Using spatial dependence to examine marketing data

The dissertation consists of two essays that use the concept of spatial dependence to explore marketing data. The essay focuses on the fact that retail chains with geographically dispersed stores would find it useful to segment their market geographically when designing programs to maintain and enhance customer satisfaction amongst their customers. Geographical segmentation of the market could be based on clustering regions (e.g., zip codes) such that each segment would be homogeneous with respect to the importance of the key drivers of satisfaction. However, implementation of segment-based strategies would be more actionable if the regions within a segment were also contiguous. This research develops and tests a methodology that spatially segments a market in terms of customer characteristics, such as the relative importance of the key drivers of satisfaction while aiming for a high level of contiguity among regions assigned to the same segment. The methodology uses the concept of spatial dependency among the regions and attempts to balance spatial contiguity and homogeneity within segments. A Monte Carlo simulation demonstrates that the SDSM outperforms competing methodologies for spatial segmentation over a wide range of conditions. The empirical application shows the benefits of the proposed methodology and discusses the managerial implications of the results. The second essay focuses on two objectives which are to identify the percentage of beds that all the hospitals in an hospital network should devote to the various types of diseases that they are catering to and to identify those regions where new hospital beds should be added to increase social welfare. Hospital beds are used as a proxy for resources that are at the disposal of the hospital like doctor specializations and medical equipment. The study uses data on various disease incidences in the greater Los Angeles region along with the current location and the size of the hospitals that currently address these disease incidences. Results from our analysis show that regions that are socio-economically backward seem to have been given relatively low importance in the determination of the current hospital network.