Visual working memory resolution defined by figural complexity in kindergarten children

Abstract

Visual working memory (VWM) allows us to store and manipulate incoming visual information briefly. Information acquisition (i.e., encoding) accuracy is critical for VWM to function properly. The accuracy of very young children's VWM encoding has not been explained adequately in previous studies. Therefore, this study clarified it by manipulating the complexity of the visual stimuli and examining kindergarten children's performance in a recognition task. Furthermore, we examined the relationship between encoding accuracy and the 4- to 6-year-old children's individual traits in a subanalysis, as individual traits (such as IQ and attention to detail—a trait of autism spectrum disorder) reportedly affect VWM capacity. The results revealed that distinguishing between target and probe stimuli becomes more difficult as stimulus and discrimination complexity increase. In addition, this study results in narrow attention (attention to detail) that could contribute to VWM capacity saving if VWM capacity is sufficient. However, if the VWM's capacity is exceeded, the relationship with IQ, such as the simultaneous processing score, is strengthened. This study clarified the degree of accuracy of information retained by preschool children aged 4 to 6 years. In addition to providing basic knowledge about VWM, we believe the findings can be useful in education and other fields.

Introduction

Visual working memory (VWM) allows us to store and manipulate incoming visual information for a short period. To recognize information as new, we need to not only retain old information, but also identify the difference between new and old information. Accurate information acquisition (i.e., encoding) is critical for the VWM to function properly. To model VWM's conceptual structure, we used change detection and delayed estimation tasks to measure VWM. Other research has explored VWM's capacity or the number of memory items it can retain. 

Previous studies have examined the factors affecting VWM capacity, including visual complexity. Among them, several used geometric figures to define visual complexity (Alvarez & Cavanagh, 2004; Awh, Barton, & Vogel, 2007; Brady & Alvarez, 2015; Eng, Chen, & Jiang, 2005; Leclercq, Maillart, Pauquay, & Majerus, 2012; Luria, Sessa, Gotler, Jolicoeur, & Dell'Acqua, 2010; Taylor, Thomson, Sutton, & Donkin, 2017), whereas others used the number of feature dimensions (such as color, shape, and line direction) in visual stimuli as a complexity index (Fougnie, Asplund, & Marois, 2010; Olson & Jiang, 2002; Wheeler & Treisman, 2002). Some studies have reported a greater burden of feature dimensions than the number of objects that consume memory capacity. The increased complexity defined by the number of features reduces the capacity and resolution of the stored representation (Fougnie et al., 2010; Olson & Jiang, 2002). 

VWM capacity is known to be affected by several factors, including age, individual fluid intelligence, and language comprehension (Alloway & Alloway, 2010; Daneman & Carpenter, 1980; Fukuda, Vogel, Mayr, & Awh, 2010; Oberauer, Schulze, Wilhelm, & Süss, 2005). Studies have estimated adults’ VWM capacity to be 4 ± 1 (Alvarez & Cavanagh, 2004; Luck & Vogel, 1997). According to research on VWM in children, relative to fluid intelligence (e.g., learning new skills, solving tasks, and acquiring new knowledge), VWM develops with age (Alloway & Alloway, 2010; Alloway, Gathercole, Kirkwood, & Elliott, 2009; Gathercole & Pickering, 2000; Gathercole, Pickering, Knight, & Stegmann, 2004; Gathercole, Service, Hitch, Adams, & Martin, 1999). Riggs, McTaggart, Simpson, & Freeman, 2006 reported that 10-year-olds performed better than 5-year-olds in a change detection task. The estimated VWM capacity (Cowan's K) was 1.52 for 5-year-olds, 2.89 for 7-year-olds, and 3.83 for 10-year-olds. Research suggests that children reach their maximum VWM capacity at the age of 10 (Astle, Nobre, & Scerif, 2012). Furthermore, VWM capacity development is related to age and attentional control (Astle & Scerif, 2011). 

