BAS
BAS copied to clipboard
merliseclyde
Mixed models
BAS makes using well-established prior types a breeze. Thanks for that!
I would very much like to see BAS.lm and BAS.glm extended to mixed models, i.e. models with random effects. This would pave the road for Bayesian (G)LMMs in JASP where this has been requested for a while: https://github.com/jasp-stats/jasp-desktop/issues/1846 and https://github.com/jasp-stats/jasp-desktop/issues/1103.
@lindeloev do you have a particular syntax or form for representing random effects? i.e. it would be helpful to piggy-back off of other package syntax for representing random effects. The independent representation in mgcv should be relatively easy to implement, but more general covariance structures need additional thought.
@merliseclyde The lme4 syntax is widespread so should ease the adoption for many. So for example:
height ~ time * gender + (1 + time | id)
It is also used in brms which is strongly gaining popularity and has shown that this framework is scalable to specifying missing data imputation, censoring, etc.. In addition, specifying the random effect structure in the same formula that specifies the fixed effects require a lot less syntax than having a separate random argument as mgcv and BayesFactor do.
Hi Dr Clyde,
I'm just wondering if you've been able to make any progress on including mixed models in your package? I agree with Jonas above, thatyou've been able to make any progress on including mixed models in your package? I agree with Jonas above, that this type of analysis has grown inutility in the last few years. Likewise, the lme4 style of syntax is certainly the most widespread, and is likely to represent the fewest barriers in encouraging users to adopt these types of models and use them from your package. I am happy to help out in any way that would be useful to you.
Wanted to check in on whether this is still on the road-map. There isn't anything out there for "default" bayes factors for mixed-model logistic regressions.
