A Methodology and Investigation of an eLoyalty Metric, Consumer Bookmarking Behavior
Jamie Murphy's hospitality background and MBA led to an international marketing career and a Ph.D. studying the Internet. His industry and academic career spans five continents and includes publications in both academic journals and leading newspapers such as The New York Times and Wall Street Journal. His research focus is effective use of the Internet for citizens, businesses and governments.
Address: Business School, University of Western Australia, Crawley, WA, 6009 Australia. Tel: + 61 8 6488 1979, fax: + 61 9 6488 1055. Web: http://www.imm.uwa.edu.au/~jmurphy/
Charles F. Hofacker's academic career includes publications in major marketing and psychology journals and lecturing in the US and Italy. His current research investigates “what makes people click?” on Web links and Web banners. He studies optimal Web page design issues and Web ad banner click through probabilities from an empirical and behavioral approach using CGI scripts and live websites to perform randomized experiments. He has also written a textbook on Internet Marketing.
Address: College of Business, Florida State University, Tallahassee, FL, USA, 32306-1110. Tel: (850) 644-7864. Fax: (850) 644-4098. Web: http://garnet.acns.fsu.edu/~chofack/
This paper presents a simple and novel method to investigate an important indicator of a website's appeal as well as the relationship between a firm and its online consumers — namely, bookmarking behavior. Analyzing data from two inter-related hospitality industry websites and a genealogy site demonstrated the applicability, validity and future research directions for this eLoyalty metric. Similar to past studies of interest in a site, bookmarking a site becomes increasingly likely as a visitor views more pages on the site. Bookmarking was more likely during non-work hours, suggesting non-work related browsing on these sites. Finally, those visitors using the latest browser version were slightly more likely to engage in bookmarking.
While it makes no claim to catalogue the entire World Wide Web, the popular search engine Google indexed over eight billion Web pages in late 2004. Faced with this massive online competition for visibility, companies seek strategies that drive traffic to their site (Drèze & Zufryden, 2004; Ilfeld & Winer, 2002). Bookmarks, a basic Web browser feature and measure of interest in a site (Nielsen, 1997), simplify revisiting a site, but there is little research on this simple tool.
This paper aids practitioners by highlighting the importance of bookmarking, summarizing past bookmarking research and offering suggestions for increasing the likelihood of bookmarking their sites. Academically, this paper adds to the small body of bookmark research and hypothesis testing of electronic customer relationship marketing by investigating factors related to bookmarking and offering future avenues for research of bookmarks.
Consumers use numerous ways to return to Web pages, mostly via the Web browser's back arrow (Cockburn, McKenzie, & JasonSmith, 2002). Browsers also have a history list for use both within and between sessions. Users can also find and click on a link to previously visited pages, or retype the page's URL. Another option is to add a bookmark (Netscape Navigator and Mozilla Firefox), or favorite (Microsoft Internet Explorer), to their browser. Once the user has added the bookmark, the user can go directly to the page's address from the Web browser menu bar.
The Internet often resembles other offline communication media but its added flexibility complicates placing bookmarks in a historical context. For example, the layout of a Web page tends to resemble a magazine page. That it is user driven makes the Internet somewhat analogous to the phone system. Since it is software-based, this tends to make the Internet more flexible than other communication media. In communication terms, what does a bookmark resemble?
Historically, media have evolved towards more end user control. Radios and televisions let users pre-program their favorite stations and a VCR can return to one's favorite show. Digital television (Lekakos & Giaglis, 2004), digital radio and digital VCRs such as Tivo (http://www.tivo.com) add even more pre- and post-programming flexibility. Telephones often have speed dial capabilities, or directories of commonly dialed numbers. In addition to analogies with radio, TV and telephones, bookmarks resemble magazine and newspaper subscriptions. In addition to giving the user more control, subscriptions give the publisher guaranteed sales and a front-loaded revenue stream. With a universe of sites to choose from, bookmarking reflects loyalty and benefits a website (Nielsen, 1997).
Regardless of the user interface, bookmarks benefit the site. Taking the concept of loyalty online, e-loyalty or site loyalty resembles the offline concept of encouraging repeat store visits (Gommans et al., 2001; Holland & Baker, 2001). The results of a study using the power law of practice suggest a correlation between repeat website visits and future purchases (Johnson, Bellman, & Lohse, 2003). Bookmarks are a valuable tool for developing relationships with loyal visitors that return to a website (Nielsen, 1997).
Even after three decades, the importance of relationship marketing (RM) continues to gain attention as marketing turns away from a view based on individual transactions and moves towards an assessment of the lifetime of value that a customer can bring to a firm. This marketing method built on relationships, networks and interactions has a dual focus: getting customers and keeping customers (Berry, 1983; Grönroos, 1994; Newell, 2000; Wang, Head, & Archer, 2000). As acquiring customers is more expensive than keeping customers, relationship marketing focuses on the latter; good customers are more profitable and easier for relationship building. There is a strong correlation between customer retention rates and profit (Newell, 2000; Reichheld, 2001; Reichheld & Shefter, 2000; Szymanski & Henard, 2001).
The Web's interactive nature helps facilitates relationship marketing and customer support to a greater degree than traditional media (Hoffman, Novak, & Chatterjee, 1996; Newell, 2000). This interactivity shifts customers away from being passive receivers of information and towards being active searchers of information (Hoffman & Novak, 1996; Kotler, Jain, & Maesincee, 2002). Thanks to this shift, the balance of power is migrating away from businesses and “to the consumers, who can now define what they want in the way of customized products and services, prices, distribution channels, and even advertising and sales promotion” (Kotler et al., 2002, p. 8). Firms must manage customer relationships as the Internet can turn customers into partners.
Given the increased competition for more active customers, the importance of relationship marketing, and that corporate websites are commonplace (Porter, 2001), businesses want customers to bookmark and return to their website. A feature on some sites is a button, text or link suggesting “bookmark this page.” Microsoft and AOL for example, charge companies for entries in the default bookmark file that comes with a newly downloaded browser. All of this suggests the value of having an entry in consumers' bookmark files.
Previous Bookmarking Research
Cognitive and computer scientists have done most of the existing bookmarking research. Cockburn and McKenzie (2001) discuss three approaches for studying bookmarks: surveys, dynamic observation such as a talk-aloud protocol analysis (Benbunan-Fich, 2001; Jones et al., 2001; Jones et al., 2002), and static observation. The third method relies upon actual behavior of the individual user (client logs) or aggregate behavior of a website (server logs). The second method observes behavior and is predominantly qualitative, while the first method relies upon stated behavior and is predominantly quantitative.
In perhaps the first survey of bookmarks, Pitkow (1996) asked users in his semi-annual “WWW Users Survey” about their bookmarking behavior. Over nine of out ten of the 6619 respondents had bookmarks, and eight out of ten respondents noted that bookmarks were a strategy for finding information. One out of three, however, noted maintaining bookmarks as a major usability problem.
A survey of 322 Web users, also in 1996, found similar results (Abrams et al., 1998). Almost 19 out of 20 respondents had bookmarks, ranging from under ten to over 300 bookmarks. An analysis of the bookmark files from client logs suggested a disorganized, growing mess. Users added a bookmark about every five days but rarely deleted bookmarks. One in three users never organized their bookmarks, leaving them in the chronological order of when added and those with less than 35 bookmarks generally had no folders for organizing their bookmarks.
Finally, the users complained that titles on their existing bookmarks failed to describe the Web page's content (Abrams et al., 1998). Titles, part of a Web page's html code, play an important role in communication and remembering the page. Titles are often the main reference to pages and used in navigation menus such as bookmark lists and history lists (Nielsen, 2000, p. 123).
An online survey this century, investigating bookmark use and website use, found several significant relationships and underscores the importance of bookmarks (Thakor, Borsuk, & Kalamas, 2004). The use and organization of bookmarks show a negative relationship with using search engines. Bookmarks use shows a positive relationship with Web experience. Finally, the number of bookmarks shows a positive relationship with Web experience, Web usage and orientation towards online shopping.
