The World Wide Web has become a ubiquitous information source and communication channel. With such an extensive user population, it is imperative to understand how web users view different web pages. Based on an eye tracking study of 30 subjects on 22 web pages from 11 popular web sites, this research intends to explore the determinants of ocular behavior on a single web page: whether it is determined by the individual differences of the subjects (including the subjects’ demographic variables and their familiarity with different web sites), different types of web sites, the order of pages being viewed, or tasks at hand. The results indicate that gender of web viewers, web page viewing order, and the interaction between page order and site type influences online ocular behavior. Task instruction and background knowledge in terms of familiarity with the web sites do not significantly affect web viewing behavior. Scanpath analysis reveals that alternate web page design influences degrees of scanpath variation of the same web page. Design implications and future research directions are discussed.
: web page, eye tracking, individual differences
, World Wide Web
The World Wide Web has increasingly become a ubiquitous information source and communication channel. With such an extensive user population, it is imperative to understand how web users view different web pages in order to provide a cognitive basis for interface design. Web pages are different from other tested visual stimuli, as they incorporate a combination of textual, pictorial, and multimedia content. Eye movement behavior involves different levels of cognitive processes, including oculo-motor and semantic processes [Henderson et al. 1999]. Only in recent years has eye-tracking been utilized in web-based environments. Studies on web use behavior largely concentrate on navigation and search with exploratory generalizations [Hsieh-Yee 2001]. Very few academic studies have been conducted on eye movement behavior on web pages, with the exception of recent research on eye movement on news sites [Stanford Poynter Project 2000], scanpath analysis on web pages [Josephson and Holmes 2002], and ocular behavior on a web portal page [Goldberg et al. 2002]. The present research investigated the determinants of ocular behavior on a single web page, both in terms of standard ocular metrics such as fixation duration, gazing time, and saccade rate, and also in terms of scanpath differences. The key research question is whether these measurements were determined by individual differences, different types of web pages, the order of viewing, or different tasks at hand. Personalization and customization have been widely promoted by interface design theories. However, it is still not clear how different web users view web pages differently. Therefore, it is essential to understand, from a behavioral perspective, how individuals view and navigate within a web page.
Although initial inquiry into eye movement research began in the early 1900s [Rayner 1998], it has only been recently that eye tracking research has moved into a web-based context. This new medium affords a number of fresh opportunities to understand the cognitive processes involved in web page viewing.
2.1 Ocular Behavior Measurements
everal behavioral definitions have been widely adopted in the study of ocular behavior, including fixations and saccades. Eye tracking research generally defines fixations as a relatively motionless gaze which lasts about 200-300 millisecond (ms), in which visual attention is aimed at a specific area of a visual display. Saccades are continuous and rapid movement of eye gazes between fixations with a velocity of 500 degrees or more. Saccades are quick eye movements to direct a viewer’s eye to a visual target. Information processing is suppressed during a saccade, though some peripheral information may be available [Rayner 1998].
While information processing is limited during saccades, fixations have been linked to intense cognitive processing. According to Viviani , at least three processes occur during an eye fixation: encoding of a visual stimulus, sampling of the peripheral field, and planning for the next saccade. Research has shown that information complexity, task complexity, and familiarity of visual display will influence fixation duration [Duchowski 1998]. Rayner  has shown that eyes are attracted to the most informative areas of a scene because they are physically distinctive and informative. Similarly, Loftus and Mackworth  have asserted that the eyes are drawn to informative areas, which can be measured using dwell time within an Area of Interest (AOI). Fitts et al. (1950) has also concluded that fixation frequency in an Area of Interests (AOIs) is an indication of the degree of importance whereas fixation duration is an indication of the complexity and difficulty of visual display. The nature of the search task also influences eye movement behavior [Rayner 1998]. Pelz et al.  has shown that the complexity of tasks influences fixation durations. Hayhoe et al.  demonstrated task-specific fixation patterns in natural environments and have shown that there is a large degree of regularity between different subjects.
The previous research demonstrated various determinants of ocular behavior, especially the effects of task difficulty, familiarity with visual stimuli, and information complexity of the stimuli on fixation durations. This research used three ocular measurements as dependent variables, including mean fixation duration, gazing time, and saccade rate. Mean fixation duration is frequently used in eye tracking studies, and taken as an indication of information complexity and task difficulty [Rayner 1998]. Gazing time has been defined as the rate of gazing (e.g. fixation) across the total observation period, which is negatively related to task difficulty [Nakayama et al. 2002]. Saccade rate (saccade occurrence rate) was defined as the number of saccades per second, which is positively related with task difficulty or mental load [Takahashi et al. 2000]. In the current research, these variables were explored as a function of subject variables (demographic variables and site familiarity), site types, task instructions, and the order of web pages within the same site.
The term scanpath was first proposed by Norton and Stark . They defined scanpath as an habitually preferred path when a subject is reexposed to a visual stimulus. The concept of scanpath has also been accepted as a sequence of fixations and saccades, alternatively defined as a movement of attention. Josephson and Holmes  used the sequences of Area of Interest (AOIs) as scanpaths and used a string-editing method to calculate the differences between any two scanpaths. However, current research has yet to develop conclusive findings relating cognitive processes to the determinants of scanpath behavior.
This research uses the second type of definition which defines scanpath as a sequence of fixations and saccades. Using the string-editing method proposed by Josephson and Holmes , the differences of scanpath among different subjects on each web page were calculated and compared with the complexity of the web pages.
2.2 Recent Research on Web Page Viewing Behavior
Several studies have recently been conducted regarding ocular behavior on web pages, including eye movement research on news web sites
, analysis of scanpaths on web pages, and ocular behavior when web users were completing tasks on a web portal page. Against the popular accepted view of “a picture is worth a thousand words”, the Stanford Poynter Project , investigating reading behavior on news web sites, concluded that text was frequently the first-entry points for a majority of online news readers. Rayner et al.  reported similar findings in which viewers of print advertisements spent more time on text than pictures. By investigating eye viewing behavior of eight university students on three different web pages, Josephson and Holmes  showed that the subjects have habitually preferred scanpaths and they also demonstrated that features of web sites and memory might also be important in determining scanpaths. Their research is mostly descriptive in nature without significance testing. Goldberg et al.  used eye tracking methods to test the performance in completing several tasks on a web portal page. Their research demonstrated that there are more regular horizontal eye movements between different portlets on the page than vertical ones
, that the headers in a portlet were not usually visited before the body, and that searches did not become more directed as screen sequence increased. Implications for improving the design of the web portal were provided in their research.
This research focuses on web viewing behavior of web pages on different types of web sites using eye movement metrics and scanpath analysis. The goal is to test the previous research on the determinants of eye movement behavior: whether it is determined by the tasks the subjects are engaged in, the demographic variables of the subjects, different types of web sites, or the order of web pages being viewed. The answers to these questions can provide a cognitive basis for designing better web interfaces.