“Experience … the fact or state of having been affected by or gained knowledge through direct observation or participation” (Merriam-Webster, 2003).

Most researchers studying online behavior use "experience" level as a way to categorize individual differences in using the Internet.  For the majority of these studies, however, experience level has been defined by time and/or frequency measurements, such as “How long have you been using the internet?”; “How much time do you spend online?”; or “How often do you use the Internet?” Two recent longitudinal studies, which report the changing online population in the United States (Cole, Suman, Schramm, Lunn & Aquino, 2003; Lenhart, Horrigan, Rainie, Allen, Boyce, Madden, & O’Grady, 2003), show that the amount of internet usage is leveling off, but that the breadth of demographic user characteristics are continuing to expand. Therefore, measuring experience level by frequency of use and longevity of use may not be truly representative. Although measures of time and frequency typically show a high correlation, the online user’s experience level cannot be adequately evaluated by considering it as a single dimension. A more descriptive measure of experience might be “how” users experience the Internet rather than “how often.”

People use the Internet for many different purposes. Computer game enthusiasts compete online and communicate with friends using Instant Messaging. Stock brokers research and monitor online stock prices and trading. Many of us pay bills and bank online, communicate via e-mail, and browse/shop online. It is also possible that each of these users could report spending roughly the same amount of time online (e.g., 4 hours a day). By most standards, all of these users would be considered “experienced” or “expert” online users. However, their specific online activities, their knowledge of computers and the Internet, and how they feel about being online could differ significantly. In order to obtain a more robust measure, researchers may want to consider how and what users think, feel, and do online when they assess level of experience. In this article we explore each of these areas and present a new measure of measuring online experience.

Three components of experience were considered in the development of the Internet Experience Scale: 1) Conative (what users “do” online) by measuring the user’s time, frequency, and activities online; 2) Cognitive (what the user “thinks” online) by measuring the user’s technical knowledge of computers and the Internet, or their “online thinking”; and 3) Affect (what the user “feels” online) by measuring their feelings and attitudes while online. A pool of questions to comprehensively assess each of these areas was collected, along with descriptive demographic and environmental queries. The questionnaire was administered to 168 graduate and undergraduate business and psychology students at Wichita State University. The following is a partial analysis of the results looking at affect and activities in the first iteration.

Our sample population consisted of 168 students – 93% full time, with 81% in undergraduate and 19% in post-graduate programs.  The mean age was 24 years, with a range of 18 to 55. The gender of the sample was 35% male and 65% female. Ninety-three percent of the sample reported accessing the Internet from their home. Forty-six percent reported using dial-up for their Internet connection, while 43% used DSL/Cable; and 3% did not know.

Conative Measures

When queried about their online usage and activities, 91% of participants reported being online that day or the previous day. Participants were asked to respond to a list of 16 various online activities on a Likert-type 5-point scale (with anchors of "hourly" and "never"), as well as time online, frequency online, and online longevity queries. Ninety-seven percent of participants reported they were online for less than four hours in their last online session. Figure 1 shows the typical daily usage. Ninety-eight percent reported being online at least a few times per week, and over 65% have been online over 5 years.

Factor Analysis of Internet Activity Variables

Principal factors extraction with Promax rotation was performed using SPSS on the 16 activity items for the sample of 168.  Results revealed three main factors (only activities loading .5 or above were retained). The results, including the interpreted factor labels, are listed below.

Table 1. Factor Analysis of Internet Activity Variables

Internet Activity Scale Reliability Analysis

In order to assess the internal consistency of the activity scale using the 13 activity variables from the factor analysis, the data were analyzed using scale reliability analysis in SPSS. A Chronbach’s Alpha of .74 was calculated reflecting adequate internal consistency of the items.

Reported Internet Activity Levels

This 13-item activity scale was sum-scored and used to divide the total pool of subjects into three groups – high activity, moderate activity, and low activity. Activity level is a score reflecting the number of different activities and the frequencies with which the participants engaged in them. Analysis found the mean score on the scale to be 33.7, within a range of scores from 16 to 58. Of the 168 cases, 60 (35.7%) were categorized as high level of activity (range 37-58), 52 (31%) moderate level of activity (range 31-36), and 56 (33.3%) were grouped as low level of activity (range 16 – 30).

