[R] ggplot stat_summary (mean_cl_boot)

David Winsemius dwinsemius at comcast.net
Wed Nov 9 22:46:55 CET 2011

On Nov 9, 2011, at 4:35 PM, Nathan Miller wrote:

> Sorry, I didn't realize I was being so obscure.
> Within ggplot it is possible to use stat_summary() to generate  
> confidence intervals about a mean. One method for generating these  
> CI assumes normality. The other uses bootstrapping to generate the  
> CI. I am using the second method which requires code like this
> stat_summary(fun.data="mean_cl_boot",  
> geom="errorbar",width=0.1,colour = "red")
> I've added some extra flourishes to make them look like errorbars,  
> alter the width and specify color.
> I would like some details regarding how this bootstrapped CI is  
> calculated. If I type "?mean_cl_boot" at the R command line I get a  
> minimal help file for "wrap_hmisc {ggplot2}" which is described   
> "wrap up a selection of Hmisc to make it easy to use with  
> stat_summary"
> I did not mean to suggest that ggplot2 calls Hmisc when I run  
> stat_summary(),

Actually it does.

> but simply that it appears that stat_summary() seems to have been  
> based upon a selection of Hmisc, hence I went looking in Hmisc to  
> try to find details regarding stat_summary(). I was unsuccessful in  
> this attempt.
> I don't believe a great deal of debugging is necessary. I am simply  
> looking for details regarding how "mean_cl_boot" works.

It doesn't. That is not the right name.

> If you don't have information regarding how it works (such as the  
> default number of resamplings) there is no need to respond.

Hadley's help files in ggplot2 are terse (or the links to outside  
resources crash my R sessions)  to the point of being too frustrating  
for me to consider using that package, so I don't know if optional  
parameters can be passed to the Hmisc functions. If they are,  then  
you should set reps=TRUE and then see what happens to the number of  
reps from the returned object ... if the wrap_hmisc function does  
happen to catch it.

 > x <- rnorm(100)
 > smean.cl.boot(x)
       Mean      Lower      Upper
-0.0211511 -0.2013623  0.1469728

 > smean.cl.boot(x, reps=TRUE)
        Mean       Lower       Upper
-0.03465361 -0.21233213  0.15178655
    [1]  0.0283330508 -0.1250784237  0.0744640779  0.1310826601  
    [6]  0.0629291714  0.0145916070 -0.0860141221  0.0549134451   
snipped pages of intervening output.
  [991]  0.1029922424  0.0613358597 -0.0645577851 -0.1664905503  
  [996] -0.0751783377 -0.0043747455 -0.1155948060 -0.0750075659   

I don't see where the number of reps is returned, but the B setting  
defaults to 1000.

> Thanks for any assistance,
> Nate
> On Wed, Nov 9, 2011 at 1:10 PM, David Winsemius <dwinsemius at comcast.net 
> > wrote:
> On Nov 9, 2011, at 2:59 PM, Nathan Miller wrote:
> Hello,
> This is a pretty simple question, but after spending quite a bit of  
> time
> looking at "Hmisc" and using Google, I can't find the answer.
> If I use stat_summary(fun.data="mean_cl_boot") in ggplot to generate  
> 95%
> confidence intervals, how many bootstrap iterations are preformed by
> default? Can this be changed? I would at least like to be able to  
> report
> the number of boot strap interations used to generate the CIs.
> I haven't been able to find "mean_cl_boot" as a function itself or
> something ressembling it in the Hmisc documentation, but it seems as  
> though
> Hmisc is wrapped up in stat_summary() and is called to compute
> "mean_cl_boot".
> You seem really, really confused (and you offer very little in the  
> way of context to support debugging efforts). You are referring to  
> ggplot functions. As far as I know there is no connection between  
> the Hmisc and ggplot (or ggplot2) packages. Al things change, I  
> know, but Frank just completed switching over to Lattice a couple of  
> years ago.
> -- 
> David Winsemius, MD
> West Hartford, CT

David Winsemius, MD
West Hartford, CT

More information about the R-help mailing list