Toward a Neurobiological Basis for Understanding Learning in University Modeling Instruction Physics Courses

Abstract

Modeling Instruction (MI) for University Physics is a curricular and pedagogical approach to active learning in introductory physics. A basic tenet of science is that it is a model-driven endeavor that involves building models, then validating, deploying, and ultimately revising them in an iterative fashion. MI was developed to provide students a facsimile in the university classroom of this foundational scientific practice. As a curriculum, MI employs conceptual scientific models as the basis for the course content, and thus learning in a MI classroom involves students appropriating scientific models for their own use. Over the last 10 years, substantial evidence has accumulated supporting MI's efficacy, including gains in conceptual understanding, odds of success, attitudes toward learning, self-efficacy, and social networks centered around physics learning. However, we still do not fully understand the mechanisms of how students learn physics and develop mental models of physical phenomena. Herein, we explore the hypothesis that the MI curriculum and pedagogy promotes student engagement via conceptual model building. This emphasis on conceptual model building, in turn, leads to improved knowledge organization and problem solving abilities that manifest as quantifiable functional brain changes that can be assessed with functional magnetic resonance imaging (fMRI). We conducted a neuroeducation study wherein students completed a physics reasoning task while undergoing fMRI scanning before (pre) and after (post) completing a MI introductory physics course. Preliminary results indicated that performance of the physics reasoning task was linked with increased brain activity notably in lateral prefrontal and parietal cortices that previously have been associated with attention, working memory, and problem solving, and are collectively referred to as the central executive network. Critically, assessment of changes in brain activity during the physics reasoning task from pre- vs. post-instruction identified increased activity after the course notably in the posterior cingulate cortex (a brain region previously linked with episodic memory and self-referential thought) and in the frontal poles (regions linked with learning). These preliminary outcomes highlight brain regions linked with physics reasoning and, critically, suggest that brain activity during physics reasoning is modifiable by thoughtfully designed curriculum and pedagogy.

Full text

Toward a Neurobiological Basis for Understanding Learning in
University Modeling Instruction Physics Courses
Eric Brewe1,*,†,‡, Jessica E. Bartley2,†,‡, Michael C. Riedel2, Vashti Sawtelle3, Taylor Salo4,
Emily R. Boeving4, Elsa I. Bravo4, Rosalie Odean4, Alina Nazareth5, Katherine L.
Bottenhorn4, Robert W. Laird2, Matthew T. Sutherland4, Shannon M. Pruden4, and Angela
R. Laird2
1Department of Physics, School of Education, Drexel University, Philadelphia, PA, United States
2Department of Physics, Florida International University, Miami, FL, United States
3Lyman Briggs College, Department of Physics and Astronomy, Michigan State University,
Lansing, MI, United States
4Department of Psychology, Florida International University, Miami, FL, United States
5Department of Psychology, Temple University, Philadelphia, PA, United States
Abstract
Modeling Instruction (MI) for University Physics is a curricular and pedagogical approach to
active learning in introductory physics. A basic tenet of science is that it is a model-driven
endeavor that involves building models, then validating, deploying, and ultimately revising them in
an iterative fashion. MI was developed to provide students a facsimile in the university classroom
of this foundational scientific practice. As a curriculum, MI employs conceptual scientific models
as the basis for the course content, and thus learning in a MI classroom involves students
appropriating scientific models for their own use. Over the last 10 years, substantial evidence has
accumulated supporting MI’s efficacy, including gains in conceptual understanding, odds of
success, attitudes toward learning, self-efficacy, and social networks centered around physics
*Correspondence: Eric Brewe, eric.brewe@drexel.edu.
†These authors have contributed equally to this work.
‡co-first author.
AUTHOR CONTRIBUTIONS
EB, JB, and AL had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the
data analysis. Study concept and design: EB, JB, MR, RL, MS, SP, and AL. Acquisition, analysis, or interpretation of data: EB, JB,
MR, VS, TS, ERB, EIB, RO, AN, KB, RL, MS, SP, and AL. Drafting of the manuscript: EB and JB. Critical revision of the
manuscript for important intellectual content: EB, JB, MR, VS, TS, ERB, EIB, RO, AN, KB, RL, MS, SP, and AL. Obtained funding:
AL, EB, and SP. Administrative, technical, or material support: EB, RL, MS, SP, and AL. Study supervision: AL, MS, SP, MR, and
EB.
Conflict of Interest Statement: The authors declare that the research was conducted in the absence of any commercial or financial
relationships that could be construed as a potential conflict of interest.
ETHICS STATEMENT
This study was carried out in accordance with the recommendations of Florida International University’s Institutional Review Board
with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of
Helsinki. The protocol was approved by the FIU IRB.
SUPPLEMENTARY MATERIAL
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fict.2018.00010/
full#supplementary-material
HHS Public Access
Author manuscript
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Published in final edited form as:
Front ICT
. 2018 May ; 5: . doi:10.3389/fict.2018.00010.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

