Background

Aphantasia describes the condition of reduced or absent voluntary imagery (Zeman, Dewar, & Della Sala, 2015), whereby imagery refers to perception-like representations without corresponding external stimuli (Pearson, Naselaris, Holmes, & Kosslyn, 2015). This means that those affected cannot, for example, imagine their parents’ faces when they are not in the same room. While first approaches to aphantasia were mainly based on self-report (Dawes, Keogh, Andrillon, & Pearson, 2020; Zeman et al., 2015, 2020), behavioural differences in objective tasks have recently been found, providing evidence that aphantasia is not merely metacognitive in nature, but the virtual inability to generate mental images. For example, Keogh and Pearson (2018) as well as Monzel, Keidel, and Reuter (2021) found that aphantasics, in contrast to non-aphantasics, cannot be primed by their own imagery. Nevertheless, it remains difficult to investigate the effects of aphantasia, since many tasks that ostensibly require visual imagery can also be solved by non-visual alternative strategies, such as the reliance on verbalization (Jacobs, Schwarzkopf, & Silvanto, 2018; Zeman et al., 2010). Lack of performance differences between aphantasics and non-aphantasics is therefore difficult to interpret, as it might be the result of compensatory processes.

One field receiving particularly much attention in aphantasia research is the association with mnestic deficits (Dawes et al., 2020; Milton et al., 2020; Zeman et al., 2015, 2020). People with aphantasia predominantly describe a severely deficient autobiographical memory (SDAM), another phenomenon which has initially been studied independently of aphantasia but was later associated with it (Palombo, Sheldon, & Levine, 2018; Watkins, 2018). Although semantic memory impairments have been reported as well (Dawes et al., 2020), mnestic deficits have not yet been confirmed in objective memory tasks (Milton et al., 2020). As mentioned earlier, this could be due to task characteristics; the tasks might have lacked sensitivity to the absence of visual imagery. Therefore, the present study aims to investigate deficits in visual as well as verbal short-term (STM) and long-term memory (LTM) in aphantasics while taking task characteristics into account. In particular, task difficulty, retrieval condition, and subcategories of stimuli will be discussed.

Visual long-term memory in aphantasics

As already mentioned, many studies have shown long-term deficits in aphantasics’ autobiographical memory: In a sample with 2,000 aphantasics and 1,288 controls, Zeman et al. (2020) found aphantasics to be more likely to report impaired autobiographical memory than controls. Milton et al. (2020) were able to replicate this effect using the Autobiographical Interview by Levine, Svoboda, Hay, Winocur, and Moscovitch (2002). In the general population, D’Argembeau and van der Linden (2006) found associations between capacity of visual imagery and richness of detail in mental time travel and Ernst et al. (2013) demonstrated that visual imagery training can improve autobiographical memory.

From a theoretical perspective, the deficits in autobiographical memory in aphantasia can be explained by the Scene Construction Theory, which assumes the same neural mechanisms for autobiographical memory and visual imagery (Hassabis & Maguire, 2007, 2009; Mullally & Maguire, 2013). This assumption has been supported by numerous studies using imaging techniques (Addis, Wong, & Schacter, 2007; Schacter et al., 2012). The Reverse Hierarchy Model (Pearson, 2019) suggests that visual imagery is, in effect, the reverse process of visual perception. While visual perception initially generates activation in the occipital lobe, which is then interpreted by frontal regions, these activations are initiated by frontal regions during visual imagery. Furthermore, the Scene Construction Theory and the Reverse Hierarchy Model share the assumption that the hippocampus is involved in generating mental images from memories (Maguire & Mullally, 2013). Evidence for the Reverse Hierarchy Model was provided by Milton et al. (2020), who demonstrated in a seed analysis that there is in fact a weaker connectivity between frontal regions and the occipital lobe in aphantasics than in hyperphantasics, that is, people with highly pronounced visual imagery. However, no empirical evidence has been provided so far, that these memory impairments in autobiographical memory generalize to visual (and more recent) LTM. On the contrary, Zeman et al. (2010) found no differences in the Visual Delayed Index of the Wechsler Memory Scale (WMS; Wechsler, 1997) and Milton et al. (2020) could not find any differences between aphantasics and non-aphantasics in several visual LTM tasks. Marks (1973) and McKelvie and Demers (1979) showed deficits in the long-term recall of pictures, at least in low imagers (= participants with relatively little imagery) in comparison to high imagers (= participants with relatively good imagery), but when McKelvie and Demers (1979) asked for recognition instead of a recall, the differences disappeared.

