Families were recruited from a database of parents who had expressed an interest in participating in research by calling back in response to an information flyer distributed locally. The Ethics Committee of the German Psychological Society only requires ethical approval for interventional studies that involve potential harm to the subject. In this study, no intervention was applied on the subjects as the task only involved looking at a computer screen, as a result no ethical approval was necessary. Informed written consent was obtained by all parents. Infants received a small toy for their participation. All infant participants were healthy and their birth week, weight, and Apgar-score reached standard values 
. Infants were tested within 10 days of their 6-month or 8-month birthday.
Experiments were performed in a darkened and sound-attenuated room, with the eye tracker screen as the only source of lighting. Infants sat in an infant-seat (Weber Babyschale, http://www.weber-products.de
) placed on their parent's lap. An EyeLink 1000 remote eye-tracking system was used (SR Research, http://www.sr-research.com
). The eye tracker camera was attached underneath a 17 inch computer screen, and recorded the reflection of an infrared light source on the cornea relative to the pupil at a frequency of 500 Hz. The experimenter controlled the stimulus presentation from a display computer in an adjacent room while monitoring the infants behavior through a video camera. The eye-tracker allowed for moderate head movements without accuracy reduction in a volume of 22 cm×18 cm×20 cm (horizontal×vertical×depth). Blink or occlusion recovery was faster than 3 ms.
During the calibration process, attractive balls (shrinking from approx. 2.5
to a point) with sound were presented in a three-point calibration sequence for infants, and a five-point-calibration was used with adults. The calibration procedure was repeated if necessary. Calibration procedures, stimulus presentation and data output were accomplished using Experiment Builder software and allowed an optimal accuracy of 0.5 degrees. The x, y coordinates (in pixel) of the three point calibration positions on a screen resolution of 1024×768 were: (511.5, 65.2), (961.6, 701.8) and (61.4, 701.8), for five point calibration they were (395.5, 287.5), (395.5, 526.1), (359.5, 48.9), (675.9, 287.5) and (43.1, 287.5). A human observer pressed a key when the participant was judged to be fixating the stimulus at each location, and an exclusion criterion was a 1.5 degree average error during validation of the calibration, which corresponded to a 1.5 cm area on the screen with a viewing distance of about 60 cm.
Animal pictures were taken from Animal Diversity Web (http://animaldiversity.ummz.umich.edu
). In Experiment 1, the size of the pictures was 13.8
horizontally; the red disc was 6.1
, and the distance between the border of the picture and the border of the disc was 10.1
. In Experiment 2, picture width was 12.4
horizontally, each disc's radius was 5.4
, and distances between edge of picture and edge of discs were 3.9
on both sides. Interest areas for eye tracking analysis were defined to exactly match position and size of the red discs and images.
Exclusion conditions were if a participant did not finish 1 min in Experiment 1 or 3 min in Experiment 2 because of fuzziness, excessive movement, sleeping, bad calibration, or software failure. Data analysis was performed with EyeLink DataViewer software and Matlab. In Experiment 1, dropout rate was 14% (5 infants). Three subjects were excluded because of bad eye tracker calibration, one because of fuzziness, and one because of a software problem. In Experiment 2, dropout rate was 35% (18 infants). Fourteen infants were excluded because they did not finish 3 min because of fuzziness, excessive movement, or sleeping, three infants because of bad calibration, and one because of a software problem. Adult dropout rate was 11% (3 adults), because of poor eye tracker calibration.
For each data sample, the eye tracking software computed instantaneous velocity and acceleration and compared these to velocity and acceleration thresholds. If either was above threshold, a saccade signal was generated. The Default Cognitive Configuration was applied. Saccade velocity threshold was 30 deg/s and saccade acceleration threshold was 8000 deg/s
. Eye position, pupil size, velocity, etc. were updated every 50 ms during a fixation.
In Experiment 1, anticipatory gaze shifts were identified as follows. Gaze shifts from the button area to the picture area could be composed of an individual saccade or a rapid sequence of two saccades. In both cases, we considerd the start time of the first saccade leaving the button area as the start time of the gaze shift. We only considered situations where the start time of the gaze shift occurred at least 200 ms after the previous image had disappeared to rule out the possibility that the gaze shift was aimed at the previous image. A gaze shift was considered anticipatory if its start occurred within 200 ms of the onset of the new image.