Leveling the playing field: Grounding learning with embedded simulations in geoscience

The pace of scientific research is accelerating and the average citizen is increasingly faced with having to understand and reason about matters of science in their everyday lives. Because it is imperative that the public be engaged in science issues that have an impact on them and their communities, it is therefore also important that the research community develop more effective methods of teaching science. More specifically, it is important that science education research focuses on methods for improving access and learning for all students including those who may find learning about science to be challenging. The findings of the current study are significant because they indicate that integrating technology, embodiment, and enactment in a science learning activity can be particularly beneficial for students who have weaker spatial skills. This finding is important because it suggests that changes in instruction can have an impact on student success and may allow students who may typically face challenges when learning about science to succeed. In a world where understanding matters of science is relevant to all our lives, it is important that all students have equal access to the opportunity to learn about science.

Spatial skills can predict learning in science, technology, engineering, and mathematics (STEM) particularly when understanding depends on the learner’s ability to visualize, manipulate, and animate spatial information (Hegarty, 2010; Uttal & Cohen, 2012). For example, developing mental representations of scientific phenomena may require spatial skills when entities or systems are not directly observable due to their spatial or temporal scale. In addition, constructing mental models of some scientific phenomena may involve representation of the configuration of multiple components, the spatial relations between components, or information about the movements or causal interactions among the components (Gentner & Stevens, 1983; Hegarty & Just, 1993; Mayer, 1989). In the case of plate tectonics, the student needs to create a mental model of a process that is too large to actually be observed. They must accurately represent the layers of the earth and the locations of tectonic plates, as well as information about how those plates move and interact with each other to cause phenomena such as earthquakes. Consistent with this analysis, prior work has demonstrated that spatial skills can predict learning in earth sciences and geology (Black, 2005; Jaeger, Taylor, & Wiley, 2016; Sanchez, 2012; Sanchez & Wiley, 2010).

Given that performance on tests of spatial skills often predicts achievement in STEM domains (Uttal & Cohen, 2012), it is important to understand the conditions that may allow students with weaker spatial skills to also succeed. Students who are low in spatial skills may struggle to create mental representations of spatial information on their own; therefore, conditions that support the representation of spatial information may be one route to improving student understanding. For example, some studies have shown that providing students with visualizations of scientific phenomena can be helpful. When benefits are seen specifically among low spatial students, it has been suggested that these external supports aid learning because these students are less able to create “runnable” mental models of phenomena on their own, and that the visualization provides a grounding or basis for mental model construction. For example, Sanchez and Wiley (2010) found that presenting relevant animations alongside an expository text about plate tectonics effectively eliminated performance differences between male and female students, even though the male students showed higher performance on tests of spatial skills. While providing visualizations to learners does not always help the low spatial students to learn more (Höffler, 2010), sometimes it fulfills a compensatory role such that low spatial students’ understanding benefits from the external support or grounding that is provided by visualizations. The present study explores whether two other types of grounding (embodying through action, and embedding in classroom space) may serve to improve learning.

One approach that has been shown to benefit learning in general, but may also reduce the need for spatial skills when learning science, is grounding student understanding via action or gesture (Lindgren & Johnson-Glenberg, 2013). Previous research has indicated that gesture is well suited to capture spatial information and is frequently produced when talking about space (Kita & Özyürek, 2003; Lavergne & Kimura, 1987). Further, because gesture is an action it may be a natural medium for aiding the creation of mental models that involve action (Goldin-Meadow & Beilock, 2010). From an embodied cognition perspective, it is suggested that students’ experiences with acting out specific events may allow them to ground their understanding within rich sensorimotor experiences (Kontra, Albert, & Beilock, 2012). For example, Kontra, Goldin-Meadow, and Beilock (2012) conducted a study in which college undergraduates learned about angular momentum either by manipulating a pair of bicycle wheels on an axle, or by receiving verbal descriptions while observing others manipulate the wheels. Although both the observation and action groups were matched on understanding at pre-test, students in the action group improved significantly at post-test, whereas students in the observation group did not. Further, action or gesture may help students to generate spatial mental models and more effectively maintain those models in memory (de Ruiter: Gesture and speech production, unpublished; Morsella & Krauss, 2004; Wesp, Hess, Keutmann, & Wheaton, 2001). de Ruiter had students verbally describe an array of shapes and lines either with the array visible or after it was removed from sight. Speakers produced more gestures when the array was no longer visible, which was taken to suggest that the gestures helped them to retrieve and maintain the mental image of the array.

Action or gesture may also help direct attention to and promote a focus on spatial information. Alibali, Spencer, Knox, and Kita (2011) showed that when students were allowed to gesture during a gear movement task they tended to use depictive strategies, modeling the movement of the gears with their gestures. When gesturing was not allowed, students generated verbal rule-based strategies. While these different strategies did not impact problem-solving success (both groups solved the same number of items correctly), gesturing biased students toward considering more spatial aspects of the problem. This suggests that the action information expressed in gesture may influence thought by adding spatial information to one’s mental representations. Other recent work has demonstrated that hand gestures can serve as an external support for maintaining spatial representations and that the frequency of gesture use is correlated with having low spatial skills (Chu & Kita, 2011).

A complementary approach to grounding learning in action is grounding learning within familiar or concrete experiences. Similar to the benefits of enactment, this approach would suggest that when a simulation takes place within a familiar space, such as the student’s classroom, that it may be easier to begin to represent spatial information than when a simulation is contrived to take place in a less familiar or more abstract other space. Simulations that are contextualized within a familiar space may be advantageous because they provide a strong and intuitive link between the elements of the real world and the elements of the hypothetical world (Fyfe, McNeil, Son, & Goldstone, 2014; Goldstone & Son, 2005; Kotovsky & Gentner, 1996). As suggested by the theory of concreteness fading, starting with a concrete context and then gradually removing context-specific elements can support learning and transfer. Specifically, more contextualized learning activities can provide a concrete context on which to ground understanding of concepts that might otherwise be more abstract and less readily comprehended, which should benefit learning on the whole, but may also be especially beneficial for low spatial students.

Taken together, instructional approaches that provide grounding for learning in two complementary ways – embodying through action, and embedding through classroom space – would seem to be a promising avenue for providing low spatial students with the support they need to learn complex scientific phenomena. The Embedded Phenomena framework is one such approach that attempts to combine embodiment and grounded experience to support learning in science (Moher, 2006). Two central components of the Embedded Phenomena framework are that technology-enhanced simulations are grounded by mapping them to the confines of the classroom itself and that simulations involve students as enactive agents. The framework has been used as the basis for a variety of hybrid simulation activities in domains spanning astronomy (HelioRoom; Thompson & Moher, 2006), population ecology (WallCology; Moher, Uphoff, Bhatt, López Silva, & Malcolm, 2008), and hydrology (AquaRoom; Novellis & Moher, 2011). While the Embedded Phenomena framework has previously been implemented in elementary classrooms, none of these investigations has specifically tested for effects of the simulation activities on domain-specific conceptual learning, or the relationship of learning with students’ spatial thinking skills. These were the goals of the current investigation.

In the current investigation the simulated scientific phenomenon was earthquakes, and students from two fifth grade classrooms learned from a unit called RoomQuake that contained both content instruction and simulations. In both embedded and non-embedded conditions, students collected earthquake data and contributed to shared aggregate representations of earthquake magnitudes and locations. In the embedded condition, the 15 earthquake events happened sporadically and unexpectedly over a 4-week period and all public data representations were persistent and present throughout. When an event occurred, students moved from their seats to go gather wave information from seismic reading stations distributed around the classroom space, and then enacted the computation of epicenters with strings and their bodies (see Fig. 1). Meanwhile, students in the non-embedded condition received the same content instruction and did the same activities, but the epicenter computations were done with maps instead of with students’ bodies. Further, the earthquake events were not embedded in classroom space, students completed the epicenter computations for all 15 events as one activity, and public data representations and traces were only present in class for those days.

The primary learning goals for the RoomQuake unit included developing an understanding of earth science concepts such as the layers of the earth, plate tectonics, and convection currents, as well as more specific concepts related to the characteristics of earthquakes, such as how they are distributed and how they are measured. Because spatial thinking skills have been shown to contribute to successful learning in STEM, and more specifically geosciences, an additional goal was to investigate the impact of different versions of RoomQuake on learning as a function of spatial thinking skills.

Because learning about earthquakes is both highly spatial and abstract (in that earthquakes involve processes too large to be actually observed) it was hypothesized that students in the embedded condition would show greater learning gains than those in the non-embedded condition. This hypothesis was motivated by the idea that embodiment and contextual grounding, as provided in the embedded condition, can be especially beneficial for developing a mental model of complex spatial information and would provide the support needed for abstraction and transfer more generally. Further, it was hypothesized that the potential effects of spatial skills on learning about plate tectonics would be reduced in the Embedded Phenomena condition. With Embedded Phenomena, the need to mentally visualize and manipulate representations of the phenomena should be reduced because the enacted activity provides a basis or grounding for the construction of mental representations. Students in the Embedded Phenomena classroom with weaker spatial skills should show equal learning outcomes to students high in spatial skills. However, in the non-embedded condition, it was expected that spatial skill would predict learning. Specifically, it was hypothesized that students with higher spatial skills would outperform students with lower spatial skills, consistent with prior work showing the effects of spatial skills on learning this content.

The current study examined whether embedding scientific phenomena within the space of the classroom would increase students’ knowledge of earth science concepts, earthquake measurement, and seismological skills beyond that of a non-embedded control condition. Furthermore, the study also investigated the role of spatial skills in learning in an Embedded Phenomenon setting. Repeated measures analysis of variance (ANOVA) was used to determine the effect of embedding a condition on various learning and skill outcomes. In order to analyze the effects of spatial skills on learning outcomes in the embedded condition and in the non-embedded condition, linear regressions were conducted and interactions were followed up with tests of simple slopes.

A repeated measures ANOVA on earth science concepts revealed an overall increase from pre-test to post-test, F(1, 42) = 11.86, MSE = 1.73, p < .01. Although there was no main effect for embedding condition, F(1, 42) = 1.90, ns, a significant interaction was observed, F(1, 42) = 11.86, MSE = 1.73 p < .01, indicating that while the conditions did not differ at pre-test, students in the embedded condition performed better at post-test than students in the non-embedded condition.

The same pattern of effects was found with a second repeated measures ANOVA on the earthquake methods items. A significant increase was seen from pre-test to post-test, F(1, 42) = 41.30, MSE = 1.77, p < .001, no main effect was seen for embedding condition, F(1, 42) = 1.77, ns, and there was again a significant interaction indicating that while there were no differences at pre-test on the earthquake methods items, students in the embedded condition performed better at post-test than students in the non-embedded condition, F(1, 42) = 7.69, MSE = 1.77, p < .01. The patterns of means for all learning measures are shown in Fig. 5 by condition.

Next, a series of linear regressions were used to analyze the effect of the Embedded Phenomena condition and spatial skills on learning outcomes. First, a linear regression examining the effects of spatial skills, condition, and their interaction on earth science concept learning was conducted. The model significantly predicted performance on the earth science concept items at post-test, R ²  = .29, F(3, 41) = 5.14, MSE = 2.10, p < .01. There was a main effect for embedding condition such that students in the Embedded Phenomenon condition learned more than students in the non-embedded condition, β = .91, t = 2.64, p < .02). There was also a main effect of spatial skills indicating that spatial skills predicted learning of earth science concepts (β = .52, t = 2.48, p < .02), and there was a marginal interaction (β = –.62, t = 1.63, p < .11 see Fig. 6, left panel). Based on a priori predictions that the embedded condition would be especially beneficial for low spatial students, the marginal interaction was followed-up with a test of simple slopes. The tests of simple slopes revealed that students with higher spatial skills showed better performance in the non-embedded condition (β = .34, t = 2.48, p < .02), but not in the embedded condition (β = .06, t = .44, ns).

A second linear regression was conducted to examine the effects of spatial skills, condition, and their interaction on earthquake methods learning. The regression model including spatial skills, condition, and their interaction significantly predicted performance on the earthquake methods items at post-test, R ²  = .26, F(3, 41) = 4.33, MSE = 4.90, p < .01 (Fig. 6, right panel). Students in the embedded condition did better on the methods items than students in the non-embedded condition, β = .87, t = 2.48, p < .02. There was also a main effect of spatial skills indicating that spatial skills predicted learning of earthquake methods (β = .63, t = 2.94, p < .01), and a marginal interaction (β = –.77, t = 1.87, p < .07). Again, to address a priori predictions that embedding would be especially beneficial for low spatial students, the marginal interaction was followed-up with a test of simple slopes. The tests of simple slopes again revealed significant effects of spatial ability on performance in the embedded condition (β = .41, t = 2.94, p < .01), but not in the Embedded Phenomena condition (β = .08, t = .56, ns).

Finally, a third linear regression was conducted including spatial skills, embeddedness condition, and their interaction and it significantly predicted performance on the seismological skills assessment, R ²  = .46, F(3, 41) = 10.84, MSE = 1.94, p < .001 (Fig. 7). Students in the embedded condition performed better on the skills assessment (M = 6.46, SD = .99) than students in the non-embedded condition (M = 4.94, SD = 2.28), β = .39, t = 3.30, p < .01. Spatial skills also predicted learning of seismological skills (β = .80, t = 4.35, p < .001), and there was a significant interaction (β = –.46, t = 2.53, p < .02). Tests of simple slopes again revealed significant effects of spatial ability on performance in the non-embedded condition (β = .52, t = 4.35, p < .001), but not in the Embedded Phenomena condition (β = .14, t = 1.20, ns).

This experiment followed two classes of students across a 6-week unit in earth science to investigate the effects of the Embedded Phenomena approach. Analyses revealed that at pre-test, students in the embedded and non-embedded conditions were matched on their knowledge about the content of the unit. In addition, there was an overall significant increase on these items from pre-test to post-test suggesting that in both conditions students did learn the content. However, the gains in the Embedded Phenomena condition were significantly larger than the gains seen in the non-embedded condition.

Furthermore, this embedded enactive simulation of earthquake phenomena particularly improved the learning of students with weaker spatial skills. Results from this study demonstrated that spatial skill did constrain learning in the non-embedded conditions, but that the dependence on spatial skills can be reduced when an activity is embedded and enacted within classroom space. Specifically, in the Embedded Phenomena condition, students with low spatial skills were performing at a level equal to students with high spatial skills. This pattern of results suggests that grounding an activity in an embodied experience may lessen the demands of mentally representing the phenomena, which may be critical for supporting improved understanding of many scientific topics. Future work should examine whether certain items drove the improvement seen in the embedded condition and, in particular, whether test items that required mentally representing spatial attributes of the phenomena showed the strongest gains.

Two salient features that were intentionally varied between the embedded and non-embedded versions of the unit were the extent to which the activity was embedded and persistent within classroom space, and the extent to which it required students to enact the epicenter computations with their bodies. However, there was also another salient feature that differed between the two activity structures: whether the epicenter activities were done in a massed fashion (one after the other within a couple of class periods in the non-embedded condition) or distributed over time (in the embedded condition). While embedding the activity in classroom space and supporting it with enactment were both intended to support spatial understanding of earthquake phenomena, the decision to distribute earthquake events unexpectedly over time was intended to help students to understand the serendipitous, slow, and patient nature of real science. Because these two different classes of feature variations were not manipulated independently in the present study, this design might be conceived as representing only two conditions of a larger 2 × 2 design that would have fully dissociated the spatial embedding/embodying features of the activity from the temporal distribution of the activities. Given the limitations of the present design, it cannot be ruled out that the advantages of the embedded condition might really be due to the distributed nature of the epicenter activities. However, although it is possible that there was a benefit to spacing or distributing the activities over time on learning (Carpenter, Cepeda, Rohrer, Kang, & Pashler, 2012), it is not clear what theory would predict that manipulating this feature should facilitate the learning of individuals with weaker spatial skills. On the other hand, the interaction that was observed is predicted well by the assumption that the embedding and embodying features were facilitating learning by helping students to ground their learning experiences.

Similarly, it could be argued that the two activity structures also differed in the extent to which they were engaging or motivating for the students. From this perspective, the greater overall learning gains seen in the embedded condition could simply be due to the fact that students were more engaged and motivated to learn. Student responses from post-unit interviews were explored in an attempt to identify any differences in student engagement or enjoyment across conditions. Overall, students from both conditions reported liking or preferring whatever condition they had experienced and that the overall length of the unit was just right. When asked about how this unit compared to other science units they had done, almost all of the students in both conditions reported that this was more fun and helped them to learn more than traditional methods such as reading textbooks or watching videos. The two points where students in the embedded and non-embedded conditions seemed to differ was in how authentic they felt their experience to be. Students in the embedded classroom felt that their experience was similar to what seismologists do in real life: such that they experience earthquakes and try to locate them. On the other hand, students in the non-embedded condition did not feel that their experience was similar to what seismologists do in real life because they did not get to experience earthquakes. In addition, students in the non-embedded condition tended to describe their task as being one of doing research on a computer; no students reported this in the embedded condition. While it is possible that increased engagement or motivation could have had an impact on student learning overall (Hampden-Thompson & Bennett, 2013), and there is some evidence from the interviews that students in the embedded condition thought their experience was more authentic, it is again not clear that this would specifically facilitate the learning of individuals with weaker spatial skills.

One factor that makes this study unusual and interesting, but also inherently limited, is that it involved implementing an immersive technology framework within a real elementary school classroom context. Some of the limitations associated with being able to conduct this unit in authentic classrooms were that only a single experienced teacher was responsible for conducting the units and, although the teacher delivered all the content, the technology was heavily supported and controlled by the research team. This poses a potential limitation for broader dissemination because naïve teachers would require training to use it, and they might also need experience to use it as effectively as the teacher in the current study. However, the results were well aligned with theories of embodied cognition and previous research indicating that low spatial students can perform as well as high spatial students when given the appropriate supports (Sanchez & Wiley, 2010, 2014; Stieff, Dixon, Ryu, Kumi, & Hegarty, 2014). These results also align well with results from Jaeger et al. (2016) which showed that spatial skills are generally related to improved learning about another geoscience phenomenon (El Niño), but that providing an interleaved analogy can help support learning for low spatial students. Like the Embedded Phenomena simulation employed in the current study, an interleaved analogy can provide low spatial students with a familiar context on which to map a spatial mental model of a phenomenon, and this in turn reduces the need to rely on spatial skills for creating a mental model on their own. Another limitation worth noting is that the teacher was highly experienced with the unit and this experience may have been important for overall learning. However, the experience level of the teacher cannot account for the pattern of results showing that low spatial students learned better from the embedded condition.

More recent research using the Embedded Phenomena framework has been conducted with the goal of further exploring shared data displays for supporting spatial representations, as well as implementing new knowledge-construction activities, and making it easier to involve new teachers and train them to use the technology themselves (Moher et al., 2015). Specifically, newer iterations of RoomQuake focused on helping students to develop an understanding of the capabilities and limitations of seismographs, in particular, their dependence on the generation of multiple waves during an earthquake. This more recent iteration of RoomQuake introduces a knowledge-building progression in which a series of challenges is presented to students where they must determine the characteristics of earthquakes using a variety of tools including seismographs, tape measures, and stop watches. One major limitation of the version of RoomQuake used in the current study was that it required extensive teacher experience as well as support for the technology. In this newer iteration of RoomQuake, the classroom still serves as the experimental space, however, the teacher controls the simulation technology. Further, rather than students working in teams to report consensus readings from specific seismographs, the new version uses mobile devices where each student reports readings from multiple seismographs. As the readings are reported, they are depicted in an aggregate public display representing a map of the room showing the distance estimates from each seismograph, reinforcing the “intersection of circles” framing that is difficult to achieve at classroom scale because of the occluding walls.

In addition, the Embedded Phenomena framework has been extended to other science domains including ecology and engineering. In a recent application called Hunger Games, students are situated as rabbits foraging among a variety of food patches within the physical space of the classroom (Gnoli et al., 2014). The learning goals of this unit include the development of student understandings of foraging behaviors as a function of food resource depletion, competition, and predation. In another application called AquaRoom, students are situated as being in a town beneath which runs a network of aquifers. Children are challenged to recommend locations for new chemical plants with the goal being to minimize the impact of potential pollutants on the underground water supply.

Previous work with Embedded Phenomena simulations has shown that they can support development of domain understandings and authentic scientific practice, facilitate positive attitudes towards science and conducting experiments, and increase student agency in finding things out through experimentation rather than from a teacher (Malcolm, Moher, Bhatt, Uphoff, & López Silva, 2008; Moher, 2008; Moher et al., 2008; Novellis & Moher, 2011). The current research extends prior findings by showing that Embedded Phenomena activities may differentially improve learning and attenuate the effects of individual differences in spatial skills on learning in science, relative to non-embedded activities. As Gibson, 1979, (pg. 223) stated, “We must perceive in order to move, but we must also move in order to perceive.” The Embedded Phenomena framework may offer unique benefits for engaging students in authentic practices, improving learning outcomes, and making science learning accessible to students with a wide variety of skill sets.