Every factor except education has a significant effect.
The model is produced from the training dataset, using the following code:
scotglm <-
vglm(formula =
cbind(offline, connected)
~ slowlines + income + employment + housing
+ health + education + crime + isolation,
family = betabinomial(zero=NULL),
data = training)
Here we use a different parametrisation of the betabinomial distribution, in terms of its mean and dispersion, as this gives a coefficient we can interpret directly as an odds ratio.