Can Humans Detect Errors in Data? Impact of Base Rates, Incentives, and Goals

There is strong evidence that data items stored in organizational databases have a significant rate of errors. If undetected in use, those errors in stored data may significantly affect business outcomes. Published research suggests that users of information systems tend to be ineffective in detecting data errors. However, in this paper it is argued that, rather than accepting poor human error detection performance, MIS researchers need to develop better theories of human error detection and to improve their understanding of the conditions for improving performance. This paper applies several theory bases (primarily signal detection theory but also a theory of individual task performance, theories of effort and accuracy in decision making, and theories of goals and incentives) to develop a set of propositions about successful human error detection. These propositions are tested in a laboratory setting. The results present a strong challenge to earlier assertions that humans are poor detectors of data errors. The findings of the two laboratory experiments show that explicit error detection goals and incentives can modify error detection performance. These findings provide an improved understanding of conditions under which users detect data errors. They indicate it is possible to influence detection behavior in organizational settings through managerial directives, training, and incentives.

• 저널MIS Quarterly · 21(2) · 169–194
• 토픽Bayesian Modeling and Causal Inference · Artificial Intelligence
• DOI10.2307/249418
• 저자Barbara Klein, Dale L. Goodhue, Gordon B. Davis