Effects of Link Arrangements on Search Efficiency
1.1 Motivation and Experimental Design Issues
The idea for this study came about while using the United States Naval Academy's web pages. It was found that the web pages were not well organized. The academy home page was too cluttered with links, making it difficult to quickly or efficiently find information. The broad purpose of our study was to discover a better method of presenting a large number (>30) of links on the home page. There were two options to pursue: decreasing the number of links on the front page while increasing the depth of links, or more effectively organizing the text using columns.
Decreasing the number of links on the homepage would have made the links easier to find since there would have been less information for the eye to scan. However, research has shown that breadth is preferred to depth on web pages. In other words, it is better to have many links on one page and a few links under that hierarchy (a broad design) than it is to have only a few links on the front page and many sub-links under that (a deep design). It is also shown to be faster for search purposes. Shneiderman (1996) writes "A higher branching factor is almost always preferred for index pages". While it may be easier to scan for information on deep designs, it is often very time consuming traversing the levels of links. Shneiderman (1996) further states that menu response times, display rates and screen size should be taken into consideration. Since we did not want to make the site any deeper than necessary, arranging the links in columns became our primary focus for improving the web page design.
The next web page design issue we had to address was scrolling. Guidelines for web page design indicate that, ideally, a minimal amount of scrolling should be necessary in order to find information on the page. Scrolling is time consuming since the user must first scan all the information currently on the screen, then move their hand to the mouse, find the appropriate controls on the scroll bar, then drag the controls. Scrolling becomes a problem in the case of our experiment because as columns are introduced, screen density decreases. When screen density decreases, less text fits in a fixed area of the screen. All of the text on the academy home page just barely fits into the area of a normal browser screen, so by decreasing the density to any degree, scrolling becomes necessary.
In designing our experiment, we did not want scrolling to skew our results. If we did not control scrolling, we would not know whether users took longer to find information because of the scrolling time or because of the column treatments. We controlled the scrolling by creating two Web pages of varying lengths for the independent variables (30- and 155-link arrangements). For the treatments, we created one-, two- and three-column formats. For the 30-link Web pages, the text fits onto a single screen (no scrolling required), while for the 155-links the text did not fit onto a single screen (scrolling required).
This 2 x 3 design yielded six experimental web page designs:
Each participant was presented with a predetermined web page which looked like this. The subject randomly selected a task list for the three task links to search for, marked the start time using a digital JAVA clock (located at the top of the page) and clicked the START button. The START button was an internal hyperlink to the lower section of the same page. The participant then searched for the first link (task) on the page. Upon finding the correct link, the participant clicked on it. The internal link then returned the participant to the START section of the page. The stop time was then recorded using the JAVA clock. Participants repeated the procedure two more times and then completed a subjective satisfaction questionnaire.
1.2 Review of Previous Research
There has been much research in the area of textual displays, especially regarding the layout of information displays such as telephone directories and control panels. However, no study has examined the effect of columns on a user's ability to search for links. A summary of previous research which is relevant follows.
Staggers (1993) studied the effect of screen density on the ability of nurses to scan for medical information on a text layout. The results indicated that increases in screen density resulted in faster performance times without reducing accuracy or subjective satisfaction. This is relevant to our study because we are asking users to scan a text layout for a link, much like the nurses were scanning for information.
Nygren (1996) studied the time it took subjects to find links on a web page. The study gave several suggestions for arranging links, such as having an ordering scheme, unique link names, and giving links different lengths. While it did not focus on column designs, there were a few findings which are relevant to this project. First, it was discovered that scanning a horizontal list of links is significantly slower than scanning vertically arranged links. Another finding was that scanning a single long vertical list is faster than scanning multiple shorter vertical lists. Long vertical lists are analogous to single-column displays and "multiple shorter vertical lists" are similar to the two and three column displays used in this experiment. However, there was not enough detail in Nygren's paper to indicate how many vertical lists were used. If there were only two or three, Nygren's experiment is very relevant. If there were several (such as five or six), the relevancy decreases.
Textual displays have been studied since before web pages existed. Tullis (1988) reviewed research and compiled four metrics for alphanumeric displays: overall density, local density, grouping, and layout complexity. Overall density deals with how much of the screen is filled up with characters. Local density measures how close the characters are to one another. The number of related or "connected" characters, such as groups of characters which form words, is indicated by the grouping metric. Layout complexity deals with the distribution of horizontal and vertical distances from a fixed point on the display. Tullis goes on to say that lower local and overall densities will tend to result in displays that are easier to use. In this experiment, overall density decreases with the addition of columns and local density increases with the incorporation of columns into the display.
Continue 2.
Experiment
Title Page for
Effects of Link Arrangement on Search Efficiency