I would like to fit a model wherein replicates belong to different groups (a multiple membership model), but importantly the replicates can differ in the number of groups to which they belong. What is the appropriate syntax to account for multiple membership when replicates belong to different numbe…

When I run dat <- data.frame( y = rnorm(100), x1 = rnorm(100), x2 = rnorm(100), g1 = sample(1:10, 100, TRUE), g2 = sample(1:10, 100, TRUE), w1 = rep(1, 100), w2 = c(1, rep(0, 99)) ) make_standata(y ~ x1 + (1|mm(g1, g2, weights = cbind(w1, w2))), data = dat) I see that the weights are correc…

(1) Is the correct solution I think. You can always achieve the desired weighting using the weights argument.

Thanks for your reply, @paul.buerkner . Do you have a sense for why (3) gives different fits depending on which group is used to replace the NAs? Taken together with the behaviour of (2), it seems like the weights argument never treats a zero weight as I had expected.

Great, thanks! This resolves my question for now. I’ll re-open if I learn anything else from continued testing with approaches (1) and (3).

Revisiting this, @wpetry , I’m wondering if you learned more about option 3, which would seemingly be useful in cases where an observed outcome should have zero weights across the board. For example, imagine a dataset in which hospital patients’ outcomes are modeled in part as a function of the nurse…

I was able to use @paul.buerkner ’s solution for my problem, and haven’t encountered a problem like the one you describe. I wonder whether zeroing out the multi-membership term would do what you describe in the patient care scenario. Wouldn’t this imply that patients with no nurse would have an outco…

Thanks, Will. For background, I’m borrowing concepts from this paper by Tranmer, Pallotti, and Lomi: The embeddedness of organizational performance: Multiple Membership Multiple Classification Models for the analysis of multilevel networks - ScienceDirect At one point, the authors write: "If the ne…

Multiple membership models when replicates belong to different numbers of groups

Interfaces brms

paul.buerkner October 31, 2019, 8:28am 4

Let me check, what zero weights actually imply. Perhaps there is a bug in that regard.

Multiple membership multilevel model - clarification questions

Topic		Replies	Views
Zero weights in multiple membership models brms	3	718	March 3, 2019
Multiple membership multilevel model - clarification questions Modeling brms	5	210	October 11, 2024
Maximum groups in multiple membership model in brms? brms specification , brms	0	323	March 17, 2023
Three-level multiple membership models in brms brms	2	670	May 11, 2020
Weights function in brms brms fitting-issues	5	1865	December 8, 2020

Multiple membership models when replicates belong to different numbers of groups

Related topics