Visual short-term memory in aphantasics

As with visual LTM, no deficits in visual STM have been found so far in aphantasics using conventional measurement methods (e.g., the Working Memory Capacity battery [Lewandowsky, Oberauer, Yang, & Ecker, 2010] in Jacobs et al., 2018 or the WMS [Wechsler, 1997] in Zeman et al., 2010). This is surprising since neuronal overlaps can also be found between visual imagery and visual STM (Albers, Kok, Toni, Dijkerman, & de Lange, 2013) and they can both be disrupted by the same interference (Borst, Ganis, Thompson, & Kosslyn, 2012). Yet, Zeman et al. (2010) found better performance of an aphantasic in the verbal version of Brook’s (1967) Matrix Task (as adapted by Salway & Logie, 1995) than in the visuo-spatial version, which is normally found reversed (Baddeley, Grant, Wight, & Thomson, 1975). After 4 weeks, the aphantasic achieved significantly better results on the visuo-spatial version than on the first attempt but still below average. This improvement was interpreted as the development of an alternative solving strategy, since the aphantasic was next asked to perform the visuo-spatial version in combination with an articulatory suppression task as well as in combination with a spatial-motor suppression task. In contrast to previous findings, in which participants performed better under articulatory suppression (Baddeley & Lieberman, 2017), the aphantasic performed better under the spatial-motor suppression task. This supports the hypothesis that he had used a verbal strategy to solve the visuo-spatial task, which he also confirmed in a self-report. In line with these findings, Jacobs et al. (2018) found no impairments in the visual STM of their aphantasic individual (AI), except for the trials with the highest difficulty. This also indicates that AI used alternative strategies in easier trials, which compensated for the lack of visual imagery but failed when the task became too difficult. Thus, when investigating differences between aphantasics and non-aphantasics, attention should be paid to using tasks in which the visual imagery cannot be compensated for by non-visual strategies.

Moreover, it should be stressed that both Zeman et al. (2010) and Jacobs et al. (2018) were single case studies with both participants scoring above average in intelligence tests (IQMX = 136, IQAI = 126), probably due to self-selection in the research process. Thus, the use of alternative strategies might have been particularly pronounced in these two participants. Besides, the aphantasic in Zeman et al. (2010) was matched with controls using the Wechsler Adult Intelligence Scale III (WAIS-III; Wechsler, 1997) which contains working memory tasks, therefore possibly nullifying memory differences in STM. Generally, the performance of these single cases cannot be interpreted as representative for an aphantasic population. Examining a sample of 32 participants, McKelvie and Demers (1979) found worse performance in short-term picture recall in low imagers than in high imagers.

Verbal long-term memory in aphantasics

According to the Dual Coding Theory, knowledge can be represented both visually and verbally, meaning that one can remember a picture of a dog and/or a verbal description of it (Paivio, 1990, 2007, 2013). Therefore, aphantasics as well as participants with SDAM should be able to compensate their deficits in visual memory by the use of verbal memory. However, participants with SDAM show deficits in the performance of visual memory tasks in comparison to controls (Palombo, Alain, Söderlund, Khuu, & Levine, 2015). Thus, the use of verbal strategies alone does not seem to be sufficient for compensation. Conversely, it can be assumed that the verbal LTM is also impaired, since the verbal strategy cannot be complemented with a visual strategy. This assumption is supported by findings that words that can be stored both verbally and visually (e.g., tree) are easier to remember than abstract words to which no concrete images can be assigned (e.g., truth; Hargis & Gickling, 1978; Paivio, 1963; Yui, Ng, & Perera-W, 2017). Similarly, the multimedia presentation of words has been found to ensure better recall performance (Brunyé, Taylor, & Rapp, 2008). This is also plausible on a neuronal level, since in the brain both, visual and verbal information, are linked together in an associative–semantic network and can support each other (Vandenberghe, Price, Wise, Josephs, & Frackowiak, 1996). It therefore seems likely that aphantasics also have deficits in verbal LTM, although they should not be as severe as for visual memory.

