Slower but more accurate mental rotation performance in aphantasia linked to differences in cognitive strategies

Mental rotation tasks are frequently used as standard measures of mental imagery. However, aphantasia research has brought such use into question. Here, we assessed a large group of individuals who lack visual imagery (aphantasia) on two mental rotation tasks: a three-dimensional block-shape, and a human manikin rotation task. In both tasks, those with aphantasia had slower, but more accurate responses than controls. Both groups demonstrated classic linear increases in response time and error-rate as functions of angular disparity. In the three-dimensional block-shape rotation task, a within-group speed-accuracy trade-off was found in controls, whereas faster individuals in the aphantasia group were also more accurate. Control participants generally favoured using object-based mental rotation strategies, whereas those with aphantasia favoured analytic strategies. These results suggest that visual imagery is not crucial for successful performance in classical mental rotation tasks, as alternative strategies can be effectively utilised in the absence of holistic mental representations.


Introduction
Mental rotation is commonly reported as the process of manipulating an internal visual representation of an object in threedimensional space to predict how that object would look from a different perspective (Shepard & Metzler, 1971;Zacks, 2008).This operation is central to many aspects of human cognition: in navigation, such as wayfinding (Malinowski, 2001); in spatial reasoning, such as transforming a chess position before it is played over the board (Saariluoma, 1991), and in simulation of movement, such as planning non-dominant hand movements that are typically done with the dominant hand (Bourrelier et al., 2015;Jones et al., 2021).Mental rotation, among other spatial abilities, has also been shown to predict career success in STEM fields (Wai et al., 2009), while some have argued that these skills (or the lack thereof) may act as a barrier to entry into STEM field expertise (Uttal & Cohen, 2012).Despite its importance, discussions about the underlying mechanisms of mental rotation, and how it relates to visual imagery, are ongoing.Whether the active use of a visual mental image is necessarily required for successful performance on different mental rotation tasks remains unknown (Nanay, 2021;Park, 2019).
Since first introduced by (Shepard & Metzler, 1971), mental rotation has been used in many studies as a behavioural paradigm to assess mental imagery ability, specifically in regards to the ability to manipulate mental images (Quee et al., 2011;Stinton et al., 2008;Urgesi et al., 2011; see Pearson et al., 2013 for a review).In a traditional mental rotation task, a pair of three-dimensional cube objects are displayed at different orientations.Participants are asked to determine whether the pair of objects are identical or mirror images of each other, despite orientation.To perform these tasks participants are assumed to create a visual mental representation of one of the objects and rotate it along the vertical or horizontal axis to determine whether it matches the other object.
A robust finding in the literature is the linear increase in response time as a function of angular disparity of the target stimulus (Cooper & Shepard, 1973;Lochhead et al., 2022;Shepard & Metzler, 1971).This increase in response time was thought to be a function of the analogue rotational 'distance' a mental object must make before it can be effectively compared with its counterpart object.Early interpretations of these findings suggested participants employ a holistic strategy, whereby the mental representation is manipulated as a whole, akin to physically rotating the object (Shepard & Metzler, 1971).However, evidence from eye fixations while performing mental rotation tasks suggest a piecemeal strategy, whereby individual features of the object are internalised and rotated for comparison until a decision can be made (Just & Carpenter, 1976, 1985).Further, research suggests strategy use varies more widely (Bilge & Taylor, 2017;Scheuer et al., 2019;Searle & Hamm, 2017) and depends on a range of factors, including stimulus figural complexity (Folk & Luce, 1987;Yuille & Steiger, 1982), stimulus type (body vs. non-body parts; Tomasino & Gremese, 2016), instructions (Sharps et al., 2016) training (Jost & Jansen, 2021;Uttal et al., 2013), age (Kail, 1985;Kaltner & Jansen, 2016), and gender (Boone & Hegarty, 2017).
Previous research has shown that a variety of different strategies can be utilised to complete classic mental rotation tasks with varying degrees of success (Bethell-Fox & Shepard, 1988;Folk & Luce, 1987;Yuille & Steiger, 1982).Strategy use has been proposed to be one source of the observed sex differences in mental rotation ability (Boone & Hegarty, 2017;Hegarty, 2018) and are distributed differently across the imagery spectrum (Bethell-Fox & Shepard, 1988;Mumaw et al., 1984).Khooshabeh et al. (2013) distinguished the strategies of good and poor imagers, with good imagers defaulting to holistic processing, whereas poor imagers take the slower approach of piecemeal processing and are pressed to change to analytic-type strategies when faced with more difficult mental rotation tasks.Another study found that good and poor imagers engage different brain regions when completing mental rotation tasks, likely reflecting individual differences in strategy usage (Logie et al., 2011).
Alternatively, some have questioned whether visual representations are used at all when performing mental rotation, postulating that only a series of point-by-point comparisons of the perceptually presented stimuli is necessary (Marks, 1999).In this case, the linear increase in reaction time as a function of angular disparity could be accounted for, not by the increased rotation required of a mental image, but rather the need to make more eye movements between the two stimuli for comparison (Just & Carpenter, 1976, 1978).Recent research argues against mental rotation tasks as a pure measure of spatial imagery ability, and suggests that strategy flexibility, being able to switch between holistic and piecemeal strategies or employ an efficient analytic strategy plays an important role in task performance (Hegarty, 2018).
A new avenue for studying the appropriateness of using mental rotation tasks to measure mental imagery has opened with research into a condition called aphantasiathose who self-report a complete absence of visual imagery (Galton, 1880;Zeman et al., 2015).
In recent years aphantasia has been objectively validated using behavioural techniques (Keogh & Pearson, 2018), skin conductance measures (Wicken et al., 2021) and now the pupil's differential response (or lack of it in aphantasia) to imagery of light vs dark objects (Kay et al., 2022).Importantly, the work using the pupillary light reflex goes further to rule out biases or simply not attempting imagery for psychogenic or other reasons, as participants with aphantasia showed a stimulus set-size pupil response, but not the differential luminance-specific response to light and dark objects that those with imagery displayed.The sets-size effect is thought to rely on cognitive load rather than the perceptual qualities of the imagined items, which provides evidence that individuals with aphantasia are actively engaging in the task.
Interestingly, intact spatial abilities and memory have been reported by those with aphantasia in recent studies (Bainbridge et al., 2021;Dawes et al., 2020;Milton et al., 2021;Pounder et al., 2022), suggesting a possible separation between spatial and object imagery (Blazhenkova & Pechenkova, 2019).Beyond visuo-spatial ability, aphantasia has been associated with a range of differences in general cognition (Dawes et al., 2020), specifically with self-reports of reduced dreaming frequency and autobiographical memory (Dawes et al., 2022;Milton et al., 2021).On the surface, performance in visual working memory tasks looks very similar between those with and without imagery (Keogh et al., 2021;Milton et al., 2021;Pounder et al., 2022).However, when participants are asked about their mnemonic strategies and the sensory characteristics of memory representations are examined, differences between the two groups emerge (Keogh et al., 2021).
A few studies have attempted to assess mental rotation in aphantasia.An early case-study of an acquired aphantasia patient MX suggested that MX took more time to respond and showed a non-linear function of response time to rotation angle, compared to controls on a 3D-block rotation task.However, despite differences in reaction times, MX could still perform the task accurately, providing evidence that their visuo-spatial ability remained intact despite the loss of visual imagery phenomenology (Zeman et al., 2010).This finding was recently replicated in patient MX several years after the original study, using the same mental rotation task (Zhao et al., 2022).Furthermore, two recent studies with aphantasia sample sizes of 24 and 20 respectively, report similar performance in two types of mental rotation tasks between those with and without imagery (Milton et al., 2021;Pounder et al., 2022), while response time deficits may only emerge in those indicating severe imagery deficits (Pounder et al., 2022).This suggests potential mental rotation performance without imagery, however, how participants with aphantasia achieve such performance (i.e.solution strategy), and if a sensory representation forms, remains unknown.
Here we aimed to assess whether being able to create and manipulate a visual representation in the mind's eye is necessary to successfully perform a range of mental rotation tasks.Building on the recent research by Milton et al., 2021 andPounder et al., 2022, we recruited large samples of individuals with aphantasia (those who lack voluntary visual imagery) online and compared their performance with controls (those with visual imagery) on two classic mental rotation tasks: the classic Shepard and Metzler (1971) three-dimensional rotation task, and a version of the human Manikin test of spatial orientation and transformation (Ratcliff, 1979) as described in Englund et al. (1987).
We further, investigated the strategies used by participants in the Manikin test to shed light on how one might perform 'mental rotation' without visually rotating a mental object.We attempted to measure three types of solution strategy: Object-based mental rotation strategy, embodied self-rotation/egocentric strategy and analytic/criteria-based strategy.Object-based mental rotation is thought to rely on visual imagery, preferentially activating the occipito-temporal-parietal network (Tomasino & Gremese, 2016).For this strategy participants create a representation of the manikin stimulus and rotate it to align with their own orientation to complete the task.Embodied self-rotation is thought to rely on motor imagery, preferentially activating the sensorimotor network (Hamada et al., 2018;Tomasino & Gremese, 2016).Here, participants might use proprioceptive reasoning to mentally align their own body or perspective with the orientation of the stimulus.Analytic based strategies are those that don't primarily rely on any imagery, but rather the memorisation of criteria and reasoning processes (Saunders & Quaiser-Pohl, 2020;Schultz, 1992).