Using a different methodology — analyzing the logs from individual's computers —Cockburn and McKenzie (2001) studied 17 users over a four-month period in late 1999 and early 2000. As was found in earlier research (Abrams et al., 1998; Pitkow, 1996), over nine out of ten users had bookmarks. They also found varied and disorganized bookmark usage. In their sample, the number of bookmarks ranged from one user with no bookmarks to two users vying for top spot with 587 and 565 bookmarks each. The average number of bookmarks was 184, stored in 18 folders. One user had 130 and another had 90 bookmarks in the top level — none of these bookmarks were in a folder.
The users added bookmarks much faster than they deleted or updated their bookmarks (Cockburn & McKenzie, 2001). Over the four months, users added an average of 28 bookmarks and deleted four. Of the existing bookmarks, one out of four pointed to a page that no longer existed, one out of 20 was a duplicate and one out of 20 bookmarks had no title. Users did, however, periodically re-organize their bookmarks by filing top-level bookmarks into folders.
Although the top three pages accounted for almost one fourth of all page visits, just six out of the 17 participants followed a bookmark to reach a top three page (Cockburn & McKenzie, 2001). The leading technique, practiced by 14 participants, was setting a popular page as the default home page. Five participants used another technique, the personal toolbar, to reach one of their top three visited pages.
This tendency to revisit Web pages rather than visiting new pages is the recurrence rate (Tauscher & Greenberg, 1997). Previous research using the same methodology on two separate data sets found a 60% recurrence rate (Cockburn & McKenzie, 2001). Applying the same technique Cockburn and McKenzie (2001) found an 80 % recurrence rate, which suggests evolving Web use. Users spent less time looking for new pages and more time looking at previously visited pages. This evolving Web use suggests that online customer loyalty and bookmarks become increasingly important (Newell, 2000)
Although from 92% (Pitkow, 1996) to 94% (Abrams et al., 1998; Cockburn & McKenzie, 2001) of users have bookmarks, using bookmarks may be a low priority. Of all possible browser actions — including reloading a page and opening a new window — opening a URL accounted for about half the actions. The other popular actions were clicking the back button about one third of the time and opening the home page about one time in twenty (Tauscher & Greenberg, 1997, p. 111-112). For opening a URL, following a link accounted for 83% of the actions followed by typing in the URL (7%) and using bookmarks (5%).
Two final studies used the third methodology, dynamic observation, filming and observing users talking aloud while performing Web tasks (Jones et al., 2001; Jones et al., 2002). Bookmarks were one of ten ways for revisiting Web pages, along with: e-mailing the URL to themselves or others, printing or saving the page, pasting the URL into a document or personal website, searching, typing in the URL and using the browser's history function. Their second study found marked differences in the ways that managers, researchers and information specialists kept Web pages (Jones et al., 2002). For example, managers always e-mailed URLs to others while information specialists never e-mailed URLs to others.
These quantitative and qualitative studies highlight academic and industry interest in bookmarking. Over nine out of ten users have bookmarks and use their bookmarks about one time in twenty to open a URL. Web browsers and Web browsing continue to evolve, with methods to revisit Web pages important. Microsoft for example highlights “Favorites” as a feature on their latest Internet Explorer, version 6 (http://www.microsoft.com/windows/ie/evaluation/overview/default.asp). Previous research, however, approaches bookmarking from the user's perspective rather than the site owner's perspective. A user making the effort to bookmark a site reflects intent to revisit the site.
It seems probable that for most sites, the more pages that a user views, the more they like the site. Adar and Huberman (2000) propose that the number of pages a visitor retrieves from the site, often referred to as site depth, expresses the utility that they derive from the site. Goldfarb (2002) shows that previous site depth increases the probability that a visitor returns to a particular portal. Other authors mention site depth as a potential goal for sites (Bucklin & Sismeiro, 2003; Holland & Baker, 2001; Novak & Hoffman, 1997). Search engine sites also want visitor to view as many pages as possible in order to sell more advertising (Hofacker & Murphy, 2000).
Hypotheses 1: The more pages within a site that a visitor views, the more likely that visitor will bookmark the site.
