[R] Hopefully an easy error bar question

[Frank E Harrell Jr]

Ben Bolker wrote:
> Martin Maechler <maechler <at> stat.math.ethz.ch> writes:
> 
>>>>>>> "FEH" == Frank E Harrell <f.harrell <at> vanderbilt.edu>
>>>>>>>     on Sat, 06 Sep 2008 07:19:33 -0500 writes:
>>     FEH> See http://biostat.mc.vanderbilt.edu/DynamitePlots for many reasons not 
>>     FEH> to use dynamite plots.
>>
>> Ah!  Very good!  Let's hope this eventually leads to progress 
>> i.e., to much less use of these..
>>
>> Thank you, Frank!
>> Martin Maechler, ETH Zurich
>>
>>     FEH> -- 
> 
>   Yes, but:
> 
> http://emdbolker.wikidot.com/blog:dynamite
> 
> (repeats Frank's criticisms, but also raises
> the point that boxplots show the range of the
> data, but not much about the statistical inferences
> on the data ...)
> 
>   comments welcome.
> 
>   Ben Bolker

Hi Ben,

I enjoyed the blog.

Depending on the sample size I prefer extended box plots (e.g., 
?panel.bpplot in the Hmisc package) with raw data superimposed.  The 
3-number summary from the box plot does not contain enough information 
when the sample size is moderate or large IMHO.  But for small samples, 
raw data points supplemented by the 3 quartiles and the mean work well.

Frank

-- 
Frank E Harrell Jr   Professor and Chair           School of Medicine
                      Department of Biostatistics   Vanderbilt University