Authors
Raymond R. Panko & Richard P. Halverson
Abstract
Will teams produce fewer errors in spreadsheet development than individuals working alone?
Working from a common task statement, 155 undergraduate business students created a spreadsheet model working alone, in a dyad (group of two), or in a tetrad (group of four).
We analyzed overall errors and errors of different types (mechanical, logic, and omission errors). Dyadic teamwork reduced overall errors by a third, but the reduction was not statistically significant. Tetradic work reduced overall errors by two-thirds, and this reduction was statistically significant.
However, tetrads were only successful in reducing certain types of errors. In addition, the tetrads did not reduce errors as much as they should have, judging from a nominal group analysis.
Overall, synchronous team spreadsheeting is promising but is expensive and does not seem to be a complete cure for spreadsheet errors.
Sample
Overall, synchronous development worked well for omission errors and for Eureka logic errors. Mechanical errors, however, often occurred too rapidly for other group members to catch. Cassandra logic error rates were relatively immune to group development. Half the error reduction between individuals and tetrads came from the reduction of omission errors.
Dyads had only somewhat lower error rates than individuals. At a doubling in cost, they reduced errors by about a third, although this difference was not statistically significant. Doubling cost again, tetrads reduced errors by about two-thirds, compared to individuals.
Another thing we noted was the wide variety of errors that occurred. Our subjects made a total of 182 errors in their 99 models. Among these were 51 different errors, even counting all pointing errors as a single type! This suggests that error making has a strong random quality. We cannot look simply at the most difficult parts of a model in debugging.
Although we focused on whether group development could reduce errors, we must not focus so much on statistical significance that we lose sight of practical importance. The simple fact is that even tetrads had error rates totally unacceptable in the real world. This study adds one more piece of evidence to existing data (Panko and Halverson, 1996) on the prevalence of spreadsheet errors.
Publication
1997, Office Systems Research Journal, Volume 15, Number 1, Spring, pages 21-32
