R devel - Dec 2004 - R stat functions do not work as stated on the mannual (PR#7419)

Dear R Developers:

I have been playing with R, release 2.0.1 for a week now and have detected that
all stat functions related to distribution probabilities have the same problem:

1.- According to the manual the log.p parameter is always the last one.
2.- When you use the software, the last parameter seems to be lower.tail

Example:
> pt (1.1, 5)
[1] 0.8392746
> pt (1.1, 5, F)
[1] 0.8392746
> pt (1.1, 5, F, T)
[1] 0.8392746
>=0D
On this example, I have used the Student T distribution. The result of this
example has been tested with the stat calculator at
http://calculators.stat.ucla.edu/.
1.- This first line shows that when this function has two arguments, the log.p
default value is false, and the lower.tail is true.
2.- The second line shows that when this function has 3 arguments, the last one
is the log.p argument and the lower.tail is taken by default to true.
3.- The third line shows that when this function has 4 arguments, the third one
keeps on being the log.p argument and the lower.tail is the last one.
4.- Acording to the mannual, the lower.tail should be the third argument and not
the last one.

Best regards,
=0D
Jos=E9 Luis Casado Mart=EDnez
------------------------------------------------------------------
European Computing Consultants
C/ Hermanos Garc=EDa Noblejas, N=BA 39, 5=AA, N 1
28037 Madrid
Telf.: 34-91-406 19 15. Fax: 34-91-406 19 16
Movil: 34-607-750 316
------------------------------------------------------------------

The last line shows that the sum of probabilities of 1.1 using 5 degrees of
freedom adding the lower tail (pt (1.1, 5, F, T)) and not adding the lower tail
( pt (1.1, 5, F, F)) is 1. This is not the case with log (p).



_____________________________________________________________________
Mensaje analizado y protegido, tecnologia antivirus www.trendmicro.es
	[[alternative HTML version deleted]]

On Tue, 14 Dec 2004 12:27:17 +0100 (CET), casadoj@ecc.es wrote:
>
>Dear R Developers:
>
>I have been playing with R, release 2.0.1 for a week now and have detected
>that all stat functions related to distribution probabilities have the
same> problem:
>
>1.- According to the manual the log.p parameter is always the last one.
Yes, but this is not what you were testing.  You seem to have missed
the ncp parameter.  Is there some documentation somewhere that
neglects to mention the ncp parameter?  Where?
>2.- When you use the software, the last parameter seems to be lower.tail
>
>Example:
>
>> pt (1.1, 5)
>[1] 0.8392746
>> pt (1.1, 5, F)
>[1] 0.8392746
Here you're setting ncp=F, not log.p=F or lower.tail=F.  ncp is the
non-centrality parameter.  If you use a logical there, it will be
treated as 0.
>> pt (1.1, 5, F, T)
>[1] 0.8392746
Here you have ncp=F and lower.tail = T.

Duncan Murdoch
>>=0D
>
>On this example, I have used the Student T distribution. The result of
this> example has been tested with the stat calculator at
http://calculators.sta>t.ucla.edu/.
>1.- This first line shows that when this function has two arguments, the
lo>g.p default value is false, and the lower.tail is true.
>2.- The second line shows that when this function has 3 arguments, the
last> one is the log.p argument and the lower.tail is taken by default to
true.
>3.- The third line shows that when this function has 4 arguments, the
third> one keeps on being the log.p argument and the lower.tail is the last
one.
>4.- Acording to the mannual, the lower.tail should be the third argument
an>d not the last one.
>
>Best regards,
>=0D
>Jos=E9 Luis Casado Mart=EDnez
>------------------------------------------------------------------
>European Computing Consultants
>C/ Hermanos Garc=EDa Noblejas, N=BA 39, 5=AA, N 1
>28037 Madrid
>Telf.: 34-91-406 19 15. Fax: 34-91-406 19 16
>Movil: 34-607-750 316
>------------------------------------------------------------------
>
>The last line shows that the sum of probabilities of 1.1 using 5 degrees
of> freedom adding the lower tail (pt (1.1, 5, F, T)) and not adding the
lower> tail ( pt (1.1, 5, F, F)) is 1. This is not the case with log (p).