However, these assumptions have not yet been confirmed by objective tests, for example by the Auditory Delayed Index of the WMS (Wechsler, 1997; Zeman et al., 2010) or several verbal LTM tasks in Milton et al. (2020). Yet, McKelvie and Demers (1979) demonstrated differences between low and high imagers in an LTM task when using concrete words but not when using abstract words.

Verbal short-term memory in aphantasics

At the level of STM, Dual Coding Theory might be comparable with intersensory facilitation. ‘Intersensory facilitation occurs if the response to a stimulus [...] from one sensory modality is in some way furthered by the concurrent stimulation of one or more other sensory modalities. The facilitation may manifest itself as (1) a speed-up of reaction time, (2) lowering of the sensory threshold for detection or discrimination of stimuli, or (3) an increase in the rate of recognition, identification, or classification of stimuli’ (Colonius & Diederich, 2012, p. 1). Since the Reverse Hierarchy Model (Pearson, 2019) assumes mental images to be weaker perceptions, it is plausible to assume that the effect is transferable to imagery modalities. This is shown, for example, in the generally faster reaction times of non-aphantasics in comparison to aphantasics in visual search tasks (Monzel et al., 2021), since non-aphantasics can use both verbal and visual representations of target stimuli. However, again, Milton et al. (2020) did not find differences between aphantasics and non-aphantasics using the LMT (immediate recall; Wechsler, 1997), and Zeman et al. (2010) did not find differences in the Auditory Immediate Index of the WMS (Wechsler, 1997). On the other hand, McKelvie and Demers (1979) could show that, in contrast to the LTM task, high imagers outperformed low imagers in a STM task, even when using abstract words, indicating that abstract words can be associated with an image at least for a short time (e.g., a pigeon as symbol for peace). Nonetheless, this association seems to decay over time and, therefore, cannot be consistently restored in a long-term recall (McKelvie & Demers, 1979).

Task characteristics

As mentioned above, the lack of objectively measurable differences between aphantasics and non-aphantasics in most of the studies does not necessarily mean that the lack of visual imagery does not impair the respective memory components but that task characteristics may allow compensating these deficits by alternative solving strategies. For this reason, the dimensions that potentially allow the use of alternative strategies are considered below.

Task difficulty

Jacobs et al. (2018) first noticed that aphantasics only performed worse in the most difficult trials of their visual STM task. It was interpreted that this was due to the failure of alternative solution strategies. Besides, the Famous Face Test (FFT) and the Recognition Memory Task (RMT) for words (Warrington, 1984) showed ceiling effects in Milton et al. (2020) and were therefore unable to identify possible differences between aphantasics and non-aphantasics. Therefore, memory tasks used to identify mnestic deficits in aphantasics should be able to differentiate in the high-performance spectrum. To increase diagnosticity, either more difficult items could be used, involving complex item construction and long validation processes, or, more economically, a time limit can be added. Crowder (2018) showed that in a Mental Rotation Task aphantasics and non-aphantasics did not differ in solution accuracy but in solution speed. Under the assumption that these differences in speed can be transferred to memory tasks because less efficient strategies are used here as well, a time limit should increase the likelihood of the memory task to detect differences between aphantasics and non-aphantasics. Essentially, all tasks mentioned above which did detect differences between aphantasics and non-aphantasics or low and high imagers, respectively, were either speed tests (Marks, 1973; McKelvie & Demers, 1979) or had other constraints to increase task difficulty (e.g., articulatory suppression in Brooks’ Matrix Task, Zeman et al., 2010; or subanalyses with the most difficult trials in Jacobs et al., 2018).

Retrieval condition

The difficulty of a task is linked to its retrieval condition. Since perceptual object recognition is generally unimpaired in aphantasics (see Bartolomeo, 2008, for a comment on the double dissociation of visual imagery and visual perception), memory recognition tasks should be easier for aphantasics than free recall or reproduction. Farah (1984) distinguishes visual imagery generation deficits (= retrieval failure; aphantasia) and long-term visual memory deficits (= storage failure), as the latter would also include limitations in object recognition. Thus, to identify differences between aphantasics and non-aphantasics, recall and/or reproduction should be used. This is in line with the studies mentioned above, since almost no study involving pure recognition showed differences between aphantasics and controls (e.g., in the RMT and FFT, Milton et al., 2020). Milton et al. (2020) were only able to show differences between aphantasics and hyperphantasics (but not controls) in the Graded Buildings Test (GBT, Evans et al., 1995), but this could be due to the more extreme differences between the groups and the inherently more difficult nature of the GBT requiring adequate LTM over several years. Besides, McKelvie and Demers’ (1979) recognition condition was the only time-limited memory task which did not show differences between the groups, further supporting the thesis of recognition tasks being too easy.

Subcategories of stimuli

As demonstrated by McKelvie and Demers (1979), the advantage of visual imagery in remembering abstract words seems to decline over time. Therefore, in visual LTM tasks, concrete words should be used predominantly in order to identify differences between aphantasics and non-aphantasics. Likewise, a subdivision can be made regarding the visual memory, since visual imagery is known to have two non-correlated facets, object and spatial imagery. While object imagery reflects the richness of detail, spatial imagery reflects the spatial relations of mental images (Blajenkova, Kozhevnikov, & Motes, 2006; Kozhevnikov, Blazhenkova, & Becker, 2010). Object imagery is correlated with vividness of mental imagery (and associated with autobiographical memory; Vannucci, Pelagatti, Chiorri, & Mazzoni, 2016), whereas spatial imagery is not (Blajenkova et al., 2006). Accordingly, aphantasics report worse object imagery, but neither worse spatial imagery nor worse spatial memory than non-aphantasics (Dawes et al., 2020). Therefore, when assessing visual memory deficits in aphantasics, the focus should be on object information specifically. In accordance with this, all tasks mentioned above which were able to identify differences between aphantasics and non-aphantasics or low and high imagers, respectively, involved either object memory (Marks, 1973; McKelvie & Demers, 1979) or concrete words (McKelvie & Demers, 1979).

Eventually, when combining all three criteria, (1) a high task difficulty, (2) recall or reproduction conditions, and (3) a focus on concrete words and object information, the probability of detecting objectively measurable memory differences between aphantasics and non-aphantasics should be maximized. And indeed, looking at the studies mentioned above, all tasks which met all three criteria were able to reveal memory differences between participants with low and high vividness of visual imagery (Jacobs et al., 2018; Marks, 1973; McKelvie & Demers, 1979). New data should be collected with tasks that also meet these criteria to check if our assumptions stand up to empirical scrutiny.

Hypothesis

While deficits in the visual LTM of aphantasics seem to be the scientific consensus and are at least suspected for the visual STM, it is yet to be clarified whether aphantasics also show impairments in the verbal STM and LTM. Based on the Dual Coding Theory, the following hypothesis is tested:

Aphantasics, as characterised by the Vividness of Visual Imagery Questionnaire (VVIQ, Marks, 1973), show worse memory performance than non-aphantasics independent of memory modality (visual vs. verbal) and memory system (STM vs. LTM).

To test this hypothesis only congenital aphantasics will be examined, since (1) acquired aphantasia (e.g., due to an accident) might more likely go along with other functional losses, (2) congenital aphantasia is particularly frequent (Zeman et al., 2015), and (3) to restrict ourselves exclusively to anterograde memory deficits.

Method Participants

Since the prevalence of aphantasia is only 2–3% (Faw, 2009; Zeman et al., 2020), an a priori power analysis was conducted (Faul, Erdfelder, Lang, & Buchner, 2007) to determine the sample size necessary to detect the expected effects. According to the self-reports of aphantasics in Dawes et al. (2020), the smallest differences were expected in semantic memory (r = .27). Hence, this effect size was used for the power calculation. Common values of .05 and .80 were chosen for alpha level and power. The calculation resulted in a required sample size of N = 78, leading to an aimed sample size of 100. One person had to be excluded due to incomplete data, resulting in N = 99 participants in total.

The final sample consisted of N = 67 congenital aphantasics (VVIQ ≤ 23; criterion according to Zeman et al., 2020) and 32 controls (VVIQ > 23). Groups were matched to avoid differences in age, gender, and education (see Table 1), since these variables could influence the performance in the memory tasks. The participants were contacted via the database of the Aphantasia Research Project. In sum, the sample is the biggest sample investigating memory effects of aphantasics with objective memory tasks so far.

