A study of discrete choice model with latent variables in apartment selection


Student thesis: Master's Thesis

View graph of relations


  • Sze Ning Luck MUI

Related Research Unit(s)


Awarding Institution
  • Hing Po LO (Supervisor)
Award date4 Oct 2004


This study investigates the deterministic factors that influence consumer's residential choice under a discrete choice framework. Purchasing a housing unit involves consideration of a number of factors. The physical characteristics of dwellings and the neighborhoods in which they are located are usually the essential consideration factors. Traditional residential models solely include observable dwelling attributes and household characteristics as explanatory variables, but relatively little attention has been directed to the significant impacts of consumer's attitudinal factors to the decision. Being different from previous residential choice models, our model integrates latent variables to measure consumer's perceived satisfaction of housing and neighborhood amenity in a hope to obtain a deeper understanding of consumer's choice behavior. Factor analysis is first carried out to estimate the psychological latent variables. The estimated latent variables are then incorporated in the utility function of the residential choice model to explicitly describe consumer's perceptions of residential satisfaction and their impacts on . . making housing choice. This dissertation presents a nested logit model describing household choice of an apartment unit among three new housing estates in the Tseung Kwan O district in Hong Kong. People are presupposed to first choose a desirable estate, and finally purchase a specific dwelling within the chosen estate. It is a hypothetically realistic behavior representation of the decision making process when consumers select an apartment. In making housing decision, the number of dwellings available to be chosen is considerably large. When the choice set gets too large, model estimation becomes cumbersome, both computationally and in data management. McFadden's (1978, 2003) sampling-of-alternatives method offers an appealing way to resolve computational constrains in model estimation when the number of conceivable alternatives is immense. Model estimation can be performed by randomly selecting alternatives from the full choice set without sacrificing the consistency property of estimates. The empirical analysis suggests that most of the dwelling attributes have statistically significant impacts on apartment selection. Also importantly, household's attitudinal factors on residential amenity are found to have strong influences on the decision. The inclusion of attitudinal factors in the residential choice model provides a more realistic representation of the choice behavior, and consequently contributes to deeper understanding of consumer's needs for planning and developing housing estates.

    Research areas

  • Housing forecasting, Mathematical models, Hong Kong, China, Residential real estate, Consumer behavior, Consumers' preferences