plot more than one column+facet_grid (ggplot)

Question

my data frame goes like this:

    NUMBER  MEDIUM  DATE    DAY TREAT   NITRAT  SILI    PHOSPHAT
1   NUT1    15/07/2016  two NBNA    20.831  20.946  0.518
2   NUT2    15/07/2016  two NBNA    26.594  25.075  1.059
3   NUT3    15/07/2016  two NBNA    47.565  42.088  7.733
4   NUT4    15/07/2016  two NBNA    61.320  53.264  6.331
5   NUT5    15/07/2016  two NBNA    64.918  52.058  4.889
6   NUT1    15/07/2016  two NBNA    19.876  22.000  0.852
7   NUT2    15/07/2016  two NBNA    29.642  29.572  1.184
8   NUT3    15/07/2016  two NBNA    37.969  35.015  4.293
9   NUT4    15/07/2016  two NBNA    56.724  45.986  6.880
10  NUT5    15/07/2016  two NBNA    66.006  53.175  5.893
11  NUT1    15/07/2016  two NBNA    21.835  20.710  0.770
12  NUT2    15/07/2016  two NBNA    25.996  26.213  1.149
13  NUT3    15/07/2016  two NBNA    48.526  46.167  3.977
14  NUT4    15/07/2016  two NBNA    56.767  47.089  6.684
15  NUT5    15/07/2016  two NBNA    64.841  54.039  6.628
16  NUT1    15/07/2016  two NBA1    13.179  18.723  0.826
17  NUT2    15/07/2016  two NBA1    18.631  23.537  0.859

I would like to put in a ggplot the values of NITRAT, SILI and PHOSPHAT(with error bars) sorted by MEDIUM. In summary, something like this(but also with error bars): to do so, I have tried the function melt:

df <- melt(nut, id.vars='MEDIUM')

however using this function does not allow me to use facet_grid afterwards

ggplot() +
  geom_bar(data=df, aes(x=MEDIUM, y=value, fill=MEDIUM),
           stat="identity", colour="black") + facet_grid(~TREAT)

Answer 1

You've got a few issues to clean up here:

1. First, only the columns that you want to "stack" (that is, use for the fill aesthetic) should not be in id.vars:

   df <- melt(nut, id.vars=c('MEDIUM','NUMBER','DATE','DAY','TREAT')
   

   Try the above code with different combinations of id.vars to get some intuition for how it works.

2. fill=variable will now give you bars for each of the three types of measurements (NITRAT, SILI, and PHOSPHAT). TREAT will still be a column in the melted data frame because it was one of the id.vars, so you can use it for facetting.

3. One other issue is that it looks like you actually want the bars to be the mean of all the values that go into that bar. So you don't want stat="identity". Instead, use stat_summary to calculate the means and plot them as bars.

4. To add errorbars, use another call to stat_summary. I've used bootstrapped confidence intervals (mean_cl_boot). For classical confidence intervals, use mean_cl_normal.

   pd=position_dodge(0.8)
   
   ggplot(df, aes(x=MEDIUM, y=value, fill=variable)) +
     stat_summary(fun.y=mean, geom="bar", width=0.7, position=pd) + 
     stat_summary(fun.data=mean_cl_boot, geom="errorbar", position=pd, width=0.3) +
     facet_grid(. ~ TREAT) +
     theme_bw()
   

For a plot like this, points might be better than bars. Here we plot the error bars first instead of second because we want the points "on top of" the bars. It won't matter here because both are the same color, but I do it out of habit as I often use different colors or transparency for the error bars:

pd=position_dodge(0.4)

ggplot(df, aes(x=MEDIUM, y=value, colour=variable)) +
  stat_summary(fun.data=mean_cl_boot, geom="errorbar", position=pd, width=0.3) +
  stat_summary(fun.y=mean, geom="point", position=pd) + 
  facet_grid(. ~ TREAT) +
  theme_bw()