new emmeans + bayestesR post

[mattansb]

I've tried to keep this practical-oriented, so it is very light on theory (This is also a rough draft right now..).

Let me know what is missing / should be polished, and feel free to add any opening or closing statements / remarks (like those in previous easystats posts).

[strengejacke]

I made some minor revisions. One sentence:

Using the fantastic emmeans package, we can explore and extract estimates from our fitted model.

Do we extract estimates in a classical sense? Wouldn't it be more clear to say "marginal estimates", or "marginal effects" or so?

[strengejacke]

Just curious, you called the priors "non-informed", though you were giving location/scale. Is there somewhere a definition when a prior is "flat", "weakly informative", "informative", "uninformed"? I thought the priors you are using were "weakly informative".

[DominiqueMakowski]

To my knowledge (non-uniform) priors located at 0 with a not-too-narrow scale are labelled as "weakly informative" and flat (uniform) priors as "non-informative"

[DominiqueMakowski]

PS: for the release plan, my suggestion is to 1) wait for confirmation that the blog is back on r-bloggers, 2) release a small "bayestestR v2" post simply advertising the new version and redirecting to the website and 3) release this post showcasing a more complete use of bayestestR

[mattansb]

Rouder et al. (2012) seem to equate "weakly informative" with "non-informative", but I'll change this for consistency with rstanarm terminology.

PS: for the release plan...

Sounds good!

I can't seem to get blogdown to work on my end - is it okay if I leave the building etc. up to you?

[DominiqueMakowski]

sure :)

[mattansb]

Okay, bayestestR is on CRAN! And the blog is back on r bloggers! Let's do this! (Can you tell I'm excited?)

[mattansb]

I light of the new emmeans integration, perhaps this post should be scrapped for now? Maybe if bayestestR has a minor 0.2.1 release I can update this post and publish it then? Or we can publish this an intended, and in the future publish a "update - now it's even easier!" post?

@DominiqueMakowski @strengejacke thoughts?

[mattansb]

Alternatively, I can update this post to actually demo the new emmeans methods, and we can publish it after the emmeans PR is completed (noting in the post that this update is available on the GitHub version, and will "soon" be available on CRAN release).

[mattansb]

@DominiqueMakowski @strengejacke I'm tending towards my last suggestion. What do you think?

[strengejacke]

I'm open to every suggestions... I just recalled a deadline you mentioned? If this is no longer important for you, go ahead!

[DominiqueMakowski]

Yup, I am okay with all alternatives as it mostly depends on your timeline :)

[mattansb]

My timeline has changes slightly 😅 I've updated the post - it is ready to build and go up when the PR is completed.

Should I add some note or footnote somewhere that support for emmeans is available on github and in will also on CRAN?

[mattansb]

Seems like I can keep my original timeline - this post is ready to go up 🎉

[DominiqueMakowski]

Tomorrow then?

[mattansb]

Yes, excellent!

-- Mattan S. Ben-Shachar, PhD student Department of Psychology & Zlotowski Center for Neuroscience Ben-Gurion University of the Negev The Developmental ERP Lab

On Wed, Jun 5, 2019, 04:40 Dominique Makowski notifications@github.com wrote:

Tomorrow then?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/easystats/blog/pull/22?email_source=notifications&email_token=AINRP6ESR4ZP4NBAHS6ATNTPY4KO7A5CNFSM4HQERC52YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODW6LCYI#issuecomment-498905441, or mute the thread https://github.com/notifications/unsubscribe-auth/AINRP6CC6RGRWQM4JG3ATATPY4KO7ANCNFSM4HQERC5Q .

[strengejacke]

We must decide on a time zone when we speak of "tomorrow" or "today". :-D

[DominiqueMakowski]

In my absolutely unbiased opinion, I would say the time zone of the majority of easystats members... How surprising that it is mine! 😄

I think Europe's time zone makes sense for everybody as it is "in the middle" (apologies for my eurocentrism :p). Midday in the easystats timezone (henceforth referred to as the easytime) is in my afternoon/evening and in the US people's morning

[strengejacke]

We could use UTC, which indeed seems to be somewhat Europe-centered ;-)

[mattansb]

So in easytime™, when @DominiqueMakowski said "tomorrow", that was referring to Thursday, yes?

[DominiqueMakowski]

Correct, how does that sound?

[mattansb]

Fantastic :)

[strengejacke]

in easytime™

We should file a PR in the lubridate package...

[DominiqueMakowski]

In fact, I don't even have to merge it... I built the site, so feel free to merge whenever you want :)

[DominiqueMakowski]

I think you could definitely link to this blogpost from the BF vignette on bayesfactor. You know in a paragraph like "More examples of usage" or something like that.

As a matter of fact, if we have more of such feature-specific posts, we might eventually want to either link them from the last tutorial vignette (become a Bayesian master), or create a separate vignette "More examples and resources" to link to these posts.

[mattansb]

The post is up + tweeted (the most important part of any publication!).

[MPLisi]

It is possible to compute ESS and R^ for each emmean comparison?

[mattansb]

@matthewlee37 emmeans doesn't save the chains separately, so you can't find rhat. But in theory it should be possible to manually compute ESS across all chains.

But I think (?) looking at these of the parameter (and note the expected means / contrasts) would be more informative anyway..