Participants
Shepard-Metzler task: A sample of 95 participants with self-described aphantasia (aged 19-82, mean age = 46.63years, 55 female) were recruited.These individuals had emailed the lab regarding their aphantasia and willingness to participate and participated in exchange for entry into a 200AUD gift card prize draw.114 control participants with imagery (aged 18 -74, mean age = 46.5 years, 60 female) were also recruited for this study.104 of these were recruited using Amazon Mechanical Turk (MTurk) and were remunerated at a rate of $12USD per hour for their participation in the study.10 were undergraduate psychology students who participated for course credit.This control (or 'general population') group was matched on mean age with our aphantasia group (mean age difference = .55years, p = .78,BF 10 = .158).
Manikin test: A separate c sample of 150 participants with self-described aphantasia (aged 18-78, mean age = 44.68,83 female) were recruited.They had also emailed the lab and participated in exchange for entry into a 200AUD gift card prize draw.164 (aged 18-74, mean age = 44.54,82 female) control participants were recruited using Prolific (135) and MTurk (29) and were remunerated at a rate of $12USD per hour for their participation.Again, the control group was matched on mean age with the aphantasia group (mean age difference = .43years, p = .67,BF 10 = .136).

Stimuli
Shepard-Metzler task: An open-source stimuli-set of 48 3D cube objects was provided by (Ganis & Kievit, 2015).Traditional Shepard and Metzler (1971) displays were formed with a 'baseline' object (0 degrees of angular disparity) on the left of screen, a 'target' object on the right and a fixation cross in the middle.The target object was either structurally identical to the baseline object ('same' condition) or was a mirror image of the baseline object ('different' condition) and was positioned at either 0, 50, 100, or 150 degrees of angular disparity along the vertical axis ('angle of rotation' conditions (see Fig. 1A for illustration)).For the experimental phase of the task, 12 cube objects were chosen at random from the stimuli-set of 48 and used for displays in the 8 conditions (2 'sameness' conditions and 4 'rotation angle' conditions), forming 96 trials in total.
Manikin test: Stimuli and instructions were based on (Englund et al., 1987).Human manikin figures holding a green circle in one hand and a red square in the other are superimposed on either a green circle or red square.The manikins have 2 vertical orientations (upright or upside-down) and 2 horizontal orientations (facing forward i.e. toward the participant, or facing away), creating four orientation conditions in total.These orientation conditions are defined as O1: upright, facing away (i.e. the same orientation as the participant), O2: upright, facing forward (i.e.rotated around the horizontal axis), O3: upside-down, facing away (i.e.rotated around the vertical axis, and O4: upside-down, facing forward (i.e.rotated around both horizontal and vertical axes).See Fig. 1B for an illustration of these stimuli.Trials were counterbalanced across the 4 orientation conditions as well as background colour/shape (green circle vs red square) and congruent hand (left vs right) creating 16 unique trial types in total.

