14 Matching Annotations
  1. Jul 2019
    1. The table 6.1 gives the mixing probabilities and the associated parametricvalues fork(number of components) = 2,3, and 4. It may be noted thatthe Log likelihood value is smaller fork= 4 (the results fork= 5 , 6 etc.are not better than that fork= 4 and hence are not given here). The fourcomponents Poisson Mixture model is given in table 6.2. It may be notedthat 58% of wards may have higher incidence/relative risk and the remainingwards have lesser/lower incidence for the Cancer disease. We computed theposterior probability for each component for each ward (see table 6.3). Eachward is assigned to a particular component so that the posterior probability islarger. These results are also given in table 6.3 Finally we present Choroplethmaps based on those results
    1. We have analysed the Cancer data of patients in 155 wards of Chennai Cor-poration by the above described method. As preliminary analyses, we havecreated the Choropleth maps for Observed counts, Population of wards, ex-pected counts for patients and SMR's.The Choropleth map for the observed counts Figure 5.2 does not show anypattern. But the Choropleth map for the expected counts Figure: 5.4 indi-cate that the inner regions of the Chennai Corporations have lower expectedcounts and the regions along the border have larger counts of patients. As ameasure of spatial heterogeneity we have computed PSH= 0:7108:Hence ofthe total spatial random variation, nearly 71% is due to spatial heterogene-ity and the remaining 28:92% is due to Poisson variation. Thus the spatialvariation is present in the data.The Choropleth map for Empirical Bayes smoothed rates Figure 5.5 re-veals that only 13 sub regions have high risk values. The wards with numbers53, 64, 67, 70, 78, 93, 100, 103, 110, 117, 122, 147 and 151 have high riskvalues. Though this information could be used by the health managers toconcentrate their work on these regions, one can look for additional covariatesin these regions for further study