File Size and Download Time

Overview

According to Jakob Nielson "Every web usability study I have conducted since 1994 has shown the same thing: Users beg us to speed up page downloads" (Nielson, 2000). In other words, file size is a crucial factor in web usability. Most of us who have spent any time on the web, particularly on a dial up modem, would have to say a big "duh" to this, given how annoying it can be to wait for a long download and how quickly we will leave a page that is not crucial, instead of waiting. In his 2000 book, Nielsen quotes a classic paper by Robert Miller (1968), which specifies a relationship between computer response time and users' perceptions: 0.1 second = limit for having the user feel the system is reacting instantaneously; 1 second = limit for user's flow of thought to remain uninterrupted; and 10 seconds = the limit for keeping user's attention focused on the dialogue.

It turns out that these "laws" were principally based on Miller's experience, not research and that, in fact, more recent research indicates that response time does not have nearly as much effect on user's performance as one might think (Butler, 1983). It appears that the main problem with long "response time", which translates into long download time on the web, is not so much a problem in interfering with tasks, but rather a problem of annoyance. So, while there is not strong evidence for a decrease in user performances, there is clear evidence for an increase in user anxiety with an increase in response time (Guynes, 1988). (This anxiety may, in fact, translate into physiological excitation, which may translate into an increase in surfing behavior, but more on that later).

Nah, 2003: Variability of Acceptable Download Time Estimates and the Effect of Providing User System Status

At the 2003 Americas Conference for Information Systems I attended a very interesting presentation by Fiona Nah, where she reviewed research on tolerable wait time on the web. In her conference proceedings paper she states the following about the varied estimates of the acceptable wait times:

Although long download time of Web pages has been a consistent problem encountered by Web users (Lightner et al., 1996; Pitkow and Kehoe, 1996; Selvidge, 1999, 2003), it is still controversial as to what constitutes an acceptable waiting time for a typical Web page download. Nielsen (1997) advocates the 10-second limit, while Zona Research (1999) recommends the 8-second rule. Selvidge’s (1999, 2003) study indicates that there is no difference in users’ frustration levels between 1-second and 20-second delay, but a difference (with 1-second delay) was observed at 30-second delay. Other researchers propose the 2-second rule (Shneiderman, 1984) and the 12-second rule (Hoxmeier & DiCesare, 2000). The conflicting evidence in the literature was also highlighted and examined by Galletta, Henry, McCoy & Polak (2002), who observed decreases in performance and behavioral intentions at 4 seconds. Interestingly and ironically, the average American users that use dial-up connections wait about 30 seconds the first time they look at a new Web page (Chen, 2002)! (Nah, 2003)

After reviewing the literature, Nah conducted an experiment of her own using a clever design that included an examination of the impact of a "feedback bar" on users' performance. This bar provided some users with information about the amount of download left to go. Such information is consistent with one of the fundamental rules of heuristics, according to Nielsen's 10 usability Heuristics. The first heuristic: Visibility of System Status, states "The system should always keep users informed about what is going on, through appropriate feedback within reasonable time." This, by the way, is consistent with an even more fundamental principle involving humans and stress, which is that our perception of stress is strongly influenced by the degree to which we can predict and control a stressful stimulus (which a web page's loading time can often be).

In order to define "acceptable" down load time Nah used a clever and somewhat devious technique. She asked students to search for specific computer hardware and software from a page that had ten links, three of which did not work. When a student clicked on the link would never connect and the user could only stop this activity by clicking a stop-loading button. The time between when the student clicked the link and clicked the browser stop button defined the wait time. For half the students a feedback bar indicated that the computer was attempting to connect (it did not show absolute progress, it just moved back and forth). Her principal findings were: 1) Wait times decreased with each time one of the infinite links was clicked (the user learned not to wait on that link); 2) Those in the feedback condition waited significantly longer across the multiple attempts with the infinite link button; and 3) Her final "acceptable waiting time" estimate was 2-4 seconds, based on the time that users without feedback waited on their third attempt to access the infinite links.