Procedure
Shepard-Metzler task: In each trial, participants were presented with a fixation cross on a blank screen for 250 ms, followed by one of the displays (pair of 3D cube objects).Participants were tasked with determining whether the target stimulus in the display was the same as, or different to (i.e. a mirror-image of) the baseline stimulus with the exception of rotation by pressing the 's' or 'd' key respectively.They were asked to respond as quickly and accurately as possible and were given a time limit of 7500 ms.These time parameters were the same as those used by (Ganis & Kievit, 2015) to validate the stimuli set.
Participants were first given visual instructions about the task and carried out 16 practice trials where they were given correct/ where the target object can be rotated along the vertical axis to align with the baseline object and 'different' trial stimuli, where alignment is not possible (because the two objects are mirror-images of each other).Here, the target objects are positioned at 50 degrees (left panel) and 150 degrees (right panel) of angular disparity.B. Examples of Manikin test stimuli in which participants must determine which hand of a manikin figure is holding the shape/colour that matches the background.Orientation conditions are defined as O1 (upright, facing away), O2 (upright, facing forward), O3(upside-down, facing away), O4 (upside-down, facing forward).In all examples depicted here, the colour/shape in the left hand of the manikin matches the background.
incorrect feedback on their responses.They then performed 2 blocks of 48 experimental trials with no feedback, separated by a break.
In each block, the 2 sameness conditions and 4 angular rotation conditions occurred equally.Trials were presented randomly with the constraint that no more than 3 sameness conditions were run consecutively.The task took 6-7 min to complete depending on response times.After finishing the experiment, participants were given a debrief on the study and were asked about the strategies they used to perform the mental rotation task.
Manikin test: In each trial, participants were presented with a fixation cross on a blank screen for 250 ms, followed by one of the manikin displays.Participants were tasked with determining if the manikin was holding the colour/shape congruent with the background (e.g.green circle) in its left or right hand by pressing 'v' or 'm' key respectively.Similar to the Shepard-Metzler task, participants were asked to respond as quickly and accurately as possible and were given a time limit of 7500 ms.
Participant were first given visual instructions about the task with example trials, then carried out 16 practice trials (one iteration of each unique trial type) and given correct/incorrect feedback on their responses.They then performed 4 blocks of 16 experimental trials with no feedback, separated by breaks.Trials were presented randomly with each block containing one iteration of each unique trial type.The task took 7-8 min to complete depending on response times.
After finishing the task, participants were given an open-ended questionnaire about the strategies they used to perform the task and which strategy worked best for them.They also responded to the 3 following statements on a 5-point scale (strongly agreestrongly disagree): • "I imagined rotating the man (stimulus) to match my own orientation to determine which hand was holding the correct colour/shape" (Object-based Strategy).
• "I imagined rotating my own body to match the orientation of the man to determine which hand was holding the correct colour/shape" (Embodied Strategy).
• "I didn't imagine rotating anything, but instead used logical criteria to determine which hand was holding the correct colour/shape" (Analytic Strategy).
These statements refer to object-based rotation, embodied self-rotation and analytic strategies respectively, and were formulated based on the wording of task instructions and questionnaires used in previous studies that categorised mental rotation strategy (Acqua et al., 2009;Hegarty, 2018;Tomasino & Gremese, 2016).
Lastly, since the study was conducted online, objective measures of visual imagery such as those using binocular rivalry (Keogh & Pearson, 2018) or pupillometry (Kay et al., 2022) were unattainable, therefore, participants completed the Vividness of Visual Imagery Questionnaire (VVIQ; Marks, 1973) as a subjective measure of imagery capacity.In addition, participants responded to the question "Do you identify as having aphantasia" on a 5-point scale from "definitely yes" to "definitely not" as a secondary check to ensure participants indeed belonged to their respective groups.A-priori cut-off of ≤32 (extremely weak to non-existent visual imagery) and >32 (average to strong visual imagery) on the VVIQ were used to confirm the aphantasia and control groups respectively.
Scripts for both tasks were provided by http://www.millisecond.com/download/library/(the Millisecond Test Library) and were run in Inquisit Web Version 5 (Millisecond Software, LLC, Seattle, WA), while the VVIQ was run on Qualtrics (Qualtrics, Provo, UT) at https://www.qualtrics.com.All data was collected online, hosted by the Millisecond and Qualtrics online servers.

Analysis
Shepard-Metzler task: Dependent variables included reaction time and accuracy (percentage of trials in which participants correctly determined whether the target object was the "same as" or "different to" the baseline object).Participants with an overall accuracy rate below 50 % were excluded from analysis (4 participants with aphantasia and 7 controls).No participants scored outside the VVIQ apriori cut-off point for their respective group (≤32 for the aphantasia group, >32 for controls).For the aphantasia group, 69 participants scored at floor on the VVIQ (16), while the other 22 participants scored between 17 and 31 (see Supplementary Fig. 1A for VVIQ score distributions of both groups post-exclusion).The two groups were still matched on mean age after exclusion (mean age difference = .05years, p = .981,BF 10 = .156).Trials without a response (i.e., the 7500 ms time limit was reached) as well as those with a response time of less than 250 ms were excluded from analysis (3.31 % of aphantasia data, 3.56 % of control data).Reaction time was calculated for correct trials only.
Manikin test: Exclusion criteria and analysis was kept consistent with that of the Shepard-Metzler task.Dependent variables of reaction time and accuracy (percentage of trials in which participants correctly determined which hand of the manikin is holding the colour/shape congruent with the background) were measured.Participants with an overall accuracy rate below 50 % were excluded from analysis (5 participants with aphantasia and 7 controls).A further 7 participants with aphantasia and 2 controls were excluded due to scoring outside of the VVIQ cut-off points for their respective groups.For the aphantasia group, 98 participants scored 16 on the VVIQ, while the other 39 participants scored between 17 and 29 (see Supplementary Fig. 1B for VVIQ score distributions of both groups post-exclusion).The two groups were still matched on mean age after exclusion (mean age difference = .64years, p = .727,BF 10 = .136).Trials without a response (7500 ms time limit reached) as well as a response time of >250 ms were excluded from analysis (2.91 % of aphantasia data, 3.29 % of control data).Reaction time was calculated for correct trials only.
All data sets met assumptions of normality using the Shapiro-Wilks test.Therefore, two-way repeated measures ANOVAs were run to compare the two groups on reaction time and accuracy for both the Shepard-Metzler task and Manikin test.Age-matching analyses were performed using R Statistical Software version 4.1.0(R Core Team, 2021) while ANOVA's and ANCOVA's were run in JASP version 0.14.1.0(https://jasp-stats.org/).Raincloud plots (Figs. 2 and 3), scatterplots (Fig. 4), and diverging bar chart (Fig. 5) were  (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)obtained using the ggplot software package in R (Wickham, 2016).Data files for this experiment are available at https://osf.io/gd94u/?view_only=ab0edd3bddc3418399d605eaa4f5636d.

