (1987). This approach models choice set generation via a constraint-based non-compensatory process and then employs utility maximisation compensatory process for the actual choice. Kaplan, et al. (2011) provide an example of such an approach in a residential choice context wherein they use real-estate website search behaviour to identify threshold which are used to determine which alternatives are retained in generation potential choice sets.

A key question we ask is how these efforts at choice set generation are affected by or mask the issues in the data in relation to the global choice set. This availability set is all houses on the market for a period of time in which the household is examining the market. Our inquiry into housing choice makes use of a complete set of transactions on residential properties for the city of Perth, Western Australia for the period of September 2015-September 2016. The data contains almost 8,000 observations of the selling price, date of transaction, attributes of the land and dwelling. The important feature of the data set is that we have the average time on the market for each of 380 suburbs of Perth, permitting a reasonable assumption on which houses were on the market at the time of the purchase as well as the changes in the availability set in the period leading up to the sale. Furthermore, the data set is enhanced by spatial characteristics of the built, social and natural environment. Our analysis tracks the availability set over a series of time windows leading up to the date of the observed transaction and test robustness of the assumptions made in terms of availability. As is generally the case, the advantages of our data are also offset by some limitations in that we do not know the characteristics of the household or where they are moving from. An additional component of our work is thus to test the impact of assumptions about choice set formation (i.e. subset of the overall choice set). In this context, we are particularly mindful of endogeneity issues as our only potential insights into the preferences of the household (which can then act as a proxy for their search criteria) is the actual choice they have made. As with other studies, we also do not observe cases where someone searches for a property but does not make a purchase. As a final component in our work, we contrast the insights we gain from the choice models to results from spatial hedonic regressions.



Kaplan, S., Bekhor, S., & Shiftan, Y. (2011). Development and estimation of a semi-compensatory residential choice model based on explicit choice protocols. The Annals of Regional Science, 47(1), 51-80.

Rashidi, T. H., Auld, J., & Mohammadian, A. K. (2012). A behavioral housing search model: Two-stage hazard-based and multinomial logit approach to choice-set formation and location selection. Transportation Research Part A: Policy and Practice, 46(7), 1097-1107.

Swait, J., & Ben-Akiva, M. (1987). Incorporating random constraints in discrete models of choice set generation. Transportation Research Part B: Methodological, 21(2), 91-102.

Conference registration is required in order to view papers.