Affective Measures

Participants were asked to respond to a list of 25 questions regarding their feelings about being online on a Likert-type 6-point scale (with anchors of "strongly disagree" and "strongly agree").  Questions were adapted from validated online surveys (Hoffman & Novak, 1997; GVU, 1997), as well as original queries.

Factor Analysis of Internet Affect Variables

Principal factors extraction with Promax rotation was performed using SPSS on the 25 affect items for the sample of 168. Results revealed four main factors (only activities loading .5 or above were retained.) The results, including the interpretative factor labels, are listed below.

Table 2. Factor Analysis of Internet Affect Variables

Internet Affective Scale Reliability Analysis

In order to assess the internal consistency of the revised affect scale, the data in this sample were analyzed using scale reliability analysis with SPSS. A Chronbach’s Alpha of .84 was calculated, reflecting good internal consistency of the items.

Discriminant Function Analysis

To determine if the users’ online activity level can be predicted by online affective measures, a direct Discriminant Function Analysis (DFA) was performed using the four factor scores calculated on the 14 affect variables as predictors of membership in the upper and lower (high and low) activity levels. To differentiate activity level more completely, the moderate activity level was not included in the analysis.

A DFA was performed on 115 cases, 59 classified as high activity level and 56 as low activity level (see Conative measures). One canonical function was used, and produced an eigenvalue of .452, canonical correlation .558. A Wilks’ Lambda of .689 was produced, indicating that there was significant discrimination between the two groups of activity levels. The four affect factors correctly classified 75.7% of the original grouped cases.

DISCUSSION

Although there is a need to sample a more diverse audience to improve the generalizability of this measurement scale, the university population does offer a representative sample of the demographic profile of the Internet user. The most recent PEW survey of Internet usage (Lenhart et al., 2003) lists the Internet user as: younger, employed, white, well-educated, suburban/urban, and parents of children living at home. This sample tapped into at least four of those demographic descriptors. A second iteration of this scale will broaden the sample to older adults, parents, and more affluent audiences.

Our initial analysis suggests that what people are doing online (activities) and how they feel (affect) while online is a more robust measure of their online experience level than simply the time and frequency of going online. Revisions of this instrument will allow us to more fully understand the conative and affective measures, as well as develop a cognitive measure to assess experience.

Cole, J. I., Suman, M., Schramm, P., Lunn, R. & Aquino, J. (2003, February).  Surveying the digital future – Year three.  Los Angeles, CA:  University of California Los Angeles, UCLA Center for Communication Policy.  Retrieved May 31, 2003, from http://www.ccp.ucla.edu/

"experience." Merriam-Webster Online Dictionary. 2003. Retrieved June 15, 2003, from http://www.merriam-webster.com/

GVU (1997).  8th WWW User Survey, Graphic, Visualization, and Usability Center (GVU) at the Georgia Institute of Technology.  Available from http://elab.vanderbilt.edu/research/data/gvusurvey/project2000.gvu8.html

Lenhart, M., Horrigan, J., Rainie, L., Allen, K., Boyce, A., Madden, M. & O’Grady, E. (2003, April).  The ever-shifting internet population:  A new look at internet access and the digital divide.  Washington, DC:  The PEW Internet & American Life Project.  Retrieved May 30, 2003, from http://www.pewinternet.org/

Novak, T. & Hoffman, D. (1997).  Measuring the flow experience among web users.  Presented at Interval Research Corporation on July 31, 1997.  Retrieved June 1, 2003, from http://elab.vanderbilt.edu/research/manuscripts/index.htm

Fonts are described in terms of their face, style, size and color. There are two main types, namely, serif and sans serif. Serif fonts have small appendages at the top and bottom of the letter. Serif fonts are the preferred fonts for large blocks of text, since the serifs are thought to help to distinguish each letter and thus, make it easier to read strings of characters. Sans serif fonts consist of only primary line strokes and are therefore simpler in shape, e.g. Arial and Futura. In standard typography these fonts are used primarily for short phrases, e.g. headings. This study compares reading performance between an ornate sans serif font (Gigi) and Times New Roman. The traditional measures of reading speed, comprehensibility, and subjective preference were employed.

