Managing Your Muse: Exploring Three Levels of Metacognitive Control in Creative Ideation

ABSTRACT

Creative ideation tasks are typically ill-defined and thus should benefit from metacognition. However, the role of metacognitive processes―especially metacognitive control―in creative ideation appears understudied. Therefore, we conducted an online study investigating the relationship between different aspects of metacognitive control and divergent thinking (DT) performance. In line with the recently proposed creative metacognition framework , we devised measures that delineate metacognitive control at the levels of task, performance, and responses: Metacognitive control at task level was assessed in terms of attention focus devoted to each task; metacognitive control at performance level was assessed in terms of the goal-directedness of the employed task strategy; metacognitive control at response level was assessed in terms of how effectively uncreative ideas were screened out. We observed substantial individual differences in all three measures as well as first evidence of their validity. Importantly, metacognitive control at all three levels independently predicted higher DT creativity. Additional analyses suggested that effects of metacognitive control may extend to creative behavior, partly mediated by DT. In sum, this study provides first empirical support for the relevance of distinguishable aspects of metacognitive control at task, performance, and response levels in creative performance.

Introduction

Creative thinking involves unique challenges. First, creative thinking tasks pose ill-defined problems that cannot be solved based on experience or in a straightforward analytical way (Pham, Magistretti, & Dell’era, 2023). To solve such problems, people either adapt known, but typically ineffective strategies or devise new ad hoc strategies (Gilhooly, Fioratou, Anthony, & Wynn, 2007). Second, creative thinking tasks, especially creative idea generation tasks (viz. divergent thinking tasks), are open-ended and have no single correct solution, but many possible solutions differing in their quality (Guilford, 1967). As a consequence, people are never done when finding an idea. Effectively navigating this large solution space may involve decision on whether to further exploit the current area or explore new ones (Hart et al., 2018), and, more generally, whether it is worth investing more effort to find even more creative ideas (Lucas & Nordgren, 2020). Third, creative thinking tasks often ask to “be creative” (Acar, Runco, & Park, 2020), which imposes an ambiguous goal as lay conceptions of creativity differ substantially (Baas, Koch, Nijstad, & De Dreu, 2015; Benedek et al., 2021). Hence, upon generating a candidate solution, it needs to be evaluated for creativity to decide whether it should be reported or not. Addressing these unique challenges in creative ideation involves considerable strategizing and reflection on one’s performance, and thus metacognition in creative performance, or, in short, creative metacognition.

The relevance of metacognition in creative thinking has long been recognized, at least implicitly, by theory and research. Models of creative ideation unequivocally refer the interplay of generation and evaluation (Benedek, Beaty, Schacter, & Kenett, 2023; Finke, 1996; Simonton, 2011). Moreover, models of the creative process have identified nuanced stages and subprocesses devoted to regulating and optimizing creative performances, which include the definition and construction of the problem, preparation, as well as implementation planning and solution monitoring (Cropley, 2006; Mumford, Medeiros, & Partlow, 2012; for a particularly comprehensive model of creative problem solving, see also Isaksen, Dorval, & Treffinger, 2011). Further theorizing explicitly acknowledged creativity as a self-regulatory, metacognitive process (Pesut, 1990) that implies knowledge about when, where, why and how to be creative (Kaufman & Beghetto, 2013), and cycles of sifting ideas (Jia, Xu, & Zhang, 2022; Puryear, 2016). Pertinent research, so far, has been especially interested in how well people recognize the creativity of their ideas and if they can accurately discern creative from less creative ones (Karwowski, Czerwonka, & Kaufman, 2020; Kenett, Gooz, & Ackerman, 2023; Rominger et al., 2022; Runco & Smith, 1992; Sidi, Torgovitsky, Soibelman, Miron-Spektor, & Ackerman, 2020; Urban & Urban, 2021). Other works identified individual differences in task strategies across various forms of creative performance such as creative problem-solving, creative ideation, and creative writing (Fleck & Weisberg, 2004; Gilhooly, Fioratou, Anthony, & Wynn, 2007; Urban & Urban, 2024). Importantly, various aspects of self-reported metacognitive skills (e.g., conditional knowledge, regulation of cognition, monitoring) were found to predict creative performance and to moderate effects of creative potential on creative behavior (Puente-Diaz, Cavazos-Arroyo, & Vargas-Barrera, 2021; Puryear, 2015).

Building on the theorizing in and empirical evidence in creativity research, as well as metacognition theory more broadly, Lebuda and Benedek (2023) have recently proposed a systematic framework of creative metacognition (CMC). The CMC framework distinguishes between core and metacognitive processes, and, within metacognition, considers three major components of metacognitive control, monitoring, and knowledge. Metacognitive control refers to the regulation of the creative process by means of strategic decisions such as whether to maintain or change the current task approach. Metacognitive monitoring represents evaluations of candidate responses and of the ongoing performance that eventually trigger the need to reconsider the current strategy. Finally, metacognitive knowledge in the context of creativity refers to experience or expertise associated with creative tasks and the concept of creativity, which allows to better calibrate judgments and make more informed decisions. Recent evidence supports the independent contributions of three main CMC components for creative performance (Lebuda & Benedek, 2024). Notably, the CMC framework outlines the dynamic interplay of the three components, which is not considered one-way, but also implies, for instance, how knowledge gets updated by monitoring, and how monitoring readjusts based on regulatory efforts.

As an important extension to other available metacognition models, the CMC framework (Lebuda & Benedek, 2023) distinguishes metacognitive control, monitoring and knowledge at three conceptual levels: task, performance, and response level. This distinction acknowledges that metacognitive processes can either refer to the task in general, the actual task performance, or to candidate responses within task performance. In the context of metacognitive control, decisions need to be made whether to engage in a task at all and how much cognitive resources to devote (task level; e.g., Plucker, Runco, & Lim, 2006); metacognitive control can further be applied to select or devise effective task strategies (performance level; e.g.; Gilhooly, Fioratou, Anthony, & Wynn, 2007), and to decide whether a candidate response should be reported, elaborated, or discarded (response level; Silvia, 2008). Metacognitive control at these three levels is informed by evaluations of the task (e.g., task difficulty; task level), monitoring of task performance (e.g., fluency of ideas; performance level), and judgments of the candidate responses (e.g., creativity evaluation; response level), respectively (Koriat & Levy-Sadot, 1999; Puente-Diaz, Cavazos-Arroyo, & Vargas-Barrera, 2021; Sidi, Torgovitsky, Soibelman, Miron-Spektor, & Ackerman, 2020; Urban & Urban, 2023; van Broekhoven, Belfi, Borghans, & Seegers, 2021). These evaluations are again calibrated by metacognitive knowledge about the task (task level), relevant strategies (performance level), and of what characterizes a creative idea (response level), respectively (Ceh, Edelmann, Hofer, & Benedek, 2022; Miroshnik & Shcherbakova, 2019; Runco & Acar, 2010). The differentiated structure of metacognitive processes across components and levels proposed by the CMC framework helped to consolidate available findings but also awaits more thorough empirical validation. Hence, it needs to be tested whether and how creative metacognitive at task, performance, and response levels, contributes to creative ideation performance. Moreover, as creative ideation (viz. divergent thinking ability) represents an important aspect of creative potential – latent abilities, traits and domain-specific expertise predicting creative activities in real-life (Benedek, 2024; Eysenck, 1995) – creative metacognition may further facilitate the engagement of creative behaviors and the production of achievements.

There have been extended discussions whether creativity relies on the spontaneous “kiss by a muse” (Gable, Hopper, & Schooler, 2019) or to what extent these “muses” are under our control (Silvia, 2015). In fact, the relevance of cognitive control for creative performance is a robust finding across different lines of research, including works on the relationship of creativity and intelligence (Gerwig et al., 2021; Miroshnik, Forthmann, Karwowski, & Benedek, 2023), the role of executive abilities in creative cognition (Benedek, Jauk, Sommer, Arendasy, & Neubauer, 2014; Gilhooly & Fioratou, 2009; Zabelina, Friedman, & Andrews-Hanna, 2019) and neuroscientific findings highlighting the relevance of the executive control network (Beaty, Benedek, Silvia, & Schacter, 2016; Ovando-Tellez et al., 2022; for reviews, see; Benedek & Jauk, 2018, 2019; Chrysikou, 2018). The CMC framework posits that the importance of cognitive control in creativity is partly due to the substantial involvement of metacognitive control in creative performance, which can apply at the levels of task, performance, and response control (Chevalier, Martis, Curran, & Munakata, 2015; Fernandez-Duque, Baird, & Posner, 2000; Roebers, 2017).

The present study

This study attempts a first empirical test of the notion that metacognitive control at different conceptual levels supports creative thinking, and eventually creative behavior, as suggested by the CMC framework (Lebuda & Benedek, 2023). To this end, we devised measures that assess metacognitive control at task, performance, and response level, and explored their contributions in creative ideation (i.e., divergent thinking tasks). This approach extends previous work that studied self-reported metacognitive skills in a trait-like, task-independent manner (e.g., Puryear, 2015) by ensuring that metacognitive assessments referred to the creative task at hand. Based on findings suggesting that the use of more complex cognitive strategies results in more creative ideas (Gilhooly, Fioratou, Anthony, & Wynn, 2007; Nusbaum & Silvia, 2011), we expected that metacognitive control at performance level increases divergent thinking (DT) performance (i.e., especially the rated creativity of ideas). We further assumed that metacognitive control at response level would support DT performance, as higher DT ability predicts higher discernment in idea evaluation which should inform accurate idea selection (Benedek et al., 2016; Grohman, Wodniecka, & Kłusak, 2006), and higher DT fluency and openness also predicts more accurate idea selection (Benedek, Mühlmann, Jauk, & Neubauer, 2013; Puente-Díaz, Cavazos-Arroyo, Puerta-Sierra, & Vargas-Barrera, 2022; Silvia, 2008). The available literature does not, however, support deriving clear hypotheses for the effects of metacognitive control at task level on DT performance. Finally, since creative ideation supports creative accomplishments (Said-Metwaly, Taylor, Camarda, & Barbot, 2022), we additionally explored relationships of metacognitive control with creative activities and achievements.

Methods

Participants

368 German-speaking participants fully completed the online study. In the course of data validity checks, 51 cases were excluded because of either missing any of the two attention checks (4.3%), overly short survey completion time (<1400 s; 2.4%), giving no response to any of the DT tasks (4.1%), or technical problems with the timing on the online assessment (9.0%). The final sample thus consisted of 317 participants (66.2% female, 33.1% male, 0.6% diverse) with an average age of 29.8 years (SD = 13.1; range = 17–73) with 59.3% being university students. Participants were recruited via the university student mailing list and via advertisements. Participants entered a raffle for 10 × 20€ gift cards; Psychology students additionally received partial course credits. The procedure has been approved by the local ethics committee.

Tasks and materials

Divergent thinking

Divergent thinking (DT) ability was assessed with three items of the alternate uses task (brick, car tire, pen), the most common task for DT assessments (Saretzki, Forthmann, & Benedek, in press), each timed for three minutes (visible timer). Participants were instructed that it is more important to come up with creative ideas than to name many ideas (i.e., be creative instructions). Each response was written in a dedicated response field; at the beginning only two response fields were visible but as soon as participants entered responses additional response fields appeared. All responses (without corrections or removals) were rated by six trained raters for creativity (i.e., a holistic judgment reflecting both the originality and effectiveness of responses; 0, not creative, to 4, highly creative). Interrater-reliability was good (ICC(3,k) = .80, .83, .83, for brick, car tire, and pen, respectively). Creative performance per item was defined as the average ratings of the three most highly rated ideas (max-3 scoring) to reduce fluency confound (Benedek, Mühlmann, Jauk, & Neubauer, 2013). DT creativity was computed as the average across the three items (ω = .70). Additionally, we computed DT fluency as the average number of responses per item (ω = .88).

Metacognitive Control (MCC)

Metacognitive Control at task level (MCC-T) was assessed with three items that asked for the level of task focus during the three DT tasks. Note that we decided not to assess task focus in terms of length of self-paced task engagement, as higher time-on-task is necessarily related to higher DT performance (Paek, Abdulla Alabbasi, Acar, & Runco, 2021).; Instead, we assessed the perceived task focus as an index of the mental resources devoted to the task at hand. Specifically, participants were told “Maybe you have been partly distracted during the previous DT tasks or may not felt like fully focusing on the task. Please indicate the percentage of time that you were focused on the preceding DT tasks (0–100%).” Internal consistency for task focus across the three DT tasks was high (ω = .86) and a total score was computed by averaging across the three items.

Metacognitive Control at performance level (MCC-P) was assessed in terms of the goal-directedness of the employed task strategy. To this end, we asked participants to briefly describe their most effective strategy for generating creative ideas (note that open-ended questions are considered more appropriate as people tend to be over-confident in their strategy use when using scales; Zepeda & Nokes-Malach, 2021) These strategy descriptions were rated by twelve raters for goal-directedness from 0 (completely spontaneous) to 3 (fully strategic). Inter-rater-reliability was very high (ICC(3,k) = .95), and a total score was computed by averaging across the ratings. Additionally, we asked participants to estimate how strategic they approached the DT tasks in general, from 0 (waited for spontaneous ideas) to 100% (I consistently used a strategic approach). Finally, they also indicated how often they used specific established strategies including the property strategy, context strategy, memory strategy, disassembly strategy (Gilhooly, Fioratou, Anthony, & Wynn, 2007), or waited for spontaneous ideas using a scale from 0 (never used) to 6 (always used this strategy). Participants could openly add other strategies, which was seized by 9.8% of the participants. Additional strategies included, for instance, to imagine what a child would have done with this object; many descriptions reflected minor twists of the listed strategies, which suggested that main strategies had been well covered.

Metacognitive control at response level (MCC-R) was assessed in terms of the applied creativity threshold for reporting responses in the DT task. Applying metacognitive control should lead to more often holding back uncreative responses either to discard them or further elaborate them. Hence, higher metacognitive control should result in a higher creativity of the least creative response, or a higher “worst performance” (Forthmann, Karwowski, & Beaty, 2023). The creativity thresholds in the three DT tasks showed moderate internal consistency (ω = .62), and a total score was computed by averaging these three scores.

Additionally, we also assessed the creativity threshold for reporting ideas in an independent response selection task. To this end, we presented participants with 15 uncreative ideas and asked them, which of those responses they would have reported as response during a DT task if they had found them (self-paced). The selected uncreative ideas (five for each of the three DT) reflected DT responses that had been collected in a previous study using the same DT tasks and had been rated one standard deviation below the mean or less. Internal consistency across the three tasks was modest (ω = .68). As selections did not refer to actual task performance, we considered a lower number of uncreative responses that would have been reported as a measure of higher creative metacognitive knowledge at response level, which may inform metacognitive control (Lebuda & Benedek, 2023).

Creative behavior

Creative activities were assessed by the average frequency of activities across nine creative domains (literature, music, handicraft, interior design, culinary arts, visual arts, performing arts science and technology, and social; cf. Benedek, Bruckdorfer, & Jauk, 2020). For each domain, people reported how often they had performed these activities within the last twelve months from 0 (never) to 5 (every day); ω = .64.

Creative achievement was assessed by the rated creativity of the three top creative achievements (Ceh, Edelmann, Hofer, & Benedek, 2022; Diedrich et al., 2018). Participants described their most creative achievements so far, which were rated by six raters from 0 (not creative – nearly everyone is doing this) to 5 (exceptionally creative – the person has gain international recognition for this achievement). Inter-rater reliability was high (ICC(3,k) = .87); ω = .68. One person did not respond to this assessment, resulting in valid data for n = 316.

Participants completed further measures that were not in the focus of this study. All materials, data, and analysis scripts of this study are provided at the OSF (https://osf.io/9qpk5/).

Procedure

The study was realized on the online survey platform LimeSurvey. Participants gave informed consent and then completed all measures in the same order. The study took on average 39 minutes (SD = 11.8). The study procedure had been approved by the local intuitional review board (GZ. 39/146/63 ex 2020/21).

Results

We first present general descriptive statistics and intercorrelations of the central study measures, followed by detailed analyses of each creative metacognitive control measure, and finally determine independent contributions of metacognitive control aspects for divergent thinking performance, as well as creative activities and achievement.

Descriptive statistics

Table 1 presents descriptive statistics and correlations of the creativity measures and metacognitive control.

Table 1. Descriptive statistics and correlations of the main study measures.

	M	SD	1	2	3	4	5	6
1 DT creativity	1.79	0.39
2 DT fluency	5.97	2.82	.59***
3 Activities	1.45	0.60	.23***	.22***
4 Achievements	1.91	0.50	.27***	.15**	.14*
5 MCC-T	74.5	18.7	.18**	.12*	−.01	.14*
6 MCC-P	1.05	0.63	.24***	.12*	.07	.26***	.11*
7 MCC-R	0.76	0.38	.16**	−.43***	.02	.06	−.04	.15**

Notes. DT = Divergent thinking; MCC-T/P/R = Metacognitive control at task/performance/response level. N = 317 except for creative achievements (n = 316).

* p < .05, ** p < .01, *** p < .001.

Metacognitive Control (MCC) measures

MCC at task level (MCC-T)

MCC-T ranged from 9 to 100% with an average of 74.5%, suggesting large individual differences in how much people focused on the DT tasks. Higher task focus was related to higher DT creativity (r = .18) and slightly higher DT fluency (r = .12; see Table 1), suggesting that higher metacognitive control at task level improved DT performance.

MCC at performance level (MCC-P)

MCC-P was assessed in terms of the rated goal-directedness of the employed strategy during DT performance (0, spontaneous to 3 fully goal-directed). People differed substantially in the goal-directedness of their strategies, ranging from relying fully on spontaneous thinking (min = 0; e.g., “I acted on a gut level” or “intuitive”) to being highly strategic about the task (max = 2.75; e.g., “I thought about different contexts and tried to place the object there”). The rated goal-directedness of strategies was also reflected in higher self-reported goal-directedness of strategies (r = .35, p < .001), indicated that people had modest insight in the goal-directedness of their approach. Importantly, higher rated goal-directedness of strategies was correlated with higher DT creativity (r = .24) and a slightly higher DT fluency (r = .12), suggesting that higher metacognitive control at performance level improved DT performance.

In side-analyses, we explored how often specific strategies were used (self-reported from 0, never, to 6, always) and how they related to the goal-directedness measures (see Table 2). The property strategy was used most frequently (3.95) followed by the context strategy (3.87) and the memory strategy (3.73), whereas the disassembly strategy was rather uncommon (2.30). Participants also frequently followed an undirected approach of waiting for spontaneous ideas (3.83). Spearman correlations showed that higher goal-directedness of strategies was related to more often using property and context strategies, and less often relying on a spontaneous approach. Using any specific strategy was only weakly related to DT performance, with only the disassembly strategy showing a significant positive association with both DT creativity and fluency. These findings replicate that DT performance involves both goal-directed and spontaneous strategies (Gilhooly, Fioratou, Anthony, & Wynn, 2007) but goal-directed strategies (such as disassembly) appear more conducive to creative performance.

Table 2. Self-reported frequency of using specific task strategies during DT performance (0-never to 6-always) and their association with the goal-directedness of the strategy and DT performance (spearman correlations).

	Use frequency		Correlation (rs) with goal-directedness of strategy		Correlation (rs) with DT
	M	Mdn	Rated	Self-assessed	Creativity	Fluency
Disassembly	2.30	2	.01	.13*	.17**	.12*
Memory	3.73	4	.01	−.01	−.11+	−.03
Property	3.95	4	.16**	.10+	.07	.03
Context	3.87	4	.21***	.21***	−.04	−.07
Spontaneous	3.83	4	−.24**	−.47***	.09	.06

+ p < .10, * p < .05, ** p < .01, *** p < .001.

MCC at response level (MCC-R)

MCC-R was measured as the individual creativity threshold applied during DT performance defined by the least creative response reported per task (i.e., worst performance). The average worst performance ranged between creativity ratings of 0 and 1.72 (on a scale of 0–4 with a mean across all responses of 1.51, and a mean of max-3 responses of 1.79), suggesting that people differed substantially in the minimal level of creativity in their responses. We found that people with a higher threshold regarding their worst performance during the timed DT tasks (i.e., their least creative idea was more creative compared to others) also selected fewer uncreative ideas in the response selection task (r = .32, p < .001). Importantly, higher worst performance (MCC-R) was associated with a lower number of responses (DT fluency; r = −.43) but a higher max-3 creativity (DT creativity; r = .16). Together, these findings support the validity of worst performance as an index of metacognitive control at response level, with higher MCC-R resulting in fewer responses and also a higher creativity of the best responses.

We observed an even higher association between worst performance and the average creativity of responses (r = .66), but average creativity, of course, is directly driven by less creative responses, whereas there is only limited overlap between worst performance and the three most creative ideas used in the max-3 DT score. People on average generated six responses and the proportion of people generating three or fewer responses ranged from 16–27% in the three DT tasks. Notably, when limiting the sample to the 196 participants who generated four responses or more in each DT task (i.e., fully avoiding overlap between worst performance, min-1, and best performance, max-3) even increased the relationship between these measures to r = .34 (p < .001), maybe by focusing on a subsample with more consistent, reliable performance. In sum, we see that higher metacognitive control at response level supports focusing on response quality over quantity and thereby increases not only the average creativity of responses but also enhances the quality of the most creative responses.

Independent contributions of MCC aspects for creativity

We examined the unique contributions of the MCC aspects for DT creativity and fluency by means of two regression analyses (see Table 3). Higher MCC at all three levels (task, performance, and response level) independently predicted higher DT creativity, with the strongest contribution observed for MCC at performance level (rated goal-directedness of task strategy). DT fluency was predicted by higher MCC-P, and lower MCC-R, whereas MCC-T did not predict DT fluency beyond the other measures.

Table 3. Regression analysis predicting DT by metacognitive control aspects.

	DT Creativity	DT Fluency
	β	β
MCC-T	.16**	.08
MCC-P	.21***	.18***
MCC-R	.13*	−.45***
	R = .31, R^2adj = .09, F(313,3) = 11.41***	R = .47, R^2adj = .21, F(313,3) = 29.76***

Note. DT = Divergent thinking; MCC-T/P/R = Metacognitive control at task/performance/response level.

* p < .05, ** p < .01, *** p < .001.

In a next step, we explored whether MCC also predicts real-life creativity in terms of creative activities and achievements in two additional regression analyses (Table 4). Given that MCC was found to predict DT, and DT is known to predict real-life creativity (Said-Metwaly, Taylor, Camarda, & Barbot, 2022; also confirmed in this study; cf. Table 1), DT fluency and creativity was entered in a second block of predictors, which allowed to explore whether DT potentially mediates the relationship between MCC and real-life creativity. For creative activities, the regression model did not offer a significant explanation of the criterion by the MCC predictors, but only by DT fluency. Creative achievement was significantly predicted by MCC at task and performance levels at step 1. Adding DT measures at step 2, improved the overall model (ΔR² = .04, p < .001), and DT creativity (but not DT fluency) significantly predicted creative achievements. At the same time, the prediction by MCC measures was attenuated with MCC at task level no longer explaining unique variance in creative achievement besides the other predictors. Formal mediation analyses (bootstrapped with 5000 samples) confirmed significant indirect effects of MCC at task level on creative achievement via DT creativity (MCC-T: β = .001, CI_95% [.000, .002], p = .004; MCC-P: β = .043, CI_95% [.018, .074], p = .003), with only a significant direct effect by MCC-P on creative achievement (β = .158, CI_95% [.076, .242], p < .001), but not by MCC-T (β = .003, CI_95% [−.001, .006], p = .161). In sum, these findings suggest that the relationship of MCC at task level with creative achievement is fully or partly mediated by DT creativity, whereas the relationship of MCC at performance level is partly mediated by DT creativity.

Table 4. Hierarchical regression analysis predicting real-life creative activities and achievements by metacognitive control aspects (MCC; step 1) and divergent thinking (DT; step 2).

	Creative Activities	Creative Achievements
	β	β
Step 1:
MCC-T	−.01	.12*
MCC-P	.07	.24***
MCC-R	.01	.03
Step 2:
MCC-T	−.05	.08
MCC-P	.01	.20***
MCC-R	.08	−.01
DT Creativity	.12	.22**
DT Fluency	.19*	−.02
	R = .27, R^2adj = .06, F(311,5) = 4.76***	R = .35, R^2adj = .11, F(310,5) = 8.45***

Note. DT = Divergent thinking; MCC-T/P/R = Metacognitive control at task/performance/response level.

* p < .05, ** p < .01, *** p < .001.

Discussion

This study examined the relevance of metacognitive control for DT performance. Following the framework of creative metacognition (CMC; Lebuda & Benedek, 2023), measures were devised to assess metacognitive control at three conceptual levels (task, performance, response levels). As a main finding, higher metacognitive control at all three levels – control of DT task engagement, use of controlled, goal-directed strategies, and response control – independently predicted higher DT creativity. These findings offer empirical support for the validity of the CMC framework with respect to the facilitating role of metacognitive control at task, performance, and response levels for DT.

Metacognitive control assessments

Some new measures were devised to assess metacognitive control at task, performance, and response levels. All measures referred directly to the DT task performance at hand to ensure that they actually captured metacognitive control (i.e., control applied to regulate and optimize task performance). The new measures revealed considerable inter-individual variance in how much metacognitive control was applied. Metacognitive control at task-level ranged from people focusing 9 to 100% at the task; metacognitive control at performance level ranged from using fully spontaneous approaches to applying sophisticated strategies (range = 0–2.64, on the goal-directedness scale from 0 to 3); metacognitive control at response level ranged from people reporting even the most uncreative ideas to imposing high creativity thresholds (range of worst responses = 0–1.72, on a creativity scale from 0 to 4, with mean creativity across all responses being just 1.51). This descriptive analysis of metacognitive measures already offered intimate insights into how substantially people differ in the extent of how much metacognitive control they apply during DT performances. The observed substantial individual differences in metacognitive control is consistent with the view that metacognitive processes – unlike core cognitive processes – can be optionally recruited to optimize task performance (Lebuda & Benedek, 2023).

Interestingly, metacognitive control measures at the three levels showed only weak intercorrelations. This finding appears to run counter to a notion of one underlying trait (e.g., rationality; Stanovich, 2012) or ability (e.g., executive control) driving how much metacognitive control is applied at each level. Yet, it also seems possible that even though people wanted to apply control at each level, this was undermined by a lack of relevant metacognitive knowledge. For example, people may not know effective strategies, or inadequately believe that creative ideas mainly result from spontaneous inspiration (Lucas & Nordgren, 2020). Similarly, they may be unclear about when an idea is creative enough to be reported (Sidi, Torgovitsky, Soibelman, Miron-Spektor, & Ackerman, 2020). Indeed, our findings showed that higher metacognitive knowledge about what responses should not be reported informed higher metacognitive response control during DT performance. It further needs to be acknowledged that the three measures relied on different sources of information, including self-report (MCC-T), ratings of verbal descriptions (MCC-P), and performance data (MCC-R). Especially, self-reported CMC depends on the ability to adequately recognize and reflect ongoing cognitive processes and strategies and thus are likely moderated by a person’s metacognitive abilities. This may also be the reason why self-reports about the goal-directedness of one’s own approach showed only a modest association with the rated goal-directedness and did not predict actual DT performance in our study. Hence, future assessments of creative metacognition should (additionally) aim for objective, behavior-related indicators of metacognition, which could include aspects of performance data (like the response thresholds used in this study), process data (e.g., how much time people spend on elaborating their ideas when entering them (Barbot, 2018), or psychophysiological data such as eye tracking that allows to track the focus of attention without cognitive interference (Jankowska, Czerwonka, Lebuda, & Karwowski, 2018).

Metacognitive control as predictor of creativity

As a central result of this study, we found that higher metacognitive control at all three levels – fully engaging in the task, using goal-directed strategies, and effectively discarding uncreative ideas – independently predicted higher creative performance. A relatively higher contribution was observed for metacognitive control at performance level, suggesting that cognitive effort spent on choosing effective task strategies is particularly well invested. Our results are consistent with previous findings from a think-aloud study which also found that using more complex strategies supports creative ideation, with the disassembly strategy being most effective in the alternate uses task (Gilhooly, Fioratou, Anthony, & Wynn, 2007; cf.; Jia, Xu, & Zhang, 2022). We further observed that metacognitive control had a larger impact on supporting the creativity of ideas than the quantity of ideas. In fact, metacognitive control at response level apparently increased DT creativity at the cost of DT fluency. This finding reflects an effective adherence to task instructions that asked to focus on producing creative ideas rather than many ideas, which is likely supported by metacognitive control at response level. Less metacognitive control at response level may be needed in more traditional DT tasks that instruct to mainly come up with many ideas (Nusbaum, Silvia, & Beaty, 2014; Saretzki, Forthmann, & Benedek, in press). In sum, we see how metacognitive control at different levels had nuanced, facilitative effects especially on the creativity of DT performance. These findings suggest that the robust positive relationship of cognitive control (Benedek & Jauk, 2019; Chrysikou, 2018) and DT performance is at least partly due to the supportive effects metacognitive control in creative ideation.

We further explored whether metacognitive control affects real-life behavior, such as via the established role of DT ability in creative activities and achievements (Said-Metwaly, Taylor, Camarda, & Barbot, 2022). Metacognitive control did not relate to how frequently people engage in creative behaviors, but it rather predicted the degree of people’s creative achievements. This appears plausible as the frequency of creative activities, unlike achievements, does not reflect the effectiveness or success of the activity. For example, intelligence (a common proxy of cognitive control) was previously found shown to be unrelated with creative activities but specifically predicted creative achievements and further moderated the effect of activities on achievements (Jauk, Benedek, & Neubauer, 2014; but also see, Lebuda, Zielinska, & Karwowski, 2021). These results are consistent with available findings showing that had demonstrated the relevance of self-reported metacognitive skills for real-life creative accomplishments (Puryear, 2015), and indicate that findings extend from trait-like measures of CMC to performance-related assessments of CMC. Specifically, higher levels of metacognitive control at task and performance levels (but not response control) predicted more creative achievements, and the effect of metacognitive control at performance level persisted even after controlling for covariation with DT ability. It seems possible that metacognitive control at performance level partly captured a more general tendency of how strategic people approach creative projects also known as creative self-regulation (Ivcevic & Nusbaum, 2017; Zielińska, Lebuda, & Karwowski, 2023). Together, these findings show that metacognitive control in DT may even be relevant for creative achievement, partly mediated by DT ability.

Limitations and future directions

Recent work has demonstrated that all three main components of CMC – metacognitive control, monitoring and knowledge – contribute independently to creative performance (Lebuda & Benedek, 2024). This study now delved into metacognitive control processes more specifically (at task, performance, and response levels), but it has not considered metacognitive monitoring and knowledge at the same level of detail. Hence, future work should also study other metacognitive components these levels of granularity and ideally also consider their interplay. Further avenues include testing how findings generalize across different tasks and populations especially from more realistic contexts (Agnoli & Mastria, 2023; Beghetto & Mangion, 2023; Kaufman, 2023).

We observed support for the validity of the new metacognitive control measures and their contribution for DT performance, but the effect sizes were quite modest, which can have different reasons. First, the correlations may have been undermined by the partly modest reliability of the measures. We further need to keep in mind that metacognition only refers to how well people reflect on and regulate their cognition, but not to how effectively they carry out the main task (i.e., core cognition). In fact, encouraging people to use a proven strategy when stuck was shown to benefit only those people with high cognitive ability (Nusbaum & Silvia, 2011), suggesting that even the best strategy is only as good as its implementation. A comprehensive prediction of DT performance thus relies on the effects both core and metacognitive processes. We further acknowledge that, although our metacognition measures clearly focused on the task performance at hand, they were not sensitive to the dynamic interplay of metacognitive processes over time. For example, metacognitive control at performance level reflects an adaptive regulation, such as to change strategies when the current approach is no longer effective. Future work thus may consider investigating how accurately people recognize when they get stuck (metacognitive monitoring of performance) and how effectively they regulate the process accordingly, such as by investing time to reconsider the strategy instead of simply trying harder. It was shown that people tend to mistake reduced fluency over time as reduced performance, although the quality of ideas was improving with time (i.e., the creative cliff illusion; Lucas & Nordgren, 2020). Future research may thus consider process data and think aloud protocols that enable time-sensitive assessments of metacognitive processes during task performance (Pringle & Sowden, 2017; Vieira, Kannengiesser, & Benedek, 2022) to unveil the dynamic interplay between core and metacognitive processes at task, performance, and response levels toward a more complete understanding of the role of metacognitive processes in creative performance.

Acknowledgments

This work was supported by the European Union’s Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement No 896518. The authors acknowledge the financial support by the University of Graz.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Materials, data, and analysis scripts of this work are provided via the Open Science Framework (https://osf.io/9qpk5/).

Additional information

Funding

The work was supported by the H2020 Marie Skłodowska-Curie Actions [896518].