Dear pymc experts,
I have one group of participants complete two experiments and I want to model these two experiments together. Note that the parameters in these two models are not related, and when I put two models into one pymc model, all the samples are divergences. However, when I separately fit the model, there is no such problem? Thank you very much.
Best,
Can you say more about what you mean by this? Do you literally just concatenate the 2 models’ code with multiple likelihoods, etc.? If so, I would suspect that maybe there is a conflict between something in the 2 model (e.g., indices into your data or something along those lines).
If this is the case, what is the motivation to combine the two into a single model? Usually the motivation is to place priors on and/or infer the joint posterior.
Hi, because recently I read a paper written by Peter.Kvam,this one. In this paper, he had participants finish the Cambridge gambling task and delay discounting task. he modeled the beta parameter in the Cambridge gambling task as a GLM, beta = beta0 + beta1 * discount rate, by doing so, he tested whether the two experiments test a same ‘impulsivity’. For the nested model, when we set beta1 = 0, models for two tasks have no correlation. I think this method is quite similar with the traditional way that you fit two models separately with non-hierarchical way and correlate the parameters. But the method has a bonus that it seems the parameter recovery is improved.
Ok, so the parameters from the two tasks are assumed to be related. Can you provide a sketch of your model?