Posted as "Sorry, Wrong Number", on the Wall Street Journal.
By Carl Bialik
My print column examines high-profile instances, both recent and old, of numerical errors, afflicting data from spacecraft orbit calculations to economic statistics.
The stakes often are high. “Decision-making today is heavily reliant on data, which we now have access to on enormous scales,” Anna Nagurney, an applied mathematician at the University of Massachusetts, Amherst, wrote in an email. “However, high-impact decisions require good data, good models and appropriate statistical and analytical methodologies, with the users having an understanding of the validity and applicability of the underlying assumptions. Plus, software in which the models and tools are implemented needs to be fully tested and ‘bug-free.’ Otherwise, we can have propagating failures with immense consequences.”
Human error becomes more likely when steps that can be automated are undertaken manually. Richard Alldritt, head of assessment for the U.K. Statistics Authority, said an investigation intoa major error in U.K. construction statistics so far has revealed that the mistake originated when a spreadsheet was updated manually. The Office for National Statistics has aimed to automate its spreadsheets but hasn’t yet completed the transition. The spreadsheet mistakenly calculated construction output for the three-month period ending in May instead of June.
The U.S. Bureau of Labor Statistics also has been trying to transition from manually updated spreadsheets, said spokesman Gary Steinberg. “It’s mostly been rooted out,” he said.
Some errors have stuck with Steinberg. “In January 2004 – I still remember this! – the producer-price-index staff discovered they couldn’t produce data at all. Their computer systems completely crashed and they couldn’t get them operating at all. That was a principal indicator that I had to tell everyone we couldn’t produce and didn’t know when we would be able to. That was a pretty busy day for me. My phone was still really smoking that day.”
Read full article and comments at http://blogs.wsj.com/numbersguy/sorry-wrong-number-1082/