The Time Course of Routine Action

Previous studies of action selection in routinized tasks have used error rates as their sole dependent measure (e.g. Reason, 1979; Schwartz et al., 1998). Consequently, conclusions about the underlying mechanisms of correct behavior are necessarily indirect. The present experiment examines the performance of normal subjects in the prototypical coffee task (Botvinick & Plaut, 2004) when carried out in a virtual environment on screen. This has the advantage of (a) constraining the possible errors more tightly than a real world environment, and (b) giving access to latencies as an additional, finer grained measure of performance. We report error data and timing of action selection at the crucial branching points for the production of routinized task sequences both with and without a secondary task. Processing branching points leads to increased latencies. The presence of the secondary task has a greater effect on latencies at branching points than at equivalent non-branching points. Furthermore, error data and latencies dissociate, suggesting that the exact timing is a valid and valuable source of information when trying to understand the processes that govern routine tasks. The results of the experiment are discussed in relation to their implication for computational accounts of routine action selection.