plotting sjPlots in a grid

sjplot.R

library(lme4)
library(sjPlot)
set.seed(101)
n <- 1e3 ## obs per group
ns <- 4  ## groups
dd <- data.frame(s = rep(1:ns, each= n),
                 f = factor(sample(1:20, size = n*ns, replace = TRUE)),
                 x = rnorm(n*ns))
sims <- lapply(1:ns,
               function(i) simulate(~x + (1|f),
                                    family = poisson,
                                    newdata = subset(dd, s==i),
                                    newparams = list(beta = c(1,1), theta = 1))[[1]])
dd$y <- unlist(sims)

fits <- lapply(1:ns,
               function(i) glmer(y~x+(1|f),
                                 family = poisson,
                                 data = subset(dd, s==i)))

plots <- lapply(fits, function(f) plot_model(f, type = "pred")$x)
cowplot::plot_grid(plotlist = plots)

Thanks. Changing it to a list and commenting out all of the options that I set produced a plot. I can try to figure the errors out from here. Thanks so much!!!

sort.est = TRUE or sort.est = FALSE causes the error "Warning: Could not access model information.Error in if (fam.info$is_linear) { : argument is of length zero"

Ok, I found the bug. sort.est is no valid argument for plot_models(), so it was captured by the ... and thereby considered as regression model object. I added a check for valid arguments and plot_models() now returns an informative message when this happens.

In short: plot_models(fits, sort.est = TRUE) is not intended to work.

There is a reason why there is no sort options, because estimates are grouped by models, and I'm assuming a case where you can also have more than 1 estimate. And I think you can't sort both by models/groups and within group by estimate in ggplot2.

library(lme4)
library(sjPlot)
library(parameters)

set.seed(101)
n <- 1e2 ## obs per group
ns <- 4  ## groups
dd <- data.frame(y = rpois(n*ns, 3),
                 f = factor(sample(1:20, size = n*ns, replace = TRUE)),
                 x1 = rnorm(n*ns),
                 x2 = runif(n*ns))
fits <- lapply(1:ns,
               function(i) glmer(y~x1+x2+(1|f),
                                 family = poisson,
                                 data = dplyr::sample_frac(dd, .6)))

plot_models(fits)

^{Created on 2022-12-17 with reprex v2.0.2}

You can still use parameters::compare_parameters(fits) to get the raw data, but then probably need to build the plot from scratch.

Thank you so much for all your help here!!!