A number of studies have considered font styles. For example, Tullis, Boynton and Hersch (1995) examined differences in reading rates for different font styles and sizes in a proof reading task carried out in a Microsoft Windows environment. Participants used Arial, MS sans serif, and MS serif at 6, 7, 8, 9 and 10-pt font sizes. Tullis et al. found no difference in reading speed between the serif and sans serif fonts; however, they found that the larger 9 and 10 point fonts elicited faster reading times. The study also found that the participants had a greater preference for the sans serif font compared to the serif font.

A study by Boyarski, Neuwirth, Forlizzi and Regli (1998) evaluated the reading speed of participants using the serif Georgia, Times New Roman, and the sans serif Verdana fonts. The fonts were all set at 10-point and the experiment involved participants completing a comprehension test (i.e. the Tinker Reading Speed test). No significant differences in reading speed were found between the fonts. However, it should be noted that the Georgia and Verdana fonts were specifically designed for on-screen reading so this may have influenced the results.

Bernard, Mills, Peterson and Storrer (2001) tested a range of fonts for effective reading speed, i.e. reading speed in conjunction with accuracy, and participants’ perception of font legibility. Twelve fonts representing sans serif, serif, and ornate styles were studied. Differences were found for reading time with Tahoma (sans serif font) being read significantly faster than Corsiva (ornate font). Perceived font legibility also showed significant differences across the 12 fonts with sans serif fonts being more legible than the ornate fonts.

A further experiment by Bernard and colleagues compared four sans serif fonts (Arial, Comic, Tahoma and Verdana), and four serif fonts (Courier New, Georgia, Century School Book and Times New Roman). They found no difference in effective reading speed between the two font types (Bernard, Lida, Riley, Hackler & Janzen, 2002). However, significant differences were found for reading times of the fonts with the serif fonts producing quicker reading times. The experimenters found that the participants perceived a difference in legibility between the fonts of which Times New Roman, Verdana and Georgia were most legible. This is interesting, as the participants did not distinguish between serif and sans serif in terms of perceived legibility. Further, Bernard et al. found that the participants associated specific personalities to the different fonts. The results showed that the ornate sans serif fonts, Bradley and Corsiva, were perceived as having a great deal of personality and being elegant, whereas Times New Roman was perceived as being "business-like." In terms of overall preference, Bernard et al. found that participants chose the sans serif fonts Arial, Verdana and Comic.

The prediction for the current study based on the evidence provided from previous research was that the serif font would be faster to read, easier to comprehend and preferred to the ornate sans serif font. It was also predicted that the sans serif font would be preferred in terms of attractiveness and overall characteristics.

Participants

Twenty-five participants (13 males and 12 females) volunteered for this study. The mean age of the participants was 27.9 years with an age range from 19.7 to 63.7 years. All participants were familiar with reading from a computer screen, and had normal or corrected to normal vision.

Task Design

Font conditions were compared by having participants read through four paragraphs of 140 words presented in the two different fonts. The paragraphs were matched for level and topic of reading as both were taken from an introductory psychology textbook. Presentation conditions were counterbalanced across participants. Passages were presented on a Winbox XLI Pentium 2 laptop computer, with a thin film transistor 13.3” screen and a resolution setting of 1024 x 768 pixels.

There were 10 substitution words used in each paragraph for testing the readability of the different fonts. The substitution words were inappropriate to the context of the passage and varied grammatically from the original words in the paragraph, e.g. the word "male" was substituted with the word "dale." The dependent variables were the time taken in seconds to read through the paragraphs and the percentage of substituted words correctly identified.

Procedure

Text was displayed in the center of the screen: the size of both the Times New Roman and the Gigi font was 12-point. Both fonts were black presented on a white background. A stopwatch was used to record the time participants took to read the paragraphs and the experimenter also noted the number of correctly identified substitution words.

