Abstract
We present novel data on the role of attention in eliciting enhanced processing of stimuli associated with self. Participants were required to make pro- or anti-saccades according to whether learned shape–label pairings matched or mismatched. When stimuli matched participants were required to make an anti-saccade, and when the stimuli mismatched a pro-saccade was required. We found that anti-saccades were difficult to make to stimuli associated with self when compared to stimuli associated with a friend and a stranger. In contrast, anti-saccades to friend-stimuli were easier to make than anti-saccades to stranger-stimuli. In addition, a correct anti-saccade to a self-associated stimulus disrupted subsequent pro-saccade trials, relative to when the preceding anti-saccade was made to other stimuli. The data indicate that self-associated stimuli provide a strong cue for explicit shifts of attention to them, and that correct anti-saccades to such stimuli demand high levels of inhibition (which carries over to subsequent pro-saccade trials). The self exerts an automatic draw on attention.
It is known that human cognition is strongly modulated by self-bias. Information encoded in relation to the self tends to be better remembered and recollected than other information (Turk, Cunningham, & Macrae, 2008); people respond faster to their own faces rather than to those of others (Ma & Han, 2010; Theeuwes, Van der Stigchel, & Olivers, 2006); responses to their own objects are faster and more accurate than responses to objects owned by others (Turk, van Bussel, Waiter, & Macrae, 2011). Recently, the effects of self-biases have been found even in simple perceptual tasks (Sui, He, & Humphreys, 2012; Sui, Sun, Peng, & Humphreys, 2014). Specifically, in a task where participants associate simple geometric shapes with personal labels (you, friend, or stranger) and then immediately judge whether subsequent label–shape pairings matched, there are faster responses and higher perceptual sensitivity to self-associations than to other-associations (Sui et al., 2012). Previous research provides strong evidence that the effects of self: (a) are stable over time (Stolte, Humphreys, Yankouskaya, & Sui, 2015); (b) mimic behavioural and neural responses generated by stimuli with high perceptual saliency (Sui, Liu, Mevorach, & Humphreys, 2015); (c) are automatic and not easily overruled even if participants set an expectation for another stimulus (friend or stranger) on the trial (Sui et al., 2014); (d) alter perceptual processing even when the shape is subsequently presented without the label (Sui, Yankouskaya, & Humphreys, 2015)
Although the effect of self-prioritization in perceptual matching is robust, it remains unknown where the effect comes from. Recent studies have shown that the self-prioritization effect may reflect attentional biases to self-related stimuli (Alexopoulos, Muller, Ric, & Marendaz, 2012; Bredart, Delchambre, & Laureys, 2006). The results in behavioual studies are also supported by neuroimaging work demonstrating that self-associated stimuli as distractors elicit increased activation over the attentional neural network (Sui, Liu, et al., 2015). There is also evidence of self-related attentional biases from event-related potential (ERP) studies. For example, researchers have reported that when an item is cued as being relevant to the self, the amplitude of P300 component was increased compared to an other-relevant cue, indicating an enhanced attentional processing to the item. This has been attributed to self-related stimuli narrowing the range of spatial attention, indexed by reduction of the amplitude of the occipital P1 component elicited by task-irrelevant probes when participants responded to self-relevant objects (Turk, van Bussel, Brebner, et al., 2011).
However, a number of other studies provide contradictory evidence (Devue, Laloyaux, Feyers, Theeuwes, & Bredart, 2009; Devue, Van der Stigchel, Bredart, & Theeuwes, 2009; Gronau, Cohen, & Ben-Shakhar, 2003; Kawahara & Yamada, 2004). For example, Devue, Van der Stigchel, et al. (2009) measured eye-tracking movements in visual search for self-faces among other faces and suggested that the self-prioritization effect reflects a difficulty in disengaging attention from highly familiar faces, rather than differential attentional engagement. Bundesen, Kyllingsbaek, Houmann, and Jensen (1997) reported that presenting the participant's own name as a distractor did not cause more interference than other names, but participants were more accurate in reporting their own name presented as targets. These authors suggested that the advantage for the participant's own name as a target was not attentional, but rather reflected a better identification of the participant's own name than of other names. Other researchers have argued that self-related stimuli capture attention only when they are presented within the participant's attentional focus (i.e., at fixation) and when the stimuli are task-relevant (Gronau et al., 2003; Kawahara & Yamada, 2004). Notably, visual search tasks typically use complex stimuli (e.g., faces), which may involve several stages of information processing, and hence these studies do not rule out the possibility that self-related stimuli can capture attention with simple displays.
