Cognitive flexibility (CF) is widely regarded as a crucial cognitive ability and a fundamental executive function, defined as the capacity to adapt one's thinking or behavior in response to unpredictable environmental demands (Barbey et al., 2013, Miyake et al., 2000). CF has also been conceptualized as thinking “outside the box” and seeing things from different perspectives, to adapt quickly to changing circumstances (Diamond, 2013). Research has demonstrated that CF is closely related to other cognitive abilities, such as attention shifting, strategic planning, perceptual speed, and working memory (Deak, 2003, Dajani and Uddin, 2015, Lucenet et al., 2014). Furthermore, CF has ecological implications, as it was found to be associated with academic achievement and the ability to navigate daily challenges, ultimately supporting optimal social and occupational functioning (Arán-Filippetti and Krumm, 2020, Braem and Egner, 2018, Highgate and Schenk, 2021, Laureiro-Martínez and Brusoni, 2018, Uddin, 2021).

The literature over the past 100 years has struggled to define the concept of CF coherently (Ionescu, 2012). While the concept seems intuitive, in practice there is a wide variety of conceptualizations, stemming mainly from the close context of CF to various everyday behaviors (e.g., Barbey et al., 2013, Cragg and Chevalier, 2012, Deák and Wiseheart, 2015). Such multiplicity of definitions has created different categorical divisions of this ability, the most common of which places CF into a single, uniform general ability, represented across a wide range of tasks, which therefore cannot be subdivided (Kercood et al., 2017, Zelazo, 2006). The second view sees CF as a modular function consisting of several subtypes (Gruner and Pittenger, 2017, Muhle-Karbe et al., 2014, Ravizza and Carter, 2008). However, within each such category, there are different definitions, which have maintained the inconsistency of CF conceptualization and the methods of assessing it (Deák and Wiseheart, 2015, Gruner and Pittenger, 2017, Smith et al., 2018).

Studies that refer to CF as representing a unified concept, exhibit diverse definitions, including attentional shifting (Casey et al., 2004), attentional flexibility (Sali et al., 2020), reversal learning (Izquierdo et al., 2017), task switching (Monsell, 2003) and set-shifting (Fuglset, 2019). Within the specific concept of set-shifting, for example, often used interchangeably with CF, it has been conceptualized as the ability to switch between cognitive sets (Dennis & Vander Wal, 2010), rules (Berner et al., 2019, Yerys et al., 2015), tasks, operations or mental sets (Fuglset, 2019, Van Autreve et al., 2013). By contrast, other authors have associated it with processes requiring attention shifts (Canas et al., 2006, Richter and Yeung, 2014, Wager et al., 2004).

On the other hand, conceptualizing CF as a modular construct suggests that it should not be confined to a singular definition, but should rather be assessed as a multidimensional and layered entity comprised of subtypes, each possessing unique characteristics that differentiate one from another, while still sharing a common component of flexibility (Dajani and Uddin, 2015, Kim et al., 2012). However, similar to the unified view, the modular view also shows a lack of consistency in conceptualizing CF and its subtypes. For example, CF has been divided into subtypes of stimulus switching, response switching, and set switching (Kim et al., 2012); perceptual switching and rule switching (Ravizza & Carter, 2008); schematic control, contextual control and sensory-motor control (Zhu & Han, 2022); simple alternation, mental flexibility, perseveration and perceptual shifts (Tchanturia et al., 2004); or attentional set-shifting and reversal learning (Wildes et al., 2014).

One model that seems to meet CF conceptions of single construct models, while also reflecting modular theoretical CF perspectives (Casey et al., 2004, Monsell, 2003, Yerys et al., 2015) is that suggested by Dajani and Uddin (2015). These model conceptions and CF subtypes corroborate parallel studies in the literature, highlighting their relevance and contribution to the discourse on CF research (Kim et al., 2012, Meiran and Marciano, 2002, Ravizza and Carter, 2008). Additionally, the model outlines the neural networks underlying the various types of CF and links it to key executive functions. Specifically, response inhibition and working memory (Diamond, 2013, Miyake et al., 2000).

The first subtype suggested is task switching, which refers to the ability to shift between different instructions in the task, such that one has to change the acquisition principle and adopt a new rule of performance. The second subtype is switching sets, which refers to the ability to switch between perceptual features of the stimulus presented (e.g., referring to the color vs. referring to the form). Here, the change is not in task rules, but in different features of the stimulus. The third subtype of CF is termed stimulus–response (S-R) mapping, which refers to shifting between two responses, while the stimulus remains the same. Specifically, upon switching, one has to respond differently to a stimulus previously matched to a different response. Although the authors suggested including switching sets and S-R mapping under the broader category of set-shifting, we have deemed both as distinct subtypes, because other researchers have identified them as separate CF sub-divisions (Kim et al., 2012, Meiran and Marciano, 2002, Schroder et al., 2012, Von Bastian and Druey, 2017).

The specific conceptualization of the three subtypes is in line with other models, which despite their use of varying terminology, have suggested similar definitions for each component. For example, Kim et al. (2012) proposed a comprehensive classification of perception, context, and response domains. Context switching corresponds to task switching, emphasizing the cognitive capacity of transition between distinct rule sets or principles. Perceptual switching aligns with switching sets, focusing on the ability to reconfigure cognitive strategies based on perceptual stimulus characteristics, and response switching parallels S-R mapping, highlighting adaptive changes in motor responses while maintaining stimulus constancy.

Additionally, Meiran and Marciano (2002) used the terms task rule shift, dimension shift, and stimulus–response shift, which are parallel to task switching, switching sets, and S-R mapping, respectively (Dajani & Uddin, 2015). Although other authors have proposed similar conceptualizations (Bunge and Zelazo, 2006, Ravizza and Carter, 2008), they did not employ confirmatory factor analysis to rigorously test the hypothesized structural relationships and discriminant validity of these CF subtypes.

Following the modular view, it has also been suggested that CF subtypes are differentiated not only by their characteristics, but also by the level of complexity required for their execution. Various studies have proposed that there exists a hierarchical organization of task flexibility representations, as elements of real-world tasks tend to be organized at different levels of difficulty (Bunge and Zelazo, 2006, Lien and Ruthruff, 2004, Von Bastian and Druey, 2017, Zhu and Han, 2022).

The hierarchical organization provides a systematic understanding of CF performance, which varies depending on the nature of the required changes. It highlights that higher levels of demand have a stronger impact on lower-level processes than lower levels have on those above them (Kleinsorge & Heuer, 1999). Such a structure facilitates the identification of shared shifting abilities and cognitive processes linked to executive functions (Von Bastian & Druey, 2017). Consequently, may improve understanding of task performance and the complexities of cognitive control in varied settings (Lien & Ruthruff, 2004).

Examples of differentiated levels of structuring include Lien and Ruthruff's (2004) division of CF into three levels, whereby judgment is placed at the highest tier, a mapping level in the middle, and a response level at the lowest tier. Von Bastian and Druey (2017) proposed organizing CF into five levels of judgment shifting, dimension shifting, S-R mapping, response set shifting, and stimulus set-shifting.

However, the organization suggested by Bunge and Zelazo (2006) appears to concur well with Dajani and Uddin's (2015) model, because the conceptualization of the two models seems to overlap. According to Bunge and Zelazo’s (2006) framework, CF consists of three hierarchical levels that reflect the complexity of flexible use in cognitive development. At the lowest level, basic rules involve straightforward S-R associations, representing initial mastery of basic rule-following, in line with the S-R mapping subtype. The next level demonstrates the ability to switch flexibly between pairs of conditional rules, applying different dimensions such as color or shape within the same task. This category corroborates with the switching sets subtype. Finally, the third and most complex level represents the ability to switch between two incompatible pairs of rules, requiring not only application of one of the rules, but also resisting interference from conflicting rules (Bunge & Zelazo, 2006). This category is associated with the task switching subtype (Dajani and Uddin, 2015). Yerys et al. (2015) adopted this hierarchical model, but changed the terminology to S-R mapping, attention, stimulus evaluation, and task evaluation.

Despite the delineation of these subtypes, Dajani and Uddin's model lacks direct validation, and although it is grounded in research literature, neither the proposed categorization nor the testing of the model's hierarchical levels of complexity has been sufficiently substantiated. Therefore, the current study’s aims are threefold: First, to test whether a modular model based on Dajani and Uddin's (2015) framework provides a better fit for defining CF, compared to a view that conceptualizes it as a single, general ability. The second aim is to test the validity of the theoretical model proposed by Dajani and Uddin (2015), by evaluating the distinction of subtypes they have defined. The third aim is to assess whether the subtypes are differentiated according to the hierarchical organization suggested by Bunge and Zelazo (2006). Such an investigation could clarify the cognitive structure underlying CF, leading to a more precise identification of the CF construct, both conceptually and operationally.