There is a connection between mental representations and VWM. Albers, Kok, Toni, Dijkerman, & de Lange, 2013 found evidence of common internal representations in early visual areas. People develop mental representations from a young age; for example, 2-year-olds often play with imaginary objects or friends. However, image representation expands after the age of 3, when pretend and creative play become more frequent, such as increased shared play scripts (e.g., roles and situations) and setting details. As children develop, their ability to represent images becomes richer, as evidenced by their interactions with imaginary people and objects. High-quality pretend play reportedly helps to develop abstract thinking, as well as social and linguistic competence (Bergen, 2002). Some studies have manipulated representational objects. For example, Frick, Hansen, & Newcombe, 2013 reported improved performance with age when comparing the percentage of correct responses in a mental rotation task among 3-, 4-, and 5-year-olds. However, the responses were biased, with many 3-year-olds providing random responses that did not exceed chance levels, suggesting that they might not have been able to manipulate mental representation sufficiently. In contrast, the responses of the 4-year-olds were above chance levels and not random, suggesting that accurate mental representation acquisition ability develops around the age of 4. 

The creation of mental representations corresponds with the role of encoding information in VWM. This is because mental representation exhibits aspects that closely resemble the encoding of VWM when internalizing information from the external world and allows for the free manipulation of retained information. The ability to freely manipulate information suggests the ability to maintain information in greater detail and accuracy. This ability is thought to develop around the age of four, as suggested by previous research (e.g., Frick et al., 2013). Although most studies on VWM in preschool children have focused on memory retention, few have examined their VWM encoding ability. Memory consists of three processes: encoding, maintenance, and retrieval. Memory requires the ability to retain broad information and the ability to retain detailed information. However, most findings on memory development have been concerned with the quantity of memory, and few have asked about its quality. 

For a better understanding of VWM development, we need to clarify the accuracy of VWM encoding during periods of altered mental representation abilities. Therefore, this study focused on visual complexity to clarify the extent to which detailed information can be captured by kindergarten children whose mental representations are newly developed and still developing. Although it is difficult to isolate the memory process clearly, we believed it would be possible to examine the degree of resolution of the preschool children's memory objects by varying the similarity of the probe stimuli presented during the recognition phase and comparing their memory performance. 

Previous studies have demonstrated that observer-related factors and stimulus characteristics influence VWM ability. Attentional orientation is necessary for the efficient use of VWM capacity. Having a well-defined attention direction should result in less distractor interference, suggesting that narrow attention directionality might enable VWM to operate more efficiently. It is well-known that attentional control develops with age. However, Richmond, Thorpe, Berryhill, Klugman, & Olson, 2013 reported that attention direction is related to age and autism spectrum disorder (ASD) traits. Individuals with ASD tend to focus on details in visual information and other forms of information, such as auditory information (Foxton et al., 2003; Haesen, Boets, & Wagemans, 2011). Therefore, they may be better at filtering out distracting stimuli so their VWM capacity can be conserved. Richmond et al. (2013) reported a positive correlation between ASD-like cognitive styles and VWM performance. Moreover, Hamilton, Mammarella, & Giofrè, 2018 used a discrimination task to assess school-aged children's ability to distinguish figure sizes and reported a positive correlation between scores on systemizing cognition and discrimination performance. 

However, previous studies have not clarified the relationship between detail-oriented attention and VWM, the foundation of attention and cognitive development in kindergarten children. In general, an ASD-like cognitive style might naturally conserve VWM capacity in young children with underdeveloped attentional control, resulting in higher VWM performance. This study determined whether the complexity of visual information affected the accuracy of kindergarten children's encoding and whether complex objects require more VWM resources than simple objects, as is the case with adults. Additionally, we examined whether ASD-like cognitive traits, particularly attention to detail, could improve VWM capacity in kindergarten children. We hypothesized that children aged 4 to 6 years who were attentive to detail could develop a high-resolution memory trace and perform better on a VWM task using complex stimuli. 

Methods

Participants

In this study, all children aged 4 to 6 years were recruited from five kindergartens in Kanazawa City, Japan. We calculated the required sample size using G*Power (version 3.1.9.6) with an expected power of 0.8 and a medium effect size of 0.25 (Cohen, 1988) to detect a significant interaction between complexity and condition in an analysis of variance (ANOVA), and that was 21. The number of participants was 31, but seven were excluded for the following reasons. A participant had visual abnormalities (poor visual acuity, strabismus, and color vision deficiency) and was, therefore, excluded. Additionally, two children were excluded because their IQs were below 70 or above 130 (2 SD away from the mean), respectively, as determined by the Mental Processing Scales in the Kaufman Assessment Battery for Children Second Edition (K-ABC II). 

