On the microstructure of learning a wordprocessor

This paper aims at studying the microstructure of computer user's actions when learning a wordprocessing system. Latencies for different actions, i.e. identifying task, planning method, selecting command, typing argument, closing command and evaluating result, during the performance of editing tasks were registered. It was found that the learning curves for latencies for all actions except identifying task, could be fitted by a power function, with a power constant around 0.90. The fit was good and the constants similar for two different commands (delete and insert). Also for the different actions observed the power constants were similar, although the curves started at very different points, ranging from 4.31 sec. to 21.72 sec. Further, a model was proposed for basic operations behind the observed actions. The model proposes a sequencial procedure, where planning and recalling precede action. This model was tested against two different commands and was found valid for all data on the 2% level for the delete command and on the 8% level for the insert command. The model was also tested for individual subjects, where it was found that it was adequate for at least 80 percent of the subjects in three out of four cases tested. A closer study of the model fit suggested that the recall of a command was often performed in parallel with the planning of a method. Three subjects refuted the model by reflecting after their actions instead of before as proposed by the model.