The overall systematic tendencies, such as the center bias and the horizontal bias are also known to be present in infants (22, 23, 24), however the existence of other systematic tendencies is largely unknown. The current study examines the leftward bias, viewing time and scan path dependencies between successive fixations and saccades which are described below in more detail.

The leftward bias is the tendency to make an initial eye movement towards the left side of the screen (25, 26, 27). This tendency is hypothesized to stem from the asymmetry of attentional control networks in the brain that are lateralized to the right hemisphere. Another common explanation is the reading direction that is left-to-right for most participants and that these learned scanning habits play a role in the leftward bias. Although the leftward bias seems universal, studies comparing left-to-right readers with not left-to-right readers report a weaker leftward bias for non-left-to-right readers (28, 29, 31). If reading direction does influence this tendency, we would not observe the leftward bias in infants, but we would in adults.

The influence of viewing time on eye movements is well-established. At the start of a trial, fixation durations are typically shorter after which they increase in duration during the trial (9,32,33). Saccade amplitudes tend to follow an opposite pattern with larger saccade amplitudes at the start of the trial that decrease in amplitude towards the end (34,35). Together these findings have been interpreted to reflect different modes of scanning. An ambient or global mode at the start of trials characterized by shorter fixation durations and saccades of longer amplitude and a focal or local mode with longer fixation durations and saccades of shorter amplitude (34). However, Follet, Le Meur, and Baccino (36) reported an initial occurrence of the focal mode followed by an ambient mode, based on their findings they suggest there is an interplay between the modes during viewing. Having different modes of viewing behavior could be interpreted as strategic viewing, a quick scan to identify the most informative regions followed by a closer inspection of this region.

Given that fixation duration and saccade amplitude depend on viewing time, we would also expect dependencies between successive fixation durations and saccade amplitudes if the ambient and focal modes exist (34). Similarly we would also expect correlations between successive fixation durations and successive saccade amplitude. Tatler and Vincent (35) did report such dependencies in which shorter fixation durations and saccade amplitudes were also followed by shorter fixation durations and saccade amplitudes and longer fixation durations and saccade amplitudes were followed by longer fixation durations and saccade amplitudes. Other scan path dependencies include the relationships between fixation durations and saccade amplitudes with the (change in) saccade direction.

The main aim of the current study is to investigate the origin of the systematic tendencies often reported in the literature. To this end we compare infants and adults on the systematic tendencies frequently observed in adults in three existing free-viewing data sets. Studying infants in comparison with adults allows to gain more insight in the mechanisms underlying the systematic tendencies. When results are similar for infants and adults it is more likely these tendencies are a result of basic mechanisms, whereas differences between adults and infants could indicate that these tendencies are a result of more elaborate cognitive strategies used by adults or are learned over time.

A secondary aim is to describe these systematic tendencies in infants, such that researchers studying infant viewing behavior can control the influence of these tendencies and avoid getting biased results. To explore the similarities between systematic tendencies in infants and adults during free scene viewing we will examine the leftward bias, the effects of viewing time and the scan path dependencies between successive fixations and saccades. For this last part we will closely follow (35) who examined these systematic tendencies in adults.

Table 1 shows the main descriptives of the participants in the three studies. Combined 157 infants (M = 9.71 month-olds, range = 3.20 - 20.53) and 88 adults (undergraduate psychology students) saw around 30 photographs of real-world scenes while their eye movements were recorded. All studies were conducted in accordance with the declaration of Helsinki and all adult participants and infant caretakers gave their informed consent.

For the center bias study a total of 30 stimuli were selected with specific requirements for three conditions (i.e. 10 stimuli in each condition). Stimuli either had saliency distributions biased towards the center, biased towards the side, or uniformly distributed saliency distributions, see Figure 1 row A. In the original study we manipulated the start position and our main interest was the first saccade. For this re-analysis the first fixation is excluded to limit the effect of the manipulation. Another large difference between the center bias study and the other two studies was the layout of the stimuli. Stimuli were presented overlaid with a circular aperture to avoid directional biases due to screen dimensions. In the horizontal bias study 28 stimuli were selected from the labelme database (38) and in the object familiarity study 29 real-world images were selected from the Object and Semantic Images and Eye-tracking (OSIE) data set (39), for examples see Figure 1 row B & C. In these studies the presentation time was 4 and 8 seconds respectively and stimuli were presented full screen (object familiarity) or with a black border around the stimuli while maintaining the aspect ratio of the screen.