learning. However, we still do not fully understand the mechanisms of how students learn physics
and develop mental models of physical phenomena. Herein, we explore the hypothesis that the MI
curriculum and pedagogy promotes student engagement via conceptual model building. This
emphasis on conceptual model building, in turn, leads to improved knowledge organization and
problem solving abilities that manifest as quantifiable functional brain changes that can be
assessed with functional magnetic resonance imaging (fMRI). We conducted a neuroeducation
study wherein students completed a physics reasoning task while undergoing fMRI scanning
before (pre) and after (post) completing a MI introductory physics course. Preliminary results
indicated that performance of the physics reasoning task was linked with increased brain activity
notably in lateral prefrontal and parietal cortices that previously have been associated with
attention, working memory, and problem solving, and are collectively referred to as the central
executive network. Critically, assessment of changes in brain activity during the physics reasoning
task from pre- vs. post-instruction identified increased activity after the course notably in the
posterior cingulate cortex (a brain region previously linked with episodic memory and self-
referential thought) and in the frontal poles (regions linked with learning). These preliminary
outcomes highlight brain regions linked with physics reasoning and, critically, suggest that brain
activity during physics reasoning is modifiable by thoughtfully designed curriculum and
pedagogy.
Keywords
modeling instruction; physics reasoning; mental models; force concept inventory; fMRI; STEM
learning; brain network; neuroeducation
INTRODUCTION
Active learning is neither a curriculum nor a pedagogy. Active learning is a class of
pedagogies and curriculum materials that strive to more fully engage students and promote
critical thinking about course material. Students learn more effectively when they engage in
investigations, discussions, model building, problem solving, and other active explorations
(National Research Council, 2012; Kober, 2014). However, typical university instruction in
physics (and other Science, Technology, Engineering, and Mathematics [STEM] fields) has
been lecture-based. While lectures can be interesting, and some students clearly have been
trained to become engaged during lectures (Schwartz and Bransford, 1998), for the majority
of students, lectures are passive activities. This mismatch between the ways that students
learn and the way many classes are taught is the primary motivation for the transformation
of STEM instruction. When classrooms are transformed, the evidence is overwhelming;
students learn more and are more likely to succeed in active learning settings (Freeman et
al., 2014).
Multiple transformative curricula and pedagogical approaches have been developed for
introductory physics to promote active learning. For example,
Peer Instruction
emerged to
enhance standard lecture-based approaches by incorporating conceptual questions for
discussion and, in turn, facilitated development of personal response systems (Crouch and
Mazur, 2001).
Tutorials in Physics
were developed to supplement standard lectures through
Brewe et al. Page 2
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

