1

I am trying dodge a pair of means and confidence intervals for two models using ggplot. I've looked into similar questions like this and this, but their code doesn't address my problem.

I have the following data frame (note that condition is a factor, values increase as you descend down the data frame):

>print(dat)
    condition        pH    lowerH    upperH model
1   Black 650 0.7863719 0.6863438 0.8548062   pdp
2   Black 650 0.8428675 0.7123000 0.9574000   ddm
3   White 650 0.8133078 0.7310262 0.8864712   pdp
4   White 650 0.8211373 0.6838000 0.9488000   ddm
5   Black 800 0.8754109 0.7908725 0.9306012   pdp
6   Black 800 0.9408192 0.8410000 0.9941000   ddm
7   White 800 0.9044102 0.8349587 0.9623462   pdp
8   White 800 0.9183600 0.8145000 0.9867000   ddm
9   Black 950 0.9003072 0.8403200 0.9601112   pdp
10  Black 950 0.9569349 0.8812000 0.9967000   ddm
11  White 950 0.9131546 0.8494812 0.9648262   pdp
12  White 950 0.9473685 0.8683000 0.9934000   ddm
13 Black 1100 0.9091797 0.8342850 0.9622412   pdp
14 Black 1100 0.9708420 0.8864000 0.9977000   ddm
15 White 1100 0.9274862 0.8636063 0.9691300   pdp
16 White 1100 0.9570067 0.8727000 0.9954000   ddm

I am trying to plot the mean (pH) and confidence interval (lowerH, upperH) for each condition as a function of the model used (model). What I'd like is for the means and CIs to be dodged so that they are not overlapping. I've written the following ggplot code:

ggplot(dat, aes(x = condition, y = pH)) +
  geom_point(size = 3, shape = 5) + #pred cond mean
  geom_pointrange(aes(ymin = lowerH, ymax = upperH), size = .2) + #95% HDI
  coord_cartesian(ylim = c(.5, 1.01)) + #adjust x-axis
  ylab("Hit Rate") + #label y axis
  theme_bw() + #black and white theme
  theme(axis.title.x = element_blank(), 
    panel.background = element_blank(), 
    panel.grid = element_blank(),
    panel.border = element_rect(colour = "black", fill = NA, size = 1))

Which creates this output

As you can see, the means and CIs for both models overlap. However, when I try to dodge them, as I've done in the code below, I only succeed in moving all data points to the left (with the model means and CIs still overlapping).

ggplot(dat, aes(x = condition, y = pH)) +
  geom_point(size = 3, shape = 5, position = position_dodge(5)) + #pred cond mean
  geom_pointrange(aes(ymin = lowerH, ymax = upperH), size = .2) + #95% HDI
  coord_cartesian(ylim = c(.5, 1.01)) + #adjust x-axis
  ylab("Hit Rate") + #label y axis
  theme_bw() + #black and white theme
  theme(axis.title.x = element_blank(), 
    panel.background = element_blank(), 
    panel.grid = element_blank(),
    panel.border = element_rect(colour = "black", fill = NA, size = 1))

What I'm looking for is a way to dodge the model means and CIs so that they are nonoverlapping. Sample code is especially helpful.

Community
  • 1
  • 1
David Johnson
  • 439
  • 1
  • 3
  • 13
  • Possible duplicate of [Dodging points and error bars with ggplot](http://stackoverflow.com/questions/25789434/dodging-points-and-error-bars-with-ggplot) – aosmith Oct 14 '15 at 03:19

1 Answers1

2

You need to add a level of group or color to the plot. Here I'm using col, you can also use group to keep the colours the same:

ggplot(dat, aes(x = condition, y = pH, col = model)) +
    geom_point(size = 3, shape = 5, position = position_dodge(1)) +
    geom_pointrange(aes(ymin = lowerH, ymax = upperH), 
                    position = position_dodge(1), size = .2)

enter image description here

jeremycg
  • 24,657
  • 5
  • 63
  • 74
  • Thanks. This solved my problem! One thing I'd like to point out is that any other data that is being used in the plot must also have columns with "x", "y", and "group", even if the data are not grouped (i.e., an observed mean that is not a function of either model--used to compare the models to). – David Johnson Oct 13 '15 at 23:26