In the present study we assessed whether stimuli associated with the self can capture attention with simple displays using a measure of explicit attention based on eye tracking. To do this we had participants carry out a perceptual matching task similar to that previously used to elicit self-bias (Sui et al., 2012), but required eye movements to be made to signal the response. Participants first associated three different shapes (e.g., circle, triangle, square) with three labels (letters represented you, friend, stranger). By having people associate the shapes with social labels, the neutral geometric shapes are “tagged” with social relevance (Sui et al., 2012). This procedure allows us study how basic perceptual processing for that shape changes when compared with other matched shapes (e.g., Sui et al., 2014). They then saw displays in which the label appeared at fixation, and the shape was presented at a peripheral location in either left or right visual field. If the label and the shape matched, the participants had to make an anti-saccade. If the label and shape mismatched, a pro-saccade had to be made. This task, which required voluntary control of attention, allows a precise assessment of the effects of self information on top-down cognitive control. Specifically, comparing the accuracy of eye-movements for conditions that involved competition between a reflexive saccadic response and the voluntary inhibition of the reflexive tendency (on anti-saccade trials) enabled us to test: (a) whether it is more difficult to make an anti-saccade to self-related than to other-related stimuli; (b) what the cost is of voluntary attentional control for self compared to friend and stranger associations.
Experimental study
Method
Participants
Thirty-four healthy participants (18 to 35 years, M = 25.18, SD = 5.32) were recruited for the study. Data from one participant were excluded from later analyses due to data lost for three out of five experimental blocks. The remaining sample of 33 participants (18 females) were right-handed, had normal or corrected-to-normal vision, and had no neurological history. This experiment was approved by Central University of Oxford Research Ethics Committee (CUREC). All participants provided informed consent.
Materials
The experiment comprised a 3 shapes × 3 labels × 2 conditions (matching, mismatching) design. The assignment of label and shape was balanced across participants following a Latin square. The letters Y, F, and S were used to represent “you”, “friend” and “stranger”, respectively, and were paired with three out of four simple geometric shapes (a square, triangle, circle, or diamond). Each display consisted of a dark grey background with a centrally presented white letter and a shape placed on the left or on the right from the letter at visual degrees of 6.8°. Both the letters and the shape were 2.8° in size. Stimulus sets, randomized across participants, consisted of pairings of three of the four geometric shapes with the letters Y, F, and S.
The experiment was implemented and run in Matlab (32-bit version R2012a; the MathWorks, Natick, MA) Psychotoolbox and were conducted with the Tobii TX300 eye tracker, running at 300 Hz on a 22″ screen (51 × 32° of visual angle, 1920 × 1080 pixels, 100 ppi). To minimize head movement and maintain the distance between eyes and the centre of monitor constant (60 cm), participants’ heads were secured in a chin-rest. The eye-tracking system was calibrated for each participant using a 9-point calibration scheme. Verification of each point calibration involved ensuring that fixation of the eye was within 2° of each calibration point. The calibration was repeated if a participant failed to fixate on less than seven out of nine points within the defined accuracy limit. Following the acceptance of the calibration, the task instructions appeared on screen.
Procedure
First, participants were asked to make associations between a shape and self (represented by the letter Y 1 ), a shape and a close friend (represented by the letter F), and a shape and a complete stranger (represented by the letter S). 2 Based on previous studies showing that participants can rapidly gain associations between the shapes and labels just after an association instruction (Sui et al., 2012, 2014), we had participants perform 24 practice trials to reinforce associations between the letters and shapes trials where they indicated their responses (“match” or “mismatch”) by pressing buttons on a keyboard. After each trial they were provided with feedback on accuracy.
After the initial associations, participants were asked to make a saccadic response (“look at the shape presented”) for mismatching pairings between a letter and a shape, and an anti-saccadic response (“look in the direction opposite the shape presented”) for matching trials. The participants then performed eight practice trials followed by five blocks of 48 trials each (24 required anti-saccadic responses) yielding a total of 240 experimental trials. The experiment lasted around 1.2 hours.
Each trial started with a black fixation cross (1° of visual angle) that appeared for 500 ms in the centre of the screen, which participants were instructed to focus on. If a participant was unable to focus on the fixation cross during this time, a display stimulus was not presented, and the trial was recorded as a miss and was added to the end of the block. This criterion was set up to ensure that participants focused at the centre of the screen before the stimuli appeared in order to accurately and consistently record eye-movements. On valid trials, when the fixation cross disappeared, the stimulus pairs were presented for 800 ms followed by a 700-ms blank screen before the start of the next trial (Figure 1). After each block, participants were given a 5-min break.
