Gaze tracking accuracy in humans: One eye is sometimes better than two

Most modern video eye trackers deliver binocular data. Many researchers take the average of the left and right eye signals (the version signal) to decrease the variable error (precision) up to a factor of root 2. What happens to the systematic error (accuracy) if the left and right eye signals are averaged? To determine the systematic error, we conducted a calibration validation in two experiments (n = 79 and n = 64). The systematic error was computed for the left eye, right eye, and version signals separately. In respectively 29.5 and 25.8% of the participants, the systematic error of a single eye signal was lower than that of the version signal at the cost of a higher variable error. If a small variable error is desirable, and the difference between the left and the right eye is not the topic of study, one should average position data from the left and the right eye (in other words, use the version signal). If a small systematic error is desirable, one should use the signal (from left eye, right eye or version) that delivers the best accuracy. In the latter case, this may cause worse precision than that of the version signal.