Bookmarking necessarily involves saving information to the local computer in front of the visitor. A home computer, typically owned by the visitor or by the household, is therefore available for storage of favorite or bookmark files. A computer used at work, however, may not be as suitable for saving bookmarks. The computer may be a shared resource or the employee may be browsing sites unrelated to work. These examples would seem especially compelling for sites unrelated to work activities.
Hypothesis 2: Bookmarking of a site occurs proportionately more often during weekends than during weekdays, as defined in the time zone of the site.
Although the Internet has rapidly diffused through much of the developed world in the last decade, there are still large numbers of people to whom the workings of this network remain a mystery. Research has shown that compared to novice users, advanced users organize their bookmarks better (Abrams et al., 1998), and employ several techniques to remember interesting sites including email to themselves and creating their own Web pages (Jones et al., 2001; Jones et al., 2002). Finally, results from the eighth GVU World Wide Web Survey suggest that compared to novice users, experienced users are more likely to switch browsers in part to upgrade their browsers (Toon, 1998). Thus compared to novice users, advanced users may have newer browser versions and tend to use bookmarks more frequently.
Hypothesis 3: The older the browser version, the less likely the visitor is to bookmark.
Past research studied consumer bookmarking behavior using surveys, dynamic observation and client logs. These methods provide useful information, especially for researching Web browser design, but they provide little information for website owners and managers. While general observations about bookmarking are important, the inherent characteristics of these previous methodologies limit addressing myriad questions about when and why visitors bookmark certain websites.
This study uses server log files, leveraging a subtlety in how Microsoft Internet Explorer requests information from a Web server. When the user bookmarks a site (or in Microsoft terminology, adds the site to their favorites list), the browser looks on the server for the file favicon.ico, a 16x16-pixel image that the browser then copies to the visitor's computer. This activity provides a small icon for the site's entry in the visitor's favorites list as well as in their browser's address bar.
Thus for Internet Explorer, which represents 96% of the browser market (Broersma, 2002), a request for favicon.ico indicates that a visitor has bookmarked the site. This method works whether or not the favicon.ico file exists as the server records the request. This method though, can only record a visitor bookmarking a site, not if a visitor uses a bookmark for subsequent site visits.
This research utilized server log files from two different content areas. The first area was two highly interlinked and related sites sponsored by a hospitality business in the southeastern United States. Both sites, online since 1996, averaged from three to eight thousand visitors per week. The third site, which offered users a variety of genealogy services, averaged over ten thousand visitors per week. This site had hundreds of content pages and many links to other genealogy sites. The look was professional yet clean — specifically, there were no animated graphics or flash introductions. Data collection ranged from July 6 through October 6, 2002 for the first two sites and from September 1 to October 20, 2002 for the third site.
Data analysis began by deleting records that originated from this paper's authors, represented an image or multimedia file, or resulted from a robot or Web spider. An exhaustive list of robots, developed during the course of this research, enabled the analysis of data from human visitors rather than automatic Web software. The remaining log records were then checked to see if they originated from the same visitor.
Our definition of a visitor resembles that proposed by Novak and Hoffman (1997), adjusting for Internet Service Providers (ISPs) who use proxy servers and dynamic Internet Protocol (IP) addresses, which complicates identifying individual computers. We used only the first half of the IP address for visitors from ISPs using proxy servers, such as American Online, Web TV, and Compuserve. To separate simultaneous visitors from the same online service, a visitor was a series of page views with a lag of no more than fifteen minutes between pages, rather than Novak and Hoffman's (1997) 30 minutes. In addition, to qualify as the same visitor, a series of page requests had to originate from the same browser and operating system version, something recorded in server log files (Murphy, Hofacker, & Bennett, 2001).
During the data collection period, 108,274 unique visits occurred on the hospitality sites for a total of 177,324 page views and 999 bookmarking events. The genealogy site recorded 90,133 visitors who viewed 583,415 pages and made 1,846 bookmarking events. Figures 1 and 2 show the relationship between the number of pages requested and the loglinear number of visitors for these sites. The loglinear nature of these functions is consistent with previous research (Adamic & Huberman, 2000; Adar & Huberman, 2000).
