This study had two components. The first was an audit of 20 spreadsheets from 10 different firms. The second part consisted of 17 interviews and 14 surveys, for a total of 31 questionnaires.
Spreadsheets ranged from 150 to 10,000 cells.
Two-thirds of the spreadsheets were beyond their first version. The spreadsheets had been in use for a median of 6 months.
They had gone through a median of 7 updates. One was in its 60th version.
All but one had been tested with data.
8 were used by others (in only two cases did these others have input to the design).
Audits were limited to an average of only about two hours. "Thus, the errors reported later are almost certainly an underestimate of the real problems." (p. 747)
Among the 20 models, 5 (25%) contained errors
In one case, a major budget item was left out because of a range specification error.
In other case, there was a serious error in a "large, strategic operations review using what-if analysis." (p. 748)
1/2 had poor layout
Only 1/3 used cell protection.
Half the spreadsheets used macros
For the 20 spreadsheets audited audited, development was informal and iterative.
For the 20 spreadsheets audited, Cragg and King described the development process as poor.
For the 20 spreadsheets audited, only one had been tested by another person, and only one had been tested with a spreadsheet auditing package.
For the 20 spreadsheets audited, there had been previous problems with 6 of the models.
In 9 of the 10 organizations, no one was responsible for spreadsheet development, and there was no formal development policy.
For the 20 spreadsheets audited, were 8 built with initial design, and this usually was rudimentary.