>Don't rely on the statistics add-in pack supplied for Excel
by Microsoft -
>IT GIVES THE WRONG ANSWERS!
>The following is extracted from an article by Jon Honeyball
in PC Pro,
>Issue 62, December 1999, pp 248-255.
> "In front of me right now is a paper entitled On
the accuracy of
>statistical procedures in Microsoft Excel 97, reprinted from
>of Computational Statistics and Data Analysis, which is a
>prestigious, refereed academic journal. I might without exaggeration
>it 'the bible of computational statistics' and there's arguably
>reference in the world. The article comes from volume 31,
issue 1, 28
> The abstract for the paper says: 'The reliability of statistical
>procedures in Excel are assessed in three areas: estimation
>and nonlinear), random number generation, and statistical
>(such as for calculating p-values). Excel's performance in
>areas is found to be inadequate. Persons desiring to conduct
>analyses of data are advised not to use Excel.' As an opening
>you must admit that it's a bit of a corker.
> The paper's authors, BD McCullough and Berry Wilson of
>Communications Commission in Washington DC, go on to describe
>detail how they applied the recently released StRD (Statistical
>Datasets) from the American National Institute of Standards
>to assess the performance of Excel in a wide range of statistical
>The results are stunningly bad, and, worse still, the paper
refers back to
>work done by Sawitski in 1994 on Excel 4 and the problems
>are still present in Excel 97. I've run some of the tests
>they're still there in Excel 2000. The paper, which can't
>argued with, is littered with phrases like 'can be judged
>'it can be deduced that Excel uses an unstable algorithm'.
>find fault with its univariate summary statistics, analysis
>linear regression, nonlinear regression, random number generation
>forth. What can I say? If you use the statistics add-on package
>ships with Excel, you really better know your stuff because
Excel may well
>come up with wrong numbers.
> Excel's statistics add-on pack is riddled with potential
>areas, and since it has been subjected to the best analysis
>the world and found to be wholly lacking, the only applicable
>'avoid' and 'plague'. Instead, you should buy yourself a
>add-on package that has numerical methods that are open to
peer review and
>whose authors know what they're doing (unfortunately, Microsoft's
>stats-pack team obviously doesn't)."