What is wrong with these people? (geek rant)
The short version:
MS Excel 2007 is not reliable for multiplication, (but I'm told they're going to fix this within the week). It has been known since 1999 that Excel 1997's statistical functions were flawed...but they made the problems worse when they tried to correct them in Excel 2000 and Excel XP. So if stats are important to your livelihood, don't use Excel. If multiplication is important to your livelihood, don't use Excel 2007 until they produce the patch to fix this new bug.
The long version:
OK. It's been known for many years (it was documented in the peer-reviewed literature as early as 1999) that MS Excel's statistical functions are not reliable. I think it'll do a decent t-test, one-way ANOVA and Chi-square test but some of the more complex problems (linear and non-linear curve-fitting) cause it to foul up. Also, its pseudo-random number generator is known to produce numbers that are NOT uniformly distributed (as they should be), and it isn't very good at calculating p-values under a number of statistical distributions. The same authors that produced the 1999 paper (which evaluated MS Excel 1997) published another in 2002 which evaluated Excel 2000 and Excel XP and found that in trying to correct the problems, MS had actually made some of them worse!
Now today I get a message from a list-serve stating that Excel 2007 can't even get multiplication right! Yes, it's correct most of the time, but there's a flaw that makes some multiplication results--there's no other way to put this--wrong. I heard through that same list-serve that MS claims it will fix the multiplication problem within a week but I have no documentation to back that up.
You'd think Microsoft could put up computational algorithms that worked the first time. And you wouldn't think they'd need eight years since first notified of the problem still to be screwing it up!
I include the message from the list-server below (which includes a link), and below that the references to the two articles that evaluated Excel's statistical prowess. If you're REALLY interested, e-mail me and I'll send you pdf's of the two articles I mentioned.
From the list-serve:
quote:
Don't know how many of you have "upgraded" to Excel 2007 yet but
Slashdot.org (tech-ish blog) is reporting a multiplication bug in excel.
Given that many of us and your students use excel as the default math
application I wanted to give everyone a heads-up on this - not sure how far this problem propagates outward.
Excel is reporting the result of the formula =850*77.1 as 100000 rather than the correct 65535. Others are reporting that this extends to any multplication/division combination that should yields 65553. (e.g. =5.1*12850 =10.2*6425, = 850/(1/77.1) and even =SUMPRODUCT(850,77.1,2,0.5) )
I have obtained similar results on my system with excel07 (with W/XP).
excel03 on the same machine is reporting correct results.
I know that stats and excel has been a perennial topic of discussion on this list since it's inception. It's too bad that now the issue becomes "normal" arithmetic.
link: http://it.slashdot.org/it/07/09/24/2339203.shtml
Articles RE: MS Excel's failing statistical routines:
McCullough, B. D. and B. Wilson (1999). "On the accuracy of statistical procedures in Microsoft Excel 97." Computational Statistics & Data Analysis 31(1): 27-37.
McCullough, B. D. and B. Wilson (2002). "On the accuracy of statistical procedures in Microsoft Excel 2000 and Excel XP." Computational Statistics & Data Analysis 40(4): 713-721.
MS Excel 2007 is not reliable for multiplication, (but I'm told they're going to fix this within the week). It has been known since 1999 that Excel 1997's statistical functions were flawed...but they made the problems worse when they tried to correct them in Excel 2000 and Excel XP. So if stats are important to your livelihood, don't use Excel. If multiplication is important to your livelihood, don't use Excel 2007 until they produce the patch to fix this new bug.
The long version:
OK. It's been known for many years (it was documented in the peer-reviewed literature as early as 1999) that MS Excel's statistical functions are not reliable. I think it'll do a decent t-test, one-way ANOVA and Chi-square test but some of the more complex problems (linear and non-linear curve-fitting) cause it to foul up. Also, its pseudo-random number generator is known to produce numbers that are NOT uniformly distributed (as they should be), and it isn't very good at calculating p-values under a number of statistical distributions. The same authors that produced the 1999 paper (which evaluated MS Excel 1997) published another in 2002 which evaluated Excel 2000 and Excel XP and found that in trying to correct the problems, MS had actually made some of them worse!
Now today I get a message from a list-serve stating that Excel 2007 can't even get multiplication right! Yes, it's correct most of the time, but there's a flaw that makes some multiplication results--there's no other way to put this--wrong. I heard through that same list-serve that MS claims it will fix the multiplication problem within a week but I have no documentation to back that up.
You'd think Microsoft could put up computational algorithms that worked the first time. And you wouldn't think they'd need eight years since first notified of the problem still to be screwing it up!
I include the message from the list-server below (which includes a link), and below that the references to the two articles that evaluated Excel's statistical prowess. If you're REALLY interested, e-mail me and I'll send you pdf's of the two articles I mentioned.
From the list-serve:
quote:
Don't know how many of you have "upgraded" to Excel 2007 yet but
Slashdot.org (tech-ish blog) is reporting a multiplication bug in excel.
Given that many of us and your students use excel as the default math
application I wanted to give everyone a heads-up on this - not sure how far this problem propagates outward.
Excel is reporting the result of the formula =850*77.1 as 100000 rather than the correct 65535. Others are reporting that this extends to any multplication/division combination that should yields 65553. (e.g. =5.1*12850 =10.2*6425, = 850/(1/77.1) and even =SUMPRODUCT(850,77.1,2,0.5) )
I have obtained similar results on my system with excel07 (with W/XP).
excel03 on the same machine is reporting correct results.
I know that stats and excel has been a perennial topic of discussion on this list since it's inception. It's too bad that now the issue becomes "normal" arithmetic.
link: http://it.slashdot.org/it/07/09/24/2339203.shtml
Articles RE: MS Excel's failing statistical routines:
McCullough, B. D. and B. Wilson (1999). "On the accuracy of statistical procedures in Microsoft Excel 97." Computational Statistics & Data Analysis 31(1): 27-37.
McCullough, B. D. and B. Wilson (2002). "On the accuracy of statistical procedures in Microsoft Excel 2000 and Excel XP." Computational Statistics & Data Analysis 40(4): 713-721.
0
Please sign in to leave a comment.
Comments
0 comments