Spreadsheet Error Detection

From Information Systems at Pitt Business

Spreadsheet applications are commonly used in the business world, due to their ease of use and the various capabilities and functionality built in to these applications. Due to the high reliance upon spreadsheets, it is important to understand the relative impact and frequency of errors in spreadsheets. This problem is compounded by the fact that error detection in spreadsheets is difficult and usually not undertaken on a regular basis.

This study reports an experiment to test the ability of reviewers to discover embedded errors in test spreadsheets. The findings indicate that it is easier to detect errors on paper rather than on the computer screen, however this tends to require more time to detect errors. Despite the increased proficiency of paper reviewing, only about 50% of the errors are detected, on average. Other similar findings are reported for spreadsheet design to aid practitioners in the develop of spreadsheet products.

[edit] Abstract

The pervasiveness and impact of electronic spreadsheets have generated serious concerns about their integrity and validity when used in significant decision-making settings. Previous studies have shown that few of the errors that might exist in any given spreadsheet are found, even when the reviewer is explicitly looking for errors. It was hypothesized that differences in the spreadsheets' presentation and their formulas could affect the detection rate of these errors. A sample of 113 M.B.A. students volunteered to search for eight errors planted in a one-page spreadsheet. The spreadsheet was presented in five different formats. A 2×2 design specified that four groups were given apparently conventional spreadsheets for comparing paper and screen and the presence or absence of formulas. A fifth group received a special printed spreadsheet with formulas visibly integrated into the spreadsheet printed in a small font directly under the resultant values As in previous studies, only about 50 percent of the errors were found overall. Subjects with printed spreadsheets found more errors than their colleagues with screen-only spreadsheets but they took longer to do so. There was no discernible formula effect; subjects who were able to refer to formulas did not outperform subjects with access to only the final numbers. The special format did not facilitate error finding. Exploratory analysis uncovered some interesting results. The special experimental integrated paper format appeared to diminish the number of correct items falsely identified as errors. There also seemed to be differences in performance that were accounted for by the subjects' self-reported error-finding strategy. Researchers should focus on other factors that might facilitate error finding, and practitioners should be cautious about relying on spreadsheets' accuracy, even those that have been "audited."

[edit] Paper Information

Authors: Dennis Galletta, Kathleen Hartzel, Susan Johnson, Jimmie Joseph, Sandeep Rustagi

Check out this paper at Spreadsheet Presentation and Error Detection: An Experimental Study

This paper was originally published in JMIS, Vol 13, Num 3, Winter 1996-1997.

[edit] Keywords

Electronic Spreadsheets, Information System Audits, Spreadsheet Validation