Keywords

1 Introduction

Educational reform efforts in numerous countries (mostly since 2015) focus on the integration of science, technology, engineering and mathematics (STEM) policies in national curricula to promote interdisciplinary learning (Zhan et al., 2022). For example, the Next Generation Science Standards (NGSS; Next Generation Science Standards Lead States, 2013) endorse coherent and interconnected content and a coherent and interconnected approach to the STEM disciplines to support the development of students’ scientific literacy. The NGSS include practices and core disciplinary ideas from engineering and the sciences that bring out the connections between the different domains, and facilitate their introduction in teaching and learning in an integrated way.

Even though inconsistencies in the definition of STEM education exist, it is nevertheless typically conceptualized as a teaching approach that incorporates skills from and knowledge of different disciplines, situated in real-world issues (Martín-Páez et al., 2019). The integration of other disciplines within STEM curricula is also a common practice in order to promote 21st century skills, such as creativity and entrepreneurship. For example, STEAM refers to the integration of arts and/or humanities in STEM curricula; STEAME additionally incorporates entrepreneurship. For the purposes of this article, the term STEM+ will be used from this point forward to represent all possible variations of STEM plus the additional disciplines that can be incorporated in STEM curricula. Educational approaches, such as inquiry-based (e.g., Pedaste et al., 2015), project-based (e.g., Capraro et al., 2013) and design-based learning (e.g., Crismond & Adams, 2012) are ideal for implementing STEM+ initiatives because they are student-centered approaches that aim at addressing real-world issues through methods and practices like those of professionals (Martín-Páez et al., 2019; Thibaut et al., 2018). These approaches can facilitate the application of STEM+ concepts and practices in contexts that are relevant and interesting to students (National Academy of Sciences, 2014).

Laboratory experimentation is a fundamental aspect of science and engineering, and hence, the aforementioned educational approaches strive to provide relevant learning experiences to students. Laboratory work provides opportunities for students to implement scientific practices and skills (e.g., observation, hypothesis generation) to test theories and understand natural phenomena (de Jong et al., 2013; de Jong et al., 2014), and also to develop practical (e.g., handling equipment) and transferable (e.g., problem-solving, time management) skills (Reid & Shah, 2007). Traditionally, experimentation took place solely in physical laboratories (PL) that allowed learners to interact with real world physical/concrete materials and apparatus in order to observe and understand natural phenomena. This direct physical experience has been reported to be of pivotal importance during experimentation (e.g., Gire et al., 2010; Kontra et al., 2015; Zacharia, et al., 2012).

However, the exponential growth of technology has led to the need to rethink the practice of laboratory experimentation (Bybee, 2009; de Jong et al., 2013; National Research Council, 2006), with many studies in the past decades focusing on the use of virtual laboratories (VL) and the consequences for teaching, learning and research. Reeves and Crippen (2021) defined VL as “technology-mediated experiences in either two- or three-dimensions that situate the student as being in an emulation of the physical laboratory with the capacity to manipulate virtual equipment and materials via the keyboard and/or handheld controllers” (p. 16). The need for VL to be optimally integrated in STEM+ education became even stronger due to the COVID-19 pandemic and the subsequent rapid shift towards online/blended learning (European Commission, 2022), which inevitably promoted VL as an alternative to PL (Bazelais et al., 2022; Radhamani et al., 2021; Raman et al., 2021). Hence, it is not surprising that there is an increasing focus on the integration of technology to support learning in STEM education research (Zhan et al., 2022).

However, given that both PL and VL are available options for experimentation purposes, the dilemma persists as to which means of experimentation should be preferred for optimizing student learning. For instance, we are still looking for definite answers to questions such as, “Which means of experimentation under what circumstances optimizes student learning across grades K-16, PL or VL?” and “Should PL and VL be used alone or in combination across grades K-16?” The objective of this paper is to synthesize the theoretical and empirical perspectives emerging from the research concerning the exploration of effects of using VL and PL on students’ learning during experimentation in order tο contribute to the efforts of the community to answer these questions and to inform the research about and practice of laboratory experimentation in STEM+ education.

2 The Theoretical Perspective

Traditional views of cognition claim that the brain is made up of abstract functions and that it is a separate entity from the body, a notion that embodied theories currently challenge (Marmeleira & Duarte Santos, 2019). Embodied cognition theories differ in the degree of the effect of sensorimotor experience on cognition that is postulated (Wellsby & Pexman, 2014), but nevertheless, their overarching notion is the same: the interaction between the environment and the body influences cognition (Clark, 2008; Pouw et al., 2014; Wilson, 2002). As Pouw et al. (2014) mentioned, learning seems to depend on “gradual internalization of sensorimotor routines” (p. 65). Hence, researchers attempt to explain human motor, perception and cognition systems as dependent on the body (Farina, 2021). The mental representation of an object involves the sensory and motor regions of the brain, which are activated when the object is perceived or interacted with (Yee & Thompson-Schill, 2016). This mental representation encompasses not only the visual properties of an object, but also relevant actions (Barsalou, 2008; Gibbs, 2005). Neurological studies have shown that memory recall activates areas of the brain associated with the sensorimotor information experienced during an episode (Kiefer & Pulvermüller, 2012). For example, during the retrieval of haptically encoded stimuli, somatosensory and motor areas of the brain are activated, whereas for visually encoded stimuli, the activation of vision-related areas is observed (Stock et al., 2009). Embodied cognition theories tend to be appealing in terms of representing the organization of conceptual knowledge because they also predict how information is obtained (through sensorimotor experiences) and how and where it is processed (in the relevant sensorimotor systems; Yee et al., 2018).

For educational research, embodied cognition theories provide the opportunity to explore the impact of action on cognition throughout development and to utilize this knowledge to scaffold the teaching and learning process (Kontra et al., 2012). Hayes and Kraemer (2017) noted that embodied cognition theories can support our understanding of how students’ STEM learning is enhanced, given the student-centered and hands-on nature of STEM education. In the study by Kontra et al. (2015), even brief physical experience with the forces related to angular momentum led to activation of the sensorimotor systems of the brain used to execute similar actions in the past, which resulted in the development of understanding about that concept. Rich sensorimotor experiences can support the presence of multimodal representations that facilitate learning (Kiefer & Trumpp, 2012) with a variety of ways to engage the body, from limited (e.g., gestures) to full-body movement (Skulmowski & Rey, 2018).

The provision of high-embodiment experiences does not necessarily guarantee a positive impact on learning, but bodily experiences that relate to the task at hand seem to do so (Johnson-Glenberg, 2019; Skulmowski & Rey, 2018). For example, in a study by Mavilidi et al. (2017), preschoolers were engaged in activities related to the solar system in three experimental conditions: (1) integration of related physical activities, (2) integration of irrelevant physical activities and (3) no physical activities. It became apparent after an immediate and a 6-week assessment, that the preschoolers participating in tasks that incorporated meaningful physical activities outperformed the rest of the groups, and even the preschoolers participating in the irrelevant physical activities outperformed the students who were not involved in any type of physical activities. Zohar and Levy (2021) investigated whether an increase in bodily engagement (movie, simulation, joystick and haptic device with force feedback) subsequently leads to an increase in understanding of the concept of chemical bonding. The movie, simulation and joystick conditions resulted in similar conceptual development, whereas the participants in the haptic device condition, which offered the highest bodily engagement, had a significantly higher increase in knowledge. In a study by Qi et al. (2021), providing force feedback in a simulation to students with limited prior understanding of forces also facilitated learning, but providing additional visual cues (i.e., abstract arrows) did not improve performance. Hence, not all bodily experiences can enhance learning in the same manner; alignment between the manipulation and the learning objective is needed.

The active and meaningful interaction with materials and apparatus during experimentation is what enhances learning, and not physicality in itself (Han, 2013; Klahr, et al., 2007; Pouw et al., 2014; Triona & Klahr, 2003; Zacharia & Olympiou, 2011). Hence, embodied cognition theories do not necessarily favor a specific mode of experimentation (Rau, 2020). However, haptic perspectives on learning do favor the haptic manipulation of materials because, when combined with visual stimuli, it can support memory retrieval, minimize the likelihood of cognitive overload and support the conceptual grounding of abstract concepts (Rau, 2020). As stated by Van Doorn et al. (2010, p. 813), “[t]he term haptic refers to a perceptual system that combines both input from receptors in the skin and kinesthetic information.” Through touch, we can gather information (e.g., about the properties of an object) and act (e.g., lift the object), but also, based on the sensory feedback received (e.g., the force used was not enough to lift the object), we attune our actions to fit our initial intentions (e.g., use more force; Reiner, 2008).

The haptic and the visual system are complementary in nature. In a study by Reiner et al. (2006), participants used a haptic interface to lift virtual cylinders marked with the labels “heavy”, “light” or “###” (neutral condition), which were compatible, incompatible or neutral with regard to the actual weight of the cylinders. It became apparent that both the reaction time and error rate were lower for the cylinders with a label that was compatible with their weight, and higher when the label was incompatible. The haptic system has a higher processing cost than the visual system and hence it will be invoked when visual information is inadequate for addressing a targeted task (Hatwell, 2003; Klatzky, et al. 1993). Given that touch has an inherent bias towards how an object “feels” (e.g., texture, material) rather than its structural properties (e.g., size, shape; Klatzky et al., 1991; Klatzky et al., 1987; Klatzky et al., 1993), the haptic system will be activated only to assist vision when exploring structural properties (Hatwell, 2003), but it can be used for discriminating objects based on their surface textures (Heller, 1989). Thus, based on haptic perspectives on learning, it is expected that haptic manipulation, as an additional available modality, will augment the development of concept-specific understanding, particularly of concepts for which haptic cues can facilitate the development of multimodal representations, such as force (Han, 2013; Zohar & Levy, 2021) and mass (Lazonder & Ehrenhard, 2014; Pavlou et al., under review; Zacharia et al., 2012). For example, in the study by Lazonder and Ehrenhard (2014) regarding free fall, students who engaged with physical materials were able to develop scientifically correct understandings because the haptic sensory feedback available facilitated the correction of students’ misconceptions and revision of mass-related beliefs. This process was not evident in the demonstration or VL groups.

Haptic sensory feedback simply cannot be offered in a virtual environment. Haptic technologies for providing sensory feedback (e.g., force feedback) in VL do exist, and studies investigating their impact on learning (e.g., Bivall et al., 2011; Han & Black, 2011; Jones et al., 2006; Magana et al., 2019; Zhuoluo et al., 2019) have reported positive findings, but the sensory feedback they provide is still quite limited in comparison to the haptic and dynamic feedback available when engaging with PL. Haptically enhanced simulations are also primarily focused on developing skills (e.g., related to surgeries; Qi et al., 2021), and the integration of such technologies, especially in formal education, is still very limited (Georgiou & Ioannou, 2019; Johnson-Glenberg, 2019; Malinverni & Pares, 2014).

As Pavlou et al. (under review) pointed out, when considering comparative studies in the field of science education from the haptic encoding and embodied cognition perspectives, findings of an equal or even negative impact of PL on learning can be attributed to a lack of significant perceptual differences between the VL and PL being compared (i.e., the perceptual stimuli and the available feedback offered in the PL did not differ significantly from the virtual environment; e.g., Han, 2013) and/or the participants’ prior experiences/knowledge of the concepts under investigation. As far as the latter is concerned, it is evident that most studies in the field of science education focus on the primary school years and onwards (Wörner et al., 2022; Zacharia, 2015). For older students, knowledge of the concepts under investigation could have been grounded in the early years and hence, embodied experiences might not be a prerequisite for those students. As Yee et al. (2018) mentioned, as development progresses, the reliance on abstract knowledge increases, and direct sensorimotor experience is not as necessary as for young children. However, PL can be more beneficial than VL in the early years of education when considering the reliance on grounded experience, especially through the haptic manipulation of objects (Pavlou et al., under review).

For example, in the comparative study by Zacharia et al. (2012), preschoolers with scientifically correct prior understanding of the concepts under investigation (the function of a balance beam) who interacted with either physical or virtual manipulatives outperformed students with incorrect prior knowledge who engaged with the virtual environment. Their findings indicated that haptic sensory feedback, which is a unique affordance of PL, is a prerequisite for learning if participants do not have any previous understanding of the concepts. The study by Pavlou et al. (under review) validated and also expanded the findings by Zacharia et al. (2012). This study compared the conceptual understanding of preschoolers engaged in VL or PL in three subject domains (balance beam, springs and sinking/floating). In the balance beam and springs domains, the mass of the objects (a property that should be multimodally grounded to support learning) is a causal factor affecting the experimental output, but for the sinking/floating domain, the idea that mass affects the object’s behavior in water is a common misconception children have (see, e.g., Havu-Nuutinen, 2005; Hsin & Wu, 2011; Pavlou et al., 2018). Preschoolers working in the balance beam and sinking/floating domains had prior understanding of the domains, but participants working in the springs domain did not. The mode of experimentation did not affect the learning outcome for preschoolers engaged in the balance beam domain, but preschoolers who engaged with PL during experimentation in the domain of springs outperformed the participants in the VL group because the haptic sensory feedback offered in the PL group seemed to facilitate the development of understanding of the causal effect of mass. However, in the sinking/floating domain, haptic sensory feedback available during experimentation with physical materials impeded the development of a scientifically correct understanding, and the mass-related idea that “heavy objects sink/light object float” was the most dominant idea used by preschoolers to explain the phenomenon both before and after experimentation. Similar findings were also found in the preliminary study (Pavlou et al., 2018). The authors concluded that although information about other object properties was available, especially through vision, and was at times more salient than mass, haptic cues related to the mass of the objects were the most dominant perceptual cues that led to the fixation/empowerment of relevant ideas. Hence, providing haptic sensory feedback during experimentation can be detrimental (for students holding relevant misconceptions, as in the sinking/floating domain), beneficial (for students with no prior understanding, as in the springs domain) or have no significant impact (as for the preschoolers with prior knowledge of the balance beam domain). To conclude, it seems that haptic sensory feedback, which is available through engagement in PL, is not always a prerequisite for learning (see also Zacharia, 2015). In other words, VL can be used under certain conditions as a means for experimentation because they can support or even augment the development of the understanding of scientific concepts.

3 The Empirical Perspective

VL carry a lot of affordances that can support learning, as highlighted by many researchers (e.g., Faulconer & Gruss, 2018; de Jong et al., 2013; Olympiou & Zacharia, 2012; Potkonjak et al., 2016; Zacharia, 2015). For example, VL provide the opportunity to experiment with unobservable phenomena (e.g., radiation), to manipulate variables (e.g., light rays) and other parameters (e.g., time, spatial dimensions, data displays) and to engage with abstract concepts (e.g., symbolic representations of light). In addition, VL provide access to multiple users, are cost- and safety-efficient and minimize trial errors. Therefore, the empirical literature initially focused on exploring whether VL can support teaching and learning and specifically, whether their impact can be similar to or even greater than that of PL. There are studies showing an advantage of VL over PL (e.g., Akpan & Andre, 2000; Bell & Trundle, 2008; Finkelstein et al., 2005) and studies showcasing opposite findings (e.g., Gire et al., 2010; Marshall & Young, 2006). However, the majority of comparative studies do not indicate that one mode of experimentation dominates over the other (e.g., Chini et al., 2012; Evangelou & Kotsis, 2019; Klahr et al., 2007; Leung & Cheng, 2021; Reece & Butler, 2017; Triona & Klahr, 2003; Zacharia & Constantinou, 2008; Zacharia & Olympiou, 2011). Inconsistencies between these studies can in part be attributed to the varying affordances carried by PL and VL that were utilized, the varying methodological approaches employed (D’Angelo et al., 2014; Faulconer & Gruss, 2018; Ma & Nickerson, 2006) and the different theoretical perspectives (or in some cases the lack thereof) adopted to predict and explain the learning outcomes (Reeves & Crippen, 2021).

Nevertheless, literature reviews conducted over the years have pointed out that VL has an effect equal to or even greater than that of PL (Brinson, 2015; D’Angelo et al., 2014; Faulconer & Gruss, 2018) or other teaching approaches (Rutten et al., 2012; Smetana & Bell, 2012). Overall, the extant empirical research studies have revealed the potential of VL experimentation for enhancing students’ learning across grades K-16 (e.g., Potkonjak et al., 2016; Triona & Klahr, 2003; van der Meij & de Jong, 2006; Zacharia & Anderson, 2003; Zacharia et al., 2008; Zacharia et al., 2012). Consequently, it can be argued that VL can provide learning experiences that are just as meaningful to students as PL and, considering the many more unique affordances carried by VL as opposed to PL, some could argue that under certain circumstances, VL could be less “messy”, easier to manage, and more flexible and expandable than PL (Klahr et al., 2008), especially if we consider some of the “disadvantages” of PL (e.g., space and time restrictions, absence of abstract representations). Hence, should one mode of experimentation be preferred over the other?

VL and PL have complementary affordances and thus their combination can support presentation of multiple representations of science concepts (de Jong et al., 2013; Olympiou & Zacharia, 2012; Puntambekar et al., 2021; Zacharia, 2007; Zacharia & de Jong, 2014; Zacharia & Michael, 2016). The meaningful integration of multiple representations of a concept (e.g., a physical and a virtual/abstract representation) can enhance learning to a greater extent than stand-alone representations (Ainsworth, 2008). For example, in a study by Wang and Tseng (2018), third-graders who engaged in the combination condition (VL and then PL) outperformed the students who engaged in stand-alone modes of experimentation. Kapici et al. (2019) also reported an advantage of the combination of VL and PL for seventh-grade students for the concept of electricity. Even though some studies have showcased that the stand-alone use of VL and PL has a greater impact on learning than their combination during experimentation (e.g., Gnesdilow & Puntambekar, 2022), overall, their combination seems to be more beneficial than stand-alone use (de Jong et al., 2013; Wörner et al., 2022). Teachers from different educational levels also seemed to support the combination of the two modes of experimentation for teaching and learning practice (Tsihouridis, et al., 2019).

Studies investigating the combination of VL and PL (e.g., Achuthan et al., 2017; Fuhrmann, et al., 2014; Olympiou & Zacharia, 2012, 2014; Trundle & Bell, 2010; Yuksel et al., 2019; Zacharia, 2007; Zacharia et al., 2008) and the different ways in which they can be combined (e.g., at the same time, blended, in sequence), have showcased their complementary nature, but nevertheless provide limited information on the preeminent affordances of each mode (Lazonder & Ehrenhard, 2014). Similarly, because most of the comparative studies focused on improving students’ learning (Reeves & Crippen, 2021), VL and PL experimentation were generally compared (e.g., as instructional approaches) without necessarily accounting for the potential effect of specific affordances. This is a vital underpinning in order to achieve optimal combinations for learning (Rau, 2020; Wörner et al., 2022; Zacharia et al., 2008). Thus, identifying these unique affordances of each mode of experimentation and their effect on student learning across grades K-16, as well as understanding when and under what conditions a particular mode of experimentation, along with its unique affordances, becomes more effective is still a critical issue. For instance, the community focusing on PL and VL has not yet distinguished the circumstances under which a unique affordance optimizes learning and whether such an effect holds true across grades K-16. For example, does providing abstract representations – a unique affordance of VL – support student learning the same way across grades K-16? The community also lacks information on how different unique affordances interact with each other and how this interaction impacts student learning. Moreover, the theoretical perspective through which the issue of unique affordances is approached affects the predictions and explanations articulated with regard to the impact of VL on learning (Rau & Herder, 2021). In the next section, we attempt to synthesize the theoretical and empirical perspective presented in this article.

4 Bridging Theory and Research with Practice in STEM+ Education

Educational approaches that facilitate STEM+ learning, such as the inquiry-based learning approach or the engineering design learning approach, call for students’ active engagement and involvement in the learning process (de Jong, 2019). Laboratory work is a focal aspect of these approaches, which can be productively enacted with virtual and/or physical means (de Jong et al., 2013; Zacharia & de Jong, 2014). The role of laboratory experimentation becomes even more crucial within STEM+ education, when science is used as the dominant discipline among all of the disciplines involved.

Given that both the theoretical and the empirical evidence support the idea that both PL and VL are viable means of experimentation for students (e.g., Finkelstein et al., 2005; Triona & Klahr, 2003; van der Meij & de Jong, 2006; Zacharia, 2015; Zacharia et al., 2008), including preschoolers (Pavlou et al., 2018, under review; Zacharia et al., 2012), the palette of experimentation possibilities available to teachers (i.e., use of PL, use of VL, use of combinations of PL and VL) needs to be expanded. In the past decades, research has shown that both VL and PL can support learning and that their meaningful combination can be conducive to developing multimodal representations of concepts (de Jong et al., 2013; Olympiou & Zacharia, 2012; Puntambekar et al., 2021; Zacharia, 2007; Zacharia & de Jong, 2014; Zacharia & Michael, 2016). However, identifying the situations in which PL and/or VL can be utilized to optimize student learning across grades K-16 is still a critical issue.

Olympiou and Zacharia (2012) developed a framework summarizing a series of considerations on how to combine/blend PL and VL. Based on their framework, contemplation of the affordances of each mode of experimentation (unique or not) in conjunction with the learning objectives of each experiment and students’ background (e.g., prior conceptions, skills) is vital. The framework was validated in studies concerning undergraduate students (Olympiou & Zacharia, 2012, 2014) and primary school students (Zacharia & Michael, 2016) that exhibited the advantages of blending the two modes of experimentation instead of using VL or PL alone. The framework can support the decision-making process when designing laboratory experimentation activities for STEM+ initiatives with the integration of VL and PL. Below, we discuss – and extend – some of the key considerations integrated in this framework, in conjunction with the relevant literature.

Research has shown that despite the overlapping affordances of VL and PL offered during experimentation (e.g., manipulation of material, perceptual grounding for abstract concepts), their unique affordances can affect learning in a different manner (for more details, see Olympiou & Zacharia, 2014; Zacharia, 2015). On the one hand, the ability to visualize abstract concepts (e.g., light rays, particles, current flow) and modify parameters, such as time and dimensions, is a unique affordance of VL. For example, in a study by Finkelstein et al. (2005), university students working with a virtual laboratory for electrical circuits in which they could manipulate parameters (e.g., voltage) and visualize current flow outperformed the students working with physical materials, to whom such affordances were not offered. A study by Zacharia and de Jong (2014) highlighted the learning benefits of using VL prior to PL for the understanding of complex circuits because the virtual environment offers the additional visualization of the current-flow. The appropriate consideration of these affordances for this subject domain (see, e.g., Zacharia & Michael, 2016) can facilitate the appropriate blending of both modes of experimentation to support development of multimodal understanding.

On the other hand, the availability of haptic sensory feedback during manipulation is a unique affordance of PL. According to haptic perspectives on learning and empirical studies (e.g., Han, 2013; Lazonder & Ehrenhard, 2014; Pavlou et al., under review; Zacharia et al., 2012; Zacharia, 2015), the presence/absence of haptic sensory feedback can affect learning; hence, the identification of the conditions under which it can be beneficial or detrimental during experimentation can support the optimal integration of VL and PL in STEM+ education. The development of understanding of concepts such as forces (e.g., Han, 2013; Zohar & Levy, 2021), mass/weight (e.g., Lazonder & Ehrenhard, 2014; Pavlou et al., under review; Zacharia et al., 2012) and magnetic fields (e.g., Reiner, 1999) seems to benefit from interaction with haptic stimuli. Haptic sensory feedback is possibly even more essential for students with no prior embodied experience/knowledge of a domain, for whom haptic cues can enhance concept-specific understanding when visual cues alone are not adequate for solving a task (e.g., Pavlou et al., under review; Qi et al., 2021; Zacharia et al., 2012; Zohar & Levy, 2021). However, it should be taken into account that haptic cues related to a students’ misconception might also hinder learning (i.e., lead to a fixation/empowerment of the misconception) as in the sinking/floating domain of the Pavlou et al. (under review) study.

In addition, based on the theoretical perspective in this article, the lack of embodied experiences with a domain seems to be a vital consideration when selecting/combining VL and PL. The initial sensorimotor grounding of scientific knowledge in STEM classrooms through hands-on approaches can support the development of abstract knowledge (Hayes & Kraemer, 2017). When prior knowledge/embodied experience with the concepts under investigation is lacking, then the use of PL during experimentation will most likely significantly improve the learning outcome (e.g., Pavlou et al., under review; Zacharia et al., 2012). For example, if students do not have any prior experience with a domain, PL can proceed VL to provide such experiences, especially when haptic manipulation (in combination with other modalities) can enhance conceptual understanding. To amplify the effect of embodiment, there should be a strong relation between the task at hand and the bodily movements enacted by the students working in a virtual or physical environment (e.g., Johnson-Glenberg, 2019; Mavilidi et al., 2017; Qi et al., 2021; Skulmowski & Rey, 2018; Zohar & Levy, 2021).

Of course, the need to provide embodied experiences (including haptic sensory feedback) might not be as vital for older students because such experiences were probably acquired during their early years through formal and/or informal learning, an argument that can even support the stand-alone use of VL (especially when considering the importance of distance/online learning). This argument agrees with empirical research that has shown an advantage or equal effect of VL on learning for a variety of domains and disciplines for students in primary school and older (Brinson, 2015; Zacharia, 2015). Nevertheless, the reliance of younger children on sensorimotor experience indicates the importance of using PL during experimentation. However, studies with preschoolers (Pavlou et al., 2018, under review; Zacharia et al., 2012) have shown that under certain conditions, the use of VL can also facilitate learning for younger children. Given that empirical research has focused on older students (for details, see Zacharia, 2015), our understanding of the situations in which VL can support learning in early childhood education is still limited.

Another aspect that should be taken into consideration when selecting a means of experimentation for STEM+ enactments is factors related to the affective domain. For example, Justo et al. (2022) found that even though VL and PL had similar effects on the learning of basic engineering concepts, students’ motivation increased with the use of physical materials. As Tsihouridis et al. (2019) also highlighted, primary students usually favor PL, whereas older students prefer VL. Hence, aspects other than learning (which were not the focus of this article, for example, the enhancement of students’ skills or attitudes) can also guide the decision about which mode of experimentation is more appropriate.

The additional value of VL in education is no longer disputed, and research seems to be turning towards addressing the question of “which one is better” in what instances, how and for whom (de Jong, 2019). As noted in this article, the theoretical underpinning guiding how these questions are being examined does matter for the instructional design and the way the findings are explained (see Rau, 2020; Rau & Herder, 2021). Combining empirical and theoretical perspectives is essential. In addition, the current literature portrays the complex interplay between aspects, such as the affordances of each mode of experimentation, the concepts under investigation (e.g., haptic cues can augment concept-specific understanding) and students’ prior embodied experiences/knowledge (including their misconceptions). However, the influence of this interplay on students’ learning and the possible dominance of one aspect over others (e.g., whether students’ misconceptions can impede concept-specific understanding in VL/PL) is yet unclear; research is still needed to develop a comprehensive framework for the optimal integration of PL and VL in STEM+ education.