Table 1. Sociodemographic data for aphantasics and controls Aphantasics Controls t/χ² p Age M 30.94 27.56 SD 11.41 6.85 1.83 .071 Gender Male (%) 65.7 87.5 Female (%) 25.4 6.3 Others (%) 9.0 6.3 5.74 .057 Education Primary school (%) 1.5 3.1 Secondary school (%) 6.0 3.1 A-levels (%) 38.8 46.9 University degree (%) 53.7 46.9 1.17 .760 Questionnaires

The ability to create visual imagery was measured by means of the Vividness of Visual Imagery Questionnaire (VVIQ, Marks, 1973) as in many other psychological studies on visual imagery (see McKelvie, 1995). The VVIQ assesses the vividness of visual imagery by presenting 16 situations which the participants have to visualize. After that, they are asked to rate the vividness of their imagery defined as the proximity to actual perception on a 5-point Likert scale ranging from No image at all, you only ‘know’ that you are thinking of the object to Perfectly clear and as vivid as normal vision.

Memory tasks

The memory tasks were constructed in the style of the German Inventar für Gedächtnisdiagnostik (Inventory for memory diagnostic; Baller, 2005; Kalbe, Baller, Brand, & Kessler, 2006), as the inventory was designed to distinguish between visual and verbal STM and LTM. Because of its applicability in the pathological range and its continuing diagnostic value in the higher performance range, it is suitable for revealing differences between aphantasics and non-aphantasics, both in case the memory of aphantasics is only weakly impaired in some memory components and in case it is severely impaired in others. The criteria of task difficulty, retrieval condition, and subcategories were also taken into account when selecting the tasks. Therefore, all tasks involve a time limit and concrete words or – at least partially – object information. The requirement for the use of recall/reproduction tasks was only lifted for the verbal LTM to keep the task as close as possible to the original. However, ongoing task difficulty was ensured by a variety of phonologically and semantically similar distractor items in a word list of 56. In contrast, in the RMT (Warrington, 1984), only one single distractor item is displayed per trial.

Verbal short-term memory

To assess verbal STM, the participants were presented with an audio recording of the following word list: apple, rabbit, coast, clock, storm, carpet, gate, paper, brain, house, table, tree, book, and barge. Afterwards, they were asked to type in the words containing at least one ‘r’. Each correct answer yielded a score of 1. The task was repeated three times in a row, resulting in a maximum total score of 21. When selecting the words for the task, care was taken to select only one- to two-syllable words that are clearly phonologically different in order to exclude word length effects and phonological similarity effects. Since concrete words seem particularly diagnostic for differences between aphantasics and non-aphantasics, particularly concrete words were selected.

Visual short-term memory

To assess visual STM, participants were presented with a screen of seven squares, in which there were lines with different orientations for a duration of 45 s. They were then given 30 s to indicate the orientation of the seven lines on a screen of seven empty squares by clicking the start and end points of the respective lines. Each correct answer yielded a score of 1. The task was repeated three times in a row, resulting in a maximum total score of 21. Since the lines had to be remembered both in correct orientation and their position in relation to each other, spatial and object memory facilitate the solution of the task (Baller, 2005), which is further supported by the convergent correlation with the figural memory subtask of the Wechsler Memory Scale – Revised (WMS-R, Wechsler, 1987).

Verbal long-term memory

To assess the verbal LTM, participants were presented with a list of 56 words. Within 150 s, they had to decide for each word whether or not it was part of the recording they had previously listened to in the verbal STM task by clicking ‘Yes’ or ‘No’. Each correct hit yielded a score of 1 while each false positive response yielded a score of minus 1, resulting in a maximum total score of 14. Distractors were either phonologically (14 items) or semantically (14 items) similar to the targets or completely unrelated (14 items).

