Now that the model is defined, the next step is to find conditional and marginal expected
values for the binomial process. Once these properties are established, they can be used
to obtain the theoretical variogram of the sample proportions, which in turn leads to the
theoretical variogram of the latent beta distribution.
The beta distribution can be parameterized in two ways: in terms of the mean and
variance of the distribution, Beta(⇡, %2
Y ); and in terms of the shape and scale parameters,
Beta(↵, #). Either parameterization is equally valid, although in some situations one might
be preferred over the other. For simplicity, the model will use the mean-variance parameterization
of the beta distribution, Yi ⇠ Beta(⇡, %2
Y ). Deriving the beta-binomial kriging
model relies on properties of the mean and variance of the latent spatial field Yi, so here it
is most convenient to use the mean-variance parameterization.
First, properties of the conditional distribution Zi|Yi are established. These follow directly
from the model specification above. If Zi|Yi ⇠ Binomial(ni, Yi), then the conditional
expectation is E(Zi|Yi) = niYi and the conditional variance is V ar(Zi|Yi) = nYi(1 ! Yi)
by the properties of the binomial distribution. Then the conditional squared expectation is
Now that the model is defined, the next step is to find conditional and marginal expectedvalues for the binomial process. Once these properties are established, they can be usedto obtain the theoretical variogram of the sample proportions, which in turn leads to thetheoretical variogram of the latent beta distribution.The beta distribution can be parameterized in two ways: in terms of the mean andvariance of the distribution, Beta(⇡, %2Y ); and in terms of the shape and scale parameters,Beta(↵, #). Either parameterization is equally valid, although in some situations one mightbe preferred over the other. For simplicity, the model will use the mean-variance parameterizationof the beta distribution, Yi ⇠ Beta(⇡, %2Y ). Deriving the beta-binomial krigingmodel relies on properties of the mean and variance of the latent spatial field Yi, so here itis most convenient to use the mean-variance parameterization.First, properties of the conditional distribution Zi|Yi are established. These follow directlyfrom the model specification above. If Zi|Yi ⇠ Binomial(ni, Yi), then the conditionalexpectation is E(Zi|Yi) = niYi and the conditional variance is V ar(Zi|Yi) = nYi(1 ! Yi)by the properties of the binomial distribution. Then the conditional squared expectation is
การแปล กรุณารอสักครู่..
