Bodong Chen

Crisscross Landscapes

Notes: Sengupta. (2015). Learning to Deflect: Conceptual Change in Physics During Digital Game Play



Citekey: @Sengupta2015

Sengupta, P., Krinks, K. D., & Clark, D. B. (2015). Learning to Deflect: Conceptual Change in Physics During Digital Game Play. Journal of the Learning Sciences, 24(4), 638–674. doi:10.108010508406.2015.1082912



The current study presents a microgenetic analysis and case study of one student’s processes of knowledge construction as he played a conceptually integrated digital game (SURGE Next) designed to support learning about Newtonian mechanics. (p. 2)

Genetic epistemology—link the way knowledge is constructed with the way someone experiences related phenomena. Inheret. (p. 2)


In a conceptually integrated game, domain-specific learning goals are integrated with the mechanics and narrative of a game. Games designed in this way can allow students to build on intuitive understandings of complex physical phenomena (p. 2)

Each game level (i.e., mission in the game) involves specific challenges that are designed to engage learners in reasoning about key concepts in Newtonian mechanics. Players navigate through the game by placing impulses on Surge’s ship so that it reaches the desired target, often avoiding obstacles on its way. (p. 3)

Conceptual integration in SURGE Next can be understood more concretely in light of two key characteristics in the design of the game levels: (a) the representational format of dot traces in each level and (b) the sequencing of levels. (p. 3)

That is, an increase in Surge’s speed results in a greater gap between successive dots, whereas a decrease in speed results in dots placed closer to one another. (p. 3)

Along the second dimension, it is noteworthy that our pedagogical approach bears deep similarities with, and builds on, previous research about learning Newtonian mechanics using microworlds, in particular, the ThinkerTools microworlds-based learning environment (White, 1984, 1993). (p. 3)

In this sense, when interacting with microworlds, learners themselves are acting simultaneously as users and designers (Edwards, 1995; Hoyles et al., 2002). In ThinkerTools, the objective is for students to construct a series of increasingly sophisticated causal models for reasoning about how forces affect the motion of objects, in a sequence of progressively more complex microworlds. (p. 4)


Two prominent theoretical approaches have tried to account for mechanisms of conceptual change in humans—theory change (also known as the coherence view) and KiP (also known as the fragmentation view). (p. 4)

concepts are embedded in theories (i.e., cognitive structures that represent a range of phenomena and the causal principles that explain them; e.g., Carey, 1985, 1999; Carey & Spelke, 1994; C. Smith, Maclin, Grosslight, & Davis, 1997; Wiser, 1995). Whereas theory change can sometimes involve the gradual change in beliefs formulated in terms of the same concepts—Carey (1988) termed this kind of change weak restructuring—in other cases, concepts in successive theories may themselves differ, and this type of change is known as strong restructuring (Carey, 1988, 1992, 1999; Carey & Spelke, 1994). (p. 4)

A complementary perspective, the KiP perspective, frames conceptual change as a gradual and continuous process that relies on bootstrapping as opposed to discarding ideas that students bring in with them to the instructional setting (Clark, 2006; diSessa, 1988, 1993; diSessa & Sherin, 1998; Gupta, Elby, & Conlin, 2014; Gupta, Hammer, & Redish, 2010; Hammer, 1996; Jeppsson, Haglund, Amin, & Strömdahl, 2013; Sengupta & Wilensky, 2009, 2011; J. Smith et al., 1993). (p. 5)

diSessa (1993) postulated that the building blocks of sense of mechanism are phenomenological primitives (p-prims). P-prims are small knowledge elements developed from repeated abstractions of familiar events. Cued on recognition of contextual cues, p-prims are used to construct intuitive understandings of the physical world. diSessa (1993) argued that conceptual change occurs through a gradual development of coherence through the alteration of structured priorities (diSessa, 1993) in relation to relevant p-prims and other knowledge elements. (p. 5)

In terms of the analysis of learning, the coherence and fragmentation views of conceptual change thus entail starkly different bootstrapping accounts (Amin, 2009). Although both acknowledge that the process of conceptual change takes time, the coherence view treats conceptual change as a gestalt shift with a great deal of consistency attributed to both the naïve and expert knowledge structures. (p. 6)

Fragmentation views argue instead that naïve understanding is highly sensitive to context and that predictions and explanations depend in subtle ways on which particular knowledge elements happen to be triggered in particular situations. diSessa (1993) argued that according to the KiP perspective, conceptual change involves a gradual increase in coherence of understanding, and also suggested a cognitive mechanism through which coherence can emerge. Recent work by Chi, Roscoe, Slotta, Roy, and Chase (2012) has also shifted away from an incompatibility stance to a more continuous one, similar to diSessa (1993), especially in the domain of mechanics. (p. 6)

Chi and colleagues (2012) defined direct schemas as intuitive explanations that involve direct causation by an agent (typically in the form of local intentional interactions of the agent with one or a few other agents or entities) and argued that sequential processes can be explained by additively “summing” or “chaining” these local events (pp. 9–11). (p. 6)