In all studies, eye movements were recorded using a remote-optics corneal reflection eye-tracker (SR EyeLink 1000), with a sampling rate of 500Hz. Visual stimuli were presented on a 17-inch monitor (1280x1024) in full color extending approximately 34° x 27° of visual angle, such that the number of pixels per degree of visual angle was approximately 38. After participants were properly seated approximately 60 centimeters from the computer monitor in either a Maxi-Cosi or on their caregiver lap, lights were dimmed and black curtains were drawn such that only the stimuli presented on the computer monitor could be seen. Caregivers were instructed not to communicate with the infant or to (re)act on the images presented on the screen. A 5-point calibration scheme was used in all studies and the experiment began once the mean error of all calibration points was smaller than 1 degree of visual angle. The calibration used colorful looming dots or cartoons accompanied with sounds to attract the infants attention. The center bias and object familiarity studies started based on a gaze-contingent attention getter between each trial. If this attention getter was not fixated 5 times in a row, there was an option to re-calibrate if the attention getter was missed due to drift from the original calibration. In the horizontal bias study the trials started with a fixation cross and no re-calibration took place during the experiment.

This manuscript is prepared in Rstudio using R (40) and the R-packages cowplot (41), gazepath (42), ggforce (43), ggplot2 (44), gridExtra (45), and papaja (46) for all analyses and visualizations.

The current study is an explorative study and as such we do not report inferential statistics or use statistical tests. The way we do present the data is using visualizations that include confidence intervals. Based on these confidence intervals conclusions can be drawn based on the similarities or differences between infants and adults. Overlapping confidence intervals implies there most likely is no significant difference, whereas non overlapping intervals indicate possible differences between infants and adults.

Fixations. For all studies the raw data is classified into fixations with the gazepath R-package (42). This method allows to identify fixations in both infant and adult data by setting individual thresholds such that noisier data results in more conservative thresholds. Since infant eye-tracking data is typically noisier than adult eye-tracking data this method is suitable to compare both groups as the same method can be used while allowing differences within individual. The top row of Figure 2 shows the densities of fixation durations for the infants and adults in the three studies. There is a clear pattern that fixation durations are longer in infants than adults.

Saccades. The gazepath method is optimized to identify fixations and not saccades. However, since the goal of this study is to get a better insight in the characteristics underlying gaze patterns, the saccades are also of interest as gaze patterns are a series of fixations and saccades. To identify the saccades the time between consecutive fixations was calculated. In a typical gaze pattern this is the duration of the saccade, however the period in between fixations can also reflect blinks or missing data instead of a saccade. As saccades typically last between 10 and 90 msec. (47), these values were used as cut-off values to identify saccades.

There are clear differences between the durations of saccades of infant and adults. Saccades of both adults and infants seems to follow a bi-modal distribution with modes around 20 and 40 msec., however the mode around 40 msec. is more prominent in adults whereas the mode around 20 msec. is more common in infants (Figure 2 row B). These differences in the durations are only to a small extent reflected in the amplitudes, Figure 2 row C. This difference may stem from the gazepath method with which the saccades were identified. The gazepath method sets individual speed thresholds based on the amount of noise in the data. In the noisier infant data the thresholds are set higher than in the less noisy adult data. This implies that the start of adult saccades is identified earlier than the start of infant saccades, whereas the end of adult saccades is identified later than the end of infant saccades. This differences in threshold may very well explain the differences in saccade duration between infants and adults and also explains why this difference is almost non-existent for the saccade amplitudes. The saccade amplitudes are calculated as the distance between fixations and those distances are not affected by the different thresholds for infants and adults. It is thus important to be careful in interpreting these differences in saccade durations as they may very well be a result of the different thresholds. On the other hand, these small differences cannot be completely ignored as they also seem to be present in the saccade amplitude data, albeit to a much smaller degree. These differences may reflect some sort of developmental pattern in which adults are more likely to make larger saccades than infants. This is in line with findings of others comparing infants (24) and children (48) with adults during a scene viewing task.

