Please help, not an expert... It's an educational study

[eddithcc]

I have a Bayesian Network whose structure was defined by expert consensus (Delphi) and whose conditional probability tables were calibrated using the EM algorithm with data from a pilot group. My goal is not to predict missing values, since all variables are fully observed, but rather to identify and compare reasoning patterns and alternative conceptions between a traditional group and an experimental group in an educational study.

In this context, what type of inference, network-based measure, or derived indicator could the Bayesian Network provide that is not already contained in the observed response trajectories themselves?

Or, if all variables are fully observed, is the Bayesian Network mainly useful as a calibrated probabilistic model, while the comparison of reasoning patterns should instead be performed directly from the observed data?

[marek [BayesFusion]]

Hi eddithcc,

I'm not sure what to advise you here -- I don't know sufficient detail and your question seems to be highly domain-dependent rather than being general about Bayesian networks/influence diagrams. What exactly would you like to compare between the experimental and control group? Would perhaps comparison of the goodness of fit of the model to the data help? I assume you have data coming from the study? Perhaps calculating the probability of the observed reasoning pattern would give you an indication of correctness of reasoning? I assume you are treating the consensus model as the gold standard?

I'm not sure this helps :-).
Cheers,

Marek