Vosniadou’s (2013) perspectives are also evolving in a manner that can be interpreted as shifting away from an incompatibility stance toward a finer grained and organic elemental account of conceptual change (Clark & Linn, 2013). (p. 6)

To this end, coherence perspectives would provide rather low-resolution accounts— replacement of incorrect ideas with correct ones. The KiP perspective provides a comparatively more mechanistic and fine-grained account (diSessa, 1993; Hammer, 1996; J. Smith et al., 1993) that aligns with the evolving trends toward fine-grained and organic elemental accounts on conceptual change across research perspectives. (p. 7)


Students often posit that forces cause motion in the direction of the force independent of prior velocity (diSessa, 1988, 1993; Halloun & Hestenes, 1985; White, 1984). (p. 7)

diSessa (1988, 1993) has shown that this form of incorrect explanation is a result of the Force as Mover p-prim being cued in the learner’s mind. (p. 7)

As the application of this p-prim illustrates, the abstracted features in this case are object, push, and result. The feature that does not get abstracted from these situations is the previous velocity of the object in motion (diSessa, 1993). (p. 7)

diSessa (1988) argued that the development of a more expert understanding raises the priority of the competing Force as Deflector p-prim. (p. 8)

From the perspective of canonical physics, whereas Force as Mover neglects the role of the momentum of an object, Force as Deflector takes momentum into consideration. (p. 8)

However, it is challenging for researchers to identify p-prims based on verbal explanations (diSessa, 1993, 2007). (p. 8)

Therefore, although our primary method of data collection focused on semistructured clinical interviews conducted at the end of each game level, we also triangulated and corroborated our analysis of students’ verbal explanations during the interviews with screen-captured videos of the students’ actual game play during the level, changes in the students’ written explanations in preand posttests, and researcher field notes. (p. 9)


● What conceptual resources does Jamal use as he plays a digital game in the domain of physics, and how do these resources manifest in the game play? (p. 10)

● How does Jamal’s use of these resources evolve as he progresses through (p. 10)

METHODS (p. 10)

Because SURGE Next is a conceptually integrated game, learners’ game play is deeply tied to the underlying concepts in physics. This means that the canonical concepts of force, speed, acceleration, and momentum are leveraged in an intuitive and qualitative manner during students’ game play, in the form of both representational elements such as dot traces as well as conceptually salient actions such as placement of impulses. (p. 11)

Research Context and Case Study Approach The setting for this study was a 100% African American high-poverty public charter school located in a metropolitan school district in the southeastern United States. (p. 12)

All of the students were Title I students (i.e., they had been identified by the state educational body as failing, or being most at risk for failing, in science, math, and reading). (p. 12)

Using Taber’s criteria of typicality and representativeness for case selection, we present the case of a single student named Jamal (a pseudonym). Jamal’s case was selected after we analyzed, coded, and compared data for all of the students. Representativeness implies that the selected cases should aptly represent key aspects of the instructional process. Jamal’s case is representative because Jamal was present every day and interviewed frequently, so his case provides an authentic representation of all of the instructional activities. Typicality implies that Jamal’s reasoning should be similar to that of majority of the students in the classroom. (p. 12)

Data analysis consisted of transcriptions of interviews, development of open coding schemes, application of codes to data to identify patterns, triangulation with other data sources (i.e., video, tests, and field notes), and selection of written and verbal excerpts to represent the data. (p. 12)

Data Collection (p. 13)

The study lasted for five consecutive days during which the students played SURGE Next for 1.5 hr per day. Data collection used the microgenetic method (Siegler & Crowley, 1991) to study short-term conceptual change. The microgenetic method requires a high density of regular observations that span the entire duration of the learning activities and a qualitative analysis of the change (Kuhn, 1995; Kuhn, Schauble, & Garcia-Mila, 1992; Siegler & Crowley, 1991). (p. 13)

In our study, these regular observations took the form of semistructured clinical interviews. (p. 13)

Each interview ranged from 1 min to approximately 10 min, and each student was interviewed several times during each class. (p. 13)

Coding for P-Prims (p. 13)

We identified p-prims based on categorization of (a) students’ actions as recorded during videos of game play and (b) utterances during their interviews in which they explained their actions. In order to categorize their actions and utterances, we used diSessa’s schematic for identifying p-prims (diSessa, 1993, pp. 217–223). (p. 13)

These observations were mainly descriptive in nature and corresponded to what Miles and Huberman (1994) termed descriptive codes. (p. 14)

We then began open coding (Strauss & Corbin, 1990). During this phase of analysis, we carefully rewatched the interview videos and read the transcripts multiple times with the goal of generating analytic codes that Miles and Huberman termed pattern codes. (p. 14)

Descriptive codes in our study corresponded to discrete events that we identified in the Camtasia screen recordings and interview transcripts. (p. 14)

