After takin' a swig o' grog, Roy Schestowitz belched out this bit o' wisdom:
> __/ [ nessuno@xxxxxxxxxxxxxxxxxxx ] on Tuesday 06 June 2006 22:19 \__
>
>> Abstract:
>> --------------------
>> "The open source spreadsheet package "Gnumeric" was such a good clone
>> of Microsoft Excel that it even had errors in its statistical functions
>> similar to those in Excel's statistical functions. When apprised of the
>> errors in v1.0.4, the developers of Gnumeric indicated that they would
>> try to fix the errors. Indeed, Gnumeric v1.1.2, has largely fixed its
>> flaws, while Microsoft has not fixed its errors through many successive
>> versions. Persons who desire to use a spreadsheet package to perform
>> statistical analyses are advised to use Gnumeric rather than Excel."
>> --------------------
>>
> http://web.archive.org/web/20040412052948/http://www.csdassn.org/software_reports/gnumeric.pdf
Interesting.
Poisson distribution:
"In Excel 20003, Microsoft obtained an accurate answer in the
central region of the distribution in exchange for inaccurate results
in the tail. This is not a good "fix". A good fix is demonstrated
by gnumeric."
Hypergeometric distribution:
"Microsoft did not fix Excel."
Standard normal distribution:
"Microsoft did not fix Excel."
Inverse normal distribution:
"Microsoft traded a weak algorithm for one that was slightly less
weak."
Inverse chi-square distribution:
"Microsoft did not fix Excel."
Inverse F distribution:
"Microsoft did not fix Excel."
Inverse t distribution:
"Microsoft did not fix Excel."
Inverse beta function:
"...Microsoft claims to have fixed the inverse beta function."
Not.
Sample standard deviation (benchmark tests):
"Excel consistently does a poor job... Microsoft did not fix the
problem in Excel."
ANOVA (benchmark tests):
"Excel's performance on this suite of tests is unacceptable."
Linear regression (benchmark tests):
"Either way, returning zero digits of accuracy is unacceptable...
Microsoft did not fix Excel."
Random number generator:
"Gnumeric RNG is based on '/dev/urandom' ... This is a very good
method to obtain seeds for pseudo-random numbers, but it is not a
good way to generate random numbers for statistical purposees... can
be judged unacceptable for statistical purposes."
"The random number generator in Excel has been examined... and found
to fail Marsaglia's DIEHARD tests."
----------------------------------
My jaw dropped reading this, as I started to wonder just how much
research is based on numbers produced by Excel's stat routines, and just
how responsive a small cadre of open-source developers was/is compared
to mighty Microsoft.
(I'm also going to take a look at my own linreg algorithm!!!)
--
/\ STOP! This post has not passed Microsoft Logo testing to verify its
/ \ compatibility with Microsoft FUD. Microsoft strongly recommends
/ !! \ you stop reading this post, and consult a poster with FUD
/______\ certification. [ Continue Anyway ] [ STOP Reading ]
|
|