Performance
Fig. 2A shows the reaction time data from the Shepard-Metzler rotation task.Both groups clearly show the classic positive slope as a function of angular disparity.Across both groups response time increased with the angle of rotation, F(3, 588) = 506.7,p < .001,η p 2 = .72.But importantly for our current enquiry, the aphantasia group, those with no mental imagery, also showed the classic response time linear increase with angular disparity.There was a main effect of group, F(1, 196) = 13.45,p < .001,Cohen's d = .48,η p 2 = .06,with the aphantasia group having slower response times on average (M = 2909 ms, SD = 664 ms) compared to controls (M = 2559 ms, SD = 676 ms).There was also a significant interaction between group and angle of rotation, F(3, 588) = 14.64, p < .001,η p 2 = .07.There was also a significant interaction between group and angle of rotation (F(3, 873) = 3.9, p = .009,η p 2 = .01).However, when using Bonferroni correction for multiple comparisons, we found the difference between groups was not significant for any orientation condition (all p > .06).Fig. 3B shows the accuracy data for the two groups across orientation conditions.There was a main effect of group (F(1, 291) = 9.55, p = .002,Cohen's d = .26,η p 2 = 03.Again, the aphantasia group had a higher percent correct (M = 91.06%, SD = 3.93 %) compared to controls (M = 87.03%, SD = 5.94 %).There was a significant main effect of orientation condition, with mean correct responses decreasing as a function of orientation F(3, 873) = 42.53,p < .001,η p 2 = 13.No significant interaction was found between group and orientation condition, F(3, 873) = 1.89, p = .13.In addition to the online Manikin test, we ran a smaller in-Lab version of the same task to ensure that participants' results were similar in a more controlled environment.8 individuals with aphantasia and 8 control participants were recruited for this in-person study.Overall, the pattern of data from the in-lab participants matched that from the online version.However, the effect of reaction time was not significant for group, F(1, 14) = .49,p = .497,but did show a significant main effect of rotation condition as before, F(3, 42) = 13.06,p = .0002.There was no significant interaction between group and rotation condition, F(3, 42) = .78,p = .513(see Supplementary Fig. 2).
To further investigate the group differences found in mental rotation performance we correlated response time and accuracy within groups.We averaged participants' results across the rotation conditions where the greatest between group differences were found (100 and 150 degrees of angular disparity for the Shepard-Metzler task, and O3 and O4 for the Manikin Test), and ran speed-accuracy correlation analyses for the aphantasia and control groups separately.The Shepard-Metzler control group data met assumptions of normality (Shapiro-Wilk test, W = .99,p = .539),therefore, Pearson's correlation coefficient was used here.All other datasets failed the Shapiro-Wilk test (Shepard-Metzler task aphantasia group: W = .94,p < .001,Manikin test aphantasia group: W = .86,p < .001,Manikin test control group: W = .97,p = .002),therefore Spearman's rank correlation coefficient was used).To control for multiple comparisons an alpha level of p = .0125was set (.05/4).
Fig. 4A shows scatterplots with speed-accuracy correlation coefficients for the aphantasia and control groups in the Shepard-Metzler task.In the aphantasia group, a significant negative correlation was found between reaction time and mean correct response rate (r s (106) = − .33,p = .001,BF 10 = 4.27), indicating moderate evidence that faster participants were also more accurate on average.Alternatively in the control group, a significant positive correlation was found between reaction time and mean correct response rate (r p (90) = .37,p < .001,BF 10 = 266.44),indicating extreme evidence of a speed-accuracy trade-off.Fig. 4B shows scatterplots with speed-accuracy correlation coefficients for both groups in the Manikin test.In both groups, a negative correlation was found between reaction time and mean correct response rate (aphantasia group: r s (137) = − .21,p = .013,BF 10 = 2.27, control group: r s (154) = − .19,p = .02,BF 10 = .41),however these are not singificant when controlling for multiple comparisons.

