Are there resources how to use the package for multilevel modeling available?

tilltnet / egor

R Package for importing and analysing ego-centered-network data.

http://egor.tillt.net

GNU Affero General Public License v3.0

23 stars 4 forks source link

Are there resources how to use the package for multilevel modeling available? #56

Closed PCK1992 closed 3 years ago

PCK1992 commented 3 years ago

The package has a lot of great features but I can't quite wrap my head around how to do MLM using the data structure and I couldn't find a reference in the documentation.

library(egor)
library(lme4)

data("alters32")
data("egos32")
data("aaties32") 

e1 <- egor(alters = alters32,
           egos = egos32,
           aaties = aaties32,
           ID.vars = list(
             ego = ".EGOID",
             alter = ".ALTID",
             source = ".SRCID",
             target = ".TGTID"))
e1

lmer(income ~  (1|.egoID), data = e1$ego)

This doesn't work and I tried different variations of this formula but perhaps a short example of how to get this to work (or how to work with this package for MLM in general) would be highly appreciated!

raffaelevacca commented 3 years ago

Hi @PCK1992, below is a reference on multilevel models for ego-networks, and you can contact me via email for the code used in the article's examples (which we've been using for some recent workshops). We're trying to incorporate all this in a vignette as soon as possible.

Vacca, R. (2018). Multilevel models for personal networks: Methods and applications. Statistica Applicata - Italian Journal of Applied Statistics, 1, 59–97. https://doi.org/10.26398/IJAS.0030-003

tilltnet commented 3 years ago

This should work for your example.

alters_egos <- as_alters_df(e1, include.ego.vars = TRUE)
fitintensity <- lmer(income ~ (1 | .egoID), data = alters_egos)

Currently egor does not provide a method for lme4::lmer(), but I can look into adding a method for that.

PCK1992 commented 3 years ago

Thank you, this is super helpful! I think in my use case it's really the only functionality that I wish was there. Thank you for your quick responses!

tilltnet commented 3 years ago

Thank you very much for the feedback @PCK1992 !

My thoughts on adding an lmer.egor() and glmer.egor() methods:

There are many ways researchers apply MLM to ego-centered network data. I think most commonly alter level variables are regressed on an ego characteristic.
Creating an interface that accommodates for the diversity of possible model types might be hard.
The methods could cover the most common case(s) and the vignette @raffaelevacca mentioned above would demonstrate how to accomplish more complex model designs.

PCK1992 commented 3 years ago

That's true, especially if you want to interface it with any package outside of lme4. I think a vignette should be enough for the most common use cases and this could be then easily extended to other packages (e.g. brms, etc.).