Plotting two lines corresponding to two different variables (and not having one of them be the legend) using plot_model or so

Question

I am performing a multilevel (logistic) regression with two predictors, B and C. B and C have similar ranges so they can reasonably be shown using the same units in the x-axis. I want to do this because I want to highlight in my figure how they have different slopes.

When I try to do so using plot_model and the code below, one of the variables gets sent to the legend and I get this. However, I would rather have this.

Can this be done?

Thanks for the help!

df_test <- data.frame('subj' = c('Joe', 'Joe', 'Joe', 'Moe', 'Moe', 'Moe'), 'A' = c(1, 0, 1, 0, 1, 1), 'B' = c(3, 2, 1, 4, 3, 3), 'C' = c(3, 3, 2, 1, 3, 0))

m = glmer(A ~ B + C + (B + C| subj), data=df_test, family='binomial')

print(plot_model(m, terms = c('B', 'C'), type='pred'))

Answer 1

You want to "overlay" two plots, one with predicted values for B, and one for C. Since sjPlot uses the ggeffects-package to compute marginal effects, I suggest you use ggeffects directly, and build your own ggplot-object.

Since B and C have different ranges, you could force the range to be the same for both (see second example below).

library(lme4)
library(ggeffects)
library(ggplot2)

df_test <-
  data.frame(
    'subj' = c('Joe', 'Joe', 'Joe', 'Moe', 'Moe', 'Moe'),
    'A' = c(1, 0, 1, 0, 1, 1),
    'B' = c(3, 2, 1, 4, 3, 3),
    'C' = c(3, 3, 2, 1, 3, 0)
  )

m <- glmer(A ~ B + C + (B + C | subj), data = df_test, family = 'binomial')
#> boundary (singular) fit: see ?isSingular

# get predictions for each predictor, combibe results
dat <- get_complete_df(ggpredict(m))

ggplot(dat, aes(x = x, y = predicted, colour = group, fill = group)) +
  geom_ribbon(aes(ymin = conf.low, ymax = conf.high), alpha = .1, colour = NA) +
  geom_line()

# get predictions for each predictor, combibe results
dat1 <- ggpredict(m, "B [0:4]")
dat1$group <- "B"
dat2 <- ggpredict(m, "C [0:4]")
dat2$group <- "C"
dat <- rbind(dat1, dat2)

ggplot(dat, aes(x = x, y = predicted, colour = group, fill = group)) +
  geom_ribbon(aes(ymin = conf.low, ymax = conf.high), alpha = .1, colour = NA) +
  geom_line()

^{Created on 2019-07-24 by the reprex package (v0.3.0)}