Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

An alarming number of scientific papers contain Excel errors

washingtonpost.com

Post reply on HN
Print view View original

1–10 of 74 posts

Re: An alarming number of scientific papers contain Excel errors

#3

Post by ramblenode »

MS Excel is absolutely unfit for most scientific and engineering problems.

The spreadsheet GUI, lack of good version tracking/history, and eagerness to coerce data types and "correct" values makes it easy to introduce errors that will go unrecognized and propagated through calculations. Unfortunately this story just keeps repeating itself.

But all of this is just a secondary concern to Excel's real trouble: it's history of incorrectly implementing numerical and statistical procedures. One could plumb the depths of this topic for hours, but here are a few highlights: regression formula accepts illegal/nonsensical inputs (e.g. collinear predictors) and gives illegal/nonsensical outputs [0], variance/standard deviation change incorrectly with sample size [0], output of a paired t-test changes when missing values are included [0], formulas are mislabeled [0], v. 2007 gives very wrong answers to 11 of 27 tests in the NIST test suite used for statistical software benchmarks [1], the random number generator was broken as late as v. 2007 [1], and calculations relying on any of 12 particular floats display an incorrect result [2]. There are plenty of other issues mentioned in the links and elsewhere; if you're interested you'll have no trouble finding them.

Remember, friends don't let friends use Excel for science. :)

[0] http://people.stern.nyu.edu/jsimonof/classes/1305/pdf/excelr...

[1] http://www.pages.drexel.edu/~bdm25/excel2007.pdf

[2] https://blogs.office.com/2007/09/25/calculation-issue-update...

Edit: clarify and add a new issue I became aware of while researching further.

Re: An alarming number of scientific papers contain Excel errors

#4

Post by hirenj »

It's funny to consider that these errors slipped past the peer review stage. It really highlights the major issue with reviewing source code published as part of an analysis.

If there aren't enough resources / skilled eyes to catch these simple errors, what are the chances they would catch errors in source code too?

Re: An alarming number of scientific papers contain Excel errors

#5

Post by krapht »

ramblenode wrote: post #3

MS Excel is absolutely unfit for most scientific and engineering problems. The spreadsheet GUI, lack of good version tracking/history, and eagerness to coerce data types and "correct" values makes it easy to introduce errors that will go unrecognized and propagated through calculations. Unfortunately this story just keeps repeating itself. But all of this is just a secondary concern to Excel's real trouble: it's hist…

I'll just quote Soustrup here: "There are only two kinds of languages: the ones people complain about and the ones nobody uses." In particular you see the spreadsheet GUI as a downfall, but I think it is a great enabler that has allowed millions of non-programmers to create useful programs.

Re: An alarming number of scientific papers contain Excel errors

#6

Post by tragomaskhalos »

ramblenode wrote: post #3

MS Excel is absolutely unfit for most scientific and engineering problems. The spreadsheet GUI, lack of good version tracking/history, and eagerness to coerce data types and "correct" values makes it easy to introduce errors that will go unrecognized and propagated through calculations. Unfortunately this story just keeps repeating itself. But all of this is just a secondary concern to Excel's real trouble: it's hist…

.... and then we get into the whole realm of simple user errors - alarmingly easy to commit in an environment that positively encourages the production of a rats' nest of essentially unauditable calculations - then then feed into important financial decisions.

Re: An alarming number of scientific papers contain Excel errors

#7

Post by cm2187 »

krapht wrote: post #5
ramblenode wrote: post #3

MS Excel is absolutely unfit for most scientific and engineering problems. The spreadsheet GUI, lack of good version tracking/history, and eagerness to coerce data types and "correct" values makes it easy to introduce errors that will go unrecognized and propagated through calculations. Unfortunately this story just keeps repeating itself. But all of this is just a secondary concern to Excel's real trouble: it's hist…

I'll just quote Soustrup here: "There are only two kinds of languages: the ones people complain about and the ones nobody uses." In particular you see the spreadsheet GUI as a downfall, but I think it is a great enabler that has allowed millions of non-programmers to create useful programs.

In fact a lot could be said of the millions of ways to shoot oneself in the foot with programming languages. Excel being visual (the formula is evaluated/tested immediatly after being entered), I'd argue a diligent user will make less mistakes with Excel than while coding.
Post reply on HN