Seeing It Both Ways: Using a Double-Cuing Task to Investigate the Role of Spatial Cuing in Level-1 Visual Perspective-Taking

John Michael University of Warwick and Central European University

Thomas Wolf and Clément Letesson Central European University

Stephen Butterfill University of Warwick

Joshua Skewes Aarhus University

Jakob Hohwy Monash University

Previous research using the dot-perspective task has produced evidence that humans may be equipped with a mechanism that spontaneously tracks others’ gaze direction and thereby acquires information about what they can see. Other findings, however, support the alternative hypothesis that a spatial-cuing mechanism underpins the effect observed in the dot-perspective task. To adjudicate between these hypotheses, we developed a double-cuing version of Posner’s (1980) spatial-cuing paradigm to be implemented in the dot-perspective task, and conducted 3 experiments in which we manipulated stimulus-onset asynchrony, as well as secondary task demands. Crucially, the 2 conflicting hypotheses generated divergent patterns of predictions across these experimental conditions. Our results support the hypothesis of an automatic perspective-taking mechanism.

Public Significance Statement

Recent research has revealed evidence that humans are equipped with a perspective-taking mechanism that spontaneously tracks others’ gaze direction and thereby acquires information about what others can see. This research has been controversial, however, with critics arguing that the evidence in question can also be interpreted by appealing to general attentional mechanisms without postulating a perspective-taking mechanism. To adjudicate between these 2 competing theoretical positions, we conducted a series of experiments for which the 2 positions generated conflicting predictions. Our findings support the hypothesis of a perspective-taking mechanism that is distinct from general attentional mechanisms.

Keywords: Level-1 visual perspective-taking, theory of mind, attention, spatial cuing, implicit processing

The ability to track others’ gaze direction and to infer what they can see (a process often referred to as Level-1 visual perspective-taking; Flavell, Everett, Croft, & Flavell, 1981) is an important component of human social cognition. It enables us to acquire information about others’ mental states (e.g., what they want or intend), and thereby helps us to anticipate their actions, and to communicate and coordinate fluently with them. Previous research using the dot-perspective task (Samson, Apperly, Braithwaite, Andrews, & Bodley Scott, 2010; Qureshi, Apperly, & Samson, 2010) has produced evidence that humans are equipped with a mechanism that spontaneously performs Level-1 visual perspective-taking.

In the dot-perspective task, participants view an image of a room with an avatar standing in the middle and facing either left or right (this is varied from one trial to the next). On each trial, anywhere from zero to three red dots are displayed on the walls of the room—sometimes all on one side, sometimes distributed between both sides. On half of the trials, the avatar can see all of the dots (e.g., the avatar is facing to the left; three dots appear on the left wall and none appear on the right wall), so the perspectives of the participant and the avatar are consistent (i.e., consistent trials). On the other half of the trials, the avatar can see some but not all of the dots (e.g., the avatar is facing to the left; one dot appears on the left and one on the right), or none of the dots (i.e., all of the dots are on the wall behind the avatar). On these trials, the perspectives of the participant and the avatar are inconsistent (i.e., inconsistent trials). Participants have the task of calculating either how many dots the avatar can see (i.e., other trials), or how many dots they themselves see (i.e., self-trials). A main finding is that participants perform worse on inconsistent self-trials than on consistent selftrials. The authors conclude that participants calculate the avatar’s perspective even on trials for which they need only calculate their own (namely, self-trials), and that computing the avatar’s perspective interferes with reporting their own (Samson et al., 2010). This is an altercentric interference effect: Another’s task-irrelevant perspective impairs performance.

In a follow-up study using the same paradigm, Qureshi et al. (2010) exposed participants to an additional cognitive load, and found that the interference from inconsistent perspectives increased. The authors interpret this as evidence that participants automatically calculated the avatar’s perspective (Level-1 perspective-taking) parallel to the calculation of what they themselves could see. In contrast, the selection of a perspective to draw upon in forming a response is a controlled process requiring executive resources, and was therefore impaired by the cognitive load manipulation. This pattern of results suggests an automatic mechanism for Level-1 visual perspective-taking.

However, as Heyes and colleagues have pointed out (Heyes, 2014; Santiesteban, Catmur, Hopkins, Bird, & Heyes, 2014), it is possible that this task does not tap a mechanism for Level-1 visual perspective-taking but, rather, spatial cuing, with the avatars serving as cues to facilitate attentional processing either on the left or the right side. Indeed, this interpretation is supported by the findings of Santiesteban et al. (2014), who replicated Samson et al. (2010)’s effect using arrows instead of avatars.

Although Santiesteban and colleagues’ (2014) findings are consistent with the hypothesis that a spatial-cuing mechanism underpins the effects observed in the dot-perspective task, their results are not decisive. First, it is possible that participants’ prior experiences with arrows lead them to interpret the arrows as indicating an implied perspective (e.g., the perspective of an agent who places an arrow and/or the perspective of an agent who looks in the direction indicated by an arrow). Second, Santiesteban and colleagues’ results do not rule out the possibility that the version of the paradigm with arrows taps a different underlying mechanism that leads to a similar pattern of findings in the particular circumstances of these experiments. To rule this out, it would be important to specify experimental conditions in which the two hypothesized mechanisms should lead to dissimilar patterns of findings. Indeed that is what we accomplished in the present study, as we explain below.

