Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Learn Bayes suggestions #82

Open
JohnnyDoorn opened this issue Nov 4, 2021 · 6 comments
Open

Learn Bayes suggestions #82

JohnnyDoorn opened this issue Nov 4, 2021 · 6 comments

Comments

@JohnnyDoorn
Copy link
Contributor

I just took a more thorough look at the Learn Bayes module, for my upcoming teaching tasks (intro Bayes first year bachelors students). I noticed two possible improvements:

  • for Binomial estimation and testing, the first menu is "Data", and after that comes "Model" or "Hypothesis". I would suggest turning these around, since ideally first we think about which models or hypotheses we want to test, and how we want to specify the prior distribution. This also makes sense for the output panel, since we need to specify a model or hypothesis for anything to show up there (whereas this is not the case for data - the output is there even if the data are left at 0 and 0).
  • I am missing the classic JASP prior/posterior plot, with both distributions, grey dots indicating the Savage-Dickey ratio, and the Bayes factor, all in one figure. I think having this figure is a lot more concise than the current way of providing many different plots (hampers the overview a lot, imo). This might be a challenge, since the idea in the module is to just have models, rather than an explicit H0 and H1 (although we could use the same approach as used under "Bayes factor" under "Predictive Performance", where you select the vs model in the dropdown menu).

What do you think @EJWagenmakers and @FBartos?

@EJWagenmakers
Copy link

Thanks for the suggestions @JohnnyDoorn
(1) About the "data" coming first -- this is consistent with the rest of JASP, right? We usually start an analysis by dragging the relevant variables to the input boxes, and then select the analyses.
(2) The required plot would go under "testing" I assume. Then it would be relevant only for a point H0 versus a composite H1. So it would be an option that could be activated only when these model specification conditions are met. I do think that could make sense. However, in terms of priorities I hope we can have a z-test first, followed by a directional test.

@AlexanderLyNL
Copy link

I wouldn't mind having the specifications of models and hypothesis before the data. That agrees with the model and hypothesis first principle. It will indeed a bit different from standard analyses such as the z-test, but there default choices are made. When learning Bayes I think that first putting forward hypotheses is quite reasonable.

@EJWagenmakers
Copy link

There is not a single analysis in JASP where we put data second, right? The data always come first in JASP. Also in Summary Stats.

@AlexanderLyNL
Copy link

In the audit module data comes later. In the standard analyses the data indeed come first, but that's because the models and default priors are chosen once you've chosen the associated analysis. In the learn Bayes module you don't only study default analyses. Considering the flow of information from prior, using data to update to posteriors. From a teaching perspective I think it's reasonable to specify priors first.

@EJWagenmakers
Copy link

OK, I see. Well I can be pursuaded. @FBartos what do you think?

@FBartos
Copy link
Collaborator

FBartos commented Nov 4, 2021

hmm, I don't have a strong emotional attachment to any order in particular, so I wouldn't mind changing it.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

4 participants