Although the use of individual speed thresholds in the gazepath method may exaggerate differences between infants and adults in saccade durations, the bi-modality of the saccade durations cannot be explained as a by-product of the saccade identification methods. The bi-modality of saccade durations in both infants and adults also exists when the standard Eyelink classification method is used, is reported by others for scene viewing tasks (47), and is also observed in other experimental tasks in our lab. This bi-modality in saccade durations may reflect different processing modes (34) in which scene exploration is focal (resulting in short and small saccades) or ambient (resulting in long and large saccades). This would imply that the bi-modality is also present in the saccade amplitude data, but this is not clearly visible by only looking at the distribution.

Figure 3 shows the histogram of the proportions first saccade directions in the top panels (row A) and the mean x-position of the fixations as a function of time in the lower panels (row B). For the center bias data set there is no leftward bias, this can both be seen in the histogram of proportion in which the saccade directions are evenly distributed as in the x-position as a function of time where the mean x-position of both infants and adults’ fixations does not deviate from the center of the screen (red line). This is a very sensible outcome as the start position was manipulated in this study, which also explains the wide range of x-positions at the start of the trial. Therefore it shouldn’t be expected that there would be a leftward in this data set.

For the horizontal bias data set there is a clear bias in the horizontal directions as can be seen from the histogram of the proportions, it can also be seen that the leftward bias is stronger in adults than infants. The initial shift of the x-position for adults is to the left, but based on the confidence interval does not seem to deviate much from the center of the screen (red line). Both infants and adults seem to have an overall rightward bias for the horizontal bias data set, which may explain why we didn’t replicate earlier findings. The object familiarity data set does show a very clear leftward bias for adults, but not for infants in both the histogram of saccade directions and the x-position. There is bias in adults to target the first saccade towards the (top) left, which results in shift in the x-position of fixations that is also biased to the left. In infants the direction of initial saccades is much more evenly distributed and as such there is also no overall leftward bias in the x-position. Overall, the pattern of results in these three studies can be taken as evidence that adults have a leftward, but only when the start position is at the center, whereas infants do not have a leftward bias.

Figure 4 shows the effect of viewing time on the fixation durations (row A) and saccade amplitudes (row B). The fixation durations of both infants and adults show a sharp initial increase after which the fixation durations stabilizes or keep increasing slightly. For adults these patterns are in line with earlier studies examining the effect of viewing time on fixation durations (9,33,34). Infants show a similar pattern as adults, which is a similar result as reported by Helo et al. (24) who also compared infants and adults. They found that fixation durations made early during the trial were shorter than fixations durations made later during the trial for older infants (> 9 months) and adults, but there was no difference for younger infants.

The effect of viewing time on saccade amplitudes is similar for infants and adults but without a clear pattern across data sets, see Figure 4B. In the center bias data set there is a sharp initial decrease which is the result of the manipulated start position after which most saccades were made towards the center. Afterwards there seems to be a small increase in saccade amplitudes with viewing time. This is similar to the pattern found in the horizontal bias data set, which also shows a slight increase, but in the object familiarity data set there is no effect of viewing time on saccade amplitudes. These mixed results are not in line with a decrease in saccade amplitude during viewing (33,34), however they do matched the adult data reported by Follet et al. (36) and are also similar to the effects Helo et al. (24) report for both infants and adult.

To get a better understanding of the relationship between successive fixations and saccades we adopt the same method as (35) and examine dependencies between the previous (N-1) and current (N) fixation duration (FD), saccade amplitude (SA), saccade direction (SD) and change in saccade direction (change SD), see Figure 5. This Figure displays a schematic scan patterns to show how the different variables are defined. In the following sections the dependencies between these variables are examined.

