Ecological and anthropogenic covariates ... - GANGAPEDIA
Ecological and anthropogenic covariates ... - GANGAPEDIA
Ecological and anthropogenic covariates ... - GANGAPEDIA
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
We compared Zero-Inflated Poisson (ZIP) <strong>and</strong> Zero-Inflated Negative Binomial<br />
(ZINB) that accounts for over-dispersion which can account for both over-dispersed<br />
<strong>and</strong> under-dispersed counts. To these models I assumed a Conditional Auto-<br />
Regressive (CAR) normal distribution as an uninformative prior distribution for<br />
spatial r<strong>and</strong>om effects. The inverse of the precision parameter (spatial variance) for<br />
the CAR normal models was calculated <strong>and</strong> compared between models. The CAR<br />
prior provides spatial smoothing of parameter estimates.<br />
The value of Gharial count per segment is influenced by the probability that it takes a<br />
value conditional upon the gharial count in the neighbouring segment. The CAR<br />
model used here has four terms: the number of neighbours of each site, (2 for all<br />
segments except terminal sites which have only one neighbour), adjacency matrix<br />
based on the IDs of neighbouring sites <strong>and</strong> spatial weights which we assign as 1 for<br />
all areas. The spatial precision (1/variance) parameter tau is the important parameter<br />
for the model, as it gives us an estimate of spatial effect. Higher the parameter tau,<br />
lower is the spatial variation or spatial effect.<br />
We used the ZINB <strong>and</strong> ZIP models to estimate the slope <strong>and</strong> intercept parameters, as<br />
well as spatial variance parameter for relationship between Gharial abundance <strong>and</strong><br />
basking site presence, as well as between abundance <strong>and</strong> depth of river channel in that<br />
segment. The parameters slope, intercept, spatial variability (variance of CAR<br />
Normal), over-dispersion parameters, <strong>and</strong> other parameters of the respective<br />
distributions (ZIP, ZINB) were estimated in each model. Deviance was compared for<br />
model selection. All statistical analyses were conducted using the software R 2.11.1<br />
(R Development Core Team 2010) <strong>and</strong> WinBUGS (Spiegelhalter et al. 2007). For<br />
22