R 2 Explained
60. When using any quantitative ET features, always test with a very
small sample before you embark on a larger study to first ensure that the technology does what you expect it to do.
ANALYZING VIDEO AND ANIMATION
As for getting a static data visualization for video you tested—just forget it. Unless you have a large, highly technical staff who can write a good program to extrapolate this information, expect to have to watch the gaze replays of videos to examine the eyetracking usability data. If you do have a slew of staff or great graduate students to make this program, please let us know when you are selling your program. We’ll gladly take a copy.
You can, of course make heatmaps of video, but they are almost always inaccurate, as they only show the static picture of one part of the video, and not all the moving pictures the users looked at. Even in the best case scenario where most of the video is an object mainly standing still, i.e., the “talking head” video, you still get
inaccurate reads in heatmaps. And even if you get past this, you can only get accurate information in the heatmap for one user at a time, negating the best reasons for using a heatmap in the first place.
In one example, we made the heatmap below from just one user. Making a heatmap with one user is not recommended for analysis, but we do this here for heatmap discussion purposes. In this image we see the user, expectedly, looking a lot at the face of the man being interviewed. If you were to take this heatmap as is and analyze it, you might come to believe that in this video, the man’s face is the most interesting thing to the user, then the tree, then the street sign, then the trashcan. You can easily explain that the user also got a bit bored with the talking head at a few points during the video segment and looked at the links to the right of the video and at the video controls at the bottom. All of these looks are understandable and expected. But what of the looks at the green sign over the man’s right shoulder, and those on the tree and trashcan. People do often look at things like signs, so that is possibly a real look and not a misrepresentation. But is a tree or a trashcan really that interesting?
The heatmap from www.cnn.com might lead one to believe that the tree and trashcan were very interesting.
When you actually watch the video you see there are some scene changes, and in fact a large segment shows the same man in a box on the right (the very location where the heatmap made it look like the user was looking at the trashcan) and the studio host in a box on the left. Where the birch tree had seemed so interesting we see it actually was not—the user was instead looking at the small box encasing the image of the female commentator. The image below only shows the time for the image depicted. In other words, we cut out the time with just the man being interviewed, so this heatmap is about right.
The heatmap from another view of the page on www.cnn.com shows that the faces were of interest, not the tree and trashcan.
And when we look at just the man, we see that his face gets the most fixations, then the controls and links, but that the street sign actually is interesting to this user.
This heatmap from www.cnn.com shows only the time when the view was of one reporter. There are more fixations on the face and the sign, and few or none on the tree and trashcan.
Advancements in Analysis of Dynamic Elements
According to Dr. Peter Brawn, the greatest challenge for eyetracking in the field also deals with analyzing the data. Peter explains that the basic output of a mobile eyetracker is a video recording of the scene, with a cross-hair overlaid to display the participant’s point of gaze. “The scene is changing all the time as the person moves around and looks at different things,” he says. “Compare this to a remote or desktop eyetracker being used for a usability study where the scene (the website) is always appearing in the same place (on-screen.)”
For field studies Peter says that AOI analysis poses even more challenges, “If we take the example of an in-store shopper study focusing on a particular category, then our scene would be an aisle containing a number of products and variants. However, each participant’s scene recording is different — a unique video recording as they navigate the store and browse the shelves of an aisle,” he says.
So there are basically two approaches for analyzing data for eyetracking field studies, according to Peter, “Use observational coding, which involves manually reviewing the footage and marking time-stamped events as they occur. For example, fixations can be calculated by recording when someone starts and stops looking at something. However, with the eyes moving so quickly and fixations occurring a number of times each second, this often requires slowing the footage for frame by
frame analysis or making multiple parses through. So depending on the complexity of the coding scheme, this could require an analysis effort of 5, 10 or even 15 times the actual length of the footage.”
Other analysis solutions allow for dynamic lookzones (i.e., lookzones that can be moved as the objects within a scene move). “However,” Peter explains, “multiple lookzones within a moving scene means this also becomes very laborious. For example, as the participant walks along an aisle looking at the shelves all of those lookzones have to move, change size and this has to be done manually. Also, because each participant’s scene video is individual, it has been impossible to perform multiple subject analysis in this way.”
The good news is that recent advancements have changed the way this analysis is performed and vastly reduced the effort involved. Peter describes a recent
advancement in this area, ASL's GazeMap software. “GazeMap uses edge detection and object recognition to map environments. For example, you can map aisles within a supermarket and overlay lookzones. The software is able to recognize them from each participant's individual recording and the lookzones adjust in size and position as the participant moves, so that they stay overlaid on the relevant objects. As well as automating this process it allows for multiple subject analysis.”
USING THE EYETRACKING ANALYSIS TOOLS IN THE BEST WAY
First of all, while you are doing the testing, always take notes. Don’t rely on the gaze replays as sometimes the system crashes and you lose the eyetracking data
completely. Also, having your notes can help you sort through the mounds of data you collect.