Errors
Incredible as it may seem, you cannot assume that a statistical package will give the right answer. Here are the results of Wilcoxon-Mann-Whitney tests on the same set of data using 11 different stats packages. Checkmarks indicate conclusions of statistical significance (p<.05). The columns show different forms of the test. As you can see, the different packages give different results for the same form of the test.
From "Different outcomes of the Wilcoxon-Mann-Whitney test from different statistics packages" by Reinhard Bergmann, John Ludbrook, Will P J M Spooren. The American Statistician. Alexandria: Feb 2000. Vol. 54, Iss. 1; pg. 72-77
Neither can you assume accurate results from a spreadsheet. Even assuming that the spreadsheet always does its basic arithmetic correctly (not always the case), there is no reason to assume more complicated comutations to be correct. Cf., "On the Accuracy of Statistical Distributions in Microsoft Excel 97."
For a review of the problem, see "Guidelines for Testing Statistical Software" by Leland Wilkinson and this blog from a programmer of Mathematica.
The bottom line: use a spreadsheet only for basic arithmetic, not for statistical functions, and if the results of a stats package seem dubious or unexpected, check them by hand.
Applications available
- Calculator
A scientific calculator with a virtual paper tape
- MathPad
Type an algebraic expression and see/save the result. Save formulae for future use.
- Numbers, NeoOffice & Excel
Spreadsheets. Daphne prefers Numbers but any will do. All three handle Excel files.
- Stata, SPSS & R
Stats packages.
Also available is StatView, which is the nicest stats package ever made for the Mac and possibly for any computer. Unfortunately, SAS bought the product and killed it in favour of JMP. It has not been updated since 1998. Although it runs under Classic on PPC machines, it will not run on the current Intel models.