Spatial autocorrelation -- the idea that objects which are nearby are more closely related than those that are further apart -- has gained wide acceptance in the field of geostatistics.  This approach lends itself very useful in modeling health service utilization patterns and the demand for health services. This is because the demand for health services is geographically clustered because communities which share similar income, education, economic status, awareness of government programs through media exposure are likely to be clustered. At the same time  factors determining the supply of health services such as the type of health facilities, access to skill-birth attendants  among other factors are also likely to more similar at the community-level.  

One limitation with using a regular regression to model community-based variables which can predict health service utilization is that it cannot capture spatial autocorrelation. This results in estimates which are biased and impacts the standard errors. Our paper seeks to develop community-based prediction map using an eigen vector spatial filtering approach. By utilizing eigen vector spatial filter, we can estimate model parameters which are devoid of spatial autocorrelation ( Griffith, Daniel). Another advantage of this approach is that these eigen vectors and their corresponding eigen values reveal distinct map patterns that reveal varying-level of spatial autocorrelation in the response variable. By using DHS data for Bangladesh, we have generated community-based prediction maps for our model. These prediction maps can serve as a useful tool for the government and health planners in Bangladesh to develop health interventions based on the specific needs of the communities.  

A major contribution of this paper will be to enhance understanding on the practical application of our model in actual health planning and policy-making.  For this purpose, we will be conducting field-level interviews of government health officials in the Central Health Ministry, those at the district-level and policy experts from non-profit and academia that are involved in the policy-making process in Bangladesh. The field work for these in-depth interviews will be conducted between May5- May 27.  Thus our quantitative results will be backed with qualitative interview results. Through this paper, we hope to gain a better understanding on the relevance, application and the barriers to using spatial-model based map patterns by the government to geographically target health interventions.  All the data analysis and the results from our field work survey will be completed by July.