There are at least two reasons that adjudication between these two competing hypotheses is important. The first is that they lead to conflicting views of the relevance of the dot-perspective task for disorders of social cognition such as autism. For example, if the task indeed taps a mechanism for automatically calculating others’ perspectives, then one may expect autistic participants not to exhibit the altercentric interference effect, at least insofar as one is persuaded by evidence from previous research suggesting that autistic persons tend not to spontaneously engage in perspectivetaking or other forms of mindreading, and must instead mobilize conscious cognitive effort to do so (Schneider, Slaughter, Bayliss, & Dux, 2013; Senju, Southgate, White, & Frith, 2009; Hamilton, Brindley, & Frith, 2009). And yet, what is interesting is that Schwarzkopf, Schilbach, Vogeley, and Timmermans (2014) did observe the altercentric interference effect in autistic participants. Thus, if the perspective-taking hypothesis is correct, then this finding indicates a need to reconsider our understanding of perspective-taking in autistic persons. If, however, the task does not tap a perspective-taking mechanism, then Schwarzkopf and colleagues’ results may not bear so directly on differences between autistic and nonautistic persons’ spontaneous perspective-taking. Instead, Schwarzkopf and colleagues’ result may, in this case, be expected to generalize to spatial-cuing paradigms.

A second reason it is important to adjudicate between these two competing hypotheses is that the dot-perspective paradigm is increasingly relied upon as a tool for investigating the nature and limitations of fast and efficient mindreading processes (Qureshi et al., 2010; Furlanetto, Becchio, Samson, & Apperly, 2016; Schwarzkopf et al., 2014), and thus also as a means of testing theories of the cognitive architecture of mindreading (Butterfill & Apperly, 2013; Christensen & Michael, 2016; Westra, 2016). If, however, the paradigm does not tap a perspective-taking process, then it may be necessary to reevaluate the uses of it.

To investigate whether the effects observed in the dotperspective task are due, at least in part, to spatial cuing, we adapted Maylor’s (1985)’s “double-cuing” task (see also Posner & Cohen, 1984). In the basic spatial-cuing task (Posner, 1980), participants are instructed to detect the appearance of a target either on the left or the right side of a screen. In the double-cuing version of this task, the appearance of the target is preceded by two simultaneous peripheral cues, one on the left and one on the right. The main finding is that target detection on either side is facilitated: Regardless of which side the target appears on, target detection is faster than in a baseline condition with no cue (Maylor, 1985, exp. 3; Posner & Cohen, 1984). The authors concluded that attention can be concurrently facilitated at two locations.

Building upon this result, we reasoned that if a spatial-cuing mechanism underlies the effect observed in the dot-perspective task, then it should also be possible for two avatars facing in different directions (left and right) to cue attention to two different locations simultaneously, and thereby facilitate attentional processing at both locations. Hence, on a version of the dotperspective task in which all test trials involve dots on both walls, participants should perform better on trials with two avatars (one facing in each direction) than on trials with just one avatar or with no avatar for the following reason: If spatial cuing underlies the effect, the leftward and rightward orientations of two avatars should facilitate attentional processing at both locations (left and right), whereas the directional orientation of one avatar would only facilitate processing of objects at one location (left or right), and a room with no avatar would provide no facilitation. Of note, this hypothesis does not generate a clear prediction about whether performance should be better on trials with one or zero avatars.

This is because when there is one avatar, the disks on one side of the room are cued, but the disks on the other side are uncued, which may inhibit processing of the uncued side. If so, it is possible that these two influences may cancel each other out. It is therefore unclear what the net effect of these two influences might be.

In contrast, if the effect observed in the dot-perspective task is driven by a mechanism for perspective-taking, then one should predict a different pattern of findings. On this account, the observed effect results from the interference of an inconsistent perspective rather than from facilitation of attentional processing. Thus, given that all trials involve dots on both walls, and assuming that the perspective-taking mechanism can compute two avatars’ perspectives, performance should be no better on trials in which there are two avatars (each with a perspective that is inconsistent with the perspective of the participant) than on trials with one avatar with just one perspective that is inconsistent with that of the participant or on trials with no avatar (i.e., no inconsistent perspective). This hypothesis also predicts that performance should be worse on trials with one avatar than on trials with zero avatars: Even just one inconsistent perspective should interfere with performance.

To test these conflicting predictions, we ran three separate experiments with different participants. In each experiment we asked the same question: When there are two avatars rather than just one in self-inconsistent trials, is performance better or worse? In Experiment 1, we created a situation most likely to reveal a cuing effect by including a delay of 800 ms after the avatar(s) appeared and before the disks appeared on the walls. The inclusion of such a stimulus-onset asynchony (SOA) has been shown to be necessary in some gaze-cuing paradigms (Friesen & Kingstone, 1998; Driver et al., 1999; Frischen, Bayliss, & Tipper, 2007; Xu, Tanaka, & Mineault, 2012), including a gaze-cuing paradigm that used the stimulus material from the dot-perspective task and in which the overall latencies were matched with the dot-perspective task (Bukowski, Hietanen, & Samson, 2015). The inclusion of an SOA may therefore be necessary to observe the facilitative effect of a spatial-cuing mechanism.

However, the standard dot-perspective task (i.e., as in Samson et al., 2010) does not include an SOA. Further, the inclusion of an SOA, as in Experiment 1, may mask the effects of a perspectivetaking mechanism by providing participants with extra time to allocate their attention in accordance with the directional information extracted from the stimuli. Thus, if the results from Experiment 1 were to reveal that performance was better with two avatars than with one (as we would expect if a spatial-cuing mechanism were at work), it would be an indication that the stimuli and the number-verification task used in the dot-perspective paradigm can elicit a spatial-cuing mechanism, but it would not settle the question of whether a spatial-cuing mechanism is responsible for the altercentric effect observed in the standard, no-SOA dotperspective task. We therefore carried out Experiment 2, which differed from Experiment 1 only in that there was no SOA. We reasoned that if cuing is responsible for how the avatar affects performance in the standard dot-perspective task, then because there is no SOA in that task, the cuing mechanism should also be activated in this version of our task with no SOA. In that case, we should observe better performance with two avatars than with one. If, on the other hand, we were to observe worse performance with two avatars than one, it would be consistent with the hypothesis of a perspective-taking mechanism.

Experiment 3 was designed to provide a more stringent test of the hypothesis of an automatic perspective-taking mechanism. We reasoned as follows: Suppose perspective-taking really does occur when there is an SOA (as in Experiment 1), but its effects are masked because the SOA provides participants with extra time to allocate their attention. In that case, it should be possible to unmask its effects by instructing participants to perform a secondary task designed to tax the executive and thereby to interfere with the operation of a spatial-cuing mechanism. That is, combining the SOA with a secondary task should mean that performance is worse with two avatars than one if the perspective-taking hypothesis is correct. Hence, the pattern of results from Experiment 3 should more closely resemble the pattern from Experiment 2 than the pattern from Experiment 1 (Tables 1a and 1b summarize the different patterns of predictions generated by the perspectivetaking hypothesis and the spatial-cuing hypothesis).

Experiment 1

Method

Participants. In determining the appropriate sample size, we used Samson et al. (2010) as our starting point. In Samson and colleagues’ study, each of the three experiments included a sample size of 16. The design we used here most closely resembles their Experiment 3, because (a) participants in that experiment (as in our design) were only ever asked to calculate how many dots they themselves could see, and to ignore the distractor in the middle of the room, and (b) they varied the type of distractor in the middle of the room (i.e., a rectangle or an avatar). Based on the effect sizes observed by Samson and colleagues in their Experiment 3 $(\upeta_{\mathrm{p}}^{2}=$ .186 and $\eta_{\mathrm{p}}^{2}=.285)$ , we determined that for $80\%$ statistical power and with an $\upalpha$ level of .05, the appropriate sample size for our study would be 20. We therefore recruited 20 participants (11 women; age range $=18-24$ , $M=20.75$ , $S D=1.71$ ) from student organizations in the Budapest area, all of whom received gift vouchers for their participation. All were naïve to the purpose of the study, reported normal or corrected-to-normal vision, and signed informed consent prior to the experiment. The experiment was conducted in accordance with the Declaration of Helsinki and was approved by the United Ethical Review Board for Research in Psychology (EPKEB; World Medical Association, 2001).

Table 1 1a. The Two Competing Hypotheses

Note. SOA $=$ stimulus-onset asynchony.

Procedure. PsychoPy software was used to control the stimulus presentation and data collection (Peirce, 2007). As in Samson et al. (2010), the stimuli consisted of a picture showing a lateral view into a room with the left, back, and right walls visible and with red dots displayed on one or two walls. Images of a human avatar were produced from the image files used by Samson et al. (2010), and were positioned in the center of the room. On one third of all trials, there was one avatar facing either the left or the right wall. On one third of all trials, there were two avatars, one facing left and one facing right. On one third of all trials, there was no avatar present (See Figure 1).

Following the display of a fixation cross for $500~\mathrm{ms}$ , a digit (0 – 4) appeared for $750~\mathrm{ms}$ , which specified a target number of dots for the participant to verify. The image of the room then appeared with the dots on the walls and with zero, one, or two avatars in the center, followed by an 800-ms SOA before the dots appeared. This image remained until a response was given (or until

2,000 ms passed). On matching trials (those with “yes” responses), the digit specifying the target number corresponded to the number of dots on the walls. On mismatching trials (those with “no” responses), the digit specified a number either one higher than or one lower than the number of dots on the walls (See Figure 2).

Female participants were presented with female avatars and male participants with male avatars. On all test trials, there was at least one dot on each wall, and never more than three on any wall, or more than four in total. As a result, the participant’s perspective was always inconsistent with that of the avatar(s): For every avatar that appeared, there was always at least one dot behind it and one in front of it.

There were 48 matching (“yes”-response) and 48 mismatching (“no”-response) trials for each condition (i.e., 0-avatar, 1-avatar, and 2-avatar conditions). On mismatching trials, the digit presented at the beginning of the trial sometimes corresponded to the number of disks visible from an avatar’s perspective (i.e., in the 1-avatar and 2-avatar conditions), making such trials particularly difficult. Because the frequency of such trials differed among the three conditions, we followed Samson and colleagues’ (2010) procedure in treating mismatching trials as fillers, and analyzed only matching trials. Thus, there were 144 test trials, 48 per condition. We also included 27 additional filler trials in which there were dots on only one wall so that “1” would sometimes be the correct response, and participants could not reliably anticipate whether there would be dots on both walls. These additional filler trials included an equal number of 0-avatar, 1-avatar, and 2-avatar trials. The trials were divided into three blocks of 105 trials (48 test trials and 57 filler trials) and were preceded by a block of 26 practice trials. The order of the trials within each block was pseudorandomized and then fixed across participants so that there were no more than three consecutive trials of the same type.

Figure 1. Examples of stimuli used in the three conditions: zero, one, or two avatars. See the online article for the color version of this figure.

Figure 2. Trial structure: For this example, the correct response was “yes.” See the online article for the color version of this figure.

Results and Discussion

To control for speed–accuracy trade-offs, reaction times (RTs) and hit rates (HRs) for correct responses were also merged into inverse-efficiency scores (IES), a combined measure which homogenizes different patterns of speed–accuracy trade-offs within a group $\mathrm{(IES~=~RT/HR}$ ; Townsend & Ashby, 1978). Because the calculation of IES entails that RTs are quasi-exponentially multiplied as the HR decreases, Bruyer and Brysbaert (2011) recommended not using the IES unless the mean HR within a group is above $90\%$ . In our sample, the mean HR was above $90\%$ in all three conditions, indicating that it was appropriate to use IES for the primary analysis. Further below, we also include analyses of the RTs and HRs.

In calculating mean RTs, response omissions due to the timeout procedure $0.31\%$ of the data) and erroneous responses $3.94\%$ of the data) were eliminated from the data set. We also removed trials with responses that were more than $2.5~S D s$ greater or less than the mean for each participant for each condition $2.83\%$ of the data).

