Consistency and usability

Abstract

This thesis is divided into two main parts. In the first a multi-component framework for usability is outlined. The components - guessability, learnability, experienced user performance, system potential, and re-usability - describe how users' performance on a task changes with experience, and each is associated with a different part of a learning curve. Definitions of the components are offered, as are examples of where each may be particularly important. The potential advantages of the framework are then discussed. An experimental methodology and criteria based analysis method for quantifying the components of usability is proposed. This is demonstrated in the evaluation of a video cassette recorder. Non-experimental alternatives to this methodology are also considered.
 
 The second part of the thesis is about the issue of consistency. A distinction between two types of consistency - set compatibility and rule compatibility - is proposed. Set compatibility is concerned with inter-task associations, whilst rule compatibility concerns the prior association of tasks and action-rules. Predictions are made about the effects of each type of consistency on the various components of usability, and these are tested in the context of a study of the invocation of menu commands. Results indicated that rule compatibility had the greater effect on early interactions, whilst set compatibility was more salient later on. A series of further studies is then reported, the aim of which was to investigate whether these effects were general across types and levels of interface, and other levels of task. Results mainly, but not entirely, indicated that they were.
 
 Data from a more `ecologically valid' usability evaluation was re-analysed, to investigate whether the effects of consistency are important outside of artificial and tightly controlled experiments. Apparently they are - almost half of the difficulties encountered during users' early interactions with a commercially available word processor could be attributed to either set or rule incompatibilities.