Page 255 - Special Topic Session (STS) - Volume 2

STS489 Danielle J.R. et al.
posterior mean estimates of the spatial effects for the different regions of the
countries.

3. Results
Based on the univariate logistic regression with 10% level of significance
for inclusion, the only independent variable not entered into the multivariable
model was the age of the head of household. The variance inflation factor (VIF)
was used to check for collinearity among the remaining continuous
independent variables and all variables had a VIF < 4 and thus it was assumed
that multicollinearity was not significantly present (Zuur et al., 2009). The non-
linear effect of all continuous variables was investigated, however the only
variable to display a significant non-linear effect on the log-odds of a child’s
anaemia status was their age in months. Thus, this was the only non-linear
effect considered in the models fitted, while the remaining independent
variables were included as linear fixed effects. Model 3 produced the lowest
DIC, and thus the results of this study are based on this model, which includes
both linear and non-linear effects as well as the spatial effects. In other words,
the model given in Equation (1) is chosen and adopted.
Table 1 displays the adjusted posterior odds ratio estimates (AOR) with
their 95% credible intervals for the linear fixed effects included in the final
geoadditive model. Figure 2 displays the non-linear effect that a child’s age in
months has on the log-odds of being anaemic as well as the 95% credible
band. There was an increase in the log-odds of anaemia from 6 to 10 months,
after which the effect declined. Figure 3 displays the estimated means of the
structured and unstructured spatial effects on the log-odds of anaemia, where
the blue regions have a negative (or lower) spatial effect and the red regions
have a positive (or higher) spatial effect, and thus are associated with an
increased risk of anaemia. The structured spatial effect, which ranges from −
0.0368 to 0.0316, is weak in comparison to the unstructured spatial effect,
which ranges from −1.3061 to 0.9780. This suggests that the prominent driver
of childhood anaemia in these countries consists of district-specific factors
that are not spatially related and that do not transcend boundaries, such as a
lack of access to good health care and poor nutrition.

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