How to Analyze and Interpret Heat Maps
In the 1950s, at the dawn of the computational revolution, scientists were presented with a kind of problem that they had rarely encountered before. It was a problem that few had seen coming, and new methods were urgently required to treat it. However, in spite of their best efforts, the problem still persists to this very day. What exactly was the problem? Too much data.
The rise of computers allowed easier and more streamlined data collection, leading to large-scale datasets that required suitably large-scale analyses. One of the outcomes from this sudden need of new analytical methods was the invention of the heatmap. Originally used with taxonomical datasets, the applications to other areas were soon obvious and embarked upon.
Around the same time, researchers had begun developing eye trackers in a similar vein to their current, modern form. It didn’t take too long for these tools to collide, and the eye tracking heat map was born. These heatmaps provide clear and accessible representations of dynamic processes, advancing our understanding of the data at hand.
But what does a heatmap really show?
In short, a heatmap shows the relative intensity of a value within an array. This means that we have a large amount of numbers, and each is given a graphical representation. Those that are highest in their value – relative to the other present numbers – will be given a “hot” color, while those that are lower in their value – again, relative to the other present numbers – will be given a “cold” color. Like a rather mundane “paint by numbers” canvas.
For eye tracking, this could be the number of times that the eyes fixated upon a certain part of an image. For each time the eye points to a pixel, the number for that pixel goes up by 1. As the number of fixations increase, so do the numbers – ultimately displayed as a “hotter” color on the heatmap.
An example of a heatmap, showing a range of numbers (from 1-10) that pertain to the intensity (“heat”) of the displayed color. This could represent a portion of an image that has been zoomed into, with each square representing a pixel.
What do we do with all the numbers?
Before we get to any conclusions, it’s good to know how we got there. Let’s dig a bit deeper and look at the principles of the data behind the heatmap.
While the heatmap is an image, we know that these colors ultimately lead back to numbers. And with numbers, we can do anything.
Now imagine the heatmap not as an array of colors, but as a grid of numbers – like a giant, confusing sudoku game. Now, for each of those numbers, we can add a label, and that label can refer to the co-ordinates. So, for the bottom left pixel on a screen, we can give it the co-ordinates (1, 1), and the one to the right of it will be (2, 1), and so on.
An example of a heatmap with numbers pertaining to the intensity of the color, and co-ordinates shown in the bottom left of each square.
We can take all these numbers and put them in a giant list. We now have a list of numbers that tell us about the intensity of values from our grid, and therefore, our heatmap.
We could even decide which co-ordinates – regions – are most important to look at (these are known as Areas Of Interest, or AOIs). Perhaps our image is of people, and we want quantify the amount that people look at the eyes of others – we could group together the co-ordinates for the eyes, and later compare those numbers against similar clusters, or other AOIs, on the image.
Of course, this process is essentially a reversal of what the computer does to produce the heatmap in the first place. It first has a number, turns it into a color, and shows it on a chart – the position of which will depend on the label attached to it.
What should we look at?
There are various ways to carry out data analysis, each of which depends on the research question and the data at hand. While there are various metrics that are captured from eye tracking experiments, we’ll focus on the ones that can be converted into (or from) heatmaps.
This refers to the continuous viewing of an image – where we are looking when presented with a visual stimulus. A heatmap of gaze data would therefore show which parts were most frequently looked at. If we wanted to know how much time each feature of the image was looked at, we should examine the fixations.
A fixation is regarded as a gaze that is maintained within the same region for more than a passing moment (in practice, this can be defined as within 1 degree radius of vision, and lasting for over 100ms). A heatmap built from fixation values therefore shows the number of times in which an individual pays focused attention to a particular part of an image.
Heatmaps in iMotions are created by default from gaze mapping data, although they can also be created from fixations – you can decide what’s best for your study. The heatmap makes for an accessible and understandable framing of the data, but if you want to know more about what underlies it, then you’ll need to export the numbers for further analysis.
A snapshot of some of the data that can be exported from an eye tracking experiment.
Analyzing the Data
Once you have the data, be it in Excel, SPSS, or another statistical program, you can start to dig deeper into the numbers. Testing the data that underlies a heatmap can ultimately be the same as with any other data analysis. To compare differences between the viewing of two images by an individual or group, a t-test could be performed on the data. This can inform you about whether or not there is a significant difference between the amount of gaze or fixation between the two images.
If comparing the results across more than two groups, you might want to consider using a statistical test called an Analysis of Variance (ANOVA), which compares the variance across groups. This is particularly useful in cases in which groups overlap.
Consider an experiment in which you want to measure participant’s responses to different pieces of art. You might have a population that can be defined by their level of interest in art, and their age. An ANOVA allows you to know if young art-lovers are more engaged by the stimuli than older people who are uninterested in art. You could also see if those grouped by a greater interest are generally more engaged, regardless of their age. Any combination of groups is essentially possible.
This provides a brief introduction in essence to the statistical tests, yet the exact applicability and usefulness of each test will depend on the research question, and the data at hand. There are of course many more statistical tests available, offering a great degree of flexibility for analysis.
What can we know from the data?
While the gaze and fixation values can’t tell you exactly what a person is thinking, it can provide insight in other ways. If the gaze or fixation data of a particular stimulus (be it image-based, video-based, or in a real-life setting) is compared to another stimulus, we can know which attracts the most attention, or which is the most salient. If you want to probe further into what an individual is thinking or feeling, then multiple methods can be employed to give you those answers.
For example, simultaneous recording of physiological arousal (such as through measurements of galvanic skin response), and recording of facial expressions can give information about the emotional valence and intensity that an individual is feeling in the presence of a stimulus. Additionally, surveys and psychometric tests can be used to help the participant reveal their thoughts, feelings, or intentions. Ultimately, the use of various biosensors can provide a complete picture of an individual’s response to a stimulus.
While the data from eye tracking studies can be converted into heatmaps, it can also be used to provide information about various other metrics, including (but not limited to!) the Time To First Fixation (TTFF), number of revisits, and the ratio of views.
A screenshot showing some of the analytical measurements available from an eye tracking experiment, including TTFF, time spent, and the ratio of views.
The TTFF relays the amount of time before a participant views a defined part of the stimulus, while the number of revisits can tell you about how many times a certain part of the stimulus is repeatedly viewed. The ratio describes the proportion of participants that actually viewed a set part of the stimulus. Overall, this data can provide a clearer picture about the level of attention that a stimulus receives.
If you’d like to know even more about eye tracking, and how to use eye tracking in your research, then have a look through our free and amazing pocket guide. It’s fantastic, and with it, you’ll soon be creating fantastic studies.