At the end of the session, all the collected data -- video, audio and proxy logs -- was reviewed and a detailed transcript of the session was produced. A re-enactment protocol was then employed to elicit from subjects the reasoning behind their actions . All the utterances of the subject and the investigator were recorded on audio cassette and later transcribed.
Analysis of this data leads us to propose a three phase model of user link activation decision-making behaviour:
``This page is on a University server and that means it should be quick because they have fastish servers.'' (Subject C)
The assessment of value process that the user undertakes prior to activating a link can therefore, be split into two parts:
Johnson has employed micro-economic cost-utility models and the notion of marginal utility to analyse Web user responses to download delays . He demonstrated that users' tolerance of delays could be manipulated by increasing their expectations of document quality. Johnson's findings are evidence for the close relationship between the assessment of value phase and the download phase; by increasing the information available to the user before they activate the link, the user is not only better informed about its nominal value, but is also better able to make an informed decision about whether the document is worth waiting for.
Some of our subjects revealed interesting beliefs about the extent to which their actions could influence the download time. For example, several subjects articulated the view that slow downloads could be speeded up by stopping and restarting them. When asked to explain this, one user drew upon a telephone analogy:
``...[it] is like a crackerly telephone line, if you ring again the quality is often better.'' (Subject E)
Others provided more technically sophisticated rationales for their behaviour:
``...when you stop and reload, it is possible to get a better route for the data, so it may be quicker.'' (Subject G)
``...there is only a certain number of connections that a server can make, by reloading a page a new connection may have become free due to a time out or other users disconnecting.'' (Subject D)
``...all sites in Australia are slow.'' (Subject A)
``...it is night time in America so all information will be transferred very fast.'' (Subject I)
It was evident from subjects' comments that the URL was often an important element in the assess-download-evaluate cycle. The URL could be used to predict the content of the referenced document, the location of the Web site and the download time in the Assessment of value phase. In the Evaluation phase, in retrospectively comparing the outcome with their initial expectations, users would try to (re)-interpret the URL in a way that was consistent with the outcome, the spatial location of the site and the time taken for data transfer.
These observations led to the design of a second study to investigate what information users can extract from the URL and how this may figure in the link activation decision. Eighteen subjects with varying degrees of Web experience and expertise in computer use were recruited. Each one was shown a list of URLs and asked to predict what he or she thought the referenced document was, and to predict its download behaviour. Subjects were then shown the actual referenced document and asked to assess the accuracy of their prediction.
The results indicated that more experienced subjects were able to
deconstruct a URL into its constituent parts (see Figure 3).
Generally, a ``full'' prediction would include inferences about geographical
location from the hostname, about content from the pathname and filename, and
about download behaviour from the
file_type and hostname parts
document type and location). In contrast, less experienced subjects were
generally unable to distinguish between the constituent parts of the URL.
Inferences about location, content and download behaviour were therefore
based upon a decontextualised recognition of strings within any part of the URL.
In Figure 4 we show extracts from transcripts of session protocols. Interviewer's and subject's comments are distinguished by (I:) and (S:) respectively. The first extract shows the kind of inferences made by a typical novice subject. The next three extracts illustrate the kinds of inferences made by more experienced subjects.
Informed by the results of these investigations, we now turn to consider in more detail how the usability objectives defined earlier may be tackled. Our discussion will focus on two broad areas in turn: documents and links.