Download

A revised and updated version of the text on this page:

Eyetracking_tips.ppt (1.2mb) or a zipped version (652k).

Introduction

In this workshop I will present some tips for doing an eyetracker experiment, including advice on experiment design, recruiting participants and processing data. This workshop is aimed at people who study or work in HCI but may be unfamiliar with eyetrackers and psychological experiments.

Participants

How many participants do I need?

• Wide variations in eye tracking perfomance from person to person, so use a within-subjects experimental design
• Within-subjects design more practical in any case (need less participants than for between-subjects design)
• Need around 30 participant's worth of results
• Recruit more than you need: (some people won't turn up, some people can't be calibrated with the eyetracker, and some data will get lost or damaged)

How will I get participants?

• If you are at university doing the experiment as part of a Master's degree, you will probably start collecting data in the summer when undergraduate students have gone home - plan ahead!
• Use existing university or work mailing lists to recruit participants
• Be prepared to beg people to participate if necessary!
• Need to pay participants at least £5 per half hour to be enough of an incentive

What will participants tolerate?

• Participants may not tolerate over-long sessions - some may get tired, bored or resentful(!)
• Will participants need to practice?

Calibration

Eye trackers can have problems calibrating the following:

• Bi-focal/tri-focal lenses
• Super-dense lenses
• Hard contact lenses
• 'Lazy eye'
• Large pupils (crescent effect *)
• Small pupils
• Low contrast between pupil and white of the eye

*If the eyelid obscures part of the pupil, it creates a "crescent" which the eyetracker cannot track

Accuracy

Ways to improve accuracy:

• Leave room for recalibration breaks in your experiment
• Work within limits of accuracy of your eyetracker (TC technologies eye tracker is accurate to within 0.45 degrees of visual angle, which at 51cm from the screen = around 13 pixels margin of error
• Don't expect pixel perfect results

Piloting

The presentation of your experiment needs to be piloted, just like any other interface:

Controls

• Choose controls wisely - test and test again!
• You may want to suppress some controls at certain times

Instruction screens

• These are vital
• Test and test again!

Feedback

• Progress indicators
• Do you need to design special error messages?
• What about feedback?

Conclusion screens

• If you don't include them, some people are confused when the presentation just disappears at the end!

Timing of trials

• Affects behaviour and results
• Boredom
• Can only determine through testing

Length of sessions

• Test to see how much they can tolerate

Use your usability evaluation skills! Piloting IS user-centred design: Design, evaluate, re-design etc...

Good instructions for users are absolutely vital! They SHOULD evolve quite a lot during piloting...

Figure 4
An instruction screen for participants.

What other data do you want to record?

• Key-logging might not be possible
• Observation, in same room

Correcting Data

Results are seldom pixel perfect, so expect a to spend time reviewing and correcting the eye-movement data

There are two types of error:

• Absolute drift
• Relative warp

Absolute drift

Easy to spot - all the points on the eye trace are good, but they have all drifted off together...

Figure 5
An example of 'absolute drift' in the eye movement record: every gaze point has drifted in the same direction by the same amount.

Just drag the trace back to where it "fits" best, using software

Figure 6
The same eye movement data from Figure 2, after the 'absolute drift' has been corrected. Note how the pattern of fixations and saccades now matches the layout of the objects on screen.

Relative warp

Hard to correct! You can drag the some of trace to where it 'fits' but other points will always miss their apparent target. The solution: concentrate on getting a good 'fit' for the data you're actually interested in - ie your areas of interest.

Figure 7
An example of 'relative warp' in the eye movement record: most gaze points match the objects on screen, while others miss their apparent target.

Filtering Data

Need to filter data by defining 'areas of interest'

This can be done by drawing boxes around your areas of interest in graphics software, and recording the box coordinates in a text file.

The text file will be used to automatically filter the data from the raw eyetracker log files so you can analyse just the relevant data.

Areas of interest and buffer zones

Buffer zones around the areas of interest exist to take into account the eyetracker's margin of error (12.8 pixels). Trace points within the buffer zones will be counted as if they were entirely within the area of interest.

You can adjust the buffer zones in the software.

Figure 8
Areas of interest defined on a web page screenshot (highlighted here by yellow borders).

Analysing Data

Understand what format and structure your data needs to be in so you can analyse it.

Ensure that the filtering software produces output that you DON'T have to manually edit before you put it into SPSS. Minimise manual data editing at all costs!!!