Using eye-tracking to study how people read

Now you have some idea how people's eyes move while reading. This is the information an eye tracker measures when we want to study how people read. In general, there are two things that the eyes do when we read:

You can feel the difference between fixations and saccades if you try to slowly look from one side of your room to the other. Look at one side (the right or the left), and try to "scan" your eyes across the room to the other side, as slowly and smoothly as possible. You will probably feel that your eyes don't actually smoothly move from right to left (or left to right); they stop somewhere, then jump, then stop, then jump, then stop, then jump... You can move across the room in very small jumps, but you can't move your eyes across the room perfectly smoothly.

When we measure people's reading, we usually are measuring their fixations. For example, when their eyes jump to one word and then look at that word for 500 milliseconds, they are making a 500-millisecond fixation on that word. The eye-tracker records that, and then we can assume that they spent 500 milliseconds to read the word. In this way, eye-tracking is just like self-paced reading: for each word, we can get a measurement of how long the person looked at that word, so we can figure out how long they needed to read each word in the sentence.

Or is it really that simple? Proceed to the reflection questions below, and we'll see if things are that simple or if there are some other problems that we need to consider...

With self-paced reading, for one word we get one result (one number telling us how long the person needed to read that word).

Will the results also work that way when we do eye-tracking, or will the results be more complicated? Think about what you saw in the previous task when you watched videos of how people's eyes move. Can you think of any situations where it will be complicated to determine how long a person took to read a word? Can you think of any situations where we will not get one result for one word (e.g., situations where we will get more than one number for one word, or situations where we will get less than one number for one word)?

Imagine I am doing an eye-tracking experiment and I want to see how long people take to read each word in the following sentence:

The key to the cabinets are sitting on the table.

Specifically, I want to know how long people take to read "cabinets".

Now let's imagine I record people's eyes moving while they read the sentence, and I observe the following pattern:

An image showing a sample pattern of eye movements. At the bottom of the image is the sentence 'The key to the cabinets are on the table'. Above this is a series of numbers showing the fixations; each number shows where a fixation was made (e.g., #1 is above 'The key', indicating that the first fixation was on these words; #2 is over 'to the'; etc. See text for detailed description of the pattern of results.

This picture tells us that the eye movements followed the following pattern:

First, the person looked at (made a fixation on) the word the.
Next, they moved their eyes (made a saccade) and then looked at the word to.
Next, they looked at the beginning of the word cabinets.
Then, they moved their eyes, but not to another word --- they just looked at another spot on the same word cabinets.
Then, their eyes jumped back, and looked at the word key!
Then, they jumped forward and looked at cabinets again.
Next, they jumped forward and looked at the phrase are sitting.
Next, they looked at on.
Next, they looked at table.
Next, their eyes jumped back again, and they looked at cabinets again!
Last, then again moved their eyes within the same word -- looking at another spot on the word cabinets.

So they made eleven fixations at different places in this sentence. For each of these fixations, the eye-tracker will also have recorded how long the fixation was (e.g., how long they spent looking at "cabinets" during fixation #3, etc.), but let's not worry about that yet.

If you wanted to know how long people spent to read the word "cabinets", what would you do? If we were doing self-paced reading we would have just one number; but with eye-tracking, they looked at this word five different times. Which of these fixations might you pay attention to if you're trying to calculate how long people spent reading the word?

As usual, there's no right or wrong answer, but I want you to think about this problem and I want to see what solution you think might be best.

Let's think again about the example from the previous question. Here is a picture of the data pattern again; after the picture I have also shown the same data pattern as a table, in case you have trouble viewing the picture.

1
	2
		3
		4
5
		6
			7	8
					9
		10
		11
The key	to the	cabinets	are sitting	on the	table.

As we discussed in the previous question, there are a lot of possible ways we could measure how long people took to read cabinets, because they looked at it five different times.

Among people who do research on eye-tracking, there are several common ways we can measure how long people took to read a word. Here are some of them:

First fixation duration: Measure how long was the first fixation on the word. In this example, that is fixation #3.
First pass duration (also called gaze duration): Add up all the fixations from the first time the person looked at this word, until the first time they looked at anything else. In other words, how much time did they spend looking at the word until they looked away. In this example, that means we'd add up fixation #3 plus fixation #4; fixation #5 is when they looked away from the word, so we wouldn't include it.
Go-past time (also called regression path duration): Add up all fixations, on this word or any other words, from the first time the person looked at this word, until the first time they looked at a later word. In other words, this is how much time they needed from the first time they looked at this word until they were able to go past this word. In this example, that means we'd add up fixations #3, #4, #5, and #6; the first time they went past the word cabinets was fixation #7, so we wouldn't include that. Crucially, we still count fixation #5, even though it's not looking at the word cabinets; maybe the time they were looking back at previous words is still time they were using to try to understand the word cabinets and figure out how it fits into the sentence!
Total time: Add up all the fixations on this word. In this example, that means we'd add up fixations #3, #4, #6, #10, and #11.

Which of these measures do you think would be most useful? Or, do you think different measures would be useful for different things? If so, what do you think would be the important differences between them? For example, what do you think one measure (such as total time) might tell us that is different from what another measure (such as first pass time) might tell us?

When you have finished these activities, continue to the next section of the module: "Practice calculating reading times".

by Stephen Politzer-Ahles. Last modified on 2021-05-14. CC-BY-4.0.

Using eye-tracking to study how people read (1hour)