search

university-of-newcastle-research / pmwg

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

Option to constrain a group-level parameter to a constant, while still allowing variability at the individual level #55

Open IMW112 opened 3 years ago

IMW112 commented 3 years ago

Is your feature request related to a problem? Please describe. It has been argued that the scaling constraints commonly imposed on LBA models (e.g. fixing standard deviation of the drift rate to 1) often result in over-constrained models and can complicate parameter interpretation. van Maanen and Miletić (2021) - https://doi.org/10.3758/s13423-020-01783-y - suggest that, in a hierarchical model, it may be preferable to satisfy the scaling constraint by fixing only a group-level parameter to a constant, while still allowing variability across individuals.

Describe the solution you'd like Option in PMwG to specify that a specific group-level parameter should be constrained to a constant value, while still allowing estimation of random effects for that parameter at the individual-subject level.

Describe alternatives you've considered Unaware of any alternative solution (other than to over-constrain the model by fixing the parameter value across all individuals)

Additional context van Maanen and Miletić (2021)'s suggestion appears in the final paragraph of their discussion.

Thanks!

gjcooper commented 3 years ago

Thanks,

This is not something that we have considered in the past, and would require some thought on how to disentangle the Gibbs step from the random effect particle generation in the package as it is currently implemented. My colleagues and I will give this some more consideration and get back to you on it.