Physiological Excitation, Surfing Behavior, and Download Time

In a particularly interesting study physiological excitation was measured as users waited for a page to load (Sundar & Wagner, 2002). Physiological excitation can be thought of as an increase in activity in the sympathetic nervous system. When the sympathetic branch of the autonomic nervous system kicks in a series of responses associated with "fight or flight" occur, including events such as increased heart rate, blood pressure, and increase respiration rate. It's important to note that these responses indicate an increase in arousal or excitation, but they do not indicate the value of the affect. In other words, the arousal could be good (e.g., excitement) or bad (e.g., fear). One of the other physiological results of this type of sympathetic activity is water on the skin, which is referred to the galvanic skin response, or skin conductance. Skin conductance was the measure of arousal used in this particular study.

Users were divided into two groups. Half viewed a page that downloaded quickly and half viewed a page that downloaded slowly. Skin conductance was measured as they watched a page download, in order to assess the degree to which download speed affected the degree of physiological arousal. This initial download served as the principal experimental manipulation. Following that, users in both groups were allowed to surf as they chose (beginning with the CNN page) with a high speed connection. The purpose of the surfing portion of the experiment was to assess the degree to which the experience with download speed transferred to subsequent intensity of physiological (skin conductance) and behavioral (surfing activity) responses.

Not surprisingly, researchers found that participants exposed to slower downloading pages had higher levels of arousal during the page download. Most of us would assume that this represents some type of negative affect like frustration. The particularly interesting finding was that those people exposed to the slow down loading site also had significantly higher levels of arousal in subsequent surfing and clicked on significantly more links. As the authors put it, “…excitation transfer of residual arousal from a slow-loading image serves to intensify physiological as well as behavioral responses to subsequent online stimuli.” This research was mentioned a number of times in the popular press as evidence that slow download can actually be good, which isn’t exactly what the study demonstrates, but it does demonstrate that slow download may have a number of different effects.

Keeping File Sizes Down

Despite the varied estimates of acceptable times, the impact of providing user feedback, and the interesting effects on intensification of subsequent behavior, it’s almost surely a good idea to try and minimize file size and, thus, download time. Nielson's rule of thumb in his 2000 book is to keep a page below 35k (which is pretty darn small, especially for an elaborate portal), but this is about 10 seconds on a 28.8 modem. This is, in reality, a pretty unrealistic estimate for a page that has much content beyond text. (For example, when I wrote the first version of this commentary, 2003, I downloaded Amazon.com’s welcome page into a folder and it was 180k and 97k of that is just the code). The main point is to take page size into consideration as you’re developing a site.

Some tricks you can use to help with user anxiety by providing feedback information is to is to attach alternative tags to graphics (which is always a good idea, anyway), so users can see these while the graphic is downloading, and to break graphics in parts, so as to give the user some notion that something is happening, and an estimate of how fast. It's also a good idea to label the size of large multimedia files on your site and estimated download times for given connections.

One trick you can use to literally decrease a file’s size is to optimize your graphics, balancing the smallest amount of colors (.gif) or least possible quality (.jpg) with the best appearance. Finally, of course, following web-coding standards can dramatically decrease the size of your code. Using structural XHTML mark up and relying on style sheets to control presentational effects can significantly reduce the amount of code required.

Impact of Broadband

One last point is to note the impact of the growing use of broadband, DSL and cable modem connections. For example, despite the fact that I live outside of town in rural Missouri, DSL was recently made available in our neighborhood. Me, being the geek that I am, immediately ordered it and now we have a fixed download rate of 1.5Mbps, and suddenly large files are much less of an issue than they used to be. However, the truth is, I still get frustrated with slow pages as most users no matter what their connection speed. So it's probably never a bad thing to try and minimize page size. Moreover, broadband connections are still far from ubiquotous. According to a Pew Internet & American Life Project report (referenced in a ClickZ article), while 55 percent of adult internet users have high-speed access either at home or work, only 39 percent have broadband at home. (This means, of course, that still close to half of those in the U.S. don't have broadband access at all, and this doesn't even take into account the rest of the world).

References