The experiment is run in three sessions. The steps of the experiment are listed in the main text in the section labelled The Reader's Task.
The documents are all presented using the Netscape browser [NS97B]. The documents are marked-up in HTML [RLHJ97]. Netscape's version of JavaScript [NS97A] is used to speed-up some of the interactions. All of the link tracking code is complete and tested. I hope to have screen shots to show at the workshop.
Subjects answer questions using HTML forms. They are asked their age, educational background and to rate: their experience with hypertext and with graphical computer user interfaces, as well as their enthusiasm for the experiment. All of those factors have been reported as significant in determining people's success with hypertext. They also answer questions about how they read text on computer screens. Because reading on screens is such an important part of the experiment it must be accounted for. It is not enough mearly to control for its effect.
After they complete each reading they rate aspects of the document using the QUIS questionnaire [QUIS94]. Selected questions from QUIS are presented in an HTML form. The form data is collected by a CGI program [CGI] and stored in a file for analysis.
Figure 1 illustrates the interface used. On the left is an illustration of the document as it is read. A large HTML frame [RLHJ97a] contains the document, while below is a small frame that holds only a password field and a form button.
Figure 1: Document Frames Used in the Experiment
(a) During reading (b) During evaluation
+-------------------------------+ +------------------------------+
| The text of the document | | The document appears |
| appears in the larger | | exactly as before. |
| frame (which is above the | | |
| small one). | | |
/ / / /
/ / / /
| | | |
+-------------------------------+ +------------------------------+
| The small frame contains only | | The small frame contains |
| a password field and submit | | the link heads and tails |
| button. | | and a ratings form. |
+-------------------------------+ +------------------------------+
Every time the user follows a link details about the link are recorded in a file. The method is complex [Blu98]. Readers with no interest in the nitty-gritty of the programming of the link tracker are advised to skip the rest of this section. Because the documents are long (over 400 Kb) and presented in one piece it is impractical to have the CGI program redirect to them after every link. Instead Netscape's version of JavaScript [NS97A] is used to allow a CGI program to run at the same time as the browser changes the display to show the destination of the link. To write to a file a CGI program must be used. To get the browser to jump to the link's head without reloading the entire document JavaScript (or some other client-side program) must be used.
Further complications arise because the CGI program must produce output otherwise there will be a server error. But if the output goes to the same frame as the document or the form then they must be reloaded every time a link is followed. The CGI program's output is therefore directed to a frame of zero size (0 columns).
In summary: links are recorded by a CGI program that is run when they are
followed.
When a link is followed a JavaScript function (called using
onClick()) runs the CGI program in a hidden frame and
redirects the browser (not the server!) to the tail of the link.
When the subject has finished reading the document, the experimenter enters the password and selects the button. A CGI program generates a list of links the subject followed while reading and they are soon loaded into the bottom frame. The links appear in the order they were selected in the document. Figure 2 shows an example of a link rating from an actual experiment. Selecting the tail (or source) of the link in the bottom frame causes the corresponding location in the document to appear in the top frame. Similarly for the head (or destination) of the link. In Figure 2 the tail is labelled L551 and the head 3.3.0.0.0.0.0.0.0.0:2.2. Once a rating has been assigned then selecting the `next' button will record the rating and change the text in the bottom frame to show the next link to be evaluated. Selecting the `next' button without making a selection provokes an informative message.
Figure 2: An Example of a link rating form
| 4. | How useful was the link from L551 to 3.3.0.0.0.0.0.0.0.0:2.2? (15 is most useful.) |
Although it may appear simple to have links in one frame affect another using the HTML 4.0 frames model [RLHJ97a] the exact behaviour depends on the configuration of the user's software. For my experiment I am using one browser in an off-line condition.
[Jump back to The Reader's Task section or to the top of the main document]
[Jump back to The Reader's Task section or to the top of the main document]
Copyright © J. Blustein, 1998.