IES analysis. We performed a three-way analysis of variance (ANOVA) for IES, which revealed a significant main effect of number of avatars, with performance being better in the 2-avatar condition, $M=661.65$ , $S D=128.38$ , than in the 0-avatar condition, $M=678.05$ , $S D=130.91_{.}$ ) and the 1-avatar condition, $M=$ $681.20S D=134.75$ ; $F(2,18)=5.30,p=.009$ , $\upeta_{\mathrm{p}}^{2}=0.218$ . This is consistent with the operation of a spatial-cuing mechanism. Planned contrast analyses revealed a significant difference between the 2-avatar condition and the 1-avatar condition, with performance being significantly better in the former than in the latter, $t(19)=3.51$ , p=.002\ , $d=0.149$ , and also between the 2-avatar condition and the 0-avatar condition, with performance being significantly better in the former than in the latter, $t(19)=$ 2.63, $p=.016$ , $d=0.127$ . Both of those results are consistent with the hypothesis of a spatial-cuing mechanism and not with the hypothesis of a perspective-taking mechanism. There was no significant difference between the 1-avatar and the 0-avatar conditions, $t(19)=0.424$ , $p=.676$ , $d=0.024$ (See Figure 3).

RT analysis. We performed a three-way ANOVA for RTs, which revealed a significant main effect of number of avatars, with performance being better in the 2-avatar condition, $M=634.35$ , $S D=117.26$ , than in the 0-avatar condition, $M=652.11$ , $S D=$ 122.94, and the 1-avatar condition, $M=649.205D=130.42,F(2,$ $18)~=~5.86$ , $p=.006$ , $\upeta_{\mathrm{p}}^{2}=0.236$ . This is consistent with the operation of a spatial-cuing mechanism (see Figure 4).

Figure 3. IES: Error bars represent the within-subject confidence intervals, following the method proposed by Cousineau, 2005; cf. Loftus and Masson, 1994. Symbols indicate significance level: $\mathrm{~\textbar{~}{~p~}~}<.05$ . $^{***}p<.01$ . $^{3\ast=\ast\ast\ast}p<$ .001. See the online article for the color version of this figure.

Figure 4. RTs: Error bars represent the within-subject confidence intervals, following the method proposed by Cousineau, 2005; cf. Loftus and Masson, 1994. Symbols indicate significance level: $\mathrm{~\textbar{~}{~p~}~}<.05$ . $^{***}p<.01$ . $^{3\ast=\ast\ast\ast}p<$ .001. $n s=$ nonsignificant. See the online article for the color version of this figure.

Planned contrast analyses revealed a significant difference between the 2-avatar condition and the 0-avatar condition, with performance being significantly better in the former than in the latter, $t(19)=3.10$ , $p=.006$ , $d=0.148$ , and also between the 2-avatar condition and the 1-avatar condition, with performance being significantly better in the former than in the latter, $t(19)=$ 2.49, $p=.022$ , $d=0.127$ . Both of those results are consistent with the hypothesis of a spatial-cuing mechanism, and not with the hypothesis that a perspective-taking mechanism, underlying the effects of the avatars on performance in this task. There was no significant difference between the 1-avatar and the 0-avatar conditions, $t(19)=0.588$ , $p=.56$ , $d=0.024$ .

Accuracy analysis. We performed a three-way ANOVA for HRs, which revealed no significant differences between the 2-avatar condition, $M=96.11\%$ , $S D=3.44\%$ , the 0-avatar condition (M=96.28\ , $S D=2.33,$ ), and the 1-avatar condition, $95.37\%$ $S D=4.28\%$ , $F(2,18)=1.33$ , $p=.275$ , $\upeta_{\mathrm{p}}^{2}=0.236$ (see Figure 5).

Planned contrast analyses revealed no significant differences among the conditions: neither between the 2-avatar condition and the 0-avatar condition, $t(19)=.377$ , p=.71\ , $d=0.058$ , nor between the 2-avatar condition and the 1-avatar condition, $t(19)=$ 1.3, $p=.211$ , $d=0.19$ , nor between the 1-avatar and the 0-avatar condition, $t(19)=.126,p=.224,d=0.263$ .

Experiment 2

The results of Experiment 1 indicate that the stimuli and the number-verification procedure used in the dot-perspective task can be used to trigger a spatial-cuing mechanism, and that attention can be facilitated at two locations with avatars oriented in opposite directions. This is consistent with the operation of a spatial-cuing mechanism. However, it would not be justified to conclude that the standard dot-perspective task taps a spatial-cuing mechanism and not a perspective-taking mechanism. This is because Experiment 1, like gaze-cuing paradigms but unlike the standard dotperspective task, included an SOA. We do not know what effect an SOA might have had on how perspective-taking processes influence judgments. To support the view that a spatial-cuing mechanism underlies performance on the standard dot-perspective task, it would be necessary to observe the same pattern of findings with no SOA. The aim of Experiment 2 was to do just that.

Method

Participants. Twenty participants (12 women; age range $=$ 20 –31, $M=25.67$ , $S D=3.16.$ ) were recruited from student organizations in the Budapest area, and received gift vouchers for their participation. A statistical power analysis for a one-way repeated measures ANOVA with three levels using $\mathrm{GPower}3.1$ (Faul, Erdfelder, Buchner, & Lang, 2009) confirmed that, based upon the effect size $(\upeta_{\mathrm{p}}^{2}=0.218)$ observed in Experiment 1 (and for $80\%$ statistical power and an $\upalpha$ level of .05), 20 was the appropriate sample size for Experiment 2. For the analyses, we excluded the data from two participants: One had an IES more than $3~S D$ greater than the mean for the group, and the other failed to complete the task. All were naïve to the purpose of the study, reported normal or corrected-to-normal vision, and signed informed consent prior to the experiment. The experiment was conducted in accordance with the Declaration of Helsinki and was approved by the EPKEB.

