Okumura's Blog

Computational Statistics and Data Analysis Volume 52, Issue 10 (2008) に Excel 2007 特集がある。

まず最初の論文 (B.D. McCullough and David A. Heiser, On the accuracy of statistical procedures in Microsoft Excel 2007, pp.4570-4578) のアブストラクト：

Excel 2007, like its predecessors, fails a standard set of intermediate-level accuracy tests in three areas: statistical distributions, random number generation, and estimation. Additional errors in specific Excel procedures are discussed. Microsoft's continuing inability to correctly fix errors is discussed. No statistical procedure in Excel should be used until Microsoft documents that the procedure is correct; it is not safe to assume that Microsoft Excel's statistical procedures give the correct answer. Persons who wish to conduct statistical analyses should use some other package.

要約：どうしょーもないから使うな。

次の A. Talha Yalta, The accuracy of statistical distributions in Microsoft® Excel 2007, pp.4579-4586 のアブストラクト：

We provide an assessment of the statistical distributions in Microsoft® Excel versions 97 through 2007 along with two competing spreadsheet programs, namely Gnumeric 1.7.11 and OpenOffice.org Calc 2.3.0. We find that the accuracy of various statistical functions in Excel 2007 range from unacceptably bad to acceptable but significantly inferior in comparison to alternative implementations. In particular, for the binomial, Poisson, inverse standard normal, inverse beta, inverse student’s t, and inverse F distributions, it is possible to obtain results with zero accurate digits as shown with numerical examples.

要約：どうしょーもないから使うな。

次の B.D. McCullough, Microsoft Excel's `Not The Wichmann-Hill' random number generators, pp.4587-4593 のアブストラクト：

Microsoft attempted to implement the Wichmann-Hill RNG in Excel 2003 and failed; it did not just produce numbers between zero and unity, it would also produce negative numbers. Microsoft issued a patch that allegedly fixed the problem so that the patched Excel 2003 and Excel 2007 now implement the Wichmann-Hill RNG, as least according to Microsoft. We show that whatever RNG it is that Microsoft has implemented in these versions of Excel, it is not the Wichmann-Hill RNG. Microsoft has now failed twice to implement the dozen lines of code that define the Wichmann-Hill RNG.

昔のExcelの乱数は無茶苦茶だった。Excel 2003でMicrosoftはやっとWichmann-Hillの乱数生成アルゴリズム（私のアルゴリズム事典にも載っている十数行で書ける簡単だがまともなアルゴリズム）に切り替えた（Microsoftの Excel 2007 と Excel 2003 の RAND 関数について 参照。ただしMicrosoftはWichmannの綴りを間違えている）。最初はバグっていて負の値を返したりしたため，パッチが出た。めでたしめでたし。じゃない。調べてみると，何かおかしい。Excel 2007になってもまだおかしい。Microsoftはたった十数行のアルゴリズムもまともにインプリメントできないんだ。

ほかにもチャートジャンクの話などがあるが，略。

追記：MicrosoftのExcelの統計関数の解説ページはみんな「統計関数を超えます」（Excel Statistical Functionsの直訳）になっている。例：統計関数を超えます。:傍受します。

2008-10-27追記：「統計関数を超えます」がなくなった！

[追記] Excel使うな：2010版を書いた。

2014-02-08追記：Excel 2007についてのCSDA誌の論文で拾い損ねていたもの：Bruce R. Hargreavesa and Thomas P. McWilliams, Polynomial Trendline function flaws in Microsoft Excel, Computational Statistics & Data Analysis, Vol. 54, No. 4, pp.1190-1196 (2010)：Excel 2003/2007の多項式フィットも駄目駄目。

2014-02-08追記：ついでに，Wichmann-Hill（1982）の現代編の論文：B. A. Wichmann and I. D. Hill, Generating good pseudo-random numbers, Computational Statistics & Data Analysis, Vol. 51, No. 3, pp.1614-1622 (2006).