The computer isn’t always right — Richard Hartung

SEPT 30 — Over a decade ago, a fraud prevention team at a bank in Malaysia missed spotting hundreds of potentially fraudulent transactions. When asked why, they said the transactions did not show up on their daily report and the report was surely correct.

A short investigation revealed that a software error caused every amount to be divided by two, and the report was wrong. The bank likely lost tens of thousands of dollars because staff did not ask why the number of transactions on a report had plummeted.

This type of data error might seem like ancient history, something that could not possibly happen in an era now when software is more robust. Yet nearly the exact same thing happened here in Singapore in September.

It turns out that the Central Narcotics Bureau (CNB) under-reported arrests because of a data error. Management assumed that reports were right, for nearly three years. Only when lower-level officers questioned the data was the error discovered.

While Today reported that the CNB "stressed that the mistake in computing the statistics did not affect its enforcement efforts", Minister of State (Home Affairs) Masagos Zulkifli was reported saying shortly after the incident that he was particularly concerned with the number of new drug abusers, which was higher than what they had previously thought.

Southeast Asia is not the only place where data errors happen, of course. In Australia, for example, The Australian reported earlier this year that New South Wales Police Minister Michael Gallacher said police computer system glitches "led to the unlawful arrest of hundreds of young people".

In Japan, The New York Times reported that continuing errors by Tokyo Electric Power Company in calculating radiation levels "added to the questions already being raised about the company's competence and credibility".

The Economic Times in India says the Central Statistical Organisation corrected massive errors in first-quarter 2010 GDP data, attributed potentially to transcription errors, within 32 hours after economists pointed out discrepancies. And, while a data error did not actually cause the disaster, the United States director of mine safety said a computer error prevented a West Virginia coal mine where 29 workers died in an explosion last year from being warned about safety violations.

These and multitudes of other examples from around the world show how widespread the practice of unequivocally accepting data in a report can be. If anything, errors may be proliferating rather than disappearing. And the errors can negatively affect everything from finances and election outcomes to public policy and safety.

The causes of many of the errors are often quite mundane.

One source is errors in formulas in spreadsheets. Membership secretary of European Spreadsheet Risks Interest Group Grenville Croll told The Telegraph, for example, that "untested spreadsheets are riddled with errors very often because the incorrect formula has been entered in the first place". Along with incorrect formulas, copying or inputting data incorrectly can also cause spreadsheet mistakes.

Another cause is software bugs due to incorrect programming or parameters that are set erroneously. Syntax errors and other mistakes can easily result in software producing wrong information.

Regardless of what causes an error, however, a more fundamental problem is that too many people accept data as being correct simply because it is in a report or comes from a computer.

The many incidents when a mistake caused a critical error show how vital it is to understand data and make sure it is correct. Everyone from clerks to CEOs needs to cast a sceptical eye on the data they are given to make sure it makes sense — then use common-sense checks to verify the data when their experience or a discrepancy with what they see on the ground makes them suspicious.

When they do spot a potential error, staff often find it less risky not to say anything and some managers may reject any questions about the data.

As organisations ranging from the CNB and bankers to electric utilities and the police have found, however, stopping to review the data when someone says it might be wrong is critical. Commending people who spot errors can help create a positive culture with better information.

In an era when data is more important than ever yet perhaps accepted more uncritically than before, it is important to go back to basics and stop to make sure that what we read is actually right. — Today

* Richard Hartung is a consultant who has lived in Singapore since 1992.

* This is the personal opinion of the writer or publication. The Malaysian Insider does not endorse the view unless specified.

The Malaysian Insider :: Side Views