Measuring the performance of spatial interaction models in practice

Spatial decision support systems are widely used throughout the retail industry. As a result of this, a competitive industry has developed in order to provide such systems. One such company is GMAP Ltd who provide information systems for companies such as WH Smith, Halifax Pic and Toyota. Due to the competitiveness of the market analysis industry it is necessary for companies such as GMAP to endeavour to continually improve their products in order to remain at the leading edge of the decision support industry. Research is required in order to discover methods of improving the performance of the spatial interaction models that GMAP have developed loi theii clients. This thesis looks at two case studies of modelling work undertaken by GMAP for clients, the WH Smith model and the model for Halifax Pic new mortgage sales and tries to improve the performance of these models through a variety of methods.
Each of the three components of spatial interaction models, demand, supply and interaction, are analysed in turn and attempts are made to improve the representation of these components in the case study models. I he demand component is investigated using the WH Smith model and different methods of estimating demand using alternative data sources are investigated in order to discover if improvements can be made to the existing demand estimation procedure used by GMAP. I he representation of the supply side is investigated by attempting to identify centre and store characteristics that influence the attractiveness of destinations and subsequently include such variables in the attractiveness calculation. Several aspects of the interaction component are investigated in order to determine if they can be improved. I he measurement of impedance and the form of the impedance function together with alternative measurements of accessibility are investigated to see whether interaction patterns can be predicted more accurately.
The conclusions that arise from the investigations undertaken are presented as a series of recommendations for GMAP Ltd that can be implemented to improve the performance of their models. The implications for the specification of spatial interaction models in other contexts are also identified.