POINT OF GAZE MEASUREMENT AND ANALYSES WITH THE EYECATCHER / THE OBSERVER COMBINATION: A VISIT TO WWW SITES ON THE INTERNET

Mooij & Associates, Oegstgeest, The Netherlands

This paper presents the possibilities of using an eye tracking system (EyeCatcher) in conjunction with a video tape analysis system The Observer). Information obtained with and from both systems can be successfully integrated. The operation of the combined systems is illustrated, using an experiment which has recently been conducted as an example.

In this experiment a test person was requested to surf' over the World Wide Web, while he was asked every now and then to find an Internet site providing information on a certain topic. Throughout the experiment the computer image (i.e. the screen of the web browser) was recorded on video tape, using a VGA-PAL converter, while EyeCatcher recorded where at any given moment the subject was looking. EyeCatcher superimposes the point of gaze' as a crosshair cursor on the video image observed with a scene camera mounted on the subject's head; the combined signal is stored on video tape. The two video tapes, both provided with time code information, were subsequently coded with The Observer.

The experiment was focused on relatively obvious occurrences, the reason being that it was set up to find out to what extent EyeCatcher and The Observer might be integrated and thus complement one another, and not to obtain extensive new data regarding a person's surfing' behaviour. A three-phase analysis was carried out with The Observer.

1. Definition of categories. Classification resulted in the definition of three main groups of categories: "computer action", "user action" and "Internet site". The various lists of records pertaining to these three groups of categories were described.

2. Data collection. The video tapes were played in slow motion and any changes in status and/or behaviour of the three main groups of categories were entered into the computer by simply clicking the mouse or by keystrokes. The time code mentioned earlier ensured input of the correct time into the computer, in spite of the slow-motion reproduction.

3. Data analysis. Analysis of the data obtained with the statistical package of The Observer resulted in a number of tables:

• The first one was a time-event table, which in chronological order demonstrated the function of the three registered systems.
• The elementary statistics showed the frequency and duration of each registered type of behaviour, including percentages, standard deviation and other statistical standards of the duration.
• Using the lag sequential analysis of the software package enabled a transition matrix to be drawn up showing which type of behaviour followed which other type of behaviour. Some trivial facts which immediately became clear from this analysis were that quite often a subject's reading and scrolling were closely related and that his waiting and browsing were equally coupled.
• Finally, the elementary statistics offered the possibility of studying nested variables, such as how long the user paused (without actually doing anything) while the system was downloading or building up a page, or what percentage of time the subject spent on reading and searching per site.

The experiment demonstrated it to be feasible to use the two systems in combination. They complement one another in a very interesting manner. This experiment was a variant to many, commonly conducted research experiments into human-computer interaction. However, it also indicates that comparable tests in different settings may very well be possible. After all, the apparent behaviour (e.g. inputs made) and viewing behaviour of a subject will be closely related in innumerable cases and knowledge of both is bound to make analysing a lot easier and less complicated and improve the quality of the outcome.