ggplot with error bars for two measurement variables and one factor

Question

I know how to create regular plots with error bars for, say, one factor (e.g. experiment) and one measurement (e.g. quality). I first summarize the data to get mean and CI using the summarySE function given on this site. For example:

   hrc_id experiment  N  quality        sd         se         ci
        0      FB_IS 77 3.584416 0.6757189 0.07700532 0.15336938
        0     FB_ACR 77 3.779221 0.6614055 0.07537416 0.15012064
        1      FB_IS 77 3.038961 0.7854191 0.08950681 0.17826826
        1     FB_ACR 77 3.129870 0.8483831 0.09668223 0.19255935
...

That way I could plot:

ggplot(d, aes(hrc_id, quality), quality, color = experiment)) + 
    geom_point(position = position_dodge(width = .5)) +
    geom_errorbar(aes(ymin = quality - ci, ymax = quality + ci), width = .5, position = "dodge")

---

However, now I have to do the same with two measurements—not just quality, but also confidence. For example, my data might look as follows:

hrc_id confidence confidence_ci  quality quality_ci
     0   3.573718    0.02068321 4.576923 0.02864818
     1   3.403846    0.03193104 1.658120 0.04441434
    10   3.160256    0.02520483 3.038462 0.04476492
...

How would I go about plotting confidence (with confidence_ci) and quality (with quality_ci) next to each other, for each hrc_id?

I thought I could melt the dataframe so that confidence and quality would be the measurement variables, but then I lose the CI values that belong to them.

Answer 1

You can convert your dataframe to long format with grouped columns in one step using reshape(...). Assuming your dataframe is df:

gg <- reshape(df,idvar="hrc_id",               # idvar: identifies cases
              times=c("confidence","quality"), # group of columns to be reshaped
              timevar="measurement",           # column name to use for grouping vars
              varying=2:5,                     # columns are to be reshaped
              v.names=c("value","value.ci"),   # column names for reshaped values
              direction="long")                # convert to long format
gg
#               hrc_id measurement    value   value.ci
# 0.confidence       0  confidence 3.573718 0.02068321
# 1.confidence       1  confidence 3.403846 0.03193104
# 10.confidence     10  confidence 3.160256 0.02520483
# 0.quality          0     quality 4.576923 0.02864818
# 1.quality          1     quality 1.658120 0.04441434
# 10.quality        10     quality 3.038462 0.04476492

As far as I know you cannot do this with melt(...) - you'd have to use the rbind approach mentioned in your comment.

Answer 2

Your data frame should eventually look like this (melt might be the right tool to use, but I don't recall the syntax at the moment):

hrc_id measurment value ci  
 0   confidence 3.573718    0.02068321 
 0   quality    4.576923    0.02864818

Then you can plot using:

p = ggplot(d, aes(x = hrc_id, y = value, color = measurment)) +
  geom_errorbar (aes(ymin = value - ci, ymax = value + ci)) +
  geom_point(position = position_dodge(width = .5))
p