This paper reports on the theoretical and empirical developments for an error prediction methodology called Task analysis For Error Identification (TAFEI). Other researchers have noted the need for theoretically-driven approaches that are able to provide practical utility in error prediction. Theoretical developments include the concept of 'rewritable routines', that describes the loop between cognitive processing, action and devices states. This has been proposed as a way of unifying ideas from systems theory and cognitive psychology. The empirical research shows that TAFEI is superior to heuristic methods, which supports the idea that structured methods assist in error prediction. The validation study shows that TAFEI reaches acceptable levels in terms of test-retest reliability and concurrent validity. It is believed that the method has reached a level of maturity after ten years of development work. This is demonstrated by the many uses that the method has been put, including that of a design tool.KEYWORDS: human error, methodology, reliability, validity, sensitivity.
Page 2Validating TAFEI 3
FOUNDATIONS OF TAFEIIn 1994, we introduced a technique for predicting human error when people use devices (Baber and Stanton 1994). The technique was called Task Analysis For Error Identification (TAFEI).The theoretical foundation for this approach was based upon the unification of ideas from general systems theory (von Bertalanffy 1950) and human problem solving (Miller et al. 1960, Hayes-Roth and Hayes-Roth 1979, Suchman 1987, Simon and Young 1988 resulting in a statespace search with boundaries between human and device elements. Initial attempts at validating the approach suggested that the errors predicted using TAFEI were relatively consistent with errors observed when people used the devices under consideration .During the past decade, our efforts have been directed towards consolidating the theory that underlies TAFEI. At the heart of this theory are the ideas of 'relevance' and 'rewritable routines ': we posit that in order to move from current to relevant states, eliminating other possible states, the user needs to retain some (temporary) record of the interaction and to have some means of assigning relevance to states . At each state, this record will be modified. We assume that the record will be held in working memory, presumably in the articulatory loop which has a limited duration (e.g., around two seconds, Baddeley 1986). This means that unless the record is updated, it will decay or be disrupted. This could result from distraction, e.g., interrupting the activity, or competition, e.g., having more than one option that appears relevant. As the record will also guide the next action, we see this as a rewritable routine. The current routine will be performed and the results used to update the record, in preparation for the next routine. To some extent this notion is similar to the 'partial provisional planning' hypothesis of Simon and Young (1988) and also resonates with the idea of a
Page 3Valid...