Mental rotation strategy
Here we wanted to investigate how individuals with aphantasia successfully complete the Manikin test if they aren't relying on visual imagery.We also wanted to determine if strategy choice in the Manikin test was an important covariate for the performance differences between the aphantasia and control groups.To do this we compared the two groups' responses to three questions aiming to tap into object, embodied and analytic strategies.Results from the strategy questionnaire are visualised in Fig. 5. Participants also gave an open-ended report of the strategy(s) they used to confirm their responses (see Table 1).
To assess the role of strategy and imagery vividness on the observed group differences in mental rotation performance, we ran linear mixed-effects models using the "lmerTest" package in R (Bates et al., 2015;Kuznetsova et al., 2017), following the practice guidelines of Meteyard and Davies (2020).We included response time and accuracy as separate dependent variables, both for which we fitted three linear mixed models: a group-only model (aphantasia vs controls), a strategies-only model (object-based rotation strategy, embodied self-rotation strategy, and analytic strategy), and a combined group and strategies model.Orientation condition was added as an additional fixed factor to all models, and participant ID was added as a random effect to account for within-subject variability.A formal model comparison was then run using Bayesian information criteria (BIC) to determine which model best explained performance variability.
A multicollinearity assessment for predictors was first conducted on the models to determine the degree of correlation between independent variables.Variance inflation factor (VIF) values indicated all independent variables had moderate levels of correlation with all other variables for the response time data (group: VIF = 1.58, object-based rotation strategy: VIF = 1.67, embodied selfrotation strategy: VIF = 1.52, analytic strategy: VIF = 1.80).This suggests that while there is some degree of overlap in the variance explained by these predictors, it is not above threshold commonly cited for severe multicollinearity concerns (VIF > 5) that would undermine the reliability of the regression coefficients or the overall validity of the models.As expected, the effect of orientation condition had no correlation with other predictors (VIF = 1.00), indicating no contribution of multicollinearity within the model.Therefore, all independent variables were kept in the analysis and the model comparison was conducted as normal.
We found that, for response time data, the strategies-only model (BIC = 12415.7)and the combined group and strategies model (BIC = 12418.7)provided a better fit than the group-only model (BIC = 18513.8).Despite the inclusion of group as an additional predictor in the combined model, its BIC value was very close to that of the strategies-only model, indicating that adding group effects does not improve explanatory power above and beyond what is provided by differences in strategy-use.However, for the accuracy data, the group-only model provided the best fit (BIC = − 1104.8)compared to the strategies-only model (BIC = − 743.2) and combined group and strategies model (BIC = − 740).This indicates that group effects were sufficient for explaining the individual differences in accuracy, and the addition of strategies did not significantly improve model fit.Overall, differences in cognitive strategy-use seem to provide the most comprehensive framework for understanding response time variability, whereas accuracy variability is best understood via group differences in imagery ability.

