In
Experiment 1 a salient object and the center of the image competed for visual selection. The three conditions differed in terms of the vertical position of the object and image center as well as the instructions provided to participants. In condition 1 participants were instructed to look at the center of an appearing image when the object was vertically displaced (
look at image center). Conversely, in condition 2 participants were instructed to look at the object while the image center was vertically displaced (
look at object). In condition 3 participants were instructed to saccade to a cued location that was vertically in between the image center and object (
look at cue).
Figure 2A–C shows saccade endpoints of all participants in all three conditions on an example image. We analyzed saccade endpoints as a function of saccadic reaction time and reconstructed individual time courses.
Figure 2D–F shows vertical endpoints as a function of saccadic reaction time for one example participant.
Figure 3A shows the reconstructed time course of vertical endpoints aggregated over all participants when participants were instructed to look at the image center. Endpoints were significantly biased toward the salient object,
p < 0.001,
t = 1756,
tcrit = 138.8, in the time window from 64–237 ms. In condition 2, when participants were instructed to look at the center of the object and the image center was vertically displaced, endpoints were systematically biased toward the image center and away from the salient object for saccades initiated in a time window between 59–172 ms after image onset (
Figure 3B),
p < 0.001,
t = 589.9,
tcrit = 162.3. A comparison between the two time courses (i.e.,
Figure 3A vs. 3B) revealed that the bias toward the salient object was stronger than the bias toward the image center,
p < 0.001,
t = 1232,
tcrit = 243.4, 79–225 ms. We found no evidence that these time courses changed with repeated exposure to the images (
Supplementary Fig. 1).
The third condition aimed to reveal which of the two biases, saliency bias or the bias exerted by the image center, dominates if they exert a pull in different directions. When asked to look at a cued location in between salient object and image center, saccade endpoints were biased toward the salient object, p < 0.001, t = 1283, tcrit = 140.8, in the time window from 79–225 ms.
We next asked whether image center and salient objects biased saccade endpoints or whether they rather captured endpoints. Salient objects in proximity of the target are said to bias endpoints continuously due to averaging responses and center-of-gravity computations in the priority map (for review, see
van der Stigchel & Nijboer, 2011). A priority map is a hypothetical retinotopic representation of space that combines bottom-up and top-down information and codes potential saccade targets by peaks of activity. Characteristics of such a map can be found in several sites along the oculomotor circuitry (for review, see
Bisley & Mirpour, 2019). Typically, the highest peak is selected as the next saccade target. Yet, a bias can occur, especially when two peaks (two potential targets) are in spatial proximity, in which case the saccade endpoint is determined by performing a weighted average of the two peaks. In contrast to this weighted average, salient objects have also been shown to capture gaze and attention (e.g.,
Theeuwes et al., 1998), thereby emphasizing a rather dichotomic mechanism of oculomotor and attentional selection. If, for example, the salient object captured gaze in the
look at image center condition, then a saccade would be erroneously directed toward the salient object. If such a capture occurred in half of the trials, the mean vertical saccade endpoints would be in between the salient object and the image center. The same mean vertical endpoint would be expected if all endpoints were biased to this location in between image center and object. Therefore, the aggregated time courses (
Figure 3) do not allow to distinguish between these two alternatives since they could be explained either by a fraction of trials in the respective time window being captured by the salient object or by a continuous shift by the entire endpoint distribution to an intermediate location. However, the two cases can be distinguished by looking at the distribution of endpoints. In case of oculomotor capture, the distribution of vertical endpoints should exhibit a bimodal profile with one peak centered at 0° and a second peak centered close to +2°. In case of center-of-gravity computations (a continuous bias), the distribution of vertical endpoints should exhibit a unimodal profile with a peak in between the two locations.
To distinguish between these two alternatives, we tested whether the distribution of endpoints can be better explained by a unimodal or a bimodal distribution (
Figures 4A & B). We fitted two models to the vertical endpoint distribution to the individual data of each condition: a Gaussian as well as the combination of two Gaussians. Importantly, we only considered trials within the identified clusters (
Figure 3). We compared the two model fits by computing information weights (
Burnham & Anderson, 2002;
Figure 4C) derived from the Bayesian Information Criterion (BIC). A higher information weight indicates a better model fit. We found that endpoint distributions of all three conditions were better explained by a unimodal distribution as indicated by higher information weights for the unimodal model: For condition 1 (
look at image center), information weights for the unimodal model were higher for 15 out of 18 participants,
Z = 3.29,
p = 0.001. For condition 2 (
look at object) this was true for all 18 participants,
Z = 3.72,
p < 0.001. For the third condition (
look at cue) information weights for the unimodal model were higher for 16 out of 18 participants,
Z = 3.2,
p = 0.0014. This shows that salient objects and the image center biased the entire endpoint distribution within a certain time window rather than a fraction of trials in that time window being captured by the salient object or image center.