headers
Adaikalavan Ramasamy | 1 Oct 2004 12:48
Picon

[Rd] same test statistic for t-test with and without equal variance assumption

Could some kindly tell me if I am supposed to be getting the same test
statistic value with var.equal=TRUE and var.equal=FALSE in t.test ?

set.seed(1066)
x1 <- rnorm(50)
x2 <- rnorm(50)

t.test(x1, x2, var.equal=FALSE)$statistic # 0.5989774
t.test(x1, x2, var.equal=TRUE)$statistic  # 0.5989774 ???

Here are my own calculations that shows that perhaps the result when
var.equal=TRUE is wrong.

n1 <- length(x1); n2 <- length(x2)
m1 <- mean(x1)  ; m2 <- mean(x2)  ; num   <- (m1 - m2)
v1 <- var(x1)   ; v2 <- var(x2)

# t-test with UNequal variance
denom1 <- sqrt( v1/n1 + v2/n2 )
num / denom1   # gives 0.5989774

# t-test with equal variance 
sp <- ( (n1-1)*v1 + (n2-1)*v2 )/(n1 + n2 - 2)
denom2 <- sp * sqrt( 1/n1 + 1/n2 )
num / denom2   # gives 0.5913777

I tested this using R-1.9.1 (21/06/2004) on Redhat Fedora Core 2 and
Windows 2000 Professional with the same results.

Any suggestions would be kindly appreciated.
(Continue reading)

Permalink | Reply | 3 comments |
headers
Peter Dalgaard | 1 Oct 2004 13:15
Picon
Favicon

Re: [Rd] same test statistic for t-test with and without equal variance assumption

Adaikalavan Ramasamy <ramasamy <at> cancer.org.uk> writes:

> Could some kindly tell me if I am supposed to be getting the same test
> statistic value with var.equal=TRUE and var.equal=FALSE in t.test ?
> 
> set.seed(1066)
> x1 <- rnorm(50)
> x2 <- rnorm(50)
> 
> t.test(x1, x2, var.equal=FALSE)$statistic # 0.5989774
> t.test(x1, x2, var.equal=TRUE)$statistic  # 0.5989774 ???
> 
> 
> Here are my own calculations that shows that perhaps the result when
> var.equal=TRUE is wrong.
> 
> n1 <- length(x1); n2 <- length(x2)
> m1 <- mean(x1)  ; m2 <- mean(x2)  ; num   <- (m1 - m2)
> v1 <- var(x1)   ; v2 <- var(x2)
> 
> # t-test with UNequal variance
> denom1 <- sqrt( v1/n1 + v2/n2 )
> num / denom1   # gives 0.5989774
> 
> # t-test with equal variance 
> sp <- ( (n1-1)*v1 + (n2-1)*v2 )/(n1 + n2 - 2)
> denom2 <- sp * sqrt( 1/n1 + 1/n2 )
> num / denom2   # gives 0.5913777
> 
>
(Continue reading)

Permalink | Reply | 2 comments |
headers
hoffman+R | 1 Oct 2004 14:42
Picon
Picon
Favicon

[Rd] bzip2 directory won't build on OSF1 due to C99 code and -std1 option (PR#7257)

Full_Name: Michael Hoffman
Version: 2.0.0-beta-20041001
OS: OSF1 V5.1
Submission from: (NULL) (193.62.199.8)

Hello. Building R 2.0.0-beta-20041001 on OSF1 V5.1 failed because the default
configure sets R_XTRA_CFLAGS to "-std1 -ieee_with_inexact." The bzip2 directory
includes C99 code that requires -std1 to not be set in order to compile. This
halts the overall build process.

______________________________________________
R-devel <at> stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
Permalink | Reply | 4 comments |
headers
Adaikalavan Ramasamy | 1 Oct 2004 14:56
Picon

Re: [Rd] same test statistic for t-test with and without equal variance assumption

Peter, thank you! I forgot the to square root in calculating sp.

sp <- sqrt( ( (n1-1)*v1 + (n2-1)*v2 )/(n1 + n2 - 2) )

For several simulation runs, the test statistics from both tests are
remarkably similar (difference is less than 10e-16). I naively assumed
that the statistics value should be slightly but visibly different too. 

It appear that the effects of equal and unequal variance assumptions are
only felt through the degrees of freedom calculation. 

