如何使用数字数据订购ggplot图例？

Question

Trying to make a plot with ggplot2 I have difficulties to order the legend as I want. For debugging I made some example data and the issue doesn't occur though data and code are similar! I am confused of in which way ggplot sorts its data.

The legend should be sorted by its values since it's numeric, as it actually is in the example data, but not in my working data.

Here's my data...

structure(list(n = c(150000, 15000, 3000, 1500, 750), estimate = c(0.0485706666666667, 
0.0454933333333333, 0.0604, 0.0413333333333334, 0.0402666666666666
), se = c(0.00230392190029327, 0.00727258789388646, 0.0163963824219692, 
0.0229426160506936, 0.0324210840623078), t.value = c(21.0811338041398, 
6.25147550973637, 3.67790396107066, 1.80003749831851, 1.23787262678884
)), .Names = c("n", "estimate", "se", "t.value"), row.names = c("150000", 
"15000", "3000", "1500", "750"), class = "data.frame")

...and code I used to make the df

# allEst <- data.frame(rbind(est1, est2, est3, est4, est5))  
# names(allEst) <- names(est1.tmp)
# rownames(allEst) <- c(150000, 15000, 3000, 1500, 750)

Code for the plot

# confidence intervals
interval1 <- - qnorm((1 - .95) / 2)  # 5% 
interval2 <- - qnorm((1 - .99) / 2)  # 1% 

# Plot
library(ggplot2)
ep <- ggplot(allEst, aes(colour=rownames(allEst)))
ep <- ep + geom_hline(yintercept=0.05, colour=gray(1/2), lty=2)
ep <- ep + geom_linerange(aes(x=n,
                                ymin=estimate - se*interval1, 
                                ymax=estimate + se*interval1),
                            lwd=2)
ep <- ep + geom_pointrange(aes(x=n, y=estimate, 
                                 ymin=estimate - se*interval2, 
                                 ymax=estimate + se*interval2),
                             lwd=1, shape=21, fill="WHITE")
ep <- ep + scale_x_log10()
ep <- ep + coord_flip() 

print(ep)

Which gives me:

And here the toy example I created:

est1 <- c(1e5, 0.0485, 0.0023, 21.08)
est2 <- c(1e4, 0.0454, 0.0072, 6.25)
est3 <- c(1e3, 0.0604, 0.0163, 3.67)
est4 <- c(1e2, 0.0402, 0.0324, 1.23)

df <- data.frame(rbind(est1, est2, est3, est4))
rownames(df) <- c(100000, 10000, 1000, 100)
df

interval1 <- - qnorm(0.025); interval2 <- - qnorm(0.005)

library(ggplot2)
ep1 <- ggplot(df, aes(colour=rownames(df)))
ep1 <- ep1 + geom_hline(yintercept=0.05, colour=gray(1/2), lty=2)
ep1 <- ep1 + geom_linerange(aes(x=X1, ymin= X2 -  X3*interval1,
                                ymax= X2 +  X3*interval1),
                            lwd=2, position=position_dodge(width=1/2))
ep1 <- ep1 + geom_pointrange(aes(x=X1, y= X2, ymin= X2 - 
                                    X3*interval2, ymax= X2 + 
                                    X3*interval2),
                             lwd=1, position=position_dodge(width=1/2),
                             shape=21, fill="WHITE")
ep1 <- ep1 + scale_x_log10()
ep1 <- ep1 + coord_flip() 

print(ep1)

...and the plot w/ perfectly ordered (!) legend:

So, what's going on? Where am I blind?

Answer 1

Reorder color legend with breaks from scale_color_discrete :

# Adding sorted rownames to breaks
ep + scale_color_discrete(breaks = sort(as.numeric(rownames(allEst))))

Result plot:

Answer 2

You pass a character vector from rownames(allEst) to the aesthetics, which are sorted as a character:

sort(rownames(allEst))
[1] "1500"   "15000"  "150000" "3000"   "750" 

If sort is used on characters, it uses alphabetic sorting, where '1' and '12' are before '2' just like 'a' and 'ab' are before 'ba' . If you look at the difference between

sort(c(1:3, 10, 20, 30)) 

which returns

[1] 1 2 3 10 20 30 

and

sort(as.character(c(1:3, 10, 20, 30))) 

which returns

1 "1" "10" "2" "20" "3" "30"

it might be clearer.

One way of changing that would be using factor(sort(as.numeric(rownames(allEst)))) . Integrated in your code:

library(ggplot2)
ep <- ggplot(allEst, aes(colour=factor(sort(as.numeric(rownames(allEst))))))
ep <- ep + geom_hline(yintercept=0.05, colour=gray(1/2), lty=2)
ep <- ep + geom_linerange(aes(x=n,
                              ymin=estimate - se*interval1, 
                              ymax=estimate + se*interval1),
                          lwd=2)
ep <- ep + geom_pointrange(aes(x=n, y=estimate, 
                               ymin=estimate - se*interval2, 
                               ymax=estimate + se*interval2),
                           lwd=1, shape=21, fill="WHITE")
ep <- ep + scale_x_log10()
ep <- ep + coord_flip() 

print(ep)

Returns: