CODEWITHSUNDEEP
rggplot2sjplot
Nate Hall
Nate Hall

Reputation: 84

specific manipulation of axis limits and scales in sjplot::plot_model of logistic glmer that generally preserves plot_model output

Sorry if this is a double post, I've seen some similar posts but haven't found the exact solution I'm looking for.

I fit a glmer model to some reaction time data of the form:

m19 <- glmer(response ~ trial_z*stim + cond  + (1 + stim | subj_idx), 
             na.action = na.exclude,  
             data = gng, family = "binomial", 
             control = glmerControl(optCtrl = list(maxfun = 50000)))

where: response is binary (correct response, incorrect response) on a standard go/no-go task.

I want to plot odds ratios of the fixed effects using the sjplot::plot_model function but don't seem to be able to get things looking quite right.

I want to plot these odds ratios for the fixed effects in my model:

library(broom.mixed)
tidy(m19,conf.int=TRUE,exponentiate=TRUE,effects="fixed")

which looks like:

# A tibble: 7 x 8
  effect term              estimate std.error statistic  p.value  conf.low conf.high
  <chr>  <chr>                <dbl>     <dbl>     <dbl>    <dbl>     <dbl>     <dbl>
1 fixed  (Intercept)      414.      123.          20.2  6.98e-91 231.       742.    
2 fixed  trial_z            0.697     0.0751      -3.35 7.97e- 4   0.564      0.860 
3 fixed  stimNoGo           0.00925   0.00257    -16.9  6.80e-64   0.00537    0.0159
4 fixed  condThreeGo        1.48      0.151        3.82 1.32e- 4   1.21       1.80  
5 fixed  condFiveGo         1.69      0.174        5.07 4.08e- 7   1.38       2.06  
6 fixed  condSevenGo        1.51      0.151        4.07 4.73e- 5   1.24       1.83  
7 fixed  trial_z:stimNoGo   1.30      0.149        2.28 2.25e- 2   1.04       1.63

The following sjplot call produces the following plot object:

gg <- plot_model(m19, type="std", show.p=TRUE, ci.lvl=.95, dot.size=5, 
                 line.size=2, vline.color = "black", axis.lim=c(.001, 2.25)) + 
  theme_bw(base_size=24) +
  theme(panel.grid.major.y=element_blank()) +
  ggtitle("") + 
  ylab("Odds Ratio")
plot(gg)

attempt 1

So I tried editing axis.limits within the call:

gg2 <- plot_model(m19, type="std", show.p=TRUE, ci.lvl=.95, dot.size=5, 
                  line.size=2, vline.color = "black", axis.lim=c(.001, 11)) + 
  theme_bw(base_size=24) +
  theme(panel.grid.major.y=element_blank()) +
  ggtitle("") + 
  ylab("Odds Ratio")

plot(gg2)

Somewhat annoyingly, bumping the second concatenated axis.lim element to anything between 1 and 10 produced my attempt 1, but then anything above 10 sets that limit in the plot to 100 (not sure why this is...)

attempt 2

So I tried, based off other posts to edit the axes post-hoc:

gg + ylim(.001, 2.25)

attempt 3

My question:

I'd like a plot that is somewhat of a combination of attempt 2 and attempt 3.

In 2 I like that the scaling below and above 1 are different and that 1 (equal odds) has a special denotion that demonstrates that CIs that overlap with 1 represent a null effect. If I want than within plot_model()'s capabilities I am stuck at how I can get the axis to the right of that 1 mark to look more like the scale in attempt 3, which wipes out the nice delineation of odds for and against and sets the scale as uniform across all y values (also, this was certainly confusing at first, given that I'm looking to manage what the user sees on the x-axis. I assume there's a coord_flip call somewhere in the function...).

If I can clarify further lmk. Thanks for your input!

Upvotes: 1

Views: 4560

Answers (1)

Daniel
Daniel

Reputation: 7832

I think this might work:

gg + scale_y_log10(limits = c(0.001, 2.25))

Example:

library(sjPlot)
library(ggplot2)
model <- glm(vs ~ mpg + cyl, data = mtcars, family = "binomial")
p <- plot_model(model)
p + scale_y_log10(limits = c(0.001, 2.5))
#> Scale for 'y' is already present. Adding another scale for 'y', which will
#> replace the existing scale.

Created on 2020-06-27 by the reprex package (v0.3.0)

Upvotes: 1

Related Questions