Visual long-term memory

To assess visual LTM, participants were presented with a screen of five complex geometric figures for a duration of 75 s. After completing the VVIQ, they were then given 90 s to select incomplete versions of the previously seen figures, each out of a set of five. Additionally, they were asked to complete these incomplete figures by clicking the corner points of the missing features (e.g., the corner points of a triangle). Each correct selection and each correct addition of a missing feature yielded a score of 1, resulting in a maximum total score of 10. Therefore, to achieve full points, recognition as well as reproduction was needed. Furthermore, it was ensured that the figures were sufficiently complex in order to make verbal encoding difficult. Thus, distractors did only differ in one feature (e.g., size, position, or orientation of one component) but never in the basic form, rendering the encoding of the basic form insufficient for recognition.

Procedure

The study was conducted from 14 April 2021 to 28 April 2021 via the online platform soscisurvey.de (Leiner, 2021). Examination language was English. The participation was anonymous and participants provided informed consent before commencing the study in accordance with the World Medical Association Declaration of Helsinki (World Medical Association, 2013). To avoid disturbances, participants were asked to ensure quiet surroundings and to focus their attention solely on the tasks to follow. The memory tasks were performed in the order listed above, with the VVIQ serving as filler task between encoding and reproduction of the visual LTM task.

Statistical analyses

Experimental groups were allocated on the basis of the VVIQ sum scores. To make the four memory tasks comparable, the scores of the individual tasks were z-standardized. Subsequently, a mixed ANOVA was calculated with group membership (aphantasics vs. non-aphantasics) as between-subject factor and memory modality (visual vs. verbal) as well as memory system (STM vs. LTM) as within-subject factors.

Results

The 2 × 2 × 2 ANOVA revealed a main effect of group, F(1, 97) = 5.92, p = .017, η2 = .06, but no significant interaction effects with memory modality, F(1, 97) = 0.68, p = .795, or system, F(1, 97) = 0.14, p = .706. Thus, non-aphantasics (M = 0.26, SE = 0.13) exhibited higher scores than aphantasics (M = −0.11, SE = 0.09), regardless of the memory component. Means and standard errors per condition are shown in Figure 1. A triple interaction effect between group, memory modality, and memory system was not significant either, F(1, 97) = 0.775, p = .381.

Memory scores dependent on group and memory component. Depicted are means ± 1 SEM.

Since some participants were not able to solve all items within the time limits and it cannot be excluded that this happened due to external distractions, we repeated the analyses after exclusion of outliers. Coincidentally, all outliers performed below average and no outlier showed above average performance. However, even after the exclusion of scores outside of three standard deviations (4 aphantasics, 1 non-aphantasic), F(1, 92) = 7.28, p = .008, η2 = .07, or even two standard deviations (13 aphantasics, 4 controls), F(1, 80) = 5.81, p = .018, η2 = .07, the main effect of group remained significant. Similarly, there were still no interaction effects that would imply that memory deficits in aphantasics differ depending on memory component.

Discussion

Despite extensive outlier analyses, no interaction effects between group, memory modality, and memory system could be found that would imply different degrees of impairment in the investigated memory components in aphantasics. Instead, a main effect of group was found, suggesting that aphantasics perform worse than non-aphantasics in all investigated memory components, that is, visual as well as verbal STM and LTM. This effect was even stable when people with particularly poor memory performance were excluded, although these were mainly aphantasics. This shows that the differences in memory performance were not only driven by a small subgroup of aphantasics, even though there seem to be individuals within the aphantasic group who are even more affected by their lack of visual imagery than others, possibly due to a lack of alternative solving strategies (cf. Jacobs et al., 2018; Zeman et al., 2010).

These results are in line with the self-reported deficits of aphantasics in all memory components (Dawes et al., 2020) as well as with early results of studies involving low and high imagers (Marks, 1973; McKelvie & Demers, 1979), but contradict the results of Jacobs et al. (2018), Milton et al. (2020), and Zeman et al. (2010) who did not find any differences between aphantasics and non-aphantasics in objective memory tasks. However, these null results can probably be attributed to task (e.g., low difficulty, recognition instead of recall, and unfavourable subcategories of stimuli) as well as sample characteristics (e.g., high intelligence), which allow the reliance on non-visual alternative solving strategies. Therefore, the results of the present study can be interpreted as confirmation that visual imagery plays an important role for memory in all its components. While Scene Construction Theory (Hassabis & Maguire, 2007, 2009; Mullally & Maguire, 2013) explains deficits in autobiographical memory due to shared neural mechanisms of visual imagery and autobiographical memory, which should be generalizable to each visual memory due to Reverse Hierarchy Model (Pearson, 2019), Dual Coding Theory (Paivio, 1990, 2007, 2013) is able to explain deficits in verbal memory due to missing complementary visual imagery strategies. This further strengthens the assumption of jointly organized semantic–associative networks of verbal and visual memory (Vandenberghe et al., 1996) and provides evidence that aphantasia might not just be a visual imagery generation deficit (Farah, 1984) but a more general information generation deficit based on the lack of visual representations. Thus, the connectivity between frontal regions, the hippocampus, and the occipital lobe (Pearson, 2019) does not only seem to play an important role in imagery but also in memory retrieval.

