Five Studies on Web Delay: A Research Program

The benefits of using the Internet are partially offset by one aspect of its usability: highly variable, intermittent, but frequent inter-page delay. For several years, the HCI literature has studied user reactions to long computer response time in clerical applications, but few studies have examined this problem in the domain of the Web. Examining the delay problem in a Web context is important, because the Web touches many more users, most of whom have little formal computer or task training.

Hence, we have examined in our labs consequences of delay, along with factors that interact with delay. Some of our experiments have been published and some are still in process. Consequences of delay that we examined include user attitudes, behavior, and performance. More recently, we have developed objective and subjective models of impacts, including stress, performance, and decision-making behaviors. Factors that we have examined for interactions include site depth, information scent (familiarity with terminology used in organizing the site), variability of the delay, and feedback (continuous and gradual filling of the screen to make it obvious that the page is indeed loading).

Experiment 1 (n = 196) provided seven levels of delay, ranging from zero to twelve seconds (in two-second increments), and discovered that ill effects began as delay exceeded two seconds. Experiment 2 (n = 160) introduced two other factors from the HCI literature, and with a 2 x 2 x 2 ANOVA, we assessed the interactions between delay, site depth, and familiarity with the terminology in the site. As predicted, we found a significant three-way interaction. Consistent with more traditional literature, we also found strong direct effects. Experiment 3 (n = 152) employed another 2 x 2 x 2 design, but along with delay we analyzed the effects of variability and feedback as interacting variables. Analysis revealed that page-loading feedback is only important when there are long delays, and variability does not seem to be important in influencing attitudes, behavior, and performance of users. Experiment 4 (n = 206) again compared reactions to the same levels of delay, but this time with Mexican subjects. It was found that Mexicans were more patient than subjects in the United States. In both studies, the outcomes differed when comparing a familiar site with an unfamiliar site, suggesting that interactions should be examined more formally.

Conclusions from the four completed studies are that user impatience is high; that the results of delay can differ with culture; and that the variables that interact with delay are familiarity with site terminology, depth of the site, and feedback (in a slow site). Also, variability does not seem to interact with delay.

The fifth study, in progress, examines only scent and delay, but breaks the model into subjective (assessed via questionnaire) and objective parts. Objective assessment of stress is provided by psycho-physiological instruments, objective assessment of web preferences is measured by choice of switching sites for remaining tasks, and objective assessment of ease of use is in observing the actual number of steps needed to complete the task. The study in process introduces psycho-physiological measures to expand our notion of web browsing outcomes. While we have barely scratched the surface in examining web delay, several implications can be drawn from the completed studies for both practitioners and researchers.