Hypothesis 1 stated the more pages viewed, the more likely a bookmarking event would occur. Figures 3 and 4 suggest that the probability of bookmarking increases with pages viewed. In Figure 3, the probability of bookmarking given nine page views appears to be an outlier due to sampling noise, which tends to be greater when there are relatively few observations.
The results of a logistic regression confirm Hypothesis 1. The findings for the hospitality sites (Figure 3) are a t-statistic with eight degrees of freedom, equal to 9.59, p < .0001. The findings for the genealogy site (Figure 4) are t(13) = 6.35, p < .0001. The R2 values for the two sets of data are .92 and .76, respectively.
Hypothesis 2 proposed that the propensity to bookmark should be greater during weekends than during weekdays. The servers for these sites are in the United States' Eastern Time Zone but determining each visitor's time zone is difficult. It is possible to determine the visitor's domain through a reverse domain name system (DNS) lookup but there is no guarantee that a visitor from a .com domain is in the United States. Furthermore, the DNS lookup taxes the server at the expense of serving Web pages.
Figures 5 and 6, which show what time individuals arrive on these sites, suggest a daily — albeit slightly different — rhythm to visiting these sites. The time of access in these figures, especially for the hospitality sites, correspond with daylight hours in the US. Visits to the hospitality sites (Figure 5) begin to peak much earlier in the day and reveal a much greater separation between high weekday access and lower weekend access. Hospitality site visits dip towards noon on weekdays. There is an equivalent dip on the weekends, about an hour later. The genealogy site, though, shows a weekday drop that reaches its lowest point at the dinner hour. This drop all but vanishes on the weekends.
Figures 7 and 8 show when individuals bookmark and these curves tend to track Figures 5 and 6. To analyze Hypothesis 2, we compared two periods within each of the two sites, calculating the number of bookmarking events per page view for weekdays and weekends. For the hospitality sites, the probability of a bookmark per page view was .0067 for weekdays and .0084 for weekends. Weekend visitors were 20% more likely to bookmark. This difference was significant by logistic regression, t(46) = 3.00, p = .0043. Although bookmarking is an uncommon event, visitors to this site bookmark significantly more often on the weekends.
The genealogy site though had a bookmarking probability of .0033 on both weekends and weekdays, thus failing to confirm the hypothesis for this site.
Hypothesis 3 proposed that older browser versions would imply less bookmarking than newer versions. The observed 86,425 page views with Microsoft Internet Explorer 5 yielded 441 bookmarkings for a .0051 proportion compared to 79,918 page views with Internet Explorer 6 that netted 474 bookmarkings for a .0059 proportion. This difference is in the hypothesized direction, but small. The improved bookmarking functions with Internet Explorer 6 may lead to more bookmarking. Yet, as Web browser technology has evolved, users tend to upgrade to the latest version less frequently (Nielsen, 1998a, 1999).
This study demonstrated a simple and novel way for website owners and managers to gauge visitor interest in returning to their site, the propensity to bookmark. Furthermore, the data for this online measure of loyalty already exist. Unlike past research, this may be the first study that approaches bookmarking from the site owner's perspective, albeit with limitations to the methodology. Bookmarks do not measure repeat visitors, rather the intention for visitors to return. Determining the visitor's time zone is also difficult.
Limitations aside, server-side data are advantageous for investigating and enhancing Web-based communication. The server automatically collects this data, which is therefore available to the site owner at zero marginal cost. In contrast, client side data require the identities of current or potential site visitors, sampling from those individuals, recruiting those visitors in a panel, and then retrieving the information from their visits.
These results confirm past studies showing that site depth is a valuable metric, this time as a gauge of loyalty. As well, the study provides several managerial implications, academic implications and fruitful avenues for future research.
This research confirms previous findings. Bookmarking is a rare event, and would be a competitive advantage for sites that promote bookmarking. Website owners and managers might overcome this low frequency by designing pages for bookmarking and reminding visitors of this neglected browser function or to “mail this page to yourself for later use.”
