[R] Why doesn't formatC( x, digits=2, format="g") doesn't always give 2 sig figs?

[Duncan Murdoch]

On 05/06/2008 8:48 PM, Peter Dunn wrote:
> Hi all
> 
> I am not a C programmer, but I am trying to understand  formatC  to 
> get consistent printing of reals to a given number of significant 
> digits.
> 
> Can someone please explain this to me?  These first three give what 
> I expect on reading ?formatC:
> 
> 
>> formatC(0.0059999, digits=2,format="fg",flag="#")
> [1] "0.0060"
>> formatC(0.59999, digits=2,format="fg",flag="#")
> [1] "0.60"
>> formatC(5.9999, digits=2,format="fg",flag="#")
> [1] "6.0"
> 
> 
> This seems consistent with what I read (but perhaps do not 
> understand) in ?formatC, where I read this:
> 
> 
> 	digits   the desired number of digits after the decimal point
> 	(format = "f") or significant digits (format = "g", = "e" or
> 	 = "fg"). 
> 
> 
> Since I am using  format="fg" and digits=2, so I am expecting two 
> significant digits to always show, which I have above. So I fail to 
> understand this:
> 
> 
>> formatC(0.000059999, digits=2,format="fg",flag="#")
> [1] "0.00006"
>> formatC(0.000059, digits=2, format="fg",flag="#")
> [1] "0.000059"
> 
> 
> I was expecting both of these to produce "0.000059".  But in the 
> first case above, I get one significant digit only.  
> 
> I'm obviously misunderstanding something; can someone enlighten me?  
> (No doubt, someone will point out a nuance of the help files I 
> didn't understand!) 
> 
> Also, since the above obviously doesn't do what I hoped 
> (consistently printing two sig figs), could someone also explain 
> how I can consistently get two significant figures in situation 
> like above?

I'll let someone else comment on whether this is a bug in formatC; it 
looks like a bug or a documentation error to me.

You can get 2 sig figs in your situation like this:

sprintf("%.6f",0.000059999)

but obviously this doesn't help with numbers on a different scale.  For 
those, I think you're going to have to put together something like

  x <- 0.0000599999
  sprintf("%.*f", -trunc(log10(x))+2, x)

(which only works for numbers less than 1).

Duncan Murdoch