Reorder legend without changing order of points on plot

I keep running into this issue in ggplot2, perhaps someone can help me.

I have a plot where the order of the variables in the legend is in reverse order to how they are displayed 开发者_开发技巧on the plot.

For example:

df=data.frame(
 mean=runif(9,2,3),
 Cat1=rep(c("A","B","C"),3),
 Cat2=rep(c("X","Y","Z"),each=3))

dodge=position_dodge(width=1)
ggplot(df,aes(x=Cat1,y=mean,color=Cat2))+
 geom_point(aes(shape=Cat2),size=4,position=dodge)+
 scale_color_manual(values=c("red","blue","black"))+
 scale_shape_manual(values=c(16:19))+
 coord_flip()

produces:

So the points are displayed on the plot as Cat2=Z,Y, then X (black diamonds, blue triangle, red circle) but in the legend they are displayed as Cat2=X, Y, then Z (red circle, blue triangle, black diamond).

How can I reverse the order of the legend without shifting the points on the plot? Reordering the factor produces the opposite problem (points on the plot are reversed).

Thanks!

To flesh out Hadley's comment, we would do something like this:

ggplot(df,aes(x=Cat1,y=mean,color=Cat2))+
 geom_point(aes(shape=Cat2),size=4,position=dodge)+
 scale_color_manual(values=c("red","blue","black"),breaks = c('Z','Y','X'))+
 scale_shape_manual(values=c(16:19),breaks = c('Z','Y','X'))+
 coord_flip()

Note that we had to set the breaks in both scales. If we did just one, they wouldn't match, and ggplot would split them into two legends, rather than merging them.

As far as i understand what you want to achieve, this simple manipulation does the trick for me:

1. define a Cat2 as a factor (with the levels in the adequate order) and
2. chage the order of colours and shapes to match the levels order (in the scale_manual commands)

Here is the code to do it:

library(ggplot2)

df=data.frame(
    mean=runif(9,2,3),
    Cat1=rep(c("A","B","C"),3),
    Cat2=factor(rep(c("X","Y","Z"),each=3), levels=c('Z', 'Y', 'X'))) 

dodge=position_dodge(width=1)
ggplot(df,aes(x=Cat1,y=mean,color=Cat2))+
    geom_point(aes(shape=Cat2),size=4,position=dodge)+
    scale_color_manual(values=c("black","blue","red"))+
    scale_shape_manual(values=c(18:16))+
    coord_flip()