Earlier I posted a general request for resources about affinity stochastic block models, introduced here: http://arxiv.org/pdf/1211.3601.pdf as
"An affinity stochastic blockmodel is a stochastic blockmodel wherein the probabilities of
connection within a block is in general larger than those between blocks. In an affinity stochastic blockmodel,a community is then a collection of nodes whose connections are more dense within the community, as compared to connections between that community and other nodes." (Priebe et al., 2014)
But I have a more specific question that does not necessarily require resources and just your ideas. So as I talked about briefly at the end of class when I talked about my project, this model allows us to talk generally about the probability p within block vs the probability q across blocks, so p is the probability of a connection within blocks on the diagonal and q is the probability across blocks on the off diagonal. So the test statistic is p-hat - q-hat and you can do a permutation test to get the null distribution. But, the concern is that the variance of q-hat will be much smaller than the variance of p-hat.
Is there is a modification with higher power that deals appropriately with the higher variances?
Earlier I posted a general request for resources about affinity stochastic block models, introduced here: http://arxiv.org/pdf/1211.3601.pdf as
"An affinity stochastic blockmodel is a stochastic blockmodel wherein the probabilities of connection within a block is in general larger than those between blocks. In an affinity stochastic blockmodel,a community is then a collection of nodes whose connections are more dense within the community, as compared to connections between that community and other nodes." (Priebe et al., 2014)
But I have a more specific question that does not necessarily require resources and just your ideas. So as I talked about briefly at the end of class when I talked about my project, this model allows us to talk generally about the probability p within block vs the probability q across blocks, so p is the probability of a connection within blocks on the diagonal and q is the probability across blocks on the off diagonal. So the test statistic is p-hat - q-hat and you can do a permutation test to get the null distribution. But, the concern is that the variance of q-hat will be much smaller than the variance of p-hat.
Is there is a modification with higher power that deals appropriately with the higher variances?