remove line and disconnect discrete points with ggeffects::plot(predict_response) and connect_lines

I'm using ggffects to plot predicted probabilities (options(ggeffects::ggeffects_margin = "marginalmeans")) of a multivariable GEE model with a binary predictor (coded as 1,2) modeled with geepack. I'm running into a new issue with some old code that I think was written under ggeffects version 1.5.0. I tried reinstalling that version without fixing the problem.

Is there a way to remove the line connecting discrete points on a plot?

Using this code

ggeffects::plot(
  predict_response(GEE model, terms = "x"),
  ci_style = "errorbar"
) +
  ggplot2::scale_x_continuous(
    labels = c("\n No",
               "\n Yes"),
    breaks = c(1, 2),
    expand = c(0,0)
  )

Produces this graph

I want to remove the line connecting the dots

when I try to remove the line connecting the discrete points using the connect_lines argument, it does not remove it, and I get the same graph as above

ggeffects::plot(
  predict_response(GEE model, terms = "x"),
  ci_style = "errorbar",
  connect_lines = FALSE
) +
  ggplot2::scale_x_continuous(
    labels = c("\n No",
               "\n Yes"),
    breaks = c(1, 2),
    expand = c(0,0)
  )

[1] stringr_1.5.1   forcats_1.0.0   patchwork_1.2.0 ggeffects_1.7.2 [5] geepack_1.3.12  gtsummary_2.0.3 ggplot2_3.4.4   dplyr_1.1.4

I also opened an issue at Github https://github.com/strengejacke/ggeffects/issues/596

One of the nice things about ggplot2 is that a ggplot object is basically a large nested list which in most cases can be manipulated, i.e. in your case we can remove the geom_line layer which adds the connecting line.

First, let's start with a minimal reproducible example of your issue based on the default example from geepack

library(geepack)
library(ggeffects)

### Example adapted from the geepack user guide
n_cluster <- 6
n_time <- 2
set.seed(1213)
timeorder <- rep(1:n_time, n_cluster)
x <- timeorder
idvar <- rep(1:n_cluster, each = n_time)
uuu <- rep(rnorm(n_cluster), each = n_time) # A 'random intercept'
yvar <- 1 + 2 * x + uuu + rnorm(length(x))
simdat <- data.frame(idvar, timeorder, x, yvar)

mod1 <- geeglm(yvar ~ x, id = idvar, data = simdat, corstr = "ar1")
###

p <- plot(
  predict_response(mod1, terms = "x"),
  ci_style = "errorbar"
) +
  ggplot2::scale_x_continuous(
    labels = c(
      "\n No",
      "\n Yes"
    ),
    breaks = c(1, 2)
  )

p

Now, to remove the connecting line one can have a look at the layers which make up the plot:

p$layers
#> [[1]]
#> mapping: group = ~.data[["group"]] 
#> geom_line: na.rm = FALSE, orientation = NA
#> stat_identity: na.rm = FALSE
#> position_identity 
#> 
#> [[2]]
#> geom_point: na.rm = FALSE
#> stat_identity: na.rm = FALSE
#> position_dodge 
#> 
#> [[3]]
#> mapping: ymin = ~.data[["conf.low"]], ymax = ~.data[["conf.high"]], shape = NULL 
#> geom_errorbar: na.rm = FALSE, orientation = NA, width = 0
#> stat_identity: na.rm = FALSE
#> position_dodge

From this output we see that there are three layers and that the first is a geom_line and that there is only one geom_line. To drop this layer we can set it to NULL:

p$layers[[1]] <- NULL

p