Using stat_summary to generate error bars for grouped data.

[dha 2001]

Hi,

I'm trying to generate y-error bars on the fly using stat_summary and
seem to be missing something. What I would like is for the error bar
calculation to be applied to each group. The following example, using
the "mtcars" data is close but seems to be generating the error bars
based on the entire column and applying them to each point.

cars <- ggplot(mtcars, aes(x=gear, y=mpg, group=as.factor(cyl),
colour=as.factor(cyl), ymin=(mpg-sd(mpg)), ymax=(mpg+sd(mpg)) )) +
stat_summary(fun.y=mean, geom="point") +
stat_summary(fun.y=mean, geom="line") +
geom_errorbar(width=0.1)
print(cars)

Thanks for any clues!

-David

[Hadley Wickham]

Hi David,

The summaries are fine:

ggplot(mtcars, aes(x=gear, y=mpg, group=as.factor(cyl),
colour=as.factor(cyl), ymin=(mpg-sd(mpg)), ymax=(mpg+sd(mpg)) )) +
stat_summary(fun.y=mean, geom="point") +
stat_summary(fun.y=mean, geom="line")

it's the error bar that's being created for every point. What are you
trying to do?

Hadley

> --
> You received this message because you are subscribed to the ggplot2 mailing list.
> Please provide a reproducible example: http://gist.github.com/270442
>
> To post: email ggp...@googlegroups.com
> To unsubscribe: email ggplot2+u...@googlegroups.com
> More options: http://groups.google.com/group/ggplot2
>

--
Assistant Professor / Dobelman Family Junior Chair
Department of Statistics / Rice University
http://had.co.nz/

[dha 2001]

I'd like the error bar for each summarized point to represent the
standard deviation of the points that were summarized. e.g. I'd like
the error bar for the point at 3 gears, in the group if 8 cylinder
cars to be +/- sd(mpg of cars with 3 gears and 8 cylinders) and
plotted at mean(mpg of cars with 3 gears and 8 cylinders).

[Joe]

Unless there's an easier within ggplot approach, what I'd do is write
a function to calculate the upper and lower ends of the error bar.
Here, I've got two that calculate 95% confidence intervals.

errorUpper <- function(x){
x.mean <- mean(x)
x.sd <- sd(x)

SEM <- x.sd / (sqrt(length(x)))

return(x.mean + (SEM*1.96))
}

errorLower <- function(x){
x.mean <- mean(x)
x.sd <- sd(x)

SEM <- x.sd / (sqrt(length(x)))

return(x.mean - (SEM*1.96))
}

Now, the ggplot code would look like this:

ggplot(mtcars, aes(x=gear, y=mpg, group=as.factor(cyl),
colour=as.factor(cyl))) +

stat_summary(fun.y=mean, geom="point") +

stat_summary(fun.y=mean, geom="line")+
stat_summary(fun.ymax = errorUpper, fun.ymin = errorLower, geom =
"errorbar")

A ribbon might look nicer

ggplot(mtcars, aes(x=gear, y=mpg, group=as.factor(cyl),
colour=as.factor(cyl))) +
stat_summary(fun.ymax = errorUpper, fun.ymin = errorLower, geom =
"ribbon", alpha = 0.6)+

stat_summary(fun.y=mean, geom="point") +
stat_summary(fun.y=mean, geom="line")

-Joe

[Luciano Selzer]

Have a look at smean.cl.normal() in Hmisc, it does what you want.
Luciano

2010/7/1 Joe <jofr...@gmail.com>

[dha 2001]

Thanks for the help. That did the right thing.

[bb]

Can you please post a snippet demonstrating this? I'm not sure how to
get the info out of the vector returned by smean.cl.normal()

Thanks

[Luciano Selzer]

Hi,here's a snippet. I hope it helps ;)

library(Hmisc); library(ggplot2)

data(diamonds)

#Mean plus 95% confidence interval (default)

price.by.cut <- ddply(diamonds, .(cut), function(df) smean.cl.normal(df$price))

qplot(cut, Mean, ymin = Lower, ymax = Upper, data = price.by.cut,

    geom = c("bar", "errorbar"))

#Mean +- 2 * SD

price.by.cut <- ddply(diamonds, .(cut), function(df) smean.sdl(df$price))

qplot(cut, Mean, ymin = Lower, ymax = Upper, data = price.by.cut,

    geom = c("bar", "errorbar"))

#Mean +- SD

price.by.cut <- ddply(diamonds, .(cut), function(df) smean.sdl(df$price, mult = 1))

qplot(cut, Mean, ymin = Lower, ymax = Upper, data = price.by.cut,

    geom = c("bar", "errorbar"))

    

#Mean +- SE

price.by.cut <- ddply(diamonds, .(cut), function(df) smean.sdl(df$price,

      mult = sqrt(length(df$price))^-1))

qplot(cut, Mean, ymin = Lower, ymax = Upper, data = price.by.cut,

    geom = c("bar", "errorbar"))

Luciano

2010/7/13 bb <bra...@peds.ucsf.edu>

smean.cl.normal()