Participants were given a few minutes to read through the instructions and then were given the opportunity to ask questions. The participants were seated at a fixed position from the computer screen at a distance of approximately 60 centimeters away. They were presented with the first paragraph on the screen and were asked to read through it silently and as quickly and accurately as possible. Participants indicated they had finished reading the paragraph by saying "stop" and the time elapsed was recorded. After a break of one minute, the second paragraph appeared on the screen. Again, participants had to read through the paragraph as before. However, if they came across words, which appeared inappropriate in the context of the paragraph, they were asked to read these words aloud and provide the correctly identified substitution words. These were recorded by the experimenter.

After the participants had completed the computer-based experiment, they completed a questionnaire to assess which font they would prefer in terms of personal use, web use, attractiveness and overall preference.

RESULTS

The reading time for the serif font Times New Roman was faster (M= 22.06 seconds, SD = 5.69) than for the ornate sans serif font Gigi (M = 26.99 seconds, SD = 6.64). This difference was found to be significant, t = -7.41, df = 24, p < 0.01. It is noted that the fast reading times were due to the relatively short lengths of the passages, and the fact that the readers were mainly university students and therefore used to reading text quickly.

Comprehensibility was also better for the serif font Times New Roman (M = 8.80, SD = 1.19) than the ornate sans serif font Gigi (M = 6.68, SD = 2.02). This difference was also found to be significant, t = 5.58, df = 24, p < 0.01.

Results from the questionnaire showed a preference of 100% for Times New Roman for personal use and a preference of 68% for Times New Roman for web use. The ornate sans serif font Gigi was rated as being more attractive by 80% of the participants. Overall preference for Times New Roman was also greater with a mean likeability score of 3, compared to Gigi with a score of 2.6 on the same 4-point Likert scale (See Figure 1).

The findings from this study support the findings of Bernard et al. (2001) that serif fonts can be read faster than ornate sans serif fonts. However, the faster reading times for serif fonts do not support the findings of Boyarski et al. (1998) that reading times do not vary between serif and sans serif fonts. However, Boyarski et al. did not use an ornate sans serif font in their study.

With reference to comprehension, one previous experiment found that serif fonts did promote greater reading comprehension than sans serif fonts. This study provides support for this finding to be extended to include the ornate sans serif fonts as well. The results from this experiment also provide contradictory evidence to Bernard et al. that sans serif fonts such as Arial are preferred to Times New Roman and other serif fonts. This experiment found that the serif font Times New Roman was preferred overall and in terms of personal usability and web usability to the ornate sans serif font.

Tullis et al. (1995) found from their experiments that there were distinct differences in the speed and accuracy with which users could read the various fonts; however, they found even stronger differences in people’s subjective preferences for a particular type of font. This was also replicated in this experiment as all the participants favored the use of the serif font for their personal use and the majority would also prefer to see the serif font used on a web site. However, the majority of the participants also strongly felt that the sans serif font Gigi was the more attractive of the two fonts. This is an important factor to consider when selecting fonts for on-screen text or for a web page as people may rate attractiveness over likeability and this may affect the extent to which a web site is used or favored over others.

As a final comment, the usability of online fonts for reading text includes factors other than font style. Factors which could affect the readability of on-screen text include the spacing between the words, the line length on the screen, the amount of white space, the use of italics, underlining and boldness. The actual characteristics of the user of the computer must also be taken into consideration: for example, the age, sex, computer experience, background and personal preferences. Hence, font selection should not be considered in isolation. A further factor concerns the origins of font styles. Most fonts being read on computer monitors today were designed to be read from paper (Boyarski et al., 1998). Perhaps further development of fonts created specifically for on-screen reading is necessary in order to find out what would be the optimum online font.