Trial sequence.
We first flagged data (0.15% across all trials) where the eye tracker lost the eye data momentarily. The flagged data were replaced with last known good value. Blink (3.1%) and missed (0.42%) trials were identified and were excluded from the analysis. Saccade onset was defined as the time point at which the eye moved 30° of visual angle or faster per second (with minimum amplitude of 1.5°). When eye velocity went down again below 30°/s, the time point immediately after this sample was regarded as the termination of the saccade. Saccadic landing positions was specified for pro-saccade as 2° window around a shape, and for anti-saccade as 2° window around mirrored shape location. We excluded 2.2% of anti-saccade trials and 1.34% of pro-saccade trials that were outside the landing positions.
The first saccade after target onset was considered the saccadic response. To exclude anticipatory responses, trials with latencies of saccadic responses less than 80 ms (2.13% of valid saccadic trials) and greater than 500 ms (0%) were excluded from consequent analysis. Each trial was classified by the condition: stimulus type and saccade type (anti-saccade, pro-saccade). There were two measures for the pro-saccade and anti-saccade trials, respectively: saccadic directional accuracy 3 (whether the initial saccade moved in the correct direction) and saccade latency (the difference in time between the presentation of the stimulus to make an eye movement and the beginning of the initial saccade). An incorrect saccade direction was defined as the first saccade after a stimulus onset towards a shape (on anti-saccade trials), or away from the shape (on pro-saccade trials).
Results
Saccadic directional accuracy for anti- and pro-saccade trials
To examine the effect of the stimulus on saccadic directional accuracy in anti-saccade and pro-saccade trials, a two-way repeated measures analysis of variance (ANOVA) was calculated with the factors being the saccade type (anti-saccade, pro-saccade) and stimulus (self, friend, stranger associations). There was a main effect of saccade type, F(1, 32) = 18.56, p < .001, Saccadic directional accuracy for anti-saccades and pro-saccades for self, friend, and stranger stimuli. Error bars represent ±1 SEM. The stars indicate significant differences between a pair of conditions (adjusted for multiple comparisons using Bonferroni correction).
We further examined anti-saccade and pro-saccade saccadic directional accuracy by testing whether correct eye-responses for self, friend, and stranger stimuli differed from chance. One-sample t tests (chance level 50%) showed that saccadic directional accuracy for all stimuli except anti-saccades for self, t(32) = 0.92, were significantly higher than chance level: friend anti-saccades, t(32) = 2.58, p = .017, dz = 0.44, 95% CI for dz [0.29, 0.52]; stranger anti-saccades, t(32) = 10.26, p < .001, dz = 1.78, 95% CI for dz [1.24, 1.93]; self pro-saccades, t(32) = 4.76, p < .001, dz = 0.82, 95% CI for dz [0.67, 1.05]; friend pro-saccades, t(32) = 9.36, p < .001, dz = 1.62, 95% CI for dz [1.47, 1.63]; stranger pro-saccades, t(32) = 6.56, p < .001, dz = 1.14, 95% CI for dz [1.03, 1.37] .
The cost of voluntary control
To verify above self-biases in saccadic directional accuracy and reduce the variations of conditions, the differences between pro-and anti-saccades in saccadic directional accuracy were contrasted for each condition as a measure of the voluntary control of saccades. There was a main effect of stimulus type, F(2, 64) = 7.87, p = .002, Saccadic directional accuracy difference between pro- and anti-saccades. Error bars represent ±1 SEM. The stars indicate significant differences between a pair of conditions (adjusted for multiple comparisons using Bonferroni correction).
