Published: Apr 25, 2003
A few years back, we conducted one of the most painful usability studies in the history of our research. We learned some really important things, but I'm not sure the users in that study will ever forgive us.
Before that particular study, we'd noticed, when searching large web sites for information, there were some sites where users always seemed to know where to find the content. No matter what content they were seeking, every user somehow knew to make a bee-line for it. Not every site worked this way and we wanted to know what made these particular sites work so well.
We had suspected that the secret was in the links. On those sites where users consistently found their target content (the content they were seeking), we had gut feel those links were helping users more than the sites where users rarely found the target content. Unfortunately, we couldn't explain why.
For example, when we put the links side-by-side, mixing the links from successful sites with those that were not helping users, we couldn't pick out the ones that were more effective. There was something that was special about those links, but we couldn't identify the magical traits. We studied the links hard, but whatever it was just kept eluding us.
That's when we decided to conduct the Lincoln study. We named it 'Lincoln' because we were studying 'Links'. (I know -- it's a dumb name. But you have to name the project something and this name just sort of stuck.)
In the Lincoln study, we looked at sites where users frequently found their target content and sites where they didn't. We wanted to know what the difference was between the links, so we compiled two questionnaires to identify key attributes for each link.
We asked each user to fill out one questionnaire before each click and another one after each click. That meant if a user clicked on 15 links to find their target content, they filled out 30 questionnaires. Fortunately, we pay our users extremely well.
Going into the study, our hypothesis was that the better sites were somehow 'telegraphing' the path to the content and users somehow knew what each link was going to bring. To test this, part of the questionnaires asked users to predict what they thought the next page would contain. On the better sites, we expected users would always know what came next.
When we compiled the results, we found users weren't any more likely to know what content was on the next page. Our theory about telegraphing was a dead-end. That wasn't the secret to good sites.
In fact, users always assumed the next page contained their target content -- no matter where they were in the site. They could be on the home page, clicking on a generic link like "Sports" or "Research" and still think the next page was going to answer their very specific question.
In our analysis of the data, we did isolate a factor we didn't expect. With each click, users told us they were more confident they would succeed on those sites where they actually did succeed. Somehow, they were predicting their success.
We measured each user's confidence with two questions. Before they clicked, we asked "Do you think clicking on this link will lead you to the info you seek?" with a 7-point scale that had the endpoints marked as "Not at all" and "Extremely Likely". After they clicked and had a quick chance to inspect the result page, we asked "Do you think this page is getting you closer to your goal?" with the same 7-point scale.
We were amazed when we discovered the answers from the first three clicks strongly predicted whether the user would eventually succeed or fail, even if the clickstream was 15 or 20 clicks long. Not only that, but as long as every subsequent click had high confidence values, the user was very likely to succeed. As soon as the confidence values dropped, so did the likelihood of the users finding their desired content.
This was the clue we needed -- the key to our research. Once we could see when the user's confidence rose and fell, we could analyze the links and determine what was contributing.
Some very clear link attributes immediately jumped out at us. First, users expect to find 'trigger words' in the links. A trigger word is a word (or phrase) that causes the user to click. When the trigger words match the user's goals, they find those words right away and the links make them more confident that they are going to find their content.
You can find a good example of trigger words on the Edmunds.com home page. People who are just starting the process of selecting a new car are likely to click on the word "New" or the phrase "Find a New Car". If they know they want a sedan or SUV, they are likely to click on one of those trigger words. If they are specifically looking for pricing of a model, with options, they are likely to choose the "Price with Options" link.
The fascinating thing is that all those links go basically to the same part of the site. The designers made sure the trigger words are all out on the surface, where users can see them.
Another finding from the Lincoln study was users expected the site to become more specific with each click. As users move through the site, they want each subsequent page to have more detail related to their goal than the page before.
If all of a sudden a page is about a general topic, the users lose confidence. For example, when we were testing the Boston.com's Red Sox page, users lost confidence when the Sports Calendar link didn't produce a schedule of the Red Sox games. Instead that link brought them to a listing of all sports activities (including paintball, sky-diving, and frisbee) in the greater Boston area. Since the users were already on the Red Sox page, they naturally assumed that any link from that page would be even more details about the Red Sox.
Because we could now use the user's confidence to tell us how well the links were working, we could start to identify other patterns. We saw that the links on many global navigation panels were sorely lacking. (What is the difference between "Products" and " Solutions"?) We could tell when graphics were helping and when they weren't. We could see when pogosticking was causing problems.
We now call the magical force that pulls users to their content the Scent of Information. (We didn't come up with the name. We heard it from Peter Pirolli and his team at PARC. But, like project names, good concept names just catch on and stick.)
We can't measure when a link has good scent, but we can measure when it gives off confidence. By looking at the confidence of the user as they move through the site, we can tell what parts are working well and what parts need rethinking.
You can measure confidence, too. As you're watching people use your site, just trying asking those two questions. Pretty soon, you'll know when users are feeling confident and when they aren't. That will give you a good sense as to where to focus your design efforts. •
Read related articles: