If I had to? I'd start with "What would I know?" And then "Why not at least ask someone in biotstats?"

@half-pass And if that didn't work, I'd start researching US biostats departments. What things do you want to rank them on?

I have a question about bayesian networks. I want to learn the parameters of bayesian network from data. During training, all variables are observed. During testing, one of the input variable is not observed, and the output is not observed. The unobserved input can be inferred from the observed input variables. In such a case, should I make that "unobserved input" a hidden variable or observed variable. I guess output has to be a hidden variable.