Regarding some trends in the data, it seems that differences between aphantasics and non-aphantasics are larger in visual STM (d = 0.53) than in verbal STM (d = 0.30), while this trend disappears in LTM (visual: d = 0.33, verbal: d = 0.37). The larger difference in visual working memory seems plausible, as aphantasics are not predominantly impaired in the use of verbal strategies (Watkins, 2018; Zeman et al., 2010), although verbal representations cannot be complemented with visual representations. In LTM, however, the larger difference may not be reflected in visual memory, since visual information might have already decayed in non-aphantasics and/or aphantasics might perform better due to our visual LTM task unfortunately being based on recognition. Although these explanations might seem plausible, no significant interaction effect between group, memory modality, and memory system was found, leading to the conclusion that in aphantasia all memory components might be impaired equally. However, since only a small proportion of higher-level memory constructs is covered by our memory tasks, statements regarding the comparable severity of impairments in these different memory components should be treated with caution.

Limitations

Notably, in the present study, task characteristics were not manipulated systematically, which is why – at this point – the criteria of task difficulty, retrieval condition, and subcategories of stimuli cannot be said to be necessary for the diagnosticity of memory deficits in aphantasics. In fact, some of the tasks we used did not fully meet the criteria and were therefore compensated otherwise, in order to change the tasks by Kalbe et al. (2006) as little as possible (e.g., by using a variety of phonologically and semantically similar distractor items in the LTM task, instead of using a recall condition to ensure task difficulty). However, the aim of the study was not to clarify the role of the individual criteria but to show that memory deficits in aphantasics might not be limited to autobiographical memory, for which the criteria have proven to be sufficient. Since aphantasics are a rare sample (Faw, 2009; Zeman et al., 2020), the authors decided not to include the criteria as further independent variables in order to keep the study as short as possible to avoid discouraging aphantasics by the length of the task. However, now that differences between aphantasics and non-aphantasics were found in all memory components, future research should investigate the criteria in greater detail, thereby drawing conclusions about alternative strategies and non-visual processing.

It could be argued that the constraints under which the deficits in aphantasics were found in our study (i.e., task characteristics) were too artificial to actually have an influence on aphantasics in their daily life. In reality, however, there are many scenarios where detailed information of all stimuli categories has to be reproduced instead of just being recognized (sometimes even under time pressure), for example remembering various device unlocking patterns or learning a large amount of vocabulary. Therefore, we assume our constraints to be ecologically valid.

Regarding our method, there are some other limitations which should be discussed. First, although the verbal targets of the memory tasks were selected to avoid word length and phonological similarity effects, it cannot be assumed that these effects were eliminated entirely. However, since both groups, aphantasics and non-aphantasics, can process stimuli verbally, these effects should affect both groups similarly. Second, it is debatable if the visual STM task actually involves object memory, although showing convergent validity to the figural memory task of the WMS-R (Baller, 2005). However, it can be assumed that object memory at least facilitates solving the task, therefore leading to the observed performance differences between aphantasics and non-aphantasics. Third, as the VVIQ was used as the filler task in the visual LTM task, the period between encoding and recall of the stimuli was not standardized in length. However, since aphantasics tend to answer each item in the VVIQ similarly (= lowest vividness scores possible) and non-aphantasics vary more in their answers, it is plausible to assume that aphantasics need less time than non-aphantasics. Still, their performance in the visual LTM task was worse than the performance of non-aphantasics. Fourth, the present study did n