use in recitation sections (McDermott and Shaffer, 2001). Other materials such as
Student
Centered Active Learning Environment with Upside-down Pedagogies
[SCALEUP]
(Beichner and Saul, 2003) and
Investigative Science Learning Environments
[ISLE] (Etkina
et al., 2006; Etkina and Van Heuvelen, 2007) implement a studio-format that integrates lab
and lecture, including greater amounts of conceptual reasoning and greater emphasis on
exploration. Modeling Instruction (MI) is an active learning approach (Brewe, 2008) similar
to SCALEUP and ISLE in that it is a complete course transformation integrating lab and
lecture components into one studio format class. However, MI is distinct from other reforms
in that it was built around an explicit epistemological theory of science, and this foundation
is one of the motivations for using functional magnetic resonance imaging (fMRI) to study
how learning physics may impact brain network development.
Hestenes (1987) avers that science by its very nature is a modeling endeavor. Science
proceeds by developing models that describe and ultimately predict phenomena. As a model
is developed, it is validated through the interplay between the predictions generated by the
model and the evidence that emerges supporting such predictions. Once a valid model has
been developed, the model is deployed to new situations. This is a process which Kuhn
(1970) called “normal science,” whereby scientists use existing prevalent models to explore
the models’ limits of applicability and search for places where the models give rise to
predictions in contrast with evidence. Ultimately, models reach their limits of applicability
and need to be revised or in some cases abandoned entirely, beginning what Kuhn called
“revolutionary science.” When this happens, a new model is proposed, and the cycle begins
anew.
The modeling theory of science is the theoretical and epistemological basis of MI. This,
however, is a theory of
science
, not a theory of
science instruction
. It translates to instruction
through the premise that, if modeling is how science proceeds and we believe students
should be engaged in authentic scientific practices, then instruction should be designed to
engage students in the process of modeling. Wells et al. (1995) describe the Modeling Cycle
as the recursive process of engaging students in model development, validation, deployment,
and revision.
In this paper, we first provide an overview of the theoretical background, development
process and critical features behind MI as a transformative curricula and model-building
endeavor. This overview serves to motivate why scientific model development in students
resulting from university instruction warrants further investigation not only at the academic
(e.g., grades) and social level (e.g., social networks) but also at the neurobiological level as a
putatively measurable phenomena that occurs within the brain. Then, we shift focus to
present results from a fMRI study in which we measured brain activity among students
engaged in physics reasoning and model use before and after they completed a MI course.
We subsequently discuss the results which show distinctive brain activity related to physics
reasoning and that instruction consistent with a Modeling theory of science modifies brain
activity from pre to post-course.
Brewe et al. Page 3
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Role of Conceptual Models in Introductory Physics Curriculum
Building instruction around modeling necessitates a working understanding of models. To
date, research in the MI context has focused on conceptual models, which are instructionally
useful, rather than mental models, which have been difficult to directly observe. Herein, we
seek to expand upon existing research by adopting neuroimaging techniques to interrogate
mental models among students receiving instruction via an explicit conceptual modeling
approach (i.e., MI). We operate from the following definition of a conceptual model:
conceptual models are purposeful coordinated sets of representations (e.g., graphs,
equations, diagrams, or written descriptions) of a particular class of phenomena that exist in
the shared social domain of discourse. This definition has several features worth elaborating.
First, it fits on a t-shirt. Second, this definition establishes the domain, purpose, and
composition of conceptual models, which we expand upon below. Finally, this definition of
conceptual models has helped us design research to look for evidence of the modeling
process in classrooms. Figure 1 illustrates the relationship between conceptual and mental
models.
Attempting to synthesize the many definitions and descriptions of models is not our purpose.
Instead, we aim to highlight some of the features of our definition that were relevant to the
development of the MI approach based on building, validating, deploying and revising
models. These features (i.e., the composition, purpose, and domain of conceptual models),
then will be used to structure the investigations into the nature of student’s mental model
formation as measured via brain-based fMRI data.
Composition—Conceptual models are composed of representations. Representations are
human inventions/constructs that stand in for the phenomena (Morgan and Morrison, 1999;
Giere, 2005; Frigg and Hartmann, 2006; Windschitl et al., 2008; Schwarz et al., 2009). In
physics, common types of representations include graphs, vector diagrams, equations,
simulations, words, and pictures (Krieger, 1987). From the MI perspective, this means that
instruction should focus on helping students to identify, use, and interpret representational
tools that are useful in describing physical systems. Instruction around model building
necessarily focuses on what representations are common to a discipline, how they are used,
and how information can be extracted from them. Further, the coordination of these
representations helps to build a more robust model, and provide a variety of ways to extract
information from the model (Hestenes, 1992; Halloun, 2004).
Purpose—Morgan and Morrison (1999) described mental models as mediators of thought,
autonomous from, but in correspondence with the system they represent. This mediating
function of models establishes the roles that models have within science as the center of
thought, explanation, and prediction (Craik, 1943; Johnson-Laird, 1983). For example, Craik
(1943) stated, “
If the organism carries a ‘small-scale model’ of external reality and of its
own possible actions within its head, it is able to try out various alternatives...
”
Instructionally, if models fill this role of mediators of thought, then models should structure
the organization of the curriculum. Models also allow students to address new phenomena
(Odenbaugh, 2005; Svoboda and Passmore, 2011; Gouvea and Passmore, 2017). This
purpose is built into the instructional modeling cycle where students are encouraged to
Brewe et al. Page 4
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

understand new phenomena by deploying existing models to extract information about and
characterize the phenomena. When existing models do not work, students
are expected to
adapt or redevelop models that can account for these new phenomena.
Domain—We propose a distinction between
scientific conceptual
and
mental model
domains and place conceptual models in the shared social domain of discourse. This
perspective differs from other conceptualizations where mental models within individuals’
minds/brains are implicitly or explicitly the center of focus (Greca and Moreira, 2000,
2001). Specifically, to infer the status of a student’s mental model, investigators typically
assess students’ actions or behaviors, such as writing, speaking, drawing, predicting, or
arguing (Halloun, 1996a; Justi and Gilbert, 2000; Lehrer and Schauble, 2006). Thus,
evidence of model-based reasoning exists external to the individual and is contingent on an
external evaluation. Instructionally, our efforts have been to help students develop models as
a distributed cognitive element. Meaning that each individual student will have an
instantiation of the shared model, but the visible elements of the model exist external to
individuals through writing, speaking, drawing, diagraming, predicting, and/or simulating.
This notion of shared models improves team performance and the learning process (Mathieu
et al., 2000). As such, the design of the MI curriculum and pedagogy focuses not on mental
models
per se
, but on the social construction of a model. In other words, we focus students
on using consistent representational tools to build models of phenomena in an interactive
team environment. Models are shared among class members and agreed upon before
deploying these models to analyze new situations. We provide a more detailed description of
the classroom setting in section “Features of MI Learning Environment” but much of class
time is spent in small groups developing models of specific phenomena on small portable
whiteboards, which are then presented at larger “board meetings.” The interplay between
smaller and larger groups provides a vehicle for students to use diagrams, equations, or
graphs to represent elements of the model.
We do not reject that individuals have internal mental models, or that these mental models
include connections between representations and concepts, or interactions between
mathematics and intuition, for example. As Rogoff (1990) points out, cognitive functions are
essential components of purposeful action. We are aligned with the notion that scientific
conceptual models are distributed cognitive elements, which are then appropriated by
individuals. During the appropriation, students construct the mental models in
correspondence with the scientific conceptual models. Rather our point is that assessing
external behaviors speaks to the conceptual model domain and assessing the mental model
domain would benefit from directly considering the brain.
Role of Conceptual Models in Instruction
For instructional purposes, models represent an appropriate and accessible level of
abstraction (Halloun, 2004). Within a larger context, models occupy the middle level of a
conceptual hierarchy (Table 1; Halloun, 2004; Matthews, 2007) which is best illustrated by a
representative example (Lakoff, 1987). Veterinarians are not likely to study the superordinate
category of animals, which is too broad a categorization to be useful. Nor are they likely to
study the subordinate category of retrievers; this is too specific to be broadly useful. Instead,
Brewe et al. Page 5
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