Furthermore, four children could not complete the recognition task; consequently, our final analysis included 24 children (12 boys and 12 girls; M = 69.29 months; SD = 9.36 months). 

The participants did not have a history of psychiatric disorders and did not take any medications regularly. 

Procedure

This study was approved by the Kanazawa University Medical Ethics Committee and conducted in accordance with the Declaration of Helsinki (#2020-308). After the provision of instructions to the caregivers, we started the experiment once written informed consent was obtained. After entering the playroom, a participant and an experimenter interacted playfully for approximately 10 minutes to establish rapport. The experimental procedure began once we confirmed that the participant initiated a spontaneous conversation with the experimenter or responded to the experimenter's question. 

We conducted a vision screening test using the Morizane dot card (Morizane & Morizane, 1990), which estimates visual acuity at 30 cm based on whether viewers can detect black dots (eyes) within the outline of a bear or rabbit's face. We used a part of the K-ABC Ⅱ to assess cognitive processing. Additionally, to ensure that the participants could count the elements used in the stimuli and understand the concept of number comparison, we checked whether they could count to 10. Subsequently, we conducted a practice session for a recognition task. Meanwhile, the caregivers in the other room completed the Japanese version of the Autism Spectrum Quotient (AQ) (Baron-Cohen et al., 2001) for children and the Social Responsiveness Scale, Second Edition (SRS-2) (Constantino & Gruber, 2012). In total, 3 hours were spent conducting the experiment. 

Recognition task

We presented the stimulus using a 13-inch (1, 920 × 1,080 pixels) laptop computer (ThinkPad X395, Lenovo Group Limited, Beijing, China) running PsychoPy (Peirce, Hirst, & MacAskill, 2022) on Windows 10 (Microsoft Corporation, Redmond, WA). Responses were collected using the laptop's built-in keyboard. We generated the visual stimuli using an online drawing tool (Iterograph, http://iterograph.laboiteatortue.com), producing iterative patterns in which two of the four simple shapes (circle, square, triangle, and star) overlapped concentrically (Figure 1). The combination of figures presented is illustrated in Supplementary Material 1. 

Figure 1.

 

Example stimuli in which figural complexity was manipulated by the number of iterations (ITs) of two simple figures, squares and circles. We used IT4, IT6, and IT8 as the memory stimuli in the recognition task.

The complexity of a geometric figure was determined by the number of iterations (ITs). We prepared 66 figures by varying the ITs from IT1 to IT11. The stimulus size was 800 × 800 pixels, and the observation distance was approximately 57 cm. Figure 2 presents the experimental procedure. The number of repetitions of the images was random, but the same images were presented at least twice. Each trial proceeded as follows. 

Figure 2.

 

Recognition task procedure. First, we presented a fixation point on the screen (700 ms). Second, during the encoding phase (5,000 ms), we presented a pair of stimuli of varying complexity. Third, we immediately followed up with the recognition phase of the experiment, conducting 45 trials, each repeated 4 times as a block (180 trials in total), with breaks lasting between 5 and 300 seconds after every 5 trials during the block.

Fixation phase

We presented a fixation point in the center of the screen for 700 ms, accompanied by a sound that attracted the participant's attention. 

Encoding phase

We presented two memory stimuli side by side (encoding phase: 5,000 ms). We used IT4, IT6, and IT8 as our memory stimuli. 

Recognition phase

Immediately after the encoding phase, we presented a probe stimulus in the center of the screen (recognition phase), showing either one of the two memory stimuli (target stimulus) or a different IT figure (novel stimulus) consisting of the same shapes used in the target stimulus but with different ITs. We manipulated the difficulty of the recognition task by increasing or decreasing the number of ITs. There was an increase or decrease of one (difficult condition) or three (easy condition). When the probe stimulus was the target, we defined the trial as the same condition. The participants were asked to press a key to indicate whether the stimulus was present or absent in the memory stimulus. We acquired the response 200 ms after the start of the recognition phase to prevent error responses (i.e., task-irrelevant responses). If a participant did not respond, we forcibly terminated the recognition phase after 2 minutes. In the difficult condition, there was always a difference of at least two between the number of IT in a novel stimulus and that in the nontarget stimulus in the encoding phase. For example, if IT6 vs IT9 was present in the encoding phase, the stimulus presented during the recognition phase was IT3 (easy condition) and IT5 or IT7 (difficult condition) (Figure 3). 

