A free comprehensive guide for evaluating site structures
The first thing we want to know is how the tree did in general:
- Did it perform well, and all we have to do is tweak a few things?
- Did it perform just OK, and some parts need reorganizing?
- Did it perform poorly, and we need to discard it and look at other options?
Overall success rate
Not surprisingly, the most important thing to look at is success rate – how many participants chose the correct answer, across all tasks?
Most tools will give us this as a rating out of 10 or 100. For example, a score of 69 means that 69% of the time, participants chose a correct answer:
Once we see a tree’s overall success rate, the natural question is “Well, is that good, bad, or just average?”
As any consultant will tell you, it depends. Mainly, it depends on two things:
- Size of the tree – All other things being equal, it’s harder to find things in a larger tree (or haystack, as the saying goes).
- Complexity for the intended user – If the topics and subtopics in the tree are challenging for participants to understand, they’re going to have a tougher time finding the right answers.
But we do need to start from somewhere. In our experience, over hundreds of tree tests, the following rough markers have emerged for trees of average size and complexity:
- 0-50 – The tree needs to be completely rethought or discarded.
Trying to tweak it will only bring it up to “mediocre”.
- 50-65 – The tree needs substantial revisions.
If our analysis reveals specific problems (and it should), and we think we can fix them, we should be able to revise this tree to perform well.
- 65+ - The tree is effective, but may need minor revisions.
Our participants are finding the correct answer at least two-thirds of the time, so the tree is doing its job well, and only needs tweaking.
What the overall success rate doesn't tell us is how much the success rate of the individual tasks varied. For example, a 60% overall score may mean that all tasks hovered around 60%, or that half our tasks were 90% and half were 30%. To find out, we need a breakdown by task, which some tools summarize in a graph like this:
In this example, we can see that a few tasks had very low success rates, and two were very high. To find out more, we need to drill down to the task level - see Task success rate later in this chapter.
Comparing tree-test scores to usability-test scores
People are often surprised that we consider 65+ to be a "good" score. Shouldn't we set the bar at 80 or 90?
Effective trees don't usually score higher than 80 because we're testing a top-down text-only tree with no other aids. Our participants are making choices without the benefit of:
- other navigation aids such as see-also links, featured links, and multi-level browsing (e.g. mega menus)
- visual design - chunking of links/content, and emphasis on more important links/content
- content that explains headings using decoration text, hover text, etc.
Once we refine our text tree to be effective (i.e. perform well in tree testing), we should then be able to add these other design elements to further improve the findability of items in our website.
Lisa Fast at Neo Insight has written an interesting article comparing tree-test scores to usability-test scores. Here's the graph of how they related in her study:
Lisa found that not only were the two scores correlated, the usability-test scores were 29% higher (on average) than the tree-test scores.
Finally, we should warn that adding other aids is no guarantee of improvement. A poor visual design, clumsy navigation, and confused content can actually make a website perform worse in usability testing than it did in tree testing. A single method can only go so far.
Overall directness (backtracking)
To get a general idea of the effectiveness of our tree, it also helps to look at how directly our participants found the right answer. Did they go straight there, or did they have to try a few different paths first?
How this is scored depends on the tool we’re using:
- Some tools treat directness as a simple yes/no measure – did the participant backtrack at all during a task? This method doesn’t care if they backtracked 1 time or 5 times during the task – it’s either yes or no.
In our experience, 70% is an average score for this method. Less than that indicates that users are having trouble finding the right path.
- Some tools try to quantify how much wandering a participant did. A single back step lowers their score a bit, but repeated meanderings through the tree lowers it much more.
~guidelines for scores using this method?
While the overall directness score gives us a rough idea of how clear and distinguishable our headings are, we’ll need to drill down to specific tasks to determine where the most backtracking happens. For more on this, see Directness – where they backtracked later in this chapter.
Overall speed (time taken)
Most tree-testing tools show us the average (or median) time taken by our participants to complete the tree test.
Comparing times between trees
If we’re testing several trees against each other, and the trees are approximately the same size (in breadth and depth), we can compare these overall times to see if some trees are “slower” than others. This suggests that participants either had to:
- think a bit longer between clicks, and/or
- click more times to get to their answer
This is a very rough measure, however, and to make sense of it, we’ll need to drill down to see which tasks (or specific areas of the tree) are slowing down our participants. For more on this, see Task speed - where they slowed down later in this chapter.
Keeping the study brief
A more practical use for the average time taken is making sure that our tree test is not taking too much of the participants’ time.
In general, we recommend an overall duration of 5 minutes for a tree test. This is typically how long it takes the average participant to do 8-10 tasks (our recommended amount) for a medium-size tree (200-500 items).
If we have a larger tree, our test time may exceed this, but we still recommend keeping it under 10 minutes to avoid participant fatigue and boredom.
If the average duration is longer than this because we are asking each participant to do a lot of tasks (say, 12 or more), we are likewise inviting participant fatigue and boredom. More importantly, our results may be skewed by the “learning effect” – see How many tasks? in Chapter 7.
A “total” score
Some tools present a single overall score, combining several measures: success rate; directness; speed; and so on. This overall score typically uses some kind of weighting, with success rate usually being the biggest factor.
This is useful when testing trees, because it makes us consider more than just the success rate itself. If people can find items in our tree, but they have to do a lot of backtracking, or they have to ponder each click, there’s something wrong and the score should reflect that.
Note that the various online tools differ in how they calculate their overall score, making it harder to compare scores between tools:
- Treejack calculates its overall score as a weighted average of success rate and directness (at a 3:1 ratio), but does not include speed in its calculations. ~no longer provides a total score?
Next: Analyzing by task