The pattern code here therefore corresponds to diSessa’s schematization of Ohm’s p-prim: increased effort or intensity of impetus leads to more result. (p. 15)

FINDINGS (p. 15)

Initial Misapplication of Force as Mover and First Appearance of Force as Deflector (p. 15)

In Level 1, Jamal began his game play using Force as Mover and completed the level without any difficulty. (p. 15)

In Level 2, students needed to maneuver Surge in two dimensions in order to reach the target. When Jamal encountered this level, he initially cued only the Force as Mover p-prim, which had been productive in Level 1 but which is not productive in Level 2. (p. 16)

On noticing Surge’s deflection (or “slant”), Jamal adopted an instrumental approach to refine his strategy. That is, instead of explicitly reasoning about how deflection emerged, he decided to use the deflection to design Surge’s trajectory. (p. 17)

Using this discovery, Jamal created a new trajectory for Surge, and his actions bear evidence that he used the Force as Deflector p-prim to do so. (p. 18)

Persistent Misapplication of Force as Mover (p. 18)

Despite his successful cuing and use of the Force as Deflector p-prim in Level 2, Jamal initially unproductively cued the Force as Mover p-prim in Level 3. (p. 18)

Use of Canceling to Correct the Misapplication of Force as Mover (p. 20)

After his unsuccessful application of Force as Mover in Level 3, Jamal revised his attempt using a downward impulse immediately before Surge encountered the right impulse. This revised strategy led to canceling the effects of the initial up impulse, thus enabling Surge to turn right by 90 degrees when it encountered the right impulse (see Figure 7).4 (p. 20)

It is important to note that Jamal’s choice of the magnitude of the impulse was incidental, or subconscious at best, in the sense that he did not explicitly reason about it. (p. 21)

In the subsequent level, Level 4, Jamal encountered a similar situation that required deflecting Surge by 90 degrees. He solved this level iteratively on his third attempt. In his initial attempt, it appears that Force as Mover was still being cued with a higher cuing priority than the Force as Deflector or Canceling pprims (p. 21)

Jamal had two 4 N impulses to move Surge to the right, however, and only one 4 N left impulse to the left. Jamal thus did not successfully cancel the rightward horizontal velocity, resulting in an unexpected diagonal deflection (see Figure 10e). (p. 23)

In contrast to Level 3, Jamal’s attempt here shows that (a) he had a canceling strategy in mind but (b) his strategy of using the default magnitude of the canceling impulse did not work. This in turn created a situation that necessitated explicitly taking into account Surge’s previous velocity—both its direction and its magnitude. (p. 23)

Toward this end, Jamal removed one of the 4 N right impulses, leaving a single 4 N right impulse to begin Surge’s motion (see Figure 11a) and the two impulses (4 N left and 4 N up) at the corner (Point S) as shown in Figures 11b and 11c. When the simulation was run, Surge made a 90-degree turn and successfully hit the target (see Figure 11d). (p. 24)

Following diSessa’s schematization (see Appendix), the situational elements salient in Jamal’s actions are the magnitude of the canceling impetus and the effect of the impetus (i.e., the horizontal speed of Surge). (p. 24)

Productive Stabilization of P-Prims for Interpreting Deflection (p. 25)

Is it fair to say that there was still no force as deflection p-prim in Jamal’s game play?
Wondering whether the game should have been designed to have grid to scaffold students’ thinking on deflection — because it is hard to construct an accurate deflection to hit the target. (p. 26)

Analytical Summary: Distributed Encoding and the Development of an Expert-Like Conceptualization of Deflection (p. 26)

Table 1 shows Jamal’s learning trajectory in a graphical form in terms of the p-prims he cued during his attempts on each level. (p. 26)

We posit that the development of reliability in the cuing of appropriate p-prims in these later levels can be explained in terms of what diSessa (1993) termed distributed encoding. (p. 26)

According to diSessa, the mechanism of conceptual change involves distributed encoding, a process in which learning to “see” (i.e., interpret) a phenomenon through canonical lenses (e.g., a physical law) involves “many intuitive contributors that each play some small role in ‘knowing the law’”(diSessa, 1993, p. 115). (p. 26)

In Jamal’s case, distributed encoding is evident in (a) Jamal’s learning to differentiate between situations involving perpendicular and diagonal deflections and (b) Jamal’s development of a progressively sophisticated sense of mechanism for dealing with perpendicular deflections. (p. 26)


Overall, the current study demonstrates the promise of designing conceptually integrated games around the KiP framework in terms of being able to foster, support, and investigate conceptual change in the domain of Newtonian mechanics. Specifically, this article makes two deeply intertwined contributions. The first contribution concerns the design of conceptually integrated games for learning Newtonian physics and beyond. The second contribution concerns analytical and methodological issues for investigating students’ learning as students interact with conceptually integrated games or games of other designs. (p. 29)