A Positive Association between Working Memory Capacity and Human Creativity: A Meta-Analytic Evidence

Creativity serves as a fountain for social and scientific development. As one of the most crucial human capabilities, creativity has been believed to be supported by the core component of higher cognitive functions—working memory capacity (WMC). However, the evidence supporting the association between WMC and creativity remains contradictory. Here, we conducted a meta-analysis using random-effects models to investigate the linear association between WMC and creativity by pooling the individual effect size from the previous literature. Further, a subgroup analysis was performed to examine whether such association is specific for different WMC categories (i.e., verbal WMC, visual–spatial WMC and dual-task WMC). The main meta-analytic results showed a significantly positive association between WMC and creativity (r = .083, 95% CI: .050–.115, p < .001, n = 3104, k = 28). The subgroup analysis demonstrated consistent results by showing a significantly positive association between them, irrespective of WMC category. We also found that cultural environments could moderate this association, and we identified a strong correlation in participants from an Asian cultural context. In conclusion, this study provides the evidence to clarify the positive association between WMC and creativity, and implies that the Asian cultural context may boost such an association.


Introduction
Creativity refers to the ability to produce novel and suitable ideas in a specific environment (Sternberg and Lubart 1999). Based on Guilford's divergent thinking test, it is defined as the composite concepts of originality, flexibility, and novelty of thinking (Guilford 1968). Creativity facilitates the generation of ideas in a problem-solving context and drives scientific discoveries and human progress (DeHaan 2009). Creativity was also found to be a phenotype associated with mental health problems, such as anxiety (Reid et al. 1959), schizophrenia (Degmečić 2018), and children's behavioral problems (Fancourt and Steptoe 2019). As one of the most crucial human-specific capabilities, creativity has been intensively studied to uncover what "cognitive cornerstones" are, with working memory being a research hot spot (Hennessey and Amabile 2010;Ovando-Tellez et al. 2022).
Growing evidence suggests that executive functions (EFs) play an important role in creativity (Zabelina et al. 2019). However, it remains unclear which EF-specific components are involved. EF refers to a series of high-order cognitive functions that are essential to ensuring physical and mental health, as well as academic and career success; EF contains three core components: inhibition, working memory, and cognitive flexibility (Diamond 2013). Among them, working memory (WM) refers to the capability to hold and manipulate information temporarily with "block-wise entities" (Baddeley 2012). To structurally quantify to pool the individual effect size from each study concerning the association between WMC and creativity. The systematic retrieval of the literature was conducted by following 2020 PRISMA pipeline in PsycINFO, Web of Science, PubMed, EMBASE, CNKI (Chinese database) and PsycARTICLES datasets on 17 June 2022. Further, to probe into the potential hierarchical factors affecting this association, we conducted an exploratory, subgroup meta-analysis by dividing comprehensive WMC into verbal WMC, visual-spatial WMC and dual-task coordination WMC. Finally, to further probe into the impact of potential confounding factors for the meta-analytic effects, we conducted a moderation analysis to examine whether the association between WMC and creativity is moderated by cultural background and age group.

Materials and Methods
To improve reproducibility and transparency as recommended, this study adhered to the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) and CHARMS pipelines (Page et al. 2021) (see Figure 1). Further, all the materials relating to the present study were deposited at the Open Science Framework (OSF) with open access. This meta-analysis mainly followed five steps: (1) developing a searching strategy for retrieval of the literature; (2) defining the inclusion and exclusion criteria; (3) screening the eligibility of the literature by using the inclusion and exclusion criteria; (4) the targeting data were extracted, coded, and assessed for evidence-based quality; and (5) statistics were estimated for pooling the individual effect size by building meta-analytic and moderation models.

Search Strategy
For completeness and accuracy of the literature search, we used a keyword-based retrieval strategy to search in Boolean logic in PsycINFO, Web of Science, PubMed, EM-BASE, CNKI and PsycARTICLES datasets. Specific Boolean expressions were as follows: ("Memory, Short-Term" OR "working memor*" OR "phonological loop" OR "visuospatial sketchpad" OR "central executive" OR "verbal working memory" OR "visuospatial working memory" OR "executive function" OR "updating") AND ("Creativity" OR "creative activit*" OR "creative thinking" OR "creative achievement*" OR "creative imagination*" OR "creative personalit*") NOT (review OR meta-analysis). To ensure data pooling completeness, reference lists of included articles, published in the last two years, were hand-reviewed.

Study Selection
According to the research objectives we predefined, the inclusion criteria were defined was as follows: (1) WMC and creativity should be measured by using standardized scales or board-certified behavioral tasks; (2) fundamental statistics (e.g., Pearson's correlation coefficient, sample size) for examining the association between WMC and creativity should be presented clearly; (3) peer-reviewed journal articles and dissertations are allowed; (4) analytic data should be self-recruited (i.e., independent dataset); (5) a sample or control group would be qualified; (6) creativity and/or WMC would be assessed without intervention; and (7) studies should be in English/Chinese language only. On the other hand, the exclusion criteria were as follows: (1) systematic reviews (with or without meta-analysis) or preprints were not be accepted; (2) non-standardized measures were used to estimate WMC or creativity; and (3) statistics were reported vaguely.

Encoding and Statistical Analysis
Meta-information was extracted from these included studies, including the author's name, publication date, sample size, age, and sample populations (nations and identity). Further, tasks for measuring WMC were extracted and coded into three domains: verbal WMC task, visual-spatial WMC task and dual WMC task. In addition, measures for quantifying creativity were extracted and coded into the following categories: Torrance Tests of Creative Thinking (TTCT), Test of Creative Thinking-Drawing Production(TCT-DP), Abbreviated Torrance Test for Adults (ATTA), Divergent thinking tests (DT), Convergent thinking tests (CT), Williams Prefer Measurement Forms (WPMF), Williams Creativity Assessment Packet (WCAP), Unusual uses task (UUT), Alternative Uses Task (AUT), Consensual Assessment Technique (CAT), Creative Achievement Questionnaire (CAQ), Associative fluency tasks (AF) and the Remote Association Test (RAT). Finally, for pooling the individual effect size into the meta-analytic model, the statistics (r value) and sample size for each included study were extracted.

Quality Analysis
To ensure the data quality, all the data that were extracted and coded from included studies were cross-validated by two independent researchers (IRs, GZ and MK). Any disagreements of data extraction and coding were solved by the third IR (CZY). Furthermore, two additional assessors (LXR and LMJ) were recruited to evaluate evidence-based quality by using a modified Newcastle-Ottawa Scale (mNOS) (Lo et al. 2014). The mNOS included five items to evaluate the risk of bias (ROB) for evidence (study) quality, with high ROB for total scores of ≤3 for each study. The specific assessment of mNOS included the following: (1) sample representativeness; (2) sample size; (3) comparability between respondents and non-respondents; (4) quantitative study quality; and (5) reporting quality for statistics.

Statistical Analysis
Comprehensive Meta-Analysis Software version 3.0 (CMAV3.0) was used to implement all the data analysis as we mentioned above (Makinde et al. 2021). To determine which statistical model is suitable in the current analysis, between-study heterogeneity, across the included studies, was estimated by using Higgins and Thompson's I 2 test (Borenstein et al. 2011) and Cochran's Q test. As recommended, the random-effects metaanalytic model is suitable to pool individual effect size by controlling high between-study heterogeneity (I 2 > 50%, p-value < 0.1). In addition to this main analysis, the sub-group meta-analysis was deployed to validate the individual meta-analytic effect for this association by three WMC tasks, including verbal, visual-spatial, and dual-task coordination. Furthermore, to examine whether the meta-analytic effect is biased by confounding factors, we built the moderation-effect models by taking the cultural background and age group into account. Finally, for quality control, publication bias was inspected by producing funnel plots and was calculated by using Egger's test and Kendall's test (Sterne and Egger 2005).

Results
Here, a total of twenty-eight papers (k = 28, the number of r statistics = 75, n = 3104) were screened and deemed eligible for generating the final data pool in the following meta-analysis. Fundamental information for all the included studies is tabulated in Table 1.

Main Meta-Analysis
The results of the heterogeneity tests revealed high between-study heterogeneity in this meta-analytic model, by showing a significantly high I 2 value (I 2 = 55%, p < .001). Thus, the random-effects models were built for the following meta-analysis.
As mentioned above, we estimated the pooled effect size by meta-analysis, concerning the r value and the sample size for the included studies. The results demonstrated a significant correlation between WMC and creativity, by pooling these individual effect sizes (r = .083, 95% confidence interval (CI): .050-.115, SE = .003, p < .001, n = 3104) (see Figure 2).

Verbal WMC Tasks and Creativity
Likewise, the heterogeneity test was conducted beforehand. Results showed a high heterogeneity for this subgroup meta-analytic model (I 2 = 39%, p = .069). Thus, the metaanalysis, using the random-effects model for investigating the association between verbal WMC and creativity, revealed that the meta-analytic effect for the positive correlation between verbal WMC and creativity reached statistical significance (r = .119, 95% CI: .072-.166, SE = .006, p < .001, n = 1733) (see Figure 3).

Visual-Spatial WMC Tasks and Creativity
Heterogeneity was found to be acceptable for the included studies in the subgroup meta-analytic analysis (I 2 = 0%, p = .58). Thus, the fix-effect model for meta-analysis was built; this demonstrated the statistically significant correlation between WMC and creativity (r = .155, 95% CI: .075-.234, SE = .006, p < .001, n = 592) (see Figure 4).

Figure 3.
Forest plot with 95% confidence intervals and weights for subgroup meta-analysis, concerning the association between verbal WMC tasks and creativity. Larger positive effect sizes indicate that increased creativity ability is related to verbal WMC.

Figure 4.
Forest plot with 95% confidence intervals and weights for subgroup meta-analysis, showing the association between visual-spatial WMC tasks and creativity.

Figure 5.
Forest plot with 95% confidence intervals and weights for subgroup meta-analysis, showing the association between dual-tasks and creativity.

Moderation-Effect Analysis
Given the high heterogeneity in the meta-analytic model, the moderation-effect analyses were performed to identify factors that may moderate the meta-analytic main effect. Here, the age and cultural contexts of the samples were reported in all the included studies and were modeled as potential moderators separately.

Figure 4.
Forest plot with 95% confidence intervals and weights for subgroup meta-analysis, showing the association between visual-spatial WMC tasks and creativity.

Figure 5.
Forest plot with 95% confidence intervals and weights for subgroup meta-analysis, showing the association between dual-tasks and creativity.

Moderation-Effect Analysis
Given the high heterogeneity in the meta-analytic model, the moderation-effect analyses were performed to identify factors that may moderate the meta-analytic main effect. Here, the age and cultural contexts of the samples were reported in all the included studies and were modeled as potential moderators separately.

Moderation-Effect Analysis
Given the high heterogeneity in the meta-analytic model, the moderation-effect analyses were performed to identify factors that may moderate the meta-analytic main effect. Here, the age and cultural contexts of the samples were reported in all the included studies and were modeled as potential moderators separately.

Moderation-Effect of WMC Type
Following the moderation analysis of the WMC tasks group, null significant findings were observed for the main meta-analytic effect (Q = 1.360, p = 0.507) ( Figure S3).

Publication Bias and Quality Assessment
A funnel plot for standard Fisher-Z scores for the included studies can be used to explore the publication bias. The scattered points showed a symmetric distribution (see Figure 6), which indicated no perceived publication bias. To quantify the risk of publication bias, the Egger's test was conducted. Results showed no prominent publication bias (e = .449, p = .350). Finally, the evidence quality for the included studies was validated to be acceptable (mean scores for mNOS = 4.65, SD = 0.56, Median = 5).
Following the moderation analysis of the WMC tasks group, null significant findings were observed for the main meta-analytic effect (Q = 1.360, p = 0.507) ( Figure S3).

Publication Bias and Quality Assessment
A funnel plot for standard Fisher-Z scores for the included studies can be used to explore the publication bias. The scattered points showed a symmetric distribution (see Figure 6), which indicated no perceived publication bias. To quantify the risk of publication bias, the Egger's test was conducted. Results showed no prominent publication bias (e = .449, p = .350). Finally, the evidence quality for the included studies was validated to be acceptable (mean scores for mNOS = 4.65, SD = 0.56, Median = 5). Figure 6. Funnel plot of this study to assess the publication bias. X--axis indicates the individual study effect estimates, and Y-axis indicates standard errors.

Discussion
The main purpose of this study was to clarify the association between WMC and creativity by synthesizing meta-analytic evidence. We found that WMC is significantly positively correlated with creativity by pooling individual effects, indicating that an increased WMC indeed supports human creativity. Furthermore, subgroup meta-analysis was conducted by dividing WMC into three categories, including verbal WMC, visualspatial WMC and dual-task coordination WMC. The results demonstrated that such associations are robust in different WMC tasks. Lastly, we conducted moderation analysis, which revealed that the correlation between WMC and creativity was moderated by cultural background, with a higher correlation for participants from Asian cultural contexts. On balance, the current study may provide weak evidence to clarify the positive correlation of WMC with creativity. In addition, such associations were found to be robust for the potential impacts of the WMC categories, and the moderating role of cultural background was revealed in this association.
One of the most crucial findings in this study is that there is a statistically significant (but weak) correlation between WMC and creativity. Both theoretical and empirical evidence supports that WMC could positively predict one's creative ability. As the most important indicator of WM, the WMC is typically described as being the limited capacity for the temporary storage and processing of information (Baddeley 2003). On the basis of controlled attention theory (CAT), creativity is theoretically argued to be a top-down cognitive process that requires considerable cognitive resource control (Beaty and Silvia 0 . 6 S t a n d a r d E r r o r F i s h e r ' s Z F u n n e l P l o t o f S t a n d a r d E r r o r b y F i s h e r ' s Z Figure 6. Funnel plot of this study to assess the publication bias. X-axis indicates the individual study effect estimates, and Y-axis indicates standard errors.

Discussion
The main purpose of this study was to clarify the association between WMC and creativity by synthesizing meta-analytic evidence. We found that WMC is significantly positively correlated with creativity by pooling individual effects, indicating that an increased WMC indeed supports human creativity. Furthermore, subgroup meta-analysis was conducted by dividing WMC into three categories, including verbal WMC, visual-spatial WMC and dual-task coordination WMC. The results demonstrated that such associations are robust in different WMC tasks. Lastly, we conducted moderation analysis, which revealed that the correlation between WMC and creativity was moderated by cultural background, with a higher correlation for participants from Asian cultural contexts. On balance, the current study may provide weak evidence to clarify the positive correlation of WMC with creativity. In addition, such associations were found to be robust for the potential impacts of the WMC categories, and the moderating role of cultural background was revealed in this association.
One of the most crucial findings in this study is that there is a statistically significant (but weak) correlation between WMC and creativity. Both theoretical and empirical evidence supports that WMC could positively predict one's creative ability. As the most important indicator of WM, the WMC is typically described as being the limited capacity for the temporary storage and processing of information (Baddeley 2003). On the basis of controlled attention theory (CAT), creativity is theoretically argued to be a top-down cognitive process that requires considerable cognitive resource control (Beaty and Silvia 2012). As an important component of cognitive control, WM has an imperative ability to upstream regulate complicated creative tasks (Lee and Therriault 2013). In other words, creativity is consistently achieved by extracting relevant knowledge from short-term memory or reconstructing it based on existing knowledge in long-term memory; this needs the substantial support of an adequate WMC. In addition, the positive association between WMC and creativity has been validated in a recent large-sample neuroimaging study (Takeuchi et al. 2020); this revealed the overlapping co-activated areas for the WM task and creativity task. Thus, it may provide robust evidence to clarify the positive association of WMC with creativity; this offers insight into addressing this long-lasting debate.
By using subgroup analysis, this study also found that the positive correlations between WMC and creativity were consistent, irrespective of the WMC category (i.e., verbal WMC, visual-spatial WMC, and dual-task coordination WMC). This finding attempted to answer whether the conflicting results, derived from previous studies, were attributed to a heterogenous WMC category. Supporting this, measuring WMC performance was argued to be comparable across different sensory pathways (e.g., visual and verbal) (Xu et al. 2017). In addition, Lee and colleagues (2011) well documented an increased activation in the related brain regions (medial temporal lobe, MTL) when working memory demand was increased, regardless of the type of stimulus (e.g., visual and auditory) (Lee and Rudebeck 2010). This evidence may lead us to draw the conclusion that the positive association between WMC and creativity is robust, or more boldly, to infer that the existing conflicting results may not be ascribed to sensory processing in WMC tasks.
To clarify the impact of potential confounding factors on the meta-analytic effects, moderation analyses were drawn for the cultural background of participants (i.e., Western culture and Asian culture) and age groups (i.e., adolescents and adults), respectively. Interestingly, the meta-analytic effect for the association between WMC and creativity was significantly moderated by cultural background. Specifically, compared to participants with Western cultural background, participants in the Asian cultural environment reported a stronger correlation between WMC and creativity. This finding could be explained partly by the relativity of the creativity theory (Guilford 1950). This theoretical framework elucidated the fact that creativity could be defined and evaluated specifically in different cultural environments, due to a lack of a practical criterion for creativity, with liberal scopes in western cultures (e.g., arts that required less deliberative cognitive process) (Hempel and Sue-Chan 2010). Conversely, measuring creativity in Asian cultures required strict executive functions (especially in WM) in creative tasks, such as problem-solving, deliberative reasoning and insight inference (Leung and Wang 2015). Rudowicz (2003) argued that the influence of culture on creativity is complex and highly interactive, involving historical, social, and personal cross-cultural factors (Rudowicz 2003). The key to the cross-cultural study of creativity is uncovering whether the definition and operationalizations of creativity from one culture can be validly applied to another one; this includes the eastern-western cultural gap or the conservative/traditional-liberal cultural gap. To provide evidence for this, some studies, comparing creativity between Westerners and Asians, demonstrated that the performance of creative activities was higher in people from cultural environments that highlighted creativity values Sternberg 2002, 2003). Furthermore, one interesting finding was that Western cultures valued individualistic, intuitive, and artistic processing in creative activities, while Asian culture stressed collectivistic, cognitive, and deliberative thoughts (Goncalo and Staw 2006). That is to say, the gap between cultures, and their required involvement of cognitive processing (i.e., WM) in creative tasks, may be a crucial factor biasing the association between WM and creativity. This study indicates that the cultural gap between participants may be a source of conflict, caused by the results of existing studies.

Limitation
Although this study clarifies the association between WMC and creativity, several limitations should be acknowledged. Due to the strict inclusion and exclusion criteria, the total sample size (n < 3500) and evidence (study, k = 28) seem to be inadequate. Therefore, the nuance of these variations in task types of WM or creativity cannot be examined currently. Thus, future studies are needed to provide neuroimaging evidence to further confirm the association between them. Additionally, extending the main conclusion of the current study is prudent, because the total effect size for such an association is relatively small (though reaching statistical significance).

Conclusions
This study provides evidence to clarify the statistically significant positive association between WMC and creativity, though it has a weak strength. Further, the present study revealed that such associations exist across different types of WMC measurement (i.e., verbal WMC task, visual-spatial WMC task and dual WMC task), indicating that the conflicting results for the association between them are not biased by measure heterogeneity. This study also demonstrates that the cultural gap may confound the association between WMC and creativity.
Supplementary Materials: The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/jintelligence11010015/s1, Figure S1: The results of the moderationeffect of culture; Figure S2, The results of the moderation-effect of age; Figure S3, The results of the moderation-effect of WMC types group.

Institutional Review Board Statement:
No ethical approval was required as the meta-analysis study does not involve original human or animal data.

Informed Consent Statement: Not applicable.
Data Availability Statement: All the materials regarding this study have been deposited in Open Science Framework (OSF): https://osf.io/3g2j9/ (accessed on 1 January 2023).