Regards, Adai

On Fri, 2004-10-01 at 12:15, Peter Dalgaard wrote:
> Adaikalavan Ramasamy <ramasamy <at> cancer.org.uk> writes:
> 
> > Could some kindly tell me if I am supposed to be getting the same test
> > statistic value with var.equal=TRUE and var.equal=FALSE in t.test ?
> > 
> > set.seed(1066)
> > x1 <- rnorm(50)
> > x2 <- rnorm(50)
> > 
> > t.test(x1, x2, var.equal=FALSE)$statistic # 0.5989774
> > t.test(x1, x2, var.equal=TRUE)$statistic  # 0.5989774 ???
> > 
> > 
> > Here are my own calculations that shows that perhaps the result when
> > var.equal=TRUE is wrong.
> >
(Continue reading)

Permalink | Reply | 1 comment |
headers
Prof Brian Ripley | 1 Oct 2004 14:58
Picon
Picon
Favicon

(PR#7257) Re: [Rd] bzip2 directory won't build on OSF1 due to C99 code and -std1 option

What `C99 code' is that?  It passes conformance tests on a very strict C89
compiler.  Please don't expect us to guess what you have in mind, for we 
haven't a clue.

Please read the section on BUGS in the FAQ, and tell us the symptoms, not 
your guess as to the diagnosis.

AFAIK, R's configure does not know about OSF1 v5.x, so it is quite likely 
that the assumptions (which are inside

case "${host_cpu}" in
  alpha*)

) are wrong for that untested platform.  The code in the bzip2 directory
has not been changed for a very long time, except for possibly allowing
inlining (and for which there is a configure test).

Note that feedback 3 days before release is too late for configuration 
changes to be made.

On Fri, 1 Oct 2004 hoffman+R <at> ebi.ac.uk wrote:

> Full_Name: Michael Hoffman
> Version: 2.0.0-beta-20041001
> OS: OSF1 V5.1
> Submission from: (NULL) (193.62.199.8)
> 
> 
> Hello. Building R 2.0.0-beta-20041001 on OSF1 V5.1 failed because the default
> configure sets R_XTRA_CFLAGS to "-std1 -ieee_with_inexact." The bzip2 directory
(Continue reading)

Permalink | Reply | 3 comments |
headers
Adaikalavan Ramasamy | 1 Oct 2004 15:17
Picon

Re: [Rd] same test statistic for t-test with and without equal variance assumption

Apologies, I made an incorrect statement. The statistics from t-test
with difference variance assumptions are the same if both groups have
the same length. Sorry for troubling everyone again.

On Fri, 2004-10-01 at 13:56, Adaikalavan Ramasamy wrote:
> Peter, thank you! I forgot the to square root in calculating sp.
> 
> sp <- sqrt( ( (n1-1)*v1 + (n2-1)*v2 )/(n1 + n2 - 2) )
> 
> For several simulation runs, the test statistics from both tests are
> remarkably similar (difference is less than 10e-16). I naively assumed
> that the statistics value should be slightly but visibly different too. 
> 
> It appear that the effects of equal and unequal variance assumptions are
> only felt through the degrees of freedom calculation. 
> 
> Regards, Adai
> 
> 
> On Fri, 2004-10-01 at 12:15, Peter Dalgaard wrote:
> > Adaikalavan Ramasamy <ramasamy <at> cancer.org.uk> writes:
> > 
> > > Could some kindly tell me if I am supposed to be getting the same test
> > > statistic value with var.equal=TRUE and var.equal=FALSE in t.test ?
> > > 
> > > set.seed(1066)
> > > x1 <- rnorm(50)
> > > x2 <- rnorm(50)
> > > 
> > > t.test(x1, x2, var.equal=FALSE)$statistic # 0.5989774
(Continue reading)

Permalink | Reply | 0 comments |
headers
Michael Hoffman | 1 Oct 2004 15:24
Picon
Picon
Favicon

Re: (PR#7257) Re: [Rd] bzip2 directory won't build on OSF1 due to C99 code and -std1 option

On Fri, 1 Oct 2004, Prof Brian Ripley wrote:

> What `C99 code' is that?

Specifically it uses the inline keyword. I do not believe that is
supported by a strict C89 compiler.

> Please don't expect us to guess what you have in mind, for we
> haven't a clue.