Inter-trial effect comparisons
Further analyses were performed to test the cost of voluntary control by examining inter-trial effects of pro-saccades and anti-saccades. We calculated saccadic directional accuracy for correct anti-saccade trials (for the self, friend, and stranger) preceding correct pro-saccade trials (averaged across self, friend, and stranger shapes), along also with saccadic directional accuracy for correct pro-saccade trials for each shape preceding correct anti-saccade trials (averaged across self, friend, and stranger stimuli). A two-way repeated measures ANOVA with preceding saccade (anti-saccade, pro-saccade) and preceding shape type (self, friend, stranger) as within-subject factors was carried out to examine the inter-trial effects. There was a main effect of preceding saccade, F(1, 32) = 76.79, p < .001,
To explore the interaction effect, two separate ANOVAs were performed for each type of preceding saccade. In the first ANOVA we examined whether anti-saccades for self, friend, and stranger on trial N – 1 had a differential effect on saccadic directional accuracy of the next pro-saccade trial (trial N). The results showed significantly lower accuracy in pro-saccade trials preceded by correct self anti-saccades than in pro-saccades preceded by correct anti-saccades to friend stimuli, t(32) = 5.9, p < .001, dz = 1.03, 95% CI for dz [0.66, 1.29], and to stranger stimuli, t(32) = 7.1, p < .001, dz = 1.23, 95% CI for dz [1.02, 1.44]; overall, F(2, 64) = 11.47, p < .001, (a) Inter-trial effects of preceding self-, friend-, and stranger-anti-saccades on the accuracy of the next pro-saccade trials; and (b) effects of preceding self-, friend-, and stranger-pro-saccades on the directional accuracy of the next anti-saccade trial. Error bars represent ±1 SEM. The stars indicate significant differences between a pair of conditions (adjusted for multiple comparisons using Bonferroni correction).
Saccade latencies
In addition to measuring saccadic directional accuracy, we calculated the latency of the first correct and incorrect eye-movement.
4
First, a two-way repeated measures ANOVA was carried out to examine the effect of saccade type (anti-saccade, pro-saccade) and stimulus (self, friend, stranger) on latency of correct responses (Figure 5). There was a main effect of saccade type, F(1, 32) = 12.95, p = .001, Latency for correct anti-saccade and pro-saccade trials. Error bars represent ±1 SEM.
Then, a similar two-way repeated measures ANOVA was performed on the latency of incorrect responses. There were no main effects of saccade type, F(1, 32) = 0.25, or stimulus, F(1, 32) = 0.404. However, the interaction between saccade type and stimulus, F(2, 64) = 3.26, p = .044, Latency for incorrect anti-saccade and pro-saccade trials. Error bars represent ±1 SEM.
Our inter-trial comparisons showed significantly lower accuracy in pro-saccade trials preceded by correct self anti-saccades than in pro-saccades preceded by other-anti-saccades (Figure 4). If the preceding correct self anti-saccade required more control (inhibition) to be generated, we may expect that the subsequent pro-saccades are also delayed. 5 To test this assumption, we calculated saccadic latency for correct anti-saccade trials (for the self, friend, and stranger) preceding correct pro-saccade trials (averaged across self, friend, and stranger shapes), along with saccadic latency for correct pro-saccade trials for each shape preceding correct anti-saccade trials (averaged across self, friend, and stranger stimuli).
A two-way repeated measures ANOVA with preceding saccade (anti-saccade, pro-saccade) and preceding shape type (self, friend, stranger) as within-subject factors was carried out to examine the inter-trial effects on saccadic latency (Figure 7). There was a main effect of preceding shape type, F(2, 64) = 4.4, p = .016, (a) Inter-trial effects of preceding self-, friend-, and stranger-anti-saccades on the latency of the next pro-saccade trials; and (b) effects of preceding self-, friend-, and stranger-pro-saccades on the latency of the next anti-saccade trial. Error bars represent ±1 SEM. The stars indicate significant differences between a pair of conditions (adjusted for multiple comparisons using Bonferroni correction).
General Discussion
In line with previous findings, our results show that anti-saccadic responses elicit more saccadic errors than pro-saccades (Barton, Raoof, Jameel, & Manoach, 2006; Hallett, 1978; Nieuwenhuis, Broerse, Nielen, & de Jong, 2004). More pertinently, however, this effect was stronger for self than friend stimuli. In addition, relative to stranger stimuli, self stimuli showed a significantly greater contrast between pro- and anti-saccade trials and anti-saccades only for self-stimuli. Furthermore, our data showed greater disruption on pro-saccades when the preceding trial required an anti-saccade to self-stimuli than with other stimuli. Taken together, both findings indicate (a) an increase in the relative difficulty of anti-saccades for self-related items, (b) larger disruption in pro-saccade responses (in both saccadic directional accuracy and latency) preceded by self anti-saccades. This latter finding may stem from the requirement to inhibit a pro-saccade to the self stimulus when an anti-saccade was required on trial N – 1, which may then make it more difficult to generate a pro-saccade on trial N. Furthermore, our data of the latency of incorrect responses showed that only for “self” condition were incorrect responses for the anti-saccade trials significantly shorter than those for pro-saccade trials. The results suggest that there are qualitative differences in the ability to generate an anti-saccade away from a stimulus associated with the self compared with other people. Note that this result is highly unlikely to reflect the difficulty of matching self-related stimuli since a strong advantage for matching self-related items has been consistently demonstrated in the literature (e.g., Sui et al., 2012).