Bernard, M.L., Lida, B., Riley, S., Hackler, T., & Janzen, K. (2002). A comparison of popular online fonts: Which size and type is best? Usability News 4.4 [Online]
../usabilitynews/41/onlinetext.htm

Bernard, M.L., Mills, M.M., Peterson, M., & Storrer, K. (2001). A comparison of popular online fonts: Which is best and when? Usability News 3.2 [Online]
../usabilitynews/3S/font.htm

Boyarski, D., Neuwirth, C., Forlizzi, J., & Regli, S.H. (1998). A study of fonts designed for screen display. In Proceedings of CHI’98 (pp. 87-94). Los Angeles, CA: ACM Press.

Tullis, T.S., Boynton, J.L., & Hersch, H. (1995). Readability of fonts in the windows environment. In Proceedings of CHI’95 (pp. 127-128). Denver, CO: ACM Press.

In our last issue of Usability News, we reported on the general usage of breadcrumb trails as a method of navigation on web sites (Lida, Hull & Pilcher, 2003).  The term “breadcrumb” derives its name from the Grimm’s fairy tale, Hansel and Gretel.  Hansel left a trail of breadcrumbs through the woods as a strategy to find his way back home. Since today’s internet user often has a need to navigate back through a website path, the cyber-version “breadcrumb trail” was named¹.  

There are three different types of breadcrumbs represented in websites – path, attribute, and location (Instone, 2003).  Path breadcrumb trails are dynamic in that any given page will show a different breadcrumb trail based on how the user reached the page.  Attribute breadcrumb trails display meta information showing many different trails representing several possible paths to reach the page. The location breadcrumb trail is a textual representation of a site’s structure, e.g. Home > Furniture > Chairs > Leather Chairs.  This representation of information allows users to link to major categories of information along a continuum of sequential order.  Regardless of how users arrive at Leather Chairs, the breadcrumb trail displayed is the same.  This study investigated the use of “location” breadcrumbs.

In general, the breadcrumb trail serves two purposes: 1) it provides information to users as to where they are located within the site, and 2) it offers shortcut links for users to “jump” to previously viewed pages without using the Back button, other navigation bars, or typing in a keyword search. Breadcrumb trails give location information and links in a backward linear manner; whereas, navigation methods, such as search fields or horizontal/vertical navigation bars, serve to retrieve information for the user in a forward-seeking approach. As suggested by Marchionini (1995), systems that support navigation by both browsing and analytical strategies are most beneficial to users since tactics associated with both types of strategies are normally used. According to Steven Krug (2000), breadcrumb trails are most valuable as an accessory to a site’s navigational scheme and are optimally located at the top of a web page in a smaller font. 

There has been speculation that a breadcrumb trail also aids the user’s “mental model” of the site’s layout to reduce disorientation within the site (Bernard, 2003); however, we have not found research to validate this assumption. It would seem logical, however, that a constant visualization of the path to the user’s current location would increase their awareness and knowledge of the site structure.  Toms (2000) suggests that users need both a stable orienting device, such as a menu, to facilitate pathways through the site, as well as a system that supports scanning to smooth the progress of the search.  Research has reported that breadcrumb navigation improves measures of site efficiency (Maldonado & Resnick, 2002; Bowler, Ng & Schwartz, 2001). Our earlier study, however, found limited use of breadcrumb trails as a navigational tool and no differences in site efficiency for two online sites, OfficeMax and Google Directory (Lida, et al. 2003). 

The purpose of this study is to further investigate breadcrumb usage by evaluating the following research questions: 

1. Do users choose to use breadcrumbs as a navigational tool?

2. Does breadcrumb usage improve navigational efficiency?

3. Does the location of the breadcrumb trail on a page effect usage?

4. Does a breadcrumb trail aid the user’s mental model of the site structure?

Participants

Forty-five participants (20 male, 25 female) with an average age of 27 (range of 18 to 64) volunteered for the usability study. The sample consisted of 60% Caucasian, 11% African American, 7% Hispanic, 20% Asian/Pacific Islander, and 2% Native American.  All participants were familiar with the web – 98% reported accessing the Internet at least once a week.

Materials/Procedure

In order to evaluate these differences, we constructed a site for gardening tools and products, The Garden Company, by adding content to a sample site in the Microsoft© Frontpage tutorial.  The site structure is shown in Figure 1.

Participants were asked to complete 21 search tasks (e.g., What is a flowering tree that tolerates wet soil?) on The Garden Company site and record their answers. The tasks were developed to traverse each of the four levels of the site and structured so that efficiency was optimized through the use of the breadcrumb trail (see Figure 2).

Three variations of the site were created, each with identical content and structure, but different in terms of the presence and position of a breadcrumb trail. The first variation displayed a breadcrumb trail at the top of the page (Figure 3); the second variation displayed a breadcrumb path under the page title, approximately 30% from the top of the page (Figure 4); and the third variation had no breadcrumb path. Other options for site navigation included a left side navigation bar (Home, Products, Class Offerings, FAQs, Specials, and Contact Us), Back button, and a breadcrumb trail for the first and second variation.  Participants were randomly assigned to one of the three sites and were given approximately 30 minutes to search for the information.  Time, mouse clicks, user satisfaction, and user “mental model” of the site was collected.

Navigational efficiency was measured by the total number of pages. Methods of page navigation such as Back clicks, Forward clicks, breadcrumb clicks, navigation bar clicks, embedded link clicks were also collected. This data was gathered by the tracking program Ergobrowser™.  Pentium 4-based personal computers, with a 60 Hz, 96dpi 17" monitor with a resolution setting of 1024 x 768 pixels on a campus network were used to access the sites.

After completing the tasks on each site, satisfaction with the site was measured by the Satisfaction User Survey (SUS) instrument, which was adapted for web usage and consisted of 10 satisfaction questions using a 1-5 Likert scale (with anchors of "Strongly disagree" and "Strongly agree"). (Brooke, 1986). 

To assess the mental model of the site, participants were asked to choose a model (Figure 5) from a selection of four graphical representations or to draw their own representation. Demographic and usage information was also collected from participants via a background questionnaire.

Of the participants that were exposed to a site with a breadcrumb trail (n=30), 40% used the breadcrumb five or more times to navigate on the site (Range = 5 - 31, n=14). However, this accounted for only 6% of the navigation overall (see Figure 6). The Back button, the main Navigation bar, and links embedded in the text content were used the majority of the time. Figure 7 shows the navigation used by the participants using the no-breadcrumb site. Without the presence of the breadcrumb, it appears as though the use of the Navigation bar, embedded links, and the Back button are relatively equal. 

As shown in Table 1, breadcrumb usage did vary based on the location on the page. There were 199 total breadcrumb clicks by the 30 participants using one of the two breadcrumb sites; of those, 82% (163) were in the site with the breadcrumb positioned under the page title (See Figure 8), and more participants clicked on this breadcrumb (under title n=10, top of page n=4 ), χ²(1, N = 30) = 4.82, p = .03.

In this study, we designed the tasks such that navigational efficiency would be optimized through the use of a breadcrumb trail. Despite this, only 6% of the page clicks were accounted for by the breadcrumb. While 40% of the participants used the breadcrumb trail, usage was lower than that of other navigational means, such as the main navigation bar, the Back button, and embedded links.

Breadcrumb users were found to use the Back button less often than users who did not use the breadcrumb; however, no differences were found in the efficiency measures of total pages visited, navigation bar clicks, embedded link clicks, or time to complete the search tasks. It is not known if all participants understood the function of the breadcrumb as a navigational tool. Future studies should investigate whether a simple understanding of the purpose of the breadcrumb trail or minimal training impacts usage and/or efficiency.

Location of the breadcrumb trail did have an effect on usage. Breadcrumb trails positioned under the page title (at eye level and closer to other links on the page) were used more than breadcrumb trails positioned at the top of the page. It is recommended, therefore, that breadcrumb trails be positioned in this location rather than at the top of the page. The results also suggest that exposure to a breadcrumb trail in a site may contribute to the type of site model formed by the user. Participants that used a site with a breadcrumb trail were more likely to choose a hierarchical model than those that used the non-breadcrumb site. This assessment of the user’s mental model requires further study.

¹ One could argue that the cyber-version "breadcrumb" name is a bit misleading in that the trail shown is not necessarily the exact path taken by the user.