Procedure. The procedure was the same as in Experiment 1, with one exception: When the image of the room with zero, one, or two avatars appeared, the dots on the walls appeared simultaneously, that is, there was no SOA. The image remained until a response was given (or until 2,000 ms passed), as in Experiment 1.

Figure 5. HRs: Error bars represent the within-subject confidence intervals, following the method proposed by Cousineau, 2005; cf. Loftus and Masson, 1994. Symbols indicate significance level: $\mathrm{~\textbar{~}{~p~}~}<.05$ . $^{***}p<.01$ . $^{3\ast=\ast\ast\ast}p<$ .001. $n s=$ nonsignificant. See the online article for the color version of this figure.

Results and Discussion

As for Experiment 1, in calculating mean RTs, response omissions due to the timeout procedure $0.17\%$ of the data) and erroneous responses $3.86\%$ of the data) were eliminated from the data set. We also removed trials with responses that were more than $2.5~S D s$ greater or less than the mean for each participant for each condition $2.78\%$ of the data).

IES analysis. We then performed a three-way ANOVA for IES, which revealed a significant main effect of number of avatars, with performance being worse in the 2-avatar condition, $M=607.87$ , $S D=120.80$ , than in the 1-avatar condition, $M=592.87$ , ${\boldsymbol{S D}}=$ 122.85, or the 0-avatar condition, $M=581.02$ , $S D=111.49$ , $(F(2,$ $16)=4.47$ , $p=.019$ , $\upeta_{\mathrm{p}}^{2}=0.208$ (see Figure 3). A planned contrast analysis revealed a significant difference between the 2-avatar condition and the 0-avatar condition, with performance being worse in the 2-avatar condition than in the 0-avatar condition, $t(17)=4.79$ , $p<$ .001, $d=0.23$ . These results are consistent with the hypothesis of a perspective-taking mechanism and are difficult to account for by appealing to the operation of a spatial-cuing mechanism. There was no significant difference between the 2-avatar condition and the 1-avatar condition, $t(17)=1.27$ , $p=.221$ , $d=0.123$ , nor between the 1-avatar and the 0-avatar conditions, $t(17)=1.639$ , $p=.118$ , $d=$ 0.1 (see Figure 3).

Reaction-time analysis. We performed a three-way ANOVA for RT, which revealed no significant difference among the 2-avatar condition, $M=582.00$ , $S D=116.66$ , the 1-avatar condition, 572.23\ $S D=114.45$ , and the 0-avatar condition, $M=568.14$ , $S D=113.15$ , $F(2,16)=2.41$ , $p=.104$ , $\upeta_{\mathrm{p}}^{2}=0.124$ (see Figure 4).

Planned contrast analyses revealed a significant difference between the 2-avatar condition and the 0-avatar condition, with performance being worse in the 2-avatar condition than in the 0-avatar condition, $t(17)=2.90$ , p=.01\ , $d=0.23$ . This result is consistent with the hypothesis of a perspective-taking mechanism and not with the hypothesis that a spatial-cuing mechanism underlies the effects of avatars on performance in this task. The difference between the 2-avatar condition and the 1-avatar condition did not reach significance, $t(17)=1.26,p=.23,$ $d=0.093$ ; nor did the difference between the 1-avatar and the 0-avatar conditions, $t(17)=0.60,p=.56,$ $d=0.04$ .

Accuracy analysis. We performed a three-way ANOVA for HRs, which revealed no significant differences among the 2-avatar condition, $M=95.94\%$ , $S D=2.65\%$ , the 0-avatar condition, $M=$ $97.76\%$ , $S D=2.84\%$ , and the 1-avatar condition, $96.68\%$ ${\boldsymbol{S D}}=$ $2.95\%$ , $F(2,16)=2.24$ , $p=.112$ , $\upeta_{\mathrm{p}}^{2}=0.236$ (see Figure 5).

Planned contrast analyses revealed a significant difference between the 2-avatar condition and the 0-avatar condition, with performance being worse in the 2-avatar condition than in the 0-avatar condition, $t(17)=3.04,p=.007,d=0.663$ . This result is consistent with the hypothesis of a perspective-taking mechanism and not with the hypothesis that a spatial-cuing mechanism underlies the effects of avatars on performance in this task. The difference between the 2-avatar condition and the 1-avatar condition did not reach significance, $t(17)=0.78$ , $p=.446$ , $d=0.262$ ; nor did the difference between the 1-avatar and the 0-avatar conditions, $t(17)=1.08$ , $p=.29$ , $d=0.375$ .

Experiment 3

In Experiment 2, the pattern of results we found in Experiment 1 was reversed. The reversal indicates that a perspective-taking mechanism, rather than a spatial-cuing mechanism, may underpin performance on standard dot-perspective tasks. However, to be at all confident in this interpretation we had to further investigate the effects of the SOA. Taken at face value, the results of Experiments 1 and 2 suggest that there may be two separate mechanisms at work, with the perspective-taking mechanism predominating earlier and the spatial-cuing mechanism predominating later. If so, it may be possible to selectively intervene on the spatial-cuing mechanism and prolong the effects of the perspective-taking mechanism.

To test whether this is indeed possible, we reintroduced the 800-ms SOA in Experiment 3, but also instructed participants to concurrently perform a secondary task. We reasoned that the secondary task might interfere with the spatial-cuing mechanism because there is evidence that attention shifts in response to central cues can be inhibited through the use of concurrent secondary tasks to increase processing demands (Jonides, 1981; Müller & Rabbitt, 1989; Frischen et al., 2007). In light of Qureshi et al.’s (2010) finding that the perspective-taking mechanism was not inhibited by the concurrent performance of a secondary task, we predicted that it would predominate in this version of the task, and that we would therefore again observe better performance with one avatar than with two. For the secondary task, we chose an auditory tone-monitoring task, and recorded verbal responses to rule out any visuospatial or motor interference with the dot-perspective task.