Figure 6 shows the relationship between successive saccades (rows A & B) and the histogram of saccade amplitudes for infants and adults. The top panels (row A) show that infants and adult differ in their saccadic behavior. In adults large amplitude saccades are more likely to be followed by a smaller amplitude saccade, whereas infants large amplitude saccades are more likely to also be followed by a larger amplitude saccade. The finding in adults are relatively consistent across data sets (expect for the center bias data) and replicate the relationship reported by (35). When the current saccade amplitude is used as predictor (row B), the relationship between infant and adult successive saccades looks similar both within and across the different data sets. There is positive relationship in which small saccades are also preceded by small saccades, this could indicate periods of local scanning (34) in both infants and adults. For larger saccades there is a trend in which larger amplitude saccades are preceded by smaller saccades in adults, replicating earlier work (35). In infants there is also an initial decline (between 3-8 degree saccades) in which larger amplitude saccades are preceded by smaller saccades, but for larger saccades this seems to shift towards a positive relationship.

Figure 7 shows the relationship between successive fixations (rows A & B) and the histogram of fixation durations (row C) for infants and adults. In adults the overall pattern is very similar between studies. Longer fixations are more likely to be followed or preceded by longer fixations. In infants the overall pattern is the same for the horizontal bias and object familiarity data sets, but there is no influence of preceding or successive fixation duration in the center bias data. Again these results are strikingly similar to the results reported by Tatler and Vincent (35).

Figure 8 shows the relationship between the current fixation duration and incoming (row A) and outgoing (row B) saccades. The top panels (row A) show a very similar pattern for infants and adults across the different data sets. Except for adults in the center bias data, both infants and adults show an inverted U-shape relationship between the current fixation duration and the incoming saccade amplitude. Shorter and longer fixation durations tend to be preceded by saccades of smaller amplitude, while fixation durations around 200 msec. in adults and 400 msec. in infants tend to be preceded by saccades of larger amplitude. These peaks of the inverted U-shape corresponds to the median fixation durations in both infants and adults. Again the relationship for adults closely matches the relationship reported by Tatler and Vincent (35).

The relationship between the current fixation duration and the amplitude of the outgoing saccade is somewhat different for both infants and adults and across data sets (row B). The 2 most similar patterns are the patterns of adult in the horizontal bias and object familiarity data sets. Saccade amplitudes are large after short fixation duration and this relationship disappears for longer fixation durations. These results are very much in line with the results reported by others (34,35). Do note that they also report a strong positive relationship for fixations durations shorter than 100 milliseconds, that we cannot assess as we only consider fixation durations of 100 milliseconds and longer. The infants in the object familiarity data set show a similar pattern as adults, although there seems to be an increase in which longer fixations tend to be followed by saccades of larger amplitude. This patterns of longer fixations followed by larger saccades is also present for infants in the other two data sets.

Figure 9 shows the relationship between the current saccade direction and current saccade amplitude (row A) and preceding fixation duration (row B). The horizontal bias and object familiarity data sets show the typical horizontal bias for infants and adults, however this bias is not present in the center bias data set, see Figure 9C. A possible explanation is the layout of stimuli in the center bias study were circular, which is known to influence the bias (50). In addition, the manipulation of the start position and the selection of the stimuli in different conditions most likely has influenced the overall horizontal bias.

There is a strong relationship between the saccade direction and saccade amplitude in both infants and adults (row A). Saccades along the horizontal axis are longer, followed by downward saccades which have an intermediate amplitude, while the upward saccades have the smallest amplitude. Again these findings are strikingly similar for both infants and adults and are very much in line with reports by others (35). The middle panels (row B) show a similar relationship between the saccade direction and the preceding fixation duration for infants and adults. Downward saccades are preceded by shorter fixations, while upward saccades are preceded by longer fixations, the saccades to the left and right fall in middle and are preceded by fixations with an intermediate duration. Interestingly these relationship seems also to be present in the adult data of the center bias data set, despite the lack of an overall horizontal bias effect. Again this effect is also reported by Tatler and Vincent (35).

Figure 10 shows the relationship between the change in saccade direction and current fixation duration (row A) and current saccade amplitude (row B). All data sets show a remarkable similar pattern for both infants and adults. Larger changes in saccade direction co-occur with longer fixation durations (row A) and larger saccade amplitudes (row B). These findings have also been reported by others (9,35) and thus seem very robust.

The author(s) declare(s) that the contents of the article are in agreement with the ethics described in http://biblio.unibe.ch/portale/elibrary/BOP/jemr/ethics.htmland that there is no conflict of interest regarding the publication of this paper.

This research was supported by the research priority area YIELD, University of Amsterdam, The Netherlands.