For each elementary school the proportion of students who are missing at least one of the
required childhood vaccinations can be expressed as a binomial random variable. Each school
has a different kindergarten enrollment, ni, and a potentially spatially correlated probability
of missing at least one vaccine, ⇡i. Then the number of children who are missing at least
one vaccination at a given school Zi, conditional on school location si, can be expressed as
Zi|si ⇠ Binomial(ni, ⇡i)
Since the primary modeling goal is to estimate the spatial structure of the underlying
vaccination probabilities, beta-binomial kriging is an ideal approach. This model works well
with changing sample sizes and provides interpretable parameter estimates. Additionally, a
prediction map for the entire state of California would be useful for identifying potential high-
115
risk areas for future outbreaks. Beta-binomial kriging provides accurate spatial predictions
for the latent probability field and is well-suited for proportional data with varying sample
sizes.