Question

I fit a count model to a vector of actual data and would now like to plot the actual and the predicted as a grouped (dodged) bar chart. Since this is a count model, the data are discrete (X=x from 0 to 317). Since I am fitting a model, I only have already-tabulated data for the predicted values.

Here is how my original data frame looks:

  actual predicted
1   3236 3570.4995
2   1968 1137.1202
3    707  641.8186
4    302  414.8763
5    185  285.1854
6    104  203.0502

I transformed the data to be plotted with ggplot2:

melted.data <- melt(plot.data)
melted.data$realization <- c(rep(0:317, times=2))
colnames(melted.data)=c('origin','count','realization')

So that my data frame now looks like this:

head(melted.data)
  origin count realization
1 actual  3236           0
2 actual  1968           1
3 actual   707           2
4 actual   302           3
5 actual   185           4
6 actual   104           5
> tail(melted.data)
       origin        count realization
631 predicted 1.564673e-27         312
632 predicted 1.265509e-27         313
633 predicted 1.023552e-27         314
634 predicted 8.278601e-28         315
635 predicted 6.695866e-28         316
636 predicted 5.415757e-28         317

When I try to graph it (again, I’d like to have the actual and predicted count –which is already tabulated in the data– by discrete realization), I give this command:

ggplot(melted.data, stat="identity", aes(x=realization, fill=origin)) + geom_bar(position="dodge")

Yet it seems like the stat parameter is not liked by ggplot2, as I don’t get the correct bar height (which would be those of the variable “count”).

Any ideas?

Thanks,

Roberto.

Answer 1

You need y-values in the aes mapping if you use stat_identity (column count). Try the following:

ggplot(melted.data, aes(x=realization, y=count, fill=origin)) + 
       stat_identity(position="dodge", geom="bar")

or

ggplot(melted.data, aes(x=realization, y=count, fill=origin)) + 
       geom_bar(position="dodge", stat="identity")