Discussion
Across two classic mental rotation tasks, as a group, individuals with self-reported aphantasia (total N = 229) were slower but more accurate when compared with controls with imagery (total N = 262).The slower reaction times are consistent with previous assessments of mental rotation performance in aphantasia (Pounder et al., 2022;Zeman et al., 2010).Further, individuals with aphantasia demonstrated the classic increase in response times as a function of stimulus orientation difference, originally thought to be due to individuals rotating an object image in their mind's eye.However, the higher levels of accuracy are not consistent with previous research, which found equivalent accuracy between groups (Milton et al., 2021;Pounder et al., 2022).It may be the case that our larger samples of individuals with aphantasia have allowed enough power to detect this group difference in performance, due to accuracy on these tasks generally being high.However, the medium to large effect sizes found when comparing group accuracy at greater degrees of angular disparity on the  for 100 and 150 degrees of angular disparity respectively) suggest that for this task, group differences at smaller samples should be detectable.Another potential explanation could come from task differences between the current study and previous research.Our Shepard-Metzler task used only mirrored 3D block stimuli to form 'different' condition trials, whereas Pounder et al. (2022) used a combination of mirrored and differently structured 3D blocks.It may be the case that accuracy on these trial types is differentially modulated by alternative strategy use in aphantasia, but this is yet to be tested.
Analysis of solution strategy in the Manikin test revealed that individuals with aphantasia report using different cognitive strategies to complete the task.While control participants with imagery predominantly favour the use of object-based holistic rotation strategies, participants with aphantasia favour analytic/criteria (feature-based, orientation independent) strategies.This supports the finding by Logie et al., 2011 that mental rotation causes differential engagement of visual brain regions for people with strong imagery vs low imagery, and the suggestion that this is caused by different strategy use.Previous research has shown that choice of strategy greatly influences performance in mental rotation tasks, comparable to sex differences observed in the literature (Hegarty, 2018;Toth & Campbell, 2019).
A linear mixed model comparison using Bayesian information criteria revealed strategy was the most important predictor of response time variability, as the strategy-only model yielded a lower BIC value than both the group-only and combined models.However, the opposite was found when modelling accuracy variability, with the group-only model providing a lowest BIC value.This may be interpreted as strategy-use and imagery ability having separate and unique impacts on mental rotation performance, although this should be considered within the context of multicollinearity among the predictors within these models.The moderate correlations found between group (imagery ability) and strategy-use, although not strong enough to significantly distort the models' outcomes, should necessitate a cautious interpretation of their differential effects on response time and accuracy (Frost, 2017).Nonetheless, these results provide evidence that both imagery ability and strategy-use play a complex role in mental rotation performance, while the presence or absence of imagery influences what strategies are available to participants and what they are predisposed to use.Future research should aim to tease apart these roles to weigh their individual impact on mental rotation performance.
When comparing group performance across tasks it is important to note that strategy use depends on the type of stimuli being rotatedanimate vs inanimate (Macintyre et al., 2013).Animate stimuli, such as those used in the Manikin task, can be more easily rotated using embodied self-rotation than the inanimate stimuli used in the Shepard-Metzler task.Here, participants map their own body to the manikin figure presented on screen, then imagine rotating their own body to complete the task.This endogenous process involves motor cortex areas, including M1 (Hamada et al., 2018;Kosslyn et al., 2001;Tomasino & Gremese, 2016), which are not known to be implicated in aphantasia.However, embodied self-rotation (also referred to as perspective taking) is ineffective for completing inanimate and abstract mental rotation tasks (Hegarty, 2018) such as the Shepard-Metzler task.Here, the aphantasia group must resort to slower analytic strategies such as counting cubes or local turns (Hegarty, 2018).Taken together, this may explain why a greater response time was observed in the aphantasia group for the Shepard-Metzler task.However, as strategy use for the Shepard-Metzler task was not collected in this study, this interpretation is based on extrapolation from the Manikin strategy results and requires further investigation to be confirmed.
It is worth noting in the Manikin test that although object-based mental rotation was used significantly less often in the aphantasia group (18 % somewhat agreed, 6 % strongly agreed) than the control group (44 % somewhat agreed, 31 % strongly agreed), it was still reportedly used by some.This is contradictory, as object-based mental rotation is believed to rely on visual imagery.One potential explanation for this is the recent argument for 'unconscious imagery' (Nanay, 2021).Here, participants may be manipulating an external representation of the stimuli without the phenomenology.This could explain why many of these participants describe the feeling of "knowing the correct answer but not knowing how they know".
In the Shepard-Metzler task, the control group exhibited a speed-accuracy trade-off, a common finding in the literature (Borst et al., 2011;Goldstein et al., 1990;Hertzog et al., 1993;Liesefeld et al., 2015).Interestingly, the aphantasia group did not show this same speed-accuracy trade-off.The opposite correlation was found, with faster aphantasia participants also being more accurate.This is consistent with the 'poor imagers' group in Khooshabeh et al. (2013) which also did not show a speed-accuracy trade-off, and further indicates an alternative strategy or mechanism at play in the aphantasia group.However, as strategy use for the Shepard-Metzler task was not collected post-test, this interpretation is postulatory and requires further investigation.In the Manikin Test, both groups exhibited similar slopes, with faster participants being relatively more accurate, although this effect.was not significant.This suggests there is a larger overlap of strategy between the groups in this task, potentially due to the changes in task stimuli and individuals from both groups accessing embodied self-rotation.Alternatively, this similarity between the groups may be reflective of the comparatively easier Manikin Test, with accuracy rates, in general, being higherspeed-accuracy trade-offs only emerge when a task is sufficiently complex (Liesefeld et al., 2015).
Overall these findings support the argument that mental representations can come in multiple formats which can be flexibly manipulated to achieve the same goal (Keogh et al., 2021;Pearson & Kosslyn, 2015;Zhao & Della Sala, 2018).

Table 1
Examples of open-ended strategy reports for each type of strategy used in the Manikin test (i.e.participants who both "strongly agreed" with using their given strategy, and "strongly disagreed" with using any other strategy).

