next up previous contents
Next: Compound Analysis by combining Up: Computational Information Technology Previous: Learning Scene and Object

Building Spatial Choice Models from Aggregate Data

Researchers: Tommi Orpana and Jouko Lampinen

This study belongs to the project ``Data-fusion and neural networks in complex models'' financed by TEKES and participating enterprises.

Traditional spatial choice modeling is based on detailed knowledge of the choices made and the prevailing conditions. Such an approach usually demands an expensive and time-consuming process of obtaining the information from questionnaires. The present study is aimed at developing spatial choice models and inverse parameter estimation methods which enable the use of aggregate (statistical) data, resulting in considerable savings in time and money.

The particular application studied is one of grocery store choice, the aim being to infer the distribution of the consumers' annual spending among the stores in an area from objectively measurable quantities, for example GIS data.

The models are of the multinomial and hierarchical logit types, being thus based on consumer utility functions which contain factors thought to be relevant to a store choice decision. The parameters reflecting the effect of the factors are estimated using maximum likelihood and Bayesian methods.


 
Figure 6
Figure 6: An example analysis of grocery store demand distribution from the Kouvola region. Based on the store locations (black dots), the population distribution (colored house icons) and the total sales of the stores, the developed system estimates the utility parameters for different store types. In the figure the red box with a white asterisk shows the largest hypermarket in the region, the colors of the house icons indicating the predicted share of money spent in that hypermarket of all the funds used in grocery stores. The width of the map area is about 70 km.


next up previous contents
Next: Compound Analysis by combining Up: Computational Information Technology Previous: Learning Scene and Object
www@lce.hut.fi