Package for contrasts and means estimations about easystats HOT 8 CLOSED

In general, aside from this and a visualisation-oriented package, I currently cannot see any more fields to fill up... which is good, as it means that easystats is getting closer to being scope-complete ☺️

from easystats.

Well I've refactored some code that I've already written here and there, and so it came together quicker than expected! here's a concept design and syntax, currently only for rstanarm models. Let me know what you think!

from easystats.

ggeffects would have been a package that fits nicely into the easyverse (and it's really easy to handle). I could of course move the package into the easystats-repo, but I think it's too late for a re-implementation now, since there's an associated publication, and ggeffects has been cited already in quite a lot of different papers now, and has a substantial user base... But in general, a package that easily creates contrasts etc. would be nice. I have also recently drafted a function that computes contrasts in ggeffects. The goal is to include this into plots, i.e. having different line/dot/whatever styles depending on if there's a significant difference between groups.

Three more set of functions come into my mind, which could be included in easystats as well, maybe in existing packages:

• Weighted statistics (see https://strengejacke.github.io/sjstats/reference/wtd_sd.html), because these functions are not readily available in R
• Reliability tests (see https://strengejacke.github.io/sjstats/reference/reliab_test.html)
• Measures of association (https://strengejacke.github.io/sjstats/reference/xtab_statistics.html)

from easystats.

Hum yeah, I also think for that it would be better for you to keep ggeffects under you, as it is indeed already very used and useful :) At the same time, I think that the current test implementation (in estimate [provisional name]) makes it a different and relatively non-competing package to ggefffects (aside from some of the inherently overlapping methods)? What do you think?

For other easystats applications, for reliability tests: This could maybe fit into performance? Or having a separate package for questionnaire data along with measures of association?

For weighted stats, I don't see how well it could fit into existing package for now... but it would be a nice feature ot have it. On a side note, I have a very novice question: I am more and more thinking about how to propagate uncertainty into further analysis, for instance the uncertainty related to a participant's level (mean) of X (reflected by the SD) into subsequent correlations or regressions. But I didnt found an easy and consistent way of doing it. Can such weighted statistics be used for this sort of uncertainty propagation?

from easystats.

1. I think estimate has it's place in the easyverse, let's see in which direction it develops.
2. Maybe performance is indeed a good place for those measures. I would skip Cramer and Phi for now - I still need some functions left for sjstats ;-)
3. Maybe we should skip this for now as well.

Can such weighted statistics be used for this sort of uncertainty propagation?

I'm not sure I fully understand your point, but from my first impression, I would say that the weighted-stuff is probably not meeting your needs here...

from easystats.

Just saw that performance is currently focusing on model performance. Should we still include reliability tests then, and enhance the focus, not limited to models?

from easystats.

So might say it depends on the definition of a model 😀 (some argue that data is already a model of the world)

The name "performance" is quite agnostic as to its object, so I guess it would not be a drastic or counterintuitive change to increase its focus

from easystats.

Also, thinking of it, what could lack is a nice small package for simulations, something "generate some data suited for a binomial model of effect size X and sample size N" :)

from easystats.