Figure 3.

 

Examples of probe conditions presented by difficulty condition. We used IT4, IT6, or IT8 as the memory stimuli in the recognition task. IT = iterations.

The target stimulus complexity (IT4, IT6, and IT8), location (left and right), and probe stimuli conditions (difficult, easy, and same) were all counterbalanced and presented in random order. 

Individual traits

We used the K-ABC II, AQ, and SRS-2 to examine the correlation between the recognition task performance and the participants’ socio-cognitive traits, particularly in relation to a visual scene in the questionnaires. The results from these analyses may provide insights into how children encode, store, and retrieve visual objects. 

K-ABC II

We used the K-ABC II to assess cognitive abilities and academic achievement. This instrument is based on the Kaufman model, which combines Luria's (1966) sequential simultaneous processing dichotomy and the Cattell–Horn–Carroll theory on the structure of human cognitive abilities. The K-ABCII can measure cognitive and academic achievement separately. This study focused on cognitive ability. This test assesses eight abilities, with Luria's four subscales measuring cognitive abilities (sequential processing, simultaneous processing, planning, and learning) and four subscales measuring academic achievement (knowledge, reading, writing, and arithmetic). We measured only four basic cognitive abilities to reduce the participants’ total time commitment. 

AQ

For children, we used the Japanese version of the AQ, which is usually used to screen for ASD. The AQ comprises five subscales: social skills, communication, imagination, attention to detail, and attention switching. In addition to the AQ total score (the sum of the five subscales), we also selected one subscale as the variable of interest, namely, attention to detail. We expected the participants with high attention to detail scores to achieve higher recognition performance since detailed encoding during the complex stimulus observation might help memory discrimination. 

SRS-2

The Japanese version of the SRS-2 for children is a social measure of the severity of social symptoms related to ASD. It comprises five subscales: social awareness, social cognition, social communication, social motivation, and restricted interests and repetitive behavior (RRB). The SRS-2 generates raw scores for the five subscales, which are then converted into T scores. The T scores are then standardized based on the respondents’ gender and age. The standardized scores range from 32 to 114, with a mean of 50 and a standard deviation of 10. Specifically, we consider the SRS-2 measure of restricted interest to be related, in a broad sense, to attention to detail. Therefore, we conducted an exploratory analysis focusing on the RRB and the attention to detail subscale of the AQ. 

Analysis

ANOVA was performed using R software, Version 3.5.2 (https://www.R-project.org), and its add-on function Anovakun, Version 4.8.6 (http://riseki.php.xdomain.jp/index.php?). We performed a repeated-measures two-way ANOVA to compare the effects of the probe (difficult, easy, or same) and complexity (IT4, IT6, or IT8) with the proportion of correct responses during the recognition task. We used Shaffer's modified sequentially rejective Bonferroni method to control the family-wise error rate on multiple comparisons in the post hoc test. The significance level was set at 5%. Using Pearson's correlation analyses, we examined the relationship between the performance of the recognition task in the difficult condition and the individual traits assessed by questionnaires. Based on our reasoning, we only reported the results of the difficult condition, which might require a greater level of attention to detail in memory trace during the recognition phase. 

