[ Jocelyn Ireson-Paine's Home Page | Publications | Dobbs Code Talk Index | Dobbs Blog Version ]
[ Previous cartoon | Next cartoon ]

Fatal Addition

Boss, holding report showing profit dropping through floor, yells at programmer 'You _have_ to use a spreadsheet! What else can we blame failure on?'

I've pinched my title from a news feature in New Scientist, 16th August 1997, in which Mark Ward reported on research into spreadsheet errors. Ray Panko, professor of IT Management at the Shidler College of Business, University of Hawai'i, had found in examining ten years of spreadsheet-error studies that every study showed a dangerously high rate of errors: he estimated that humans make around one error per hundred cells.

Accountants Coopers and Lybrand had found errors in 90% of the spreadsheets they looked at.

And the Computer Audit Unit of HM Customs and Excise — who fight smuggling, collect taxes on "bad habits" such as alcohol and tobacco, and collect duties on imported goods — also found errors. In fact, a fairly low error rate of only 11%. But some errors ran into millions of pounds. The study is online as Is This Spreadsheet a Tax Evader?

I'm summarising the errors part of the New Scientist feature because, although it is online, the full text is open only to subscribers, so you probably won't be able to read it. Until 2001, features on New Scientist's site were open to everyone, and a wonderful resource this was. But then Reed, the magazine's owners, stopped that.

Anyway, people still make spreadsheet errors. The Customs and Excise study was done by Ray Butler, one of the three founders of the European Spreadsheet Risks Interest Group. (The other two being Pat Cleary and David Chadwick; Ray explains in Three Men in a Bar Found a Spreadsheet Society how they met, with Ray Panko's aid.) It's ten years since EuSpRIG was founded, and perhaps EuSpRIG has helped a higher proportion of spreadsheet programmers realise how error-prone spreadsheets are.

The fault, of course, is with the human brain, never evolved for worlds where small errors can lie unnoticed for years, then return and bite you a millionfold. We can do nothing about that, but we can do something about spreadsheets. Martin Erwig, in this year's EuSpRIG keynote address, Software Engineering for Spreadsheets: Challenges and Opportunities, suggests two fruitful research topics to be type checking and debugging. And he pointed out to me that spreadsheets contain a great deal of information in the way data and calculations are embedded in space. Too much research ignores this.