Please accept my apologies for an inadequate bug report. This might
make things easier to understand:

mh5 <at> ecs4d /ecs4/scratch2/ensembl/mh5/sandbox/R-2.0.0/src/extra/bzip2
$ make
making blocksort.d from blocksort.c
making bzlib.d from bzlib.c
making compress.d from compress.c
making crctable.d from crctable.c
making decompress.d from decompress.c
making huffman.d from huffman.c
making randtable.d from randtable.c
gmake[1]: Entering directory `/ecs4/scratch2/ensembl/mh5/sandbox/R-2.0.0/src/extra/bzip2'
gmake[1]: `Makedeps' is up to date.
gmake[1]: Leaving directory `/ecs4/scratch2/ensembl/mh5/sandbox/R-2.0.0/src/extra/bzip2'
gmake[1]: Entering directory `/ecs4/scratch2/ensembl/mh5/sandbox/R-2.0.0/src/extra/bzip2'
cc -I. -I. -I../../../src/include -I../../../src/include -I/usr/local/include -DHAVE_CONFIG_H
-std1 -ieee_with_inexact  -fast -arch ev6 -c blocksort.c -o blocksort.o
cc: Error: blocksort.c, line 79: Missing ";". (nosemi)
void fallbackSimpleSort ( UInt32* fmap,
(Continue reading)

Permalink | Reply | 2 comments |
headers
Prof Brian Ripley | 1 Oct 2004 15:32
Picon
Picon
Favicon

Re: (PR#7257) Re: [Rd] bzip2 directory won't build on OSF1 due to C99 code and -std1 option

I believe this occurs because the configure test for `inline' is not using
the R_XTRA_CFLAGS, which is a pervasive design problem (we have noted it
occurs with PIC flags as well).  If you set -std1 in CFLAGS then it should 
be used in the configure test which should conclude `inline' is not 
allowed.

[I use CFLAGS for making my test compilers as picky as possible.]

On Fri, 1 Oct 2004, Michael Hoffman wrote:

> On Fri, 1 Oct 2004, Prof Brian Ripley wrote:
> 
> > What `C99 code' is that?
> 
> Specifically it uses the inline keyword. I do not believe that is
> supported by a strict C89 compiler.
> 
> > Please don't expect us to guess what you have in mind, for we
> > haven't a clue.
> 
> Please accept my apologies for an inadequate bug report. This might
> make things easier to understand:
> 
> mh5 <at> ecs4d /ecs4/scratch2/ensembl/mh5/sandbox/R-2.0.0/src/extra/bzip2
> $ make
> making blocksort.d from blocksort.c
> making bzlib.d from bzlib.c
> making compress.d from compress.c
> making crctable.d from crctable.c
> making decompress.d from decompress.c
(Continue reading)

Permalink | Reply | 1 comment |
headers
Michael Hoffman | 1 Oct 2004 15:49
Picon
Picon
Favicon

Re: (PR#7257) Re: [Rd] bzip2 directory won't build on OSF1 due to C99 code and -std1 option

> If you set -std1 in CFLAGS then it should be used in the configure
> test which should conclude `inline' is not allowed.

OK, I will do as you suggest. I should note that cc (version info
below) does support inline although not with -std1 so I'd prefer to
have any performance benefits of inline, if possible.
--

-- 
Michael Hoffman <hoffman <at> ebi.ac.uk>
European Bioinformatics Institute

Compaq C V6.5-011 on Compaq Tru64 UNIX V5.1B (Rev. 2650)
Compiler Driver V6.5-003 (sys) cc Driver

______________________________________________
R-devel <at> stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
Permalink | Reply | 0 comments |
headers
Prof Brian Ripley | 1 Oct 2004 15:55
Picon
Picon
Favicon

Re: [Rd] (PR#7254)Documentation: Reference Index (.pdf) -- setOldClass

Have you looked at the 2.0.0 beta version?

It is hard for us to know what you are looking at, but I believe this was 
fixed a while ago (but post 1.9.1).  Better PDF cross-refs is one of the 
changes in 2.0.0-to-be.

On Thu, 30 Sep 2004, Pfaff, Bernhard wrote:

> Apologies for cross postings or manual work caused by resending; the
> intially sent bug report through the web-interface bounced, however.
> 
> 
> Full_Name: Bernhard Pfaff
> Version: 1.9.1
> OS: Windows NT
> Submission from: (NULL) (193.194.7.77)
> 
> 
> Dear core team members / maintainers,
> 
> in the pdf version of the reference index I observed two entries of
> "setOldClass" in the bookmark section of Adobe Acrobat Reader. Furthermore,
> the
> second bookmark entry points to **4.1 setOldClass** which is type setted out
> of
> the context.
> This ill-typesetting does however not occur in the html-versions; I have
> checked
> my locally one installed version (i.e. help.start()), as well as the R-devel
> version at:
(Continue reading)

Permalink | Reply | 0 comments |

Gmane