Meta issue for FAQ suggestions

[athowes]

In PR #181 we added an FAQ section to the website. As the package is used more we should look to answer a wider range of questions / provide more tips.

Here we can:

1. Collect any potential tips (please contribute if you have suggestions!)
2. Consider how we might systematically get FAQ questions from users

For 2. we might think about recording these in some kind of bug hunt / test demo / presentation / ... -- perhaps worth thinking about which venues we could present the package or get early testing from users?

[athowes]

Oops! Intended to put this here:

Deal with Sam's comment about "Document an approach to deal with stochastically failing models." -- I'm unclear on exactly what this means. As in what to do if your model fails to fit? And how to document this would be a vignette? Or write in the epidist roxygen that you can call epidist_diagnostics and try to debug? Let me know.

Comment from https://github.com/epinowcast/epidist/issues/163#issuecomment-2232922382.

[athowes]

In #371 @seabbs writes:

Add priorsense and loo_epred to the FAQ as useful user facing tool options

https://bsky.app/profile/avehtari.bsky.social/post/3l5ls7qjcpq26

https://bsky.app/profile/avehtari.bsky.social/post/3l5haushmjh2a

[athowes]

In #281 @athowes writes:

Taken from appendix of https://github.com/epinowcast/epidist/pull/52: how does this rvars stuff work?

obs_prep |>
  as.data.frame() |>
  add_predicted_rvars(fit) |>
  select(.prediction) |>
  head()

Should we be recommending ahead of the draws versions?

[athowes]

In #302 @athowes writes:

There is also https://marginaleffects.com/

Originally posted by @seabbs in https://github.com/epinowcast/epidist/issues/238#issuecomment-2305087347

I think if we could use marginaleffects somewhere in one of the vignettes it might be nice to demonstrate its use as an alternative to tidybayes. Perhaps it has some new functionality as well, I'm not sure.

[athowes]

Add documentation in FAQ covering how to do proper prediction (see #218, like the Ebola vignette):

This was discussed F2F. From my POV my understanding is that:

1. The Ebola vignette does demonstrate how to do this with a fitted model when you use tidybayes (and you can set pwindow and swindow and relative_obs_time)

2. Hence I am unsure what more needs to be done here to close this. I recall some kind of suggestion of additional documentation?

   • Note that in future we might have functions to help users more here (Functionality to support users with modelr::data_grid into newdata #280)

   • I can see putting something about this into the FAQ sooner? Am I right that that's the suggestion here?

   • Perhaps include something about primarycensoreddist?