A useful page title for all Web pages is mandatory since browsers label bookmarks based upon the page's title. Nielsen (2000, p. 123) compares titles to billboard slogans and recommends that each title is a pearl of clarity. Maintaining rather than deleting old content is also important, as users may have bookmarked these pages (Nielsen, 1998b)
Finally, the methodology in this study could help improve site or page design. Using favicon.ico and two versions of some design elements or content, management can test which version visitors are more likely to bookmark. Furthermore, the methodology profits from the high internal validity of using actual Web visits (Drèze & Zufryden, 1997)
A wealth of Internet marketing and communication literature is emerging, such as the 2002 Journal of Academy of Marketing Science special issue (Parasuraman & Zinkhan, 2002). A broad theme in that issue was “drivers of customer attraction and retention” (p. 287), yet this may be the first communication study addressing a simple tool adopted by about nineteen out of twenty Internet users for returning to a website (Abrams et al., 1998).
The results, although exploratory, are intriguing. Hypothesis 1 drew upon economic research to claim that increased page views would lead to increased bookmarking (Adar & Huberman, 2000; Goldfarb, 2002), which happened. This relationship might also work in exactly the opposite manner. If a site is too difficult to navigate and visitors lose interest (Hoffman & Novak, 1996; Nielsen, 2000), perhaps users would not bookmark it and stay away in droves. While this failed to happen with these two sites, such a phenomenon could exist.
Hypothesis 2 left us with the greatest mystery. We predicted heavier bookmarking activity during weekends than weekdays. The results from the hospitality sites confirmed this hypothesis but not the results from the genealogy site. Figures 5 and 6 show a subtle difference in the timing of the usage of these sites. There are at least two possible explanations. Individuals might access these sites at different times of the day, or it might be that individuals access the sites — most likely the genealogy site — from outside of the site's time zone. In any case, these descriptive online data suggest time-related differences in browsing and bookmarking particular sites.
This study argues that bookmarking activity on a website reflects an early stage of relationship building and e-loyalty, thereby adding another metric of online relationship marketing. This study also demonstrates a methodology for measuring consumer bookmarking on a site and supports relationship-building by showing that the more pages a consumer visits, the more likely the probability of bookmarking the website. Finally, this exploratory paper offers a future research agenda for this electronic customer relationship tool.
As previously noted, our methodology utilized the Web server to track bookmarking on a site. A complementary approach would be to survey consumers in order to understand what drives customers to use bookmarks on a particular site or category of sites. Jones and colleagues (2002) for example, found differences in bookmarking based upon one's job function. Demographic, technographic and psychographic characteristics may also relate to how individuals tend to bookmark websites, organize their bookmarks and use their bookmarks. Individual differences in cognitive problem solving (Alba & Hutchinson, 1987) also merit investigating.
On a macro scale, visitors from different domains may be more or less likely to bookmark. Using global domains, are .edu visitors or .org visitors more or less likely to bookmark than .com visitors? Research has found cultural differences in website content (Zhao et al., 2003) as well as how consumers use and perceive websites (Chua et al., 2002). Using country domains, are visitors from Western cultures such as .au for Australians or .uk for Britons less or more likely to bookmark than visitors from Eastern cultures such as .sg for Singaporeans or .cn for Chinese? Among other things, such an investigation might shed additional light on temporal differences in site log data.
Domain names may also play a role in bookmarking behavior. Research suggests that websites with shorter domain names tend to survive longer than websites with longer domain names (McMillan, 2002). Yet given the difficulty remembering and typing long addresses (Hanson, 2000; Ries & Ries, 2000), do visitors tend to bookmark websites with longer domain names?
Finally, this study focused on a particular set of sites. Does consumer bookmarking behavior vary across different sites and different site categories? The motivation to use some sites is relatively hedonic or ritualistic, while others visit a site for instrumental purposes to achieve a particular goal (Hoffman & Novak, 1996; Holland & Baker, 2001; Novak, Hoffman, & Duhachek, 2003). Future research should compare bookmarking for these two motivations to visit a site.