Results

Recognition task

Figure 4 presents the mean proportion of correct responses in the recognition task. The ANOVA revealed a significant main effect of probe and complexity, F (2,46) = 17.18, p < 0.01, η² = 0.28; F (2,46) = 7.18, p < 0.01, η² = 0.01. Additionally, it indicated a significant interaction between the effects of probe and complexity, F (4,92) = 13.02, p < 0.01, η² = 0.06. Our post hoc analysis revealed the results of the multiple comparison tests between the levels of complexity across each probe condition. In the difficult condition, IT4 had the highest mean proportion of correct responses, t (23) = 3.64, p < 0.01 for IT4 to IT6, and t (23) = 4.13, p < 0.01 for IT4 to IT8. However, we found no significant differences between IT6 and IT8, t (23) = 0.51. p = 0.62. In the easy condition, we found significant differences between IT4 and IT8, t (23) = 4.56, p < 0.01, IT6 and IT8, t (23) = 2.41, p = 0.02, as well as IT4 and IT6, t (23) = 2.31, p = 0.03. In the same condition, the IT4 condition demonstrated significantly lower performance than the IT6 and IT8 conditions, with t (23) = 4.08, p < 0.01 for IT4 to IT8 and t (23) = 3.33, p < 0.01 for IT4–IT6. Moreover, the post hoc tests revealed the results of the multiple comparison tests between the types of probe stimulus across each complexity condition. In the IT4 condition, the easy condition had the highest mean proportion of correct responses, t (23) = 5.57, p < 0.01 for easy–difficult, and t (23) = 2.47, p = 0.02 for easy–same. However, we found no differences between the difficult and same conditions, t (23) = 1.28, p = 0.21. In the IT6 condition, the difficult condition showed significantly lower performance than the other difficult conditions, t (23) = 9.17, p < 0.01 for the easy–difficult and t (23) = 5.11, p < 0.01 for same–difficult. We found no differences between the easy and same conditions, t (23) = 0.71, p = 0.48. In the IT8 condition, the difficult condition demonstrated significantly lower performance than the other difficult conditions, t (23) = 7.57, p < 0.01 for easy–difficult, t (23) = 5.41, p < 0.01 for same–difficult and t (23) = 2.23, p = 0.04 

Figure 4.

 

Mean proportion of correct responses for different complexities and probes. An error bar indicates ±1 SD. Asterisks denote statistically significant differences (*p < 0.05, ** p < 0.01) from the result of the multiple comparisons following analysis of variance (n = 24). IT = iterations.

Discrimination bias

We then conducted one-sample t tests to examine discrimination bias split by chance level. Because the participants were asked whether they had previously seen the stimuli in the recognition phase, the interpretation of a correct response rate changed depending on whether it was above or below the chance level of 0.5. When the correct response rate is less than 0.5 in either the easy or difficult conditions, the participants made a confident decision that the target and the probe stimuli are the same even though they are different. In contrast, when the response rate is less than 0.5 in the same condition, it indicates that participants are actively judging the stimuli to be different. 

Consequently, the difficult condition did not exceed the chance level in IT4, t (23) = 0.30, p = 0.77 , and the easy condition IT 8, t (23) = 1.88, p = 0.07. However, in the other conditions, the chance level exceeded and the responses were biased (IT4), easy condition t (23) = 11.92, p < 0.01, same condition t (23) = 2.93, p < 0.01, (IT6) difficult condition t (23) = −3.10, p < 0.01, easy condition t (23) = 4.08, p < 0.01, same condition t (23) = 6.33, p < 0.01, and (IT8) difficult condition t (23) = −3.45, p < 0.01, same condition t (23) = 6.95, p < 0.01 (Figure 4). 

Exploration of individual differences: Correlation coefficients between recognition and individual traits

VWM capacity is also related to individual factors. Therefore, we calculated correlation coefficients in an exploratory manner. This is because we did not have sufficient statistical power (0.56) to perform a correlation analysis in this dataset due to the post-hoc analysis. 

We investigated Pearson's correlation coefficients to determine the relationship between the proportion of correct responses to the recognition task and individual traits assessed using the K-ABC Ⅱ, AQ, and SRS-2 in the difficult condition. Table 1 presents the correlation coefficients. 

Table 1.

 

Correlations between the proportion of correct responses at each level of complexity and individual traits. Note. We Derived the Correlation Coefficients from Pearson's Correlation (n = 24). AQ = autism spectrum quotient; IT = iterations; RRB = restricted interests and repetitive behavior; SRS-2 = Social Responsiveness Scale, Second Edition.

 

In the IT4 condition, weak correlations were observed with age, simultaneous processing measures, and attention to detail on AQ. In the IT6 condition, weak to moderate correlations were observed with two IQ measures (sequential and simultaneous processing measures), RRB on SRS-2, and age. Finally, in the IT8 condition, a weak correlation was observed between simultaneous processing measures and attention to detail on AQ. Scatter plots of each variable are shown in Supplementary Material 2. 

Discussion

This study focused on the accuracy of VWM encoding during early childhood by manipulating the figural complexity of stimuli. Furthermore, we explored the relationship between individual differences (i.e., IQ, ASD-like cognitive styles) and VWM capacity in kindergarten children. 

Our two-way ANOVA showed that the mean proportion of correct responses in the difficult condition was lower than in the other conditions. The probe stimuli used under the difficult condition differed slightly from the target stimuli. This result might indicate that the encoded memory traces of the target stimuli were not clear enough to be discriminated from the difficult probe stimuli. Hence, the difficult condition decreased the performance of the recognition task. In addition, a significant main effect on complexity after multiple comparisons showed a decrease in performance according to the degree of complexity defined by the IT within an item. The complexity of the visual stimuli reduced task performance, even though there were always two memories. This finding is consistent with a previous report, which stated that more complex objects consume more memory capacity than simpler objects (Alvarez & Cavanagh, 2004). Thus, we confirmed that VWM resources were allocated according to item complexity. This study showed that preschool children performed significantly worse in the difficult condition. This result represents a discrimination threshold for memory, and perceptual discrimination thresholds may differ. According to Newhall, Burnham, & Clark, 1957, performance on discrimination tasks varies depending on the target stimuli's presentation method (simultaneous or sequential). Therefore, preschoolers are predicted to show better discrimination in this task when discrimination thresholds are measured using the simultaneous matching method. We considered that preschoolers’ discrimination abilities also include the ability to internalize external stimuli. Therefore, it is considered necessary to investigate the discrimination abilities of preschool children using simultaneous matching methods in the future. 

Furthermore, we found significant interactions. The mean proportion of correct responses was lower when the IT was greater than six (IT6) in the easy and difficult conditions. Thus, the participants increasingly viewed stimuli as familiar instead of novel, suggesting that IT6 was at or near the limit of the kindergarten children's encoding capacity and that the memory trace of the figure was ambiguous. In other words, the high rate of correct responses in IT6 and IT8 in the same condition was not because they discriminated against the target stimuli but because they could not distinguish between the target and probe stimuli. However, it may be a consequence of conventional VWM capacity. In contrast, when the correct response rate is considered in terms of encoding accuracy, IT4 or lower is the capacity limit. 

As VWM capacity reached the limit of information retained, the information became ambiguous. Thus, our recognition task under the IT4 condition was within the appropriate level of difficulty for children aged 4 to 6 years; however, the recognition tasks under IT6 and IT8 conditions exceeded their ability to discriminate. 

The recognition task had two phases: encoding and recognition. When considering recognition performance, it is important to distinguish between encoding and matching errors. We assumed that the probe types were reflected in errors during the recognition phase in difficult conditions. This assumption is supported by the result that the participants’ judgments were already at the chance level in the IT4 condition of the difficult condition. If the encoding error occurred more often in the IT4 condition, recognition performance should have been lower (or chance level) even in the easy condition; however, it was not the case. In other words, IT6 exceeded the participants’ VWM capacity. Nevertheless, the ability to match retained information with new information and detect differences was limited within IT4. 

The correct responses were at the chance level at the IT4 in the difficult condition. Some participants could discriminate between stimuli, although others were not able to do so, and it is the threshold for discrimination of VWM in preschool children. To elucidate how individual differences affect accuracy, we calculated correlation coefficients to examine the relationship between VWM capacity and individual abilities in difficult conditions where individual differences are large. 

The results suggest that the individual factors associated with performance vary depending on the level of complexity. In the IT4 condition, weak correlations were between age, simultaneous processing IQ, and attention to detail, which is a subscale of AQ. Next, in the IT6 condition, weak to moderate correlations were found between sequential processing IQ, simultaneous processing IQ, age, and RRB on the SRS-2 scale. Finally, in the IT8 condition, a weak correlation was found between simultaneous processing IQ and attention to detail. As a trend, factors such as simultaneous processing IQ and attention to detail of ASD traits are associated with efficiently operating VWM. It should be noted that the tests used to measure simultaneous processing IQ with K-ABC II relied on visual processing, which may explain the observed relationship with VWM. 

Conversely, the relationship between attention to detail and VWM capacity is consistent with the findings of Richmond et al. (2013) and Hamilton et al. (2018). In other words, Children with greater attention to detail may be efficient when encoding information into VWM. Furthermore, it seems that, as complexity increases, the correlations with cognitive traits (such as IQ and attention to detail) persist more strongly than the factor of age. 

Thus, VWM capacity maximized by the individual trait of attention to detail might help to provide the correct answer when the task difficulty approaches the individual's VWM capacity limit. 

Many studies have reported that children's VWM capacity increases with age as the number of items increases to four (Alloway & Alloway, 2010; Alloway et al., 2009; Gathercole & Pickering, 2000; Gathercole et al., 2004; Gathercole et al., 1999). Moreover, studies have demonstrated that kindergarteners can usually retain up to two items in their memory; more items exceed the average VWM capacity of children aged 4 to 6 years. Our results suggest that detailed attentional orientation may serve as a factor in the efficient use of VWM capacity. 

In this study, a weak correlation was observed with the RRB score, a subscale of the SRS-2 only in the IT6 condition. 

These results suggested that the RRB subscale in the SRS-2 primarily measured repetitive behavior (such as stressed children often exhibit oddly stubborn and inflexible behavior patterns), and AQ may emphasize cognitive aspects more. 

However, it is important to note that these are merely descriptive statical interpretations and may be specific to this data set. From a future perspective, it is important to consider that the results in this study warrant further statistical examination with a larger sample size. 

Conclusions

Our findings suggest that figural complexity within a stimulus affected kindergarteners’ VWM capacity, even with a fixed number of items. We found a capacity limit of IT4 to IT6. Further, ASD-like cognitive styles might help kindergarteners to effectively encode complex figures to conserve VWM capacity. Furthermore, the distracter stimuli were used in this study to adjust the task's difficulty. However, the difference in IT between probe and distracter stimuli varied across conditions. Therefore, it would be necessary in the future to examine memory performance when the difference between probe and distractor stimuli is the same across conditions. 

This study determined kindergarten children's VWM based on reversing the mean proportion of correct responses in the same and easy conditions. However, we cannot exclude the possibility that the children were able to match the same stimulus in the same condition correctly. We obtained our data from only 24 kindergarten children in Japan. Therefore, our results cannot be generalized to different ages or children from other countries without further research. Further studies must examine the development of VWM capacity and children's ages. Our study used AQ as a measure of the degree of detail-oriented attention. However, none of the children had been diagnosed with ASD. In our opinion, children with ASD and higher AQ scores should perform better during this recognition task using complex stimuli; however, further research is needed to confirm this hypothesis. 

This study provides insight into children's cognitive development by examining VWM accuracy at the ages of 4 to 6 years when mental representations are newly developed or developing. We expect that the results of this study will contribute to the field of education by promoting the learning of visual information according to age and traits. In addition, the suggestion of a correlation among attention to detail, an ASD-like cognitive style, and the rate of correct responses indicates the possibility of identifying ASD subtypes with various symptom forms by increasing the accuracy of the tasks used in this study. 

Acknowledgments

The authors thank the participants, their parents, and the kindergarten for their cooperation in the experiment. We also thank Keiko Takiguchi of the Faculty of Education, Kanazawa University, and Haruyuki Kojima, Faculty of Human Sciences, Kanazawa University, for their cooperation in this research. 

Supported by grants from the Meiji Yasuda Foundation for Mental Health, JSPS KAKENHI Grant Number 19K12728, and the Moonshot Research and Development Program Grant Number JPMJMS2297 from Japan Science and Technology 

Author contribution statements using CRediT: Momoka Suda: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Visualization, Writing—original draft. Takashi Ikeda: Conceptualization, Funding acquisition, Investigation, Methodology, Resources, Visualization, Writing—review and editing. Mitsru Kikuchi: Funding acquisition, Project administration, Supervision, Writing—review and editing. 

Data availability: None of the data or materials for the experiments reported here is available, and none of the experiments was preregistered. Please contact the first author if you wish to view or use the data used in this study. 

Commercial relationships: none. 

Corresponding author: Mitsuru Kikuchi. 

Email: [email protected]. 

Address: Research Center for Child Mental Development, Hamamatsu University School of Medicine, Hamamatsu, Japan.