Method

Participants. A statistical power analysis for a one-way repeated measures ANOVA with three levels using $\mathbf{G}^{*}$ Power 3.1 (Faul et al., 2009) confirmed that, based upon the effect size $(\boldsymbol{\upeta}_{\mathrm{p}}^{2}=$ 0.218) observed in Experiment 1 (and for $80\%$ statistical power and an $\upalpha$ level of .05), 20 was the appropriate sample size for Experiment 3. Twenty participants (10 women; age range $=21-$ 30, $M=24.77$ , $S D=2.63.$ ) were therefore recruited from student organizations in the Budapest area, and received gift vouchers for their participation.

All were naïve to the purpose of the study, reported normal or corrected to normal vision, and signed informed consent prior to the experiment. The experiment was conducted in accordance with the Declaration of Helsinki and was approved by the EPKEB.

Procedure. The procedure for the dot-perspective task was the same as in Experiment 1. In addition, however, participants performed an additional practice block of trials for the auditory tone-monitoring task, and in all nonpractice trials, they concurrently performed that task. For the auditory task, two tones were presented over a pair of headphones during the $800~\mathrm{{ms}}$ . The first tone was presented $100~\mathrm{{ms}}$ after the image of the room appeared, and the second tone was presented $150~\mathrm{ms}$ later, that is, $250~\mathrm{ms}$ after the appearance of the image of the room with the avatar(s). On half of the trials, the tones were presented at the same pitch. On the other half of the trials, one of the tones was presented at a high pitch and the other at a low pitch. Participants were instructed to give a verbal response (by saying “same” into a microphone) if the two tones were the same, and otherwise to give no response. They were instructed to make their responses (if at all) as quickly as possible, and in any case, before the disks on the wall appeared. Participants first performed one practice block of the dotperspective task, as in Experiments 1 and 2. Then they performed one practice block of the tone-monitoring task alone, without the dot-perspective task. Next, they performed a practice block of the dot-perspective task in conjunction with the tone-monitoring task. After these three practice blocks, they proceeded to the test blocks.

Results and Discussion

As for Experiments 1 and 2, in calculating mean RTs, response omissions due to the timeout procedure $1.12\%$ of the data) and erroneous responses $7.05\%$ of the data) were eliminated from the data set. We also removed trials with responses that were more than $2.5\:S D\mathrm{s}$ greater or less than the mean for each participant for each condition ( $3.18\%$ of the data).

IES analysis. We then performed a three-way ANOVA for IES, which revealed a significant main effect of number of avatars, with performance being worse in the 2-avatar condition, $M=$ 773.56, $S D=261.83$ , than in the 1-avatar condition, $717.37~S D=$ 188.80, or the 0-avatar condition, $M=714.20$ , $S D=190.60$ , $F(2,$ $18)~=~4.71$ , $p~=~.015$ , $\upeta_{\mathrm{p}}^{2}=0.199$ (see Figure 3). A planned contrast analysis of performance in the 2-avatar condition and the 1-avatar condition revealed a marginally significant difference, with performance in the 2-avatar condition being worse than in the 1-avatar condition, $t(19)=2.08$ , $p=.051$ , $d=0.246$ . A planned contrast of performance in the 2-avatar condition and the 0-avatar condition revealed a significant difference between the 2-avatar condition and the 0-avatar condition, with performance being worse in the 2-avatar condition than in the 0-avatar condition, $t(19)~=~2.89$ , $p~=~.009$ , $d=0.259$ . There was no significant difference between the 1-avatar and the 0-avatar conditions, $t(19)=0.194$ , $p=$ .849, $d=0.017$ (see Figure 3).

On the auditory monitoring task, the overall HR for the group was $78.49\%$ . Mean RT (from the initiation of the second tone) was 354.83 ms \langle S D=210.91\ ). We decided not to exclude any participants on the basis of their performance on the secondary task, because it was not possible to infer how participants were distracted from the primary task (i.e., some will have found the task more difficult than others), and the purpose of this exercise was to distract participants from the dot-perspective task.

RT analysis. We performed a three-way ANOVA for RTs, which revealed no significant difference among the 2-avatar condition, $M=655.51$ , $S D=160.30$ , the 0-avatar condition, $M=646.13$ , $S D=153.39$ , and the 1-avatar condition, $M=$ 648.11 $S D=144.46$ , $F(2,18)=0.66$ , $p=.525$ , ${\eta_{\mathrm{p}}^{2}=0.033}$ (see Figure 4).

Planned contrast analyses revealed no significant difference between the 2-avatar condition and the 0-avatar condition, $t(19)=$ 3.10 $\jmath,p=.006$ , $d=0.148$ , nor between the 2-avatar condition and the 1-avatar condition, $t(19)=2.49$ , $p=.022$ , $d=0.127$ , nor between the 1-avatar and the 0-avatar conditions, $t(19)=0.588$ , $p=.56$ , $d=0.024$ .

Accuracy analysis. We performed a three-way ANOVA for HRs, which revealed a significant main effect of number of avatar, with performance being worse in the 2-avatar condition, $M=$ $87.10\%$ , $S D=10.27\%$ , than in the 0-avatar condition, $M=$ $91.27\%$ , $S D=6.49\%$ , and the 1-avatar condition, $91.49\%$ ${\boldsymbol{S D}}=$ $8.02\%$ , $F(2,18)=6.17$ , $p=.005$ , $\eta_{\mathrm{p}}^{2}=0.250$ (see Figure 5).

Planned contrast analyses revealed a significant difference between the 2-avatar condition and the 0-avatar condition, with performance being significantly worse in the former than in the latter, $t(19)=2.83$ , $p=.011$ , $d=0.485$ , and also between the 2-avatar condition and the 1-avatar condition, with performance being significantly worse in the former than in the latter, $t(19)=$ 2.69, $p=.015$ , $d=0.476$ . These result are consistent with the hypothesis of a perspective-taking mechanism and not with the hypothesis that a spatial-cuing mechanism underlies the effects of avatars on performance in this task. There was no significant difference between the 1-avatar and the 0-avatar conditions, $t(19)=0.213,p=.83,d=0.031$ .

