search

university-of-newcastle-research / pmwg

Repository for code for the Samplers Team at UoN
3 stars 3 forks source link

Option to fit models to multi-group data simultaneously #54

Open IMW112 opened 3 years ago

IMW112 commented 3 years ago

Is your feature request related to a problem? Please describe. I am analysing data from a two-group design, and it'd be great to be able to fit a model to all participants simultaneously whilst respecting the multi-level hierarchy (Population -> Group A or Group B -> Individuals within each group)

Describe the solution you'd like Option to specify multi-group models in PMwG with group-level hyper-parameters estimated to facilitate subsequent between-group comparisons

Describe alternatives you've considered Separate modelling of each group in isolation and subsequent between-group comparison based on the posterior distributions. Issues: seems a shame not to be able to take advantage of fitting all data simultaneously (each group is fairly small in isolation), and, depending on the scaling constraint used, parameters may not be directly comparable across groups/models.

Additional context Multi-group designs are common in applied research, so I would think it might find quite widespread use.

Thanks!

gjcooper commented 3 years ago

Thank you for your message @IMW112, this is something that we and our collaborators have certainly considered in the past, but is not trivial to add to the package in it's current form.

It will probably require a long process of consideration and is unlikely to be added soon, but we will definitely keep this in mind for future additions to the package.