casadoj@ecc.es wrote:
> Dear R Developers:
> 
> I have been playing with R, release 2.0.1 for a week now and have detected
> that all stat functions related to distribution probabilities have the
same>  problem:
> 
> 1.- According to the manual the log.p parameter is always the last one.
> 2.- When you use the software, the last parameter seems to be lower.tail
> 
> Example:
> 
> 
>>pt (1.1, 5)
> 
> [1] 0.8392746
> 
>>pt (1.1, 5, F)
> 
> [1] 0.8392746
> 
>>pt (1.1, 5, F, T)
> 
> [1] 0.8392746
> 
>>=0D
> 
> 
> On this example, I have used the Student T distribution. The result of
this>  example has been tested with the stat calculator at
http://calculators.sta> t.ucla.edu/.
> 1.- This first line shows that when this function has two arguments, the
lo> g.p default value is false, and the lower.tail is true.
> 2.- The second line shows that when this function has 3 arguments, the
last>  one is the log.p argument and the lower.tail is taken by default to
true.
> 3.- The third line shows that when this function has 4 arguments, the
third>  one keeps on being the log.p argument and the lower.tail is the last
one.
> 4.- Acording to the mannual, the lower.tail should be the third argument
an> d not the last one.

This is your bug!!!
Please do read the documentation exactly and do think twice before 
posting bug reports!

According to the manual, pt() has 5 (!!!) arguments:
   pt(q, df, ncp=0, lower.tail = TRUE, log.p = FALSE)

If you do want to use more arguments, why not naming them? It's always 
much less irritating for yourself to name arguments rather than wild 
guessing what the hell the 7th argument of foo() was intended to do in 
your code.

See also the corresponding sections in the docs on argument matching.

Uwe Ligges

> Best regards,
> =0D
> Jos=E9 Luis Casado Mart=EDnez
> ------------------------------------------------------------------
> European Computing Consultants
> C/ Hermanos Garc=EDa Noblejas, N=BA 39, 5=AA, N 1
> 28037 Madrid
> Telf.: 34-91-406 19 15. Fax: 34-91-406 19 16
> Movil: 34-607-750 316
> ------------------------------------------------------------------
> 
> The last line shows that the sum of probabilities of 1.1 using 5 degrees
of>  freedom adding the lower tail (pt (1.1, 5, F, T)) and not adding the
lower>  tail ( pt (1.1, 5, F, F)) is 1. This is not the case with log (p).
> 
> 
> 
> _____________________________________________________________________
> Mensaje analizado y protegido, tecnologia antivirus www.trendmicro.es
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-devel@stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel

> 1.- According to the manual the log.p parameter is always the last one.
> 2.- When you use the software, the last parameter seems to be lower.tail
AFAIK, R has no concept of "last parameter" in a function call.  It
has the concept of "first parameter", "second paramter",
etc., but not
"last".
>> pt (1.1, 5)
> [1] 0.8392746
>> pt (1.1, 5, F)
> [1] 0.8392746
>> pt (1.1, 5, F, T)
> [1] 0.8392746
>>=0D
> 4.- Acording to the mannual, the lower.tail should be the third argument
an> d not the last one.
No. According to ?pt or args(pt), pt can take five parameters, the
third of which (when they aren't named) is `ncp', not `lower.tail'.

What happens in your calls is that your third argument F is converted
to 0, which is the default for ncp (and then the T is given to
`lower.tail', which is the default value for that parameter).

It is often a very good practice to name the arguments explicitly in
function calls.  Your calls above would be equivalent to

pt(q = 1.1, df = 5)
pt(q = 1.1, df = 5, ncp = F)  # or ncp = 0
pt(q = 1.1, df = 5, ncp = F, lower.tail = T)  # or ncp = 0

You will find that

pt(q = 1.1, df = 5, lower.tail = F)
and
pt(q = 1.1, df = 5, lower.tail = F, log.p = T)
give quite different answers.

-- 
Bj?rn-Helge Mevik