General Discussion

The results of Experiment 1 revealed a significant effect of the number of avatars, with participants performing better on trials with two avatars than on trials with one or zero avatars. This is consistent with the operation of a spatial-cuing mechanism. In contrast, they are difficult to account for by appealing to the operation of a perspective-taking mechanism. This confirms that the stimuli and the number-verification procedure used in the dot-perspective task can be used to trigger a spatial-cuing mechanism. However, it would be hasty to draw any conclusions about the mechanisms underpinning the findings from the original dotperspective task, because Experiment 1 differed from the original dot-perspective task in that it included an SOA of $800~\mathrm{{ms}}$ .

For this reason, we conducted Experiment 2, which included no SOA. Here we observed the reverse pattern: Participants performed better on trials with one avatar than on trials with two avatars. This is consistent with the hypothesis that the dotperspective task taps a perspective-taking mechanism, which is distinct from spatial cuing. On the perspective-taking hypothesis, the presence of any avatar with an inconsistent perspective should interfere with the task rather than facilitate it. Thus, given that all trials involved disks on both walls, performance should have been worse on trials in which there were two avatars (each with a perspective that is inconsistent with the perspective of the participant) than on trials with one avatar (with just one perspective that is inconsistent with that of the participant) or on trials with no avatar (in which there is no inconsistent perspective).

The findings from Experiments 1 and 2, taken together, suggest that there may be two separate mechanisms at work, with the perspective-taking mechanism predominating at earlier time points and the spatial-cuing mechanism predominating at later time points. If this is correct, it may be possible to selectively intervene on the spatial-cuing mechanism and prolong the effects of the perspective-taking mechanism. Because the spatial-cuing mechanism depends upon executive function (Jonides, 1981; Müller & Rabbitt, 1989; Frischen et al., 2007), it might, therefore, be suppressed through the imposition of a demanding secondary task. By contrast, there is evidence that the perspective-taking mechanism operates automatically (Qureshi et al., 2010), and would thus be robust under dual-task conditions. Therefore, we conducted Experiment 3, which had an SOA of $800~\mathrm{{ms}}$ , as in Experiment 1, but in which participants were also asked to concurrently perform a secondary task. We predicted that the secondary task would interfere with the spatial-cuing mechanism, and that the perspectivetaking mechanism would therefore predominate on this version of the task, which should have resulted in better performance with zero or one avatar than with two. The results confirm this prediction, with performance being worse with two avatars than with one or zero (see Tables 1a and 1b for a summary of the different patterns of predictions generated by the perspective-taking hypothesis and the spatial-cuing hypothesis).

Taken together, the results of Experiments 1, 2, and 3 suggest that a spatial-cuing mechanism may be engaged in the dotperspective task if an SOA is involved, but that this is unlikely to explain the findings from standard versions of the paradigm that do not include an SOA, which is consistent with the findings of Marotta, Lupiánez, Martella, and Casagrande (2012). Marotta et al. systematically varied not only the locations of targets, but also the objects (i.e., rectangular figures) in which those targets appeared, and were thereby able to show that faces, unlike arrows, trigger a pure location-based cuing effect, whereas arrows, unlike faces, trigger a pure object-based cuing effect. They interpreted this finding as evidence that faces and arrows engage qualitatively different (i.e., location-based vs. object-based) orienting mechanisms.

In sum, our results strongly suggest that the perspective-taking mechanism engaged in the standard dot-perspective task is distinct from the spatial-cuing mechanism. This finding justifies the use of the dot-perspective paradigm as a tool for investigating the nature and limitations of fast and efficient mindreading processes in both neurotypical (Furlanetto et al., 2016; Qureshi et al., 2010) and autistic (Schwarzkopf et al., 2014) populations, and thus also as a means of testing theories of the cognitive architecture of mindreading (Butterfill & Apperly, 2013; Christensen & Michael, 2016; Westra, 2016). Further researchers might investigate how, if at all, this perspective-taking mechanism is modulated by and/or integrated with attentional mechanisms.

