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
Syntactic changes/additions: p_* and bf_* families #258
Comments
1. SplittingDom, is that you??? Is it this the same person who originally wanted all the Bayes factor functions to be under one single Within the
Internally, there is only one function, so we would essentially be making wrap-functions to split the functionality up? If we do this, I would suggest (as I originally did) adding (not replacing) only two functions, by the null type:
On the other hand / additionally - Now the
I feel triggered.... 2. RenamingWould the |
|
Sorry, right, only |
Agree with the aliasing. If I understand you, we'd have:
and |
Looks good, and we should somehow make this "overview" prominent in the docs/vignettes. Once users grasp the conceptional differences, it's easier to follow the decision for naming functions. |
@DominiqueMakowski as I noted above, I think we can do with only 2 So hope about:
|
to make room for new p_rope #258 - It's an ugly name, but since this index is not relly used/advertised (it's mainly there for legacy reasons now), I guess it's okay (til eventually find some use for this thing)
Why? Would It's just that "savage-dickey" is not really transparent, whereas above I was leaning toward a new framework of
I am not sure about it. I think the theoretical overlaps and discrepancies would be more obvious with this new design. Currently, it seems like the BFs are completely different from the proportion tests, whereas in fact it appears to me now that it is all a matter of perspective (method: BF or posterior proportions) on a given hypothesis (something vs. something else). (not talking here about the other BFs such as inclusions and restricted) Reflecting this in the functions could make things easier, and it was probably because this wasn't clear to me at first that I thought it would be a good idea to have only one bf() function Bayesian tests
|
Okay, you broke me - I will add
They are compactly different! They ask completely different questions - the BF asks questions about the data (what did we learn from the data) whereas posterior estimates (and their proportion tests) ask questions about the posterior. Their relationship to the priors are also different (as we will soon show in our upcoming paper). But all this is somewhat besides the point here (kind of)... The main reason we shouldn't have separate functions for Additionally, I think it is a disservice to users, to whom having there "separate functions" makes it seem like these are different things (and might also encourages bf-hacking, by running all functions and seeing which is the "best"). Honestly, the more I think about it, even |
True...
good point... Although I am afraid if people want to hack they will find a way no matter what, we cannot be the ultimate gatekeepers of good practices here Again, I favour leaving all options open for the users, but accompanied with documentation of good practices, - for them to understand why some methods are worse than others-, rather than enforcing some practices and restricting freedom, because the user will find what he's looking for anyway through digging in or in other packages.
we opened the pandora's box 🙉 Okay, let's add bf_pointnull (why not ZERO then?? 😅) and alias its counterpart p_pointnull, and we'll leave it at that? |
Alright 😁 - I've replaced |
Good point!
Replaced might to brutal, maybe we could either alias it and for now just deprecate bf_sd? |
|
then I agree, let's remove! (sorry I'm out of sync 😵, I was quite busy with some other stuff recently) |
I would not remove |
Can anyone check the state of this issue? Has everything been addressed in the dev-branch? |
All is done on the BF end 👍 |
@DominiqueMakowski bump |
maybe I should use some animations here to gain more attention... |
don't you see I'm busy creating some awesome animations?! don't have time for minor details such as the API of the package 😁😁 But I think everything's in order, we just need indeed to summarise this in the vignette #262 |
Currently, we have:
p_map
p_direction
p_significance
rope
which are similar in the sense that they represent some sort of proportion/probability.
First question, should we add an alias to
rope
asp_rope
, to make thep_*
family consistent and complete?Also, we currently have the
bayesfactor_*
family, withbayesfactor_parameters
being able to compute BFs against 0 (in a way, kinda close top_map
), against rope and potentially against direction and significance.Hence, rather than having these options available as arguments of
bayesfactor_parameters
, what about exposing them as mirrors to thep_*
family, in abf_*
family, withbf_direction
,bf_rope
,bf_significance
and lastlybf_map
(the "basic" BF against a point-null), which might not the most appropriate name (maybe something likebf_zero
).Thoughts?
(PS: I know that initially, I favoured
bayesfactor_*
overbf_*
, but it seems like the latter shortcut has become super popular and accepted so I think I am changing my mind 😁)The text was updated successfully, but these errors were encountered: