Embodied learning theory (Stolz, 2015; Smyrnaiou et al., 2016), as a pedagogical approach rooted in embodied cognitive science, seeks to expand the application of embodied cognition into education. Embodiment is experiencing and interacting with the world through our bodies, suggesting that mind and body are linked (Wilson, 2002; Kilteni et al., 2012; Smyrnaiou et al., 2016). Therefore, in contrast to traditional cognitive science, embodied cognition explains how body and environment are related to cognitive processes (Barsalou, 1999; Barsalou, 2008; Shapiro, 2007; Shapiro, 2014; Skulmowski and Rey, 2018). Embodiment is rooted in human perception and motor systems and through the body’s interaction with the world rather than only relying on abstract symbolic and internal representations (Barsalou, 1999; Wilson, 2002; Waller and Greenauer, 2007; Shapiro, 2007, Shapiro, 2014). In recent years, the design of embodied interfaces, including immersive experiences, has captured the attention of researchers in different fields in an attempt to improve embodied learning (e.g., Dalgarno and Lee, 2010; Johnson-Glenberg et al., 2014; Clifton et al., 2016; Yeonhee, 2018; Czerwinski et al., 2020). To conceptualize embodiment in the context of virtual reality, we should define how SOE is constructed based on embodied mental representations. SOE is a psychological response to being situated in the space in relation to other objects and the self. A virtual interface can be an extension of human senses linking the human to the virtual environment (Biocca, 1999; Kilteni et al., 2012). In other words, SOE in VR can be defined as the integration of our senses with our technology extended bodies (Biocca, 1999).

Among research studies focused on embodiment in VR, some have focused on defining different contributing factors to the embodiment. For instance, Kilteni et al. (2012) define the sense of embodiment as a result of the sense of self-location, the sense of agency, and the sense of body ownership. Some researchers (e.g., Gonzalez-Franco and Peck, 2018) focus on the role of the body as an avatar and its effect on the sense of body ownership and agency. In another example, Southgate (2020) conceptualizes embodiment in virtual learning from different angles focusing on various representations of the body such as cyborg body, naturalistic body, political body, etc. Furthermore, several research studies are focusing on the role of bodily engagement on SOE in VR (e.g., Johnson-Glenberg, 2018; Skulmowski and Rey, 2018; Johnson-Glenberg et al., 2020; Johnson-Glenberget al., 2021). Johnson-Glenberg et al. (2020) defined two affordances for designing VR for learning: 1) the sensation of presence, and 2) embodiment and the agency linked with manipulating objects in 3D. They define embodiment as a meaningful interaction with the learning content through bodily engagement. In another study by Johnson-Glenberg et al. (2016), they found that embodiment and sensorimotor feedback can increase knowledge retention in some types of knowledge. Johnson-Glenberg et al. (2021) compared passive learning (watching a video) vs. active learning through embodied interactions on a 2D platform and an immersive VR (Oculus Go). In all conditions, users sit. In the active learning scenario, using a mouse on a 2D desktop and controllers in an immersive VR platform is highly embodied. Watching a video on both platforms is considered low embodied. Therefore, the user has the same level of bodily engagement both in VR and a 2D desktop when assigned to active learning. They found a significant main effect for embodiment regardless of the platform. Participants in high embodied conditions learned the most. Zielasko and Riecke (2021) carry out a systematic analysis with VR experts in a workshop to find out the effect of body posture and embodied interactions on various VR experiences such as engagement, enjoyment, comfort, and accessibility. They also found higher embodied locomotion cues for walking rather than sitting. Among other research studies focusing on interaction techniques, locomotion, and embodiment (e.g., Zielasko et al., 2016; Weise et al., 2019; Di Luca et al., 2021), Lages and Bowman (2018) focused on the effect of manipulating objects vs physically walking in the virtual environment on performance in demanding visual tasks. They found that in designing the learning environments, the creator should consider the user controller experience, past gaming experience, and spatial ability of the user.

In a desktop VR, hands movement and a mouse or a keyboard simulate bodily engagement at a lower level, giving the user the sense of being situated in the virtual environment while sitting in front of a 2D interface. We consider this form of SOE as the lower level of bodily engagement than immersive VR, where the whole body can be moved and engaged. In this article, instead of comparing the degree of embodiment, we investigate the design choices for bodily engagement in two web-based desktop VR with the same level of embodiment. We posit that different design choices for interaction techniques would affect learning experience and performance. We hypothesize that various interaction techniques can generate different levels of agency over the learning materials and result in different learning outcomes in terms of knowledge gain. Two main interaction techniques with the learning contents introduced in the literature are 1) gesture, and 2) object manipulation (Paas and Sweller, 2012). Several studies have explored the role of gesture as an effective bodily engagement technique in learning spatial information and offloading mental tasks to the surrounding environment (e.g., Hostetter and Alibali, 2008; Lindgren and Johnson-Glenberg, 2013; Plummer et al., 2016; Johnson-Glenberg, 2018). We propose to add a third interaction technique, 3) to move the user in space. This interaction technique creates a sense of embodied locomotion and gives the user the ability to control the rotation of the viewpoint by either stepping back in x,y,z direction and seeing an overview of the 3D objects or moving closer to inspect the 3D objects in greater detail. We are interested in examining the role of object manipulation and moving the user in space as interaction techniques contributing to bodily engagement in enhancing learning and the associated sense of agency.

Spatial thinking is a fundamental part of many fields of science. One of the ways students can gain a better understanding of a spatial phenomenon is through visual-spatial thinking (Mathewson, 1999). Adequate visualization helps students to understand the spatial representation of information better. Spatial representations can be either extrinsic (e.g., locations) or intrinsic (e.g., shapes) to objects. One of the important spatial transformations related to intrinsic characteristics of objects is the ability to visualize penetrative views and to switch between two-dimensional and three-dimensional views. The ability to understand spatial relations inside an object and transform 3D data into a 2D profile is called penetrative thinking or cross-sectioning (Ormand et al., 2014; Newcombe and Shipley, 2015; Hannula, 2019). Figure 1 shows a penetrative thinking ability test to test students’ ability on mental slicing of a 3D geologic structure in a block diagram (Ormand et al., 2014).

In domains such as geosciences, students usually visualize the 3D structure of objects presented on 2D interfaces (e.g., desktop computers) and then extract 2D profiles from the 2D representation of the data. For instance, phenomena and observations related to plate tectonics are inherently three-dimensional, yet are often plotted on 2D maps. In introductory geoscience courses, students are often trained to visualize 3D data by learning how to read 2D maps and block diagrams. For instance, this method of representation makes it difficult for some students to visualize the depth, extent, and geometry of earthquakes as they have different levels of penetrative thinking abilities. A 3D representation of the data can aid in better understanding the extent, shape, and cross-sections of the data. As an example, Figure 2 shows the cross-section of earthquakes and volcanoes across South America. Drawing a cross-section based on a 3D visualization of data can be much easier than seeing the 2D representation of data, imagining the 3D visualization, and then extracting the 2D profile.

This research examines whether penetrative thinking as a topic in spatial learning can benefit from embodied learning. We incorporate embodied interactions with the 3D visualization of the data (earthquakes, volcanoes, and plate boundaries) to enhance students’ ability in visualizing penetrative views and better understand the cross-section or profile of the data in different regions around the world. To evaluate the role of bodily engagement through different interaction techniques introduced in Section 2.1, object manipulation and moving the user in space, in a penetrative thinking exercise, we compared the two design choices by providing two VLEs in the form of web-based desktop VR applications. These VLEs are designed to create an interactive environment to support penetrative thinking in an introductory physical geology course to facilitate visualization of the distribution and depth of earthquakes around the world. Full bodily engagement and a higher level of embodiment can be achieved in an immersive VR using Head-Mounted Displays (HMDs). In a web-based desktop VR application, a lower level of bodily engagement can be created through hand movements and the use of a device like a mouse or a keyboard.

In the first condition, where bodily engagement is induced through object manipulation, students do not actively move in the environment. They move and manipulate all the 3D objects together by dragging, rotating, or zooming in/out. This manipulation technique helps the students to get closer to a specific location along x,y,z direction, where they can observe a specific subduction zone. In this condition, students have complete control over manipulating all 3D objects at the same time. They can switch between different datasets but they cannot manipulate each object individually (i.e, individual earthquake locations or volcanoes). We refer to this visualization as the drag and scroll condition (Supplementary Video S1). This interaction technique is similar to what is experienced in conventional 3D editors or geoscience software programs such as ArcScene ⁴ .

In the second condition, where the bodily engagement is induced through moving the user in space and creating a sense of locomotion, students rotate the viewpoint to the desired direction (along x,y,z axes) and move farther and closer to the 3D objects to inspect their spatial arrangement and their associated information. In this condition, the user can move in space and change the direction of the viewpoint in the virtual environment in a natural way (similar to what is experienced in conventional first-person camera views in games). In this condition, we manipulate the position and rotation of the first-person camera in VR to create a sense of egocentric movement in space. The first-person camera manipulation is designed based on the rotation of the camera using the mouse for determining the direction of the viewpoint and the arrow keys on the keyboard to translate in that direction. we refer to this condition as the first-person condition (see the Supplementary Video S1). This type of interaction technique in a web-based desktop VR is the closest type of simulation that we could create to induce the sense of locomotion compared to physical walking in an immersive VR using HMDs. Based on these definitions, the main difference between these two interaction techniques is the design choice of moving the 3D objects or moving the user.

The data used to realize the visualizations in both conditions is the USGS Centennial Earthquake Catalog, which is a global catalog of well-located earthquakes from 1900 to 2008 that allows for the investigation of the depth and lateral extent of seismicity at plate boundaries (Coffin et al., 1997). To complement the earthquake locations and further connect the exercise to plate tectonics and plate boundary zones, maps of the current plate boundaries and the location of Holocene (i.e., < 10,000 years) volcanoes are also provided. Figure 5 shows the information provided in both conditions: 1) the three main plate boundary types; 2) horizontal scale in km; 3) the depth of the earthquakes: depth is less than 35 km; depth is between 35 and 70 km; depth is between 70 and 150 km; depth is between 150 and 350 km; depth is between 350 and 550 km; depth is between 550 and 720 km; 4) volcanoes: in subduction zones, in rift zones, and intraplate settings. The original format of the USGS Centennial Earthquake Catalog was a text file and for the Holocene Volcano, the original format was an Excel XML, both containing several values including X, Y, Z. The coordinates stored in the tables were imported into ArcGIS Pro ⁶ as XY point data using the XY Table to Point tool.

The shapefiles were imported into Blender (Community, 2018) using a Blender importer called BlenderGIS ⁷ . Then they were imported to Unity3D® ⁸ as FBX files. The earthquakes and volcanoes were visualized in the form of point clouds and were properly georeferenced. To overcome the performance limitation of rendering a large dataset (a total of 13,077 points for earthquakes) in VR, we used the particle system of Unity3D to generate points to have a more efficient and performant experience. Plate boundaries were visualized in the form of lines overlaid on the world map. Using these datasets, students can examine different subduction zone plate tectonics in terms of the locations and depths of the earthquakes.

The two different interaction techniques (one per condition) with the datasets were implemented in Unity3D. In both conditions, the users can switch between the earthquake and volcanoes datasets or enable both at the same time. Furthermore, they can access the label and other information of the data by opening showing/hiding a legend of the dataset. There is a scale bar next to the map to help users with the perception of distances. In the drag and scroll condition, the view of the users (i.e., the camera) orbits around a pivot point (starting at the center of the scene) using a common drag and movement functionality with the right mouse button, allowing the user to rotate the viewpoint. In addition, the pivot point can be moved within the 3D space of the scene along the X, Y, and Z axes using the drag and movement functionality with the left mouse button. Doing so would enable the users to move along these axes, and consequently orbit around the new pivot position. In the first-person condition, the users will use a combination of mouse and keyboard to perform a smooth translation along the X, Y, and Z axes using the WASD (or arrow) keys on the keyboard, while changing the direction of the movement based on the rotation of the camera using the mouse (i.e., steering which direction to move to with the mouse while the force is applied to that direction via the keyboard keys). The locomotion techniques in the conditions are very similar in nature (virtual travel and view point manipulation), but the two conditions are different in the mechanics of interaction used for locomotion. The drag and scroll condition simulates the interaction mechanics in software like ArcGIS, and the first-person condition simulates the interaction mechanics found in typical first-person shooter game.

236 students from two separate sections of an introductory physical geology course were invited to participate in this study in the Fall of 2020. The experience was embedded into the course as a lab assignment. Using a web-page, students selected whether they would like to take part in the research or only do the exercise as a lab assignment. From the 177 students who agreed to participate in the study, 96 students were randomly assigned to the drag and scroll condition and 81 students to the first-person condition. The section enrollment of participants was anonymized during the condition assignment to control for the environmental factors. All participants were compensated with extra course credit for their participation. 29.94% of the participants were female, 67.79% male, and less than 3% declared were non-binary or gender-nonconforming. The average age of the students was 19.45, with a maximum age of 21 and a standard deviation of 0.83. Also, 73.44% of the participants were majoring in Engineering.

To measure learning experience and knowledge gain, two types of questions were used in this study: 1) standardized measures, and 2) knowledge tests. Several existing standardized measures were incorporated into the pre-, and post-study questionnaires. Except for the demographic and background questions, all measures were of the type Likert-scale (ranging from 1 to 5 with 5 being the most positive), open-ended or multiple choice.

The pre-study questionnaire was comprised of the following measures:

• Demographics and background-related questions about gender, age, major and minor fields of study, and the year of study.

• A self-report measure of individual differences in terms of visual imagery: using the Visual Spatial Imagery (VSI) from MEC Spatial Presence questionnaire (Vorderer et al., 2004), with each item measured on a 1 to 5 Likert-scale. VSI is one of the spatial abilities that measures the ability to create clear spatial images and later access them from memory. People with higher VSI ability find it easier to access those spatial images from their memory (Wirth et al., 2007).

The post-study questionnaire was used to assess the learning experience of participants in light of the sense of presence and the sense of agency. Furthermore, the perceived learning experience of participants was measured.

• For measuring the sense of presence, we used the 6-item metric of Spatial Situation Model (SSM) from the MEC Spatial Presence Questionnaire (Vorderer et al., 2004). According to Wirth et al. (2007), a sense of presence can be built based on the Spatial Situation Model (SSM).

For the knowledge tests, a pre-study and a post-study test were designed. Besides, a test that measured the participants’ mental slicing and penetrative thinking ability was used:

• The pre-study knowledge test contained six multiple-choice questions that tested students’ pre-knowledge of subduction zones and plate boundaries before going through the main experience.

In both conditions, students filled out the pre-study questionnaire and then answered the pre-study knowledge test to establish their prior knowledge about the learning topic. Then, they were given information on the types of datasets they were going to explore in the VR experience and instructions on what areas to focus on. Figure 6 shows the area of interests including boxes 1–4 and cross-section A-B.

Region 1: South America

Students were asked to explore and pay attention to the distribution of the earthquakes and volcanoes, and the depth range of the earthquakes in these regions while reflecting on the following questions: What do you observe with respect to these different subduction zones? Are the geometries of the subducting oceanic lithosphere the same (i.e., the distribution and geometry of the earthquakes) or are they different? Now, look specifically at the western margin of the South American Plate (Region 1). Is the Wadati-Benioff zone (i.e., the zone of seismicity that defines the subducting plate) the same north to south along the margin? They were informed that after the experience, they will be asked to answer several questions about these regions and the cross-section. In both conditions, they were given 15 min to explore the datasets and memorize the distribution of earthquakes in the defined regions. A two-dimensional guide map on the lower right side of the screen showed the position and the direction of the user in the world map. A timer on the upper left side reminded them of the remaining time (Figure 7). In both conditions, students could hide/show legend and instructions.

After the experience, students first answered the post-study questionnaire, and then the penetrative thinking ability test. Finally, they answered the post-study knowledge test. Placing the post-study knowledge test at the end introduced a period between the experience and the post-study knowledge test. This way, we could test the effect of various embodied interactions on knowledge retention. The session, from start to end, took around 40 min.

For the learning experience assessment, we first identified the outliers using the Interquartile Range (IQR) method and carefully checked the dataset for removing any outliers. Then, we used Welch’s two-sample t-tests to compare the first-person condition with the drag and scroll condition based on the learning experience measures. For the learning performance measures, when Z-scores of the pre-, and post-study knowledge tests were compared regardless of the condition, Welch’s two-sample t-test was calculated. When we compared the post-study grades among the conditions, since the grades were ranked data, the Wilcoxon signed-rank test was used. To predict students’ sense of presence based on Visual Spatial Imagery and post-study grades based on their penetrative thinking ability, regression equations were calculated. As the number of participants in the two groups was different, Hedges’ g (Hedges and Olkin, 2014) was calculated instead of Cohen’s d for the calculation of effect size. A qualitative analysis of the two open-ended questions was performed to gain a better understanding of the participants’ opinions and experiences. Based on the approach proposed by Schreier (2012), two independent coders went over the responses of participants and inductively generated codes that would capture their content. Followed by consensus meetings, the codes were then grouped or rearranged into the final schema. Inter-rater reliability tests based on Cohen’s Kappa were then calculated for the finalized results.

Table 1 presents an overview of the mean, standard deviation, p-value, and effect size of the experience measures in the drag and scroll and the first-person conditions. As mentioned in the measures section, we measured the sense of presence, sense of agency, and perceived learning experience. There was no significant difference between the two conditions in terms of the sense of presence. Therefore, the first hypothesis (students in the first-person condition experience a higher sense of presence) is rejected. In terms of sense of agency, we measured possible actions, immediacy of control, perceived ease of use, and control and active learning, introduced in Section C. There is a significant group difference in the ease of use scores between the first-person (M = 3.14, SD = 0.63) and the drag and scroll (M = 3.32, SD = 0.50) conditions in favor of the drag and scroll condition, [t (153.12) = −1.98, p = 0.04]. The immediacy of control measures the students’ agency to change the view position and manipulate spatial objects. The difference for immediacy of control is very close to significant [t (174) = −1.77, p = 0.07] in favor of the drag and scroll condition (M = 4.07, SD = 0.88). We could not find any significant difference between the two conditions in terms of possible actions, and control and active learning. Based on these results, we have found some evidence in favor of the second hypothesis: students in the drag and scroll condition have higher control over the learning materials and as a result experience more agency. However, we could not find significant differences in all measures related to this affordance and as a result, we cannot conclude that the second hypothesis can entirely be accepted. In terms of perceived learning, students in the drag and scroll condition (M = 3.41, SD = 0.56) were significantly more satisfied [t (146) = 1.76, p = 0.04] than in the first-person condition (M = 3.20, SD = 0.75). We could not find any significant difference between the conditions in terms of reflective thinking and perceived learning effectiveness. Therefore, the only evidence that we could find in favor of the third hypothesis (students report a higher level of perceived learning in the drag and scroll condition) was satisfaction. Subsequently, we cannot conclude that the third hypothesis can be entirely accepted.

To conclude briefly, based on the discussed results, students in the drag and scroll condition had a better learning experience in terms of ease of use, immediacy of control, and satisfaction.

A simple linear regression was calculated to predict the effect of Visual Spatial Imagery (VSI) as a spatial ability on the sense of presence (SSM). Independent of the condition, a significant regression equation was found [F (1,175) = 53.04, p < 0.001] with an adjusted R² of 0.228. Students’ sense of presence has increased by 0.64 for each unit of VSI. Therefore, hypothesis 4 can be accepted: students with a higher level of VSI experience a higher sense of presence. Figure 8 shows that in both the drag and scroll and the first-person conditions, the level of presence is dependent on the VSI spatial ability. A significant regression equation was found for the first-person condition [F (1,79) = 28.64, p < 0.001] with an adjusted R² of 0.256. Students’ sense of presence has increased by 0.66 for each unit of VSI. For the drag and scroll condition, the significant regression equation is [F (1,94) = 22.83, p < 0.001] with an adjusted R² of 0.186. Students’ sense of presence has increased by 0.61 for each unit of VSI.

Before going through the experience, students answered six questions about subduction zones to test their knowledge of the subject in terms of the ability to understand the extent and geometry of the subduction zones based on their interpretation of the earthquakes, volcanoes, and plate boundaries. The total possible score was 8; The result of the test indicated an average score of 3.21 (SD = 1.24) with a minimum score of 1 and a maximum score of 7. 56.49% of the students who attended the study obtained a score that is less than the average score. This indicates that 56.49% of the students who attended the study had lower knowledge of the field compared to average performance. The post-study knowledge test contained seven questions with a total possible score of 14. The post-study knowledge test examined the same knowledge concepts with different types of questions to evaluate whether students’ understanding of the subject has improved after going through the experience. The result of the test indicated an average score of 7.9 (SD = 2.32) with a minimum score of 3 and a maximum score of 13.

Comparing the Z-scores of the pre-, and post-study knowledge tests, regardless of the condition, shows that students’ performance has improved by 0.05. However, the difference is not statistically significant: [t (176) = 0.55, p = 0.58]. We were under the impression that we can detect the presence or absence of students’ knowledge gain by studying the whole sample size. However, students with higher prior knowledge have a different level of improvement than students with lower knowledge of the field. Subsequently, we decided to analyze the learning performance of students with low prior knowledge of the subject compared to the average performance (pre-test Z-score ≤ 0). Based on our analysis, the performance of students with low prior knowledge of the field improved significantly regardless of the conditions: [t (167) = −5.86, p < 0.001]. For the drag and scroll condition [t (52) = −3.34, p < 0.001] and for the first-person condition, [t (46) = −5.41, p < 0.001]. Therefore, hypothesis 5 is accepted: both conditions have a significantly positive effect on students with low prior knowledge of the subject and the exposure to the VLEs improved their learning performance in terms of understanding earthquakes’ distribution and depth. In other words, when students with low prior knowledge of the field were exposed to the 3D representation of the epicenters of earthquakes from the USGS Centennial Earthquake catalog and locations of Holocene volcanoes, they understood the locations, depth, and geometry of the earthquakes in subduction zones better in different regions through 3D visualization. Yet, we could not find a significant difference between the conditions in terms of knowledge gain in students with low prior knowledge of the field: [t (97.2) = 0.94, p = 0.34].

We also analyzed the impact of the immediacy of control as one of the important measures of the sense of agency on students’ learning performance (post-study grades). The students’ post-study grades were dependent on their evaluation of the immediacy of control in both conditions. In both conditions, the higher a student felt to be in control, the higher their post-study grades were (Figure 9). In the drag and scroll condition, a significant non-linear regression equation was found [F (1,92) = 3.406, p = 0.02] with an adjusted R² of 0.07. In the first-person condition, a significant regression equation was found [F (1,77) = 3.007, p = 0.03] with an adjusted R² of 0.069. Although the adjusted R² for both equations are incredibly low and show that the immediacy of control is not a strong contributing factor, it is worth mentioning that there is a significant correlation. Respectively, hypothesis 6 is accepted: a higher level of control positively affects students’ learning performance.

To measure students’ penetrative thinking ability, students were asked to take the Geologic Block Cross-sectioning Test (GBCT) (Ormand et al., 2014). A simple linear regression was calculated to predict the result of the post-study grades based on the GBCT score (penetrative thinking ability). Independent of the condition, a significant regression equation was found [F (1,175) = 21.87, p < 0.001] with an adjusted R² of 0.106. Therefore, hypothesis 7 about learning performance is accepted: students with higher penetrative thinking ability show higher learning performance. This shows that for students who understand the spatial relations between the objects, this penetrative thinking ability enables them to understand the location, direction, and shape of earthquake events around the world better. In terms of students with lower penetrative thinking ability (hypothesis 8), there is a significant difference between the post-study knowledge grades of the first-person condition (M = 7.84, SD = 2.07) and the drag and scroll condition (M = 6.63, SD = 2.5) in favor of the first-person condition, [t (67) = 2.36, p = 0.02]. We can conclude that the first-person condition with the freedom of moving in space and inspecting earthquake locations by moving closer to the objects in a first-person view has a positive effect on students with a low penetrative thinking ability. Therefore, hypothesis 6 is accepted: students with lower penetrative thinking ability perform better in the first-person condition. Interestingly, in the drag and scroll condition, there is a significant difference between the pre-, and post-study grades (Z-scores) of students with a low penetrative thinking ability (mean of the differences = 0.45) [t (37) = 2.11, p = 0.04]. Although students with a low penetrative thinking ability in the drag and scroll condition had lower post-study grades compared to the first-person condition they had a significant improvement from their pre-study grades. This result indicates that even though the drag and scroll condition is not as effective as the first-person condition in terms of the knowledge gained in students with a low penetrative ability, it still is an effective medium and has improved students’ knowledge gain after being exposed to the VR experience.

Two open-ended questions were asked from the participants about their experiences as part of the post-study questionnaire:

Q1: If you could have changed something in the experience what would it have been and why?

Along with the quantitative analysis, the conducted qualitative analysis provides insights into the experiences of users after going through each condition. The extracted codes, capturing the content of the comments by participants, the percentage of participants talking about a code, and Cohen’s Kappa inter-rater reliability coefficient are reported in Table 2. Some of the codes are generally applicable to the experience regardless of the conditions and some are specific to the design choices based on the condition.

For the first question, the most frequent code was requiring more comprehensive information. Examples of this code include requesting an interactive legend (i.e., audio feedback), detailed description of features, and adding more features (i.e., mountains or continent names). In terms of accessibility to learning objectives in VR, before the experience, instructions and learning objectives were given to the students, including the highlighted areas to focus on and questions to have in mind while exploring the datasets. However, many students felt the need to see these learning objectives in the VR experience, being able to turn on and off the highlighted areas, and receiving more educational explanations of various subduction zones in the form of audio or text instead of self-exploration. Some students felt that there is no need for a change in the application whereas others mentioned difficulty in navigating in the space, negative learning experience, and difficulties running the app. In the first-person condition, some suggested different movement mechanics be designed to improve the experience. They suggested that instead of using the mouse as defining the direction of movement, two keys on the keyboard should allow for up and down movement. No one in the first-person condition complained that the experience was too long while three people in the drag and scroll condition complained about the length of the experience. Since the method of interaction in the first-person condition was new and students were not familiar with this method of movement in space, they might have used a considerable amount of time learning how to navigate in space and did not feel the time passing. Whereas the method of interaction in the drag and scroll condition is similar to geoscience software programs that many are familiar with.

In response to the second question, almost half of the students found this method of teaching superior to the classical methods of teaching. The advantages counted for this experience included learning at your own pace, being easier, and indicating that the 3D representation and interactive features improved their understanding of the concepts. 11.04% of the students in the drag and scroll condition declared that this method helped them to learn at their own pace, while only 0.02% in the first-person condition felt that way. As mentioned in the analysis of the first question, students in the first-person condition might have used a considerable amount of time learning how to navigate in space and that might have affected their learning pace. On the other hand, 19.7% of the students in the first-person condition found this type of experience to have no advantages over classical methods of teaching while only 0.09% in the drag and scroll condition felt that way. This indicates that although 40.7% of the students in the first-person condition found this method advantageous, 19.7% disagreed. One of the negative feedback about this method of teaching was the absence of active Q&A with the instructor while learning.

Overall, the insights from the first question show that students enrolled in the physical geology course are not used to memorization tasks, they typically would plot the locations and depths of the earthquakes by directly observing the data. In the exercise we designed, they first observed the data and then recalled the cross-sections based on memorization and memory trace. The second question gives insight that half of the students are open to technology-integrated teaching methods. Perhaps by improving their experience regarding the issues mentioned in the coding of the first question, more students might be open to this method of teaching.

Our quantitative evaluation of the learning experience utilized established self-reported measures. We were anticipating a significant difference in the sense of presence between the two conditions. Although the sense of presence is not significantly different among the two conditions, we found that students with higher Visual Spatial Imagery (VSI) ability experience a higher sense of presence in both conditions. In terms of perceived learning, we found that students are significantly more satisfied with the drag and scroll condition but we could not find any difference in other measures related to perceived learning. Concerning the sense of agency, students reported that the drag and scroll condition is significantly easier to use than the first-person condition. They also found the drag and scroll condition to have a higher level of immediacy of control compared to the first-person condition. This evaluation indicates that students are more comfortable and familiar with the interaction method data manipulation which is dragging, rotating, and zooming in/out of the 3D data. This made us curious to see if declaring the drag and scroll condition as an easier interaction technique would translate into superior knowledge gain as well. The results of the structured content analysis show that almost the same percentage of students in both conditions felt that the 3D representation and the method of interaction have improved their understanding of the subject.

Overall, all students gained some knowledge by going through the experience but we aimed to investigate the impact of interaction techniques on knowledge gain for students with low prior knowledge of the field. Our analysis showed that knowledge gain in students with low prior knowledge of the field improved significantly after going through the virtual experience in both conditions. We also found that when students felt more in control, in both conditions, they significantly performed better in terms of knowledge gain. This demonstrates that having control can be a contributing factor in knowledge gain. This shows that some students are more comfortable with moving in the three-dimensional environment and inspect objects based on changing their viewpoint whereas some students are more comfortable with data manipulation. With this result in mind, we looked into the penetrative thinking ability of the students to find out whether it would play a role in knowledge gain in different conditions. In the next section, we discuss our findings regarding students with lower penetrative thinking ability.

Weise et al. (2019) advise that the characteristics of the users should be considered in choosing an interaction technique. They suggest that users’ abilities can affect the performance and usability of the interaction technique. In this study, We used the Geologic Block Cross-sectioning Test (GBCT) to evaluate students’ penetrative thinking ability. We assessed whether this ability might affect their performance using either of the interaction techniques. Regardless of the conditions, we observed that the higher the penetrative thinking ability of the students, the higher the knowledge gain was. We hypothesized that students with higher spatial ability would better understand spatial relations of 3D objects and would perform better in either condition. One goal of designing interactive and embodied VLEs in 3D is to help students with lower spatial abilities, to help them visualize data in 3D, and better understand spatial relations between 3D objects. We found that students with a lower penetrative thinking ability benefited more from the interaction of the first-person condition. They had a significantly higher knowledge gain than students with a lower penetrative thinking ability in the drag and scroll condition. This result indicates that students with lower penetrative thinking ability benefit from active movement in space that facilitates adjusting their viewpoints. In other words, manipulating objects and trying to rotate them to get the desired viewpoint might be complex for students with lower penetrative thinking ability than naturally moving in space. Even though students with a lower penetrative thinking ability performed significantly higher in the first-person condition in terms of knowledge gain, students with a low penetrative thinking ability in the drag and scroll condition improved significantly compared to their pre-test Z-Score. This result suggests that even though the drag and scroll condition is not as ideal as the first-person condition for these students in terms of post-study knowledge gain, being exposed to a 3D representation of the data and interacting with the data would improve students’ penetrative view and result in a higher understanding the locations and depths of earthquakes when they have low penetrative thinking ability.