Page 306 - Contributed Paper Session (CPS) - Volume 2

CPS1871 Natividad J.M. et al.
to convert the shapefile to S-PLUS format needed in GeoBUGS and GeoDa to
produce maps.
Deviance Information Criterion (DIC is commonly used in Bayesian model
selection. Lawson (2013) stated that the lower value of DIC is preferred in
comparing and selecting the best Bayesian model to fit the data.

3. Result
Seven cities in NCR have SIR greater than 1 implying that more HIV
incidence occurred than expected in these areas. Areas with SIR greater than
1 are Mandaluyong (2.2953) with the highest relative risk of infection followed
by Makati (1.8925), Pasig (1.4432), Pasay (1.4114), Marikina (1.1584), Manila
(1.0849), and San Juan (1.0233), respectively. The remaining cities, on the other
hand, experienced less HIV cases than expected as shown in Table 1.
The association between the SIR of HIV cases and the spatial units in
condition to the covariate, percentage of sexually active male population, is
shown in Figure 1. The conditional map shows that spatial units with higher
percentages of sexually active male individuals are expected to experience
higher HIV incidences. Specifically, the cities of Mandaluyong and Makati were
observed to have high reported incidences of HIV infection while exhibiting
high percentages of sexually active male individuals as well.

Figure 1. Conditional plot map for the SIRs of HIV based on quartiles of sexually
active male population

Global Moran’s I was used to see if there is a spatial autocorrelation which
looks into how related the values of the sexually active male population are
based on the locations where they are measured. Results show that the
Moran’s I p-value (0.0206) is statistically significant at α = 0.05. This concluded
that there is a sufficient evidence to say that there is a positive spatial
autocorrelation in the values of the covariate. In support of this, the value of
the Moran’s I (0.314714) exceeds the expected value which is - 1 (n=17)
−1
equal to -0.0625 and therefore Bayesian CAR model can be appropriately used.
Posterior relative risk estimates of HIV for each city/municipality based on
the Bayesian CAR and Poisson-gamma models were obtained as shown in
Table 1.

295 | I S I W S C 2 0 1 9