References

Bruyer, R., & Brysbaert, M. (2011). Combining speed and accuracy in cognitive psychology: Is the inverse efficiency score (IES) a better dependent variable than the mean reaction time (RT) and the percentage of errors (PE)? Psychologica Belgica, 51, 5–13. http://dx.doi.org/10 .5334/pb-51-1-5 Bukowski, H., Hietanen, J. K., & Samson, D. (2015). From gaze cueing to perspective taking: Revisiting the claim that we automatically compute where or what other people are looking at. Visual Cognition, 23, 1020 –
1042. http://dx.doi.org/10.1080/13506285.2015.1132804 Butterfill, S. A., & Apperly, I. A. (2013). How to construct a minimal theory of mind. Mind & Language, 28, 606 – 637. Christensen, W., & Michael, J. (2016). From two systems to a multisystems architecture for mindreading. New Ideas in Psychology, 40,
48 – 64. Cousineau, D. (2005). Confidence intervals in within-subject designs: A simpler solution to Loftus and Masson’s method. Tutorials in Quantitative Methods for Psychology, 1, 42– 45. http://dx.doi.org/10.20982/ tqmp.01.1.p042 Driver, J., IV, Davis, G., Ricciardelli, P., Kidd, P., Maxwell, E., & Baron-Cohen, S. (1999). Gaze perception triggers reflexive visuospatial orienting. Visual Cognition, 6, 509 –540. http://dx.doi.org/10.1080/
135062899394920 Faul, F., Erdfelder, E., Buchner, A., & Lang, A.-G. (2009). Statistical power analyses using GPower 3.1: Tests for correlation and regression analyses. Behavior Research Methods, 41, 1149 –1160. http://dx.doi.org/
10.3758/BRM.41.4.1149 Flavell, J. H., Everett, B. A., Croft, K., & Flavell, E. R. (1981). Young children’s knowledge about visual perception: Further evidence for the Level-1–Level-2 distinction. Developmental Psychology, 17, 99 –103. http://dx.doi.org/10.1037/0012-1649.17.1.99 Friesen, C. K., & Kingstone, A. (1998). The eyes have it! Reflexive orienting is triggered by nonpredictive gaze. Psychonomic Bulletin & Review, 5, 490 – 495. http://dx.doi.org/10.3758/BF03208827 Frischen, A., Bayliss, A. P., & Tipper, S. P. (2007). Gaze cueing of attention: Visual attention, social cognition, and individual differences. Psychological Bulletin, 133, 694 –724. http://dx.doi.org/10.1037/0033- 2909.133.4.694
Furlanetto, T., Becchio, C., Samson, D., & Apperly, I. (2016). Altercentric interference in Level-1 visual perspective taking reflects the ascription of mental states, not submentalizing. Journal of Experimental Psychology: Human Perception and Performance, 42, 158 –163. http://dx.doi .org/10.1037/xhp0000138
Hamilton, A. F., Brindley, R., & Frith, U. (2009). Visual perspectivetaking impairment in children with autistic spectrum disorder. Cognition, 113, 37– 44. http://dx.doi.org/10.1016/j.cognition.2009.07.007
Heyes, C. (2014). Submentalizing: I’m not really reading your mind. Perspectives on Psychological Science, 9, 131–143. http://dx.doi.org/10 .1177/1745691613518076
Jonides, J. (1981). Voluntary versus automatic control over the mind’s eye’s movement. In J. B. Long & A. D. Baddeley (Eds.), Attention and performance IX (pp. 187–203). Hillsdale, NJ: Erlbaum.
Loftus, G. R., & Masson, M. E. (1994). Using confidence intervals in within-subject designs. Psychonomic Bulletin & Review, 1, 476 – 490. http://dx.doi.org/10.3758/BF03210951
Marotta, A., Lupiáñez, J., Martella, D., & Casagrande, M. (2012). Eye gaze versus arrows as spatial cues: Two qualitatively different modes of attentional selection. Journal of Experimental Psychology: Human Perception and Performance, 38, 326 –335. http://dx.doi.org/10.1037/ a0023959
Maylor, E. A. (1985). Facilitatory and inhibitory components of orienting in visual space. In M. I. Posner & O. S. M. Martin (Eds.), Attention and performance XI (pp. 184 –204). Hillsdale, NJ: Erlbaum.
Müller, H. J., & Rabbitt, P. M. (1989). Reflexive and voluntary orienting of visual attention: Time course of activation and resistance to interruption. Journal of Experimental Psychology: Human Perception and Performance, 15, 315–330. http://dx.doi.org/10.1037/0096-1523.15.2.315
Peirce, J. W. (2007). PsychoPy—Psychophysics software in Python. Journal of Neuroscience Methods, 162, 8 –13. http://dx.doi.org/10.1016/j .jneumeth.2006.11.017
Posner, M. I. (1980). Orienting of attention. The Quarterly Journal of Experimental Psychology, 32, 3–25. http://dx.doi.org/10.1080/ 00335558008248231
Posner, M. I., & Cohen, Y. (1984). Components of visual orienting. Attention and performance X: Control of language processes, 32, 531– 556.
Qureshi, A. W., Apperly, I. A., & Samson, D. (2010). Executive function is necessary for perspective selection, not Level-1 visual perspective calculation: Evidence from a dual-task study of adults. Cognition, 117, 230 –236. http://dx.doi.org/10.1016/j.cognition.2010.08.003
Samson, D., Apperly, I. A., Braithwaite, J. J., Andrews, B. J., & Bodley Scott, S. E. (2010). Seeing it their way: Evidence for rapid and involuntary computation of what other people see. Journal of Experimental Psychology: Human Perception and Performance, 36, 1255–1266. http://dx.doi.org/10.1037/a0018729
Santiesteban, I., Catmur, C., Hopkins, S. C., Bird, G., & Heyes, C. (2014). Avatars and arrows: Implicit mentalizing or domain-general processing? Journal of Experimental Psychology: Human Perception and Performance, 40, 929 –937. http://dx.doi.org/10.1037/a0035175
Schneider, D., Slaughter, V. P., Bayliss, A. P., & Dux, P. E. (2013). A temporally sustained implicit theory of mind deficit in autism spectrum disorders. Cognition, 129, 410–417. http://dx.doi.org/10.1016/j.cognition .2013.08.004
Senju, A., Southgate, V., White, S., & Frith, U. (2009). Mindblind eyes: An absence of spontaneous theory of mind in Asperger syndrome. Science, 325, 883– 885. http://dx.doi.org/10.1126/science.1176170
Schwarzkopf, S., Schilbach, L., Vogeley, K., & Timmermans, B. (2014). “Making it explicit” makes a difference: evidence for a dissociation of spontaneous and intentional level 1 perspective taking in highfunctioning autism. Cognition, 131, 345–354.
Townsend, J. T., & Ashby, F. G. (1978). Methods of modeling capacity in simple processing systems. In J. Castellan & F. Restle (Eds.), Cognitive theory (Vol. III, pp. 199 –239). Mahwah, NJ: Erlbaum.
Westra, E. (2016). Spontaneous mindreading: A problem for the twosystems account. Synthese, 1–23. http://dx.doi.org/10.1007/s11229-016- 1159-0
World Medical Association. (2001). World Medical Association Declaration of Helsinki. Ethical principles for medical research involving human subjects. Bulletin of the World Health Organization, 79, 373.
Xu, B., Tanaka, J., & Mineault, K. (2012). The head turn cueing effect is sustained at longer SOA’s in the presence of an object distractor. Journal of Vision, 12, 396 –396.