R help - Jan 2009 - NA and NaN question

Hi all,

I ran into a problem in some of my code that could be traced back to
'mean'
sometimes returning NA and sometimes NaN, depending on the value of na.rm:
> mean(c())
[1] NA
> mean(c(NA),na.rm=T)
[1] NaN

However, I don't understand the reasoning behind this and would appreciate
and
explanation. 

I understand that the mean of an empty vector is not definied, but I don't 
understand why it matters whether the vector was empty from the beginning or 
only after removing the NAs.

Pascal Niklaus

Pascal A. Niklaus wrote:
> Hi all,
> 
> I ran into a problem in some of my code that could be traced back to
'mean'
> sometimes returning NA and sometimes NaN, depending on the value of na.rm:
> 
>> mean(c())
> [1] NA
> 
>> mean(c(NA),na.rm=T)
> [1] NaN
> 
> However, I don't understand the reasoning behind this and would
appreciate and
> explanation. 
> 
> I understand that the mean of an empty vector is not definied, 
Not so, it is well-defined as 0/0 = NaN.
> but I don't 
> understand why it matters whether the vector was empty from the beginning 
You didn't try that case:  mean(numeric(0)) is also NaN.  The issue is that

 > typeof(c())
[1] "NULL"

is not numeric (not evan a vector), and so mean() of it is undefined.

 > or only after removing the NAs.

Speculation (and wrong).

> Pascal Niklaus
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

Pascal A. Niklaus wrote:
> Hi all,
>
> I ran into a problem in some of my code that could be traced back to
'mean'
> sometimes returning NA and sometimes NaN, depending on the value of na.rm:
>
>   
>> mean(c())
>>     
> [1] NA
>
>   
>> mean(c(NA),na.rm=T)
>>     
> [1] NaN
>
> However, I don't understand the reasoning behind this and would
appreciate and
> explanation. 
>   
note the types:

typeof(c())
typeof(c(NA))
typeof(c(NA)[-na.omit(c(NA))])

now,

mean(NULL)
mean(logical(0))

mean(c())
# NA, because you take the mean of a vector of non-{numeric,logical}
type (see the warning message)
mean(c(NA), na.rm=TRUE)
# NaN, because you take the mean of a zero-length logical vector
mean(c(NA), na.rm=FALSE)
# NA, because you take the mean of a logical vector containing an NA


you can argue that ?mean underspecifies this (it doesn't say anything
about the value for a zero-length logical, numeric, or complex vector,
though you can guess it will be the value of 0/0).

vQ