Human-computer interaction task learning: an empirical investigation of interface quality

Abstract

This research examines learnability measures derived from the all-or-none stochastic learning model. The all-or-none model represents the paired-associate learning that occurs during the initial learning of human-computer interaction (HCI) tasks. The ability of the model to quantify interface quality was the primary focus of the research. The quantitative measure of interest was a parameter of the all-or-none model referred to as a learning rate. Model validation was investigated by comparing learning rates to two other traditional measures of interface quality, mean task completion time and mean number of errors. The HCI environment selected for study was SuperCard , a Macintosh project utility that allows a user to create various applications. Thirty-two college students who TI were experienced Macintosh users but inexperienced SuperCard users participated in the experiment. They were required to learn 16 basic SuperCard T" tasks involving windows, menus, and resources. Sample path data was collected to estimate the model learning rates. Video tapes were analyzed to collect task completion times and numbers of errors. Subjective ratings of user proficiency were also collected. The subjective ratings were originally intended to be used as another comparison measure. However, the uniform distribution of the data and model residuals made any conclusions invalid. The results of this research indicated that learning rates derived from the all-or-none model are valid indicators of interface quality. Correlations between the learning rates and the other two measures were significant. Analyses of variance were conducted for the three measures. The learning rates identified seven SuperCardl" tasks for potential reevaluation and improvement: Name Resource, Window Type, Open/Close Resource, Name Menu, Size Window, Locate Window, and Resource ID. The Size Window and Locate Window tasks were identified by the mean task completion time measure and the Size Window task was identified by the mean number of errors measure. Additionally, user type was not a significant effect for the all-or-none model. Whether a user is a frequent or occasional user was a significant effect for mean task completion time and approached significance for mean number of errors.