Current views on anti-saccadic responses (e.g., Barton et al., 2006; Munoz & Everling, 2004) propose an involvement of two processing stages that are linked to two independent mechanisms: (a) suppressing an automatic pro-saccade eye-movement and (b) generating instead a voluntary saccade in the opposite direction. Presumably, the higher cost of voluntary control for self-associated responses may reflect either or both stages together. If the cost of voluntary control is linked to the suppression of reflexive saccades (e.g., if the inhibition of pre-target activity for the self takes longer than that for others due to stronger attentional capture), but the generation of a voluntary saccade remains the same for all stimuli, then we should expect longer latencies for self anti-saccades than for others. However, we found no significant differences in correct anti-saccade latency between the conditions. On the other hand, there may need to be more inhibition of the pro-saccade to self stimuli, even if the time to generate effective inhibition (e.g., to enable an anti-saccade to be made) does not differ for the contrasting stimuli. It would then follow that it may be more difficult to generate a subsequent pro-saccade, since the pro-saccade will be strongly inhibited. This is what we observed. Our results on inter-trial effects support the idea that there is greater suppression of pro-saccades to make an anti-saccade to a self-related stimulus. We also observed longer latency on pro-saccades preceded by self anti-saccades. This finding further supports the idea that the correct anti-saccades for self-associated stimuli require more control (inhibition) to be generated.
In contrast to the effects of self-related stimuli, anti-saccades to friend-related stimuli were easier to make than those to stimuli related to a stranger. This result is interesting since it indicates a dissociation between effects of self-association (disrupting anti-saccades) and effects of familiarity (the association to a friend). We propose that familiar stimuli (friend associations) are processed more efficiently than unfamiliar stimuli (stranger associations) and exert a reduced cost on difficult responses (anti-saccades). Self-related stimuli, on the other hand, cue attention to themselves, disrupting anti-saccades. Our results provide strong support for the attentional draw of self-associated stimuli and suggest that voluntary saccadic responses away from a self-related item demand greater inhibition than is the case for stimuli related to other people.
In addition, a recent eye-tracking study (Siebold, Weaver, Donk, & van Zoest, 2016) using an oculomotor visual search paradigm did not find the effects of self on an early perceptual level. In this study, participants were required to form associations between two orientation lines (right- and left-tilted) and two labels (“you” and “stranger”), and then make a speeded eye-movements to one of the two lines without any task instruction (Experiment 1), to dot probe target located on one of the two lines (Experiment 2), or to the line that was cued by its associated label (Experiment 3). Similar to our finding, there was no effect of self-prioritization on involuntary eye-movements. Furthermore, the study suggests that top-down information affects detection responses, but not localization of socially salient stimuli, providing indirect support for our conclusion that once a “self-cued” stimulus is detected, it is more difficult to control involuntary eye-movement.
The results of the present study raise an important question for further studies: what the relationship is between individual differences in the magnitude of self-biases (i.e., the response time advantage for stimuli associated with self compared to other) and the cost of voluntary control of saccades for self. If the magnitude of the facilitation effect for self-associated stimuli can predict the cost of self-related attentional control, this may have important implications in neuropsychological studies—for example, in patients with lesions affecting brain regions shown to be related to self-biases (van den Bos & Güroğlu, 2009) and the frontal eye fields (Machado & Rafal, 2004), and in patients suffering from depression (Koenigs & Grafman, 2009).
Disclosure statement
No potential conflict of interest was reported by the authors).
Funding
This work was supported by an Advanced Investigator grant from the 10.13039/501100000781 European Research Council [grant number Pepe: 323883] to the G.W. Humphreys.
Footnotes
1.
Our previous work showed that there were no significant differences in the magnitude of self-biases elicited by an association to a shape with a label “you” or a label “me” or “myself”.
2.
For each participant we carefully checked whether the participant's initials might interfere with the labels. Two out of 34 participants’ surnames started with the letter “S”, and one participant's first name started with the letter “F”. We found no evidence for interference.
3.
We emphasize here the term “saccadic directional accuracy” to avoid confusion with saccadic accuracy, which typically refers to the spatial error between intended and actual saccade landing point.
4.
We also examined the effect of stimuli on the amplitude of saccadic responses, but no significant differences were found between the conditions.
5.
Thanks to the anonymous reviewer who pointed this out.