Object-based Mental Rotation strategy:
"I imagined rotating or flipping him to be in the same position as me (i.e.facing away from me as if we were in a queue)" (Control participant) "I tried to imagine rotating the man so that he was facing away from me, though I could only rotate him for a second or so since I can't 'see' it."(Aphantasia participant) Embodied Self-rotation strategy: "I imagined myself in the same position as the figure, or rather, I mentally mapped my body-sense onto that position" (Control participant) "I 'imagined' my body rotating into the position of the person in the picture" (Aphantasia participant) "Although i was not picturing images, i felt like i was using proprioception to 'feel' the correct orientation…" (Aphantasia participant) Analytic strategy: "Used logic to determine left or right hand, if the man was facing towards or away from me; upside-down or rightside-up; and then used that criteria to solve each problem."(Control participant) "I established rules about how many rotations (front/back, upside-down/rightside-up) it would take (zero: same side; one: other side; two: same side)."(Aphantasia participant) L. Kay et al.Our findings are parallel to arguments made in recent studies measuring sex differences in mental rotationthat the difference in performance observed between sexes is a result of different strategy choice rather than mental rotation ability (Hegarty, 2018).Raabe et al., 2006 provided evidence that women are predisposed to efficiently employ an analytic strategy while men are predisposed to efficiently use a holistic rotation strategy.
One limitation of this study is the differences in the recruitment process between the groups.Those with aphantasia had contacted the lab regarding their willingness to participate in research and thus may have been more engaged and motivated compared to controls.Potential differences in engagement and motivation between the groups cannot be ruled out as a possible driving factor of accuracy and reaction time differences.Future studies should aim to include individuals with hyperphantasia, recruited in the same manner as those with aphantasia, as a third group to eliminate the potential confounding variable of the recruitment process (for example, see Milton et al., 2021).Recruitment bias may have also impacted the participant strategy reports.For example, those with aphantasia may have been biased towards agreeing with using an analytic strategy because the Likert probe included the words "I didn't imagine anything", despite not actually using the strategy.Future aphantasia strategy questionnaires should aim to be as neutral as possible and refrain from using any words directly related to visual imagery to avoid these biases.Lastly, future studies could also use the newly developed eye-tracking approach to identify mental rotation strategies and cross-reference this with post-test questionnaire data to achieve a more thorough and accurate analysis of solution strategy between groups (Saunders & Quaiser-Pohl, 2021).

Conclusion
In conclusion, these data suggest that classical mental rotation tasks that depend on accuracy and reaction time performance, can be completed in a variety of different ways and therefore do not necessarily measure mental imagery ability.Much like recent findings for visual working memory (Keogh et al., 2021), the presence of different cognitive tools (i.e.having the ability to visualise or not) seems to dictate what kind of strategy an individual will use during mental rotation.From this point of view any compound performance-based task should include a measure of task strategy and it should not be assumed participants are using the same mental processes to arrive at the same correct answer.In light of these findings, it also might be worth looking back over the past four decades of research that utilised mental rotation to assess mental imagery abilities and check the interpretation and influence of such work.We recommend that to assess imagery ability in future work, researchers specifically ask about imagery vividness or use one or all of the number of objective or behavioural methods that measure the impact of sensory representations in the brain, such as the bias on binocular rivalry (Keogh & Pearson, 2018;Pearson et al., 2008) or pupil response to visual imagery (Kay et al., 2022) rather than using mental rotation as an index of imagery ability.

Funding
We would like to thank the Australian Research Council grants (DP220100183) and Future Fellowship FT220100388 JP.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Fig. 1 .
Fig. 1.Examples of Shepard-Metzler rotation stimuli and Manikin test stimuli.A. Examples of 'same' trial stimuli in the Shepard-Metzler task,where the target object can be rotated along the vertical axis to align with the baseline object and 'different' trial stimuli, where alignment is not possible (because the two objects are mirror-images of each other).Here, the target objects are positioned at 50 degrees (left panel) and 150 degrees (right panel) of angular disparity.B. Examples of Manikin test stimuli in which participants must determine which hand of a manikin figure is holding the shape/colour that matches the background.Orientation conditions are defined as O1 (upright, facing away), O2 (upright, facing forward), O3(upside-down, facing away), O4 (upside-down, facing forward).In all examples depicted here, the colour/shape in the left hand of the manikin matches the background.

L
.Kay et al.

Fig. 2 .Fig. 3 .
Fig. 2. Group comparison of performance on the Shepard-Metzler rotation task. A. Mean reaction times for aphantasia (blue diamonds) and control (red circles) groups shown as a function of angle of rotation (small data points for individual participants; filled surfaces for smoothed density plot).B. Mean correct response rate for aphantasia (blue diamonds) and control (red circles) groups shown as a function of angle of rotation (small data points for individual participants; filled surfaces for smoothed density plot).*p < .0001.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 4 .
Fig. 4. Correlations between reaction time and mean correct response, with separate plots for group and task. A. Shepard-Metzler task correlations for the aphantasia group (left) and control group (right).Results are averaged across 100 and 150 degrees of angular disparity conditions.B. Manikin test correlations for the aphantasia group (left) and control group (right).Results are averaged across orientation conditions O3 and O4.Blue trendlines represent correlation coefficients.Shaded areas indicate 95 % confidence intervals.Individual data points represent one participant.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 5 .
Fig. 5. Results from the Manikin test strategy questionnaire.Diverging bars show the percentage of participants responding to the 5-point Likert questions with "Stongly disagree" (dark red), "Somewhat disagree" (light red), "Neither agree or disagree" (grey), "Somewhat agree" (light blue), or "Strongly agree" (dark blue).Aphantasia and control groups are compared for each question.***p < .0001.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)