get_predicted() for new models

[strengejacke]

• glm.nb
• plm::plm() (see Error when using ggpredict() with plm() model strengejacke/ggeffects#121 (comment))
• panelr::wbm() (see Non-conformable errors wbm-models strengejacke/ggeffects#266)

[strengejacke]

problem for plm: get_modelmatrix() (resp. the underlying model.matrix() and model.frame() from the plm package) don't return model matrix for a pre-defined data frame:

library(plm)
data("Produc", package = "plm")

m <- plm(pcap ~ hwy + water + hwy * water, data = Produc, effect = "twoway", index = c("state", "year"), model = "within")
d <- insight::get_datagrid(m, c("hwy", "water"), range = "grid", include_response = TRUE)
insight::get_modelmatrix(m, data = d) |> dim()
#> [1] 816   3

m <- lm(pcap ~ hwy + water + hwy * water, data = Produc)
d <- insight::get_datagrid(m, c("hwy", "water"), range = "grid", include_response = TRUE)
model.matrix(m, data = d) |> dim()
#> [1] 30  4 # this is what we also want for plm

^{Created on 2022-09-11 with reprex v2.0.2}

[strengejacke]

@vincentarelbundock if you have any ideas, let me know - you seem to work more frequently with panel-regression-models.

[vincentarelbundock]

Are you just hoping to make get_modelmatrix() work, or is the intention to use it in ˋget_predicted`? I'm overworked and won't be able to look into this seriously for a few weeks.

My initial reaction is that we should be super careful here, because plm models use a lot of complex de-meaning strategies, and it's not clear to me whether stock methods will return the design matrices we actually need to compute standard errors around fitted values. Until someone really investigated the issue, it would be safer to rely on the default predict methods, which work well since the last CRAN release of ˋplm`