dogs are likely to be the level of focus. This level is referred to as the “basic” level and is
considered the ideal focus for instruction (Halloun, 2004).
In the MI classroom, building basic conceptual models begins with considering a specific
phenomenon to be described. Once a target phenomenon is established, the next step is to
characterize the phenomena through relevant representational tools. For example, using
velocity vs. time graphs to represent the motion of a moving object. As students create
representations of the object’s motion, a model of this specific phenomenon is being
developed, or what we call a
specific model
. These specific models are not generally
applicable, they pertain to the specific details of the situation being considered. By necessity,
specific models are predecessors to
basic models
. Specific models are made more robust as
additional representational tools are introduced and integrated with existing ones.
Introduction of representational tools and the subsequent negotiation of their use and
interpretation are motivated by specific phenomena to be modeled, so the models created are
always specific models.
However, a desirable scientific skill is to reason based on
general models
(Nersessian, 1995,
2002a,b). As such, the MI curriculum and pedagogy is specifically designed to facilitate the
students’ transition from specific to basic models. Basic models, which are general and
represent entire classes of phenomena (such as a constant acceleration model), are abstracted
from a collection of specific models (Halloun, 1996b, 2004). For example, the general
features of a basic constant acceleration model can be abstracted from specific models of
objects undergoing constant acceleration, such as objects in free fall, or uniformly slowing
down. This is achieved in the MI classroom by having students consider a number of
specific models, and then identifying the features that are similar to all such models. For
example, all constant acceleration models include a linear velocity time graph. These similar
features are then compiled into one model that can be used for all situations, a basic model.
Basic models are useful because they are not tied to a specific phenomenon, much like the
Standard Model is a basic model built up and abstracted from the specific models of atomic
collisions, particle interactions, etc. Basic models are essential in science as they promote
abstract reasoning about novel phenomena (Nersessian, 1995); when physicists seek to
understand interactions of atomic particles they start by using the Standard Model.
Once a basic model is established, students deploy the model in a variety of settings. This
deployment phase is most aligned with the standard problem solving that happens in physics
classes. The purpose is to develop skill at adapting the representations that make up the
model to new situations and extracting information about the situation from the
representations.
The final stage in the MI instructional cycle is revision. Revision of a basic model happens
when students encounter a phenomenon that does not fit with the model’s assumptions. An
example often encountered comes when students attempt to generate a specific model of
two-dimensional motion on the basis of a one-dimensional constant acceleration model. The
one-dimensional case is inadequate without modification to understand motion in two
dimensions, and thus must be revised. In some cases, revision involves a simple modification
Brewe et al. Page 6
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

of the representational tools, and in other cases, it requires starting with an entirely different
model.
In summary, the modeling cycle of MI describes the progression of course content. In
addition, MI also interweaves social interactions designed to facilitate discourse in the
service of building conceptual models. Next, we more fully describe the precise aspects of
the MI learning environment that support the development, validation, deployment, and
revision of models.
Features of MI Learning Environment
Basic conceptual models are often well-developed for scientists and course instructors, yet
these models are not well-developed for the students in introductory physics courses.
Accordingly, the first contextual feature of the MI classroom is to support students in re-
developing constituent basic models within their own learning environment. The MI
instructor’s role is thus to guide students through the development of these basic conceptual
models by establishing activities and providing scaffolding to manage student discourse and
promote model building and deployment. In this way, the MI curriculum and pedagogy can
be considered a guided inquiry approach. Students are not expected to discover physical
laws without strong instructor guidance who chooses activities, introduces representational
tools, and guides students toward their appropriate use and interpretation. In this way, the
instructor is a guide to the disciplinary norms and tools.
Student Participation in a Model-Centered Learning Environment—
Accomplishing this fundamental re-development of basic conceptual models requires
students to be active and engaged participants in the learning environment. Accordingly,
there are specific ways MI students are expected to participate in the re-development of
basic conceptual models. First, students are expected to be involved in identifying the way
that tools such as pictures, diagrams, graphs, and equations are used to represent
phenomena. They are not expected to invent or discover these tools, but instead to determine
with instructor guidance how these tools are used and how to interpret these representations.
For example, how does a vector representation of forces describe interactions the object is
involved in, and what do these forces allow us to infer about the current state of the object
and its future behavior? Second, students are expected to be involved in the interpretation of
these representational tools and drawing inferences from them as they pertain to physical
laws. Third, students are expected to then deploy these established basic conceptual models
by extending them to novel situations. Finally, students are expected to communicate basic
conceptual models. This promotes greater expertise with the models when presenting to
others and facilitates competence in scientific communication skills.
Studio Format—MI is designed for implementation in a studio-format classroom. In
studio physics classrooms students are able to flexibly engage in various types of activities,
which may include labs, conceptual reasoning, or problem-solving activities. At Florida
International University (FIU), the MI classroom integrates both the lecture and lab
components of the introductory physics course and meets for a total of 6 h per week across 3
days. Typically, students work in small groups of three to complete in-class activities. This
Brewe et al. Page 7
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

small group work is summarized on small portable whiteboards. These whiteboards are then
presented in larger group “board meetings” where all students in the class actively
participate.
Small Group Participation—During the small group component, students work on
model-building activities. In these groups, students begin the process of reaching consensus
by creating whiteboards for sharing or “publishing” their lab results and/or solutions to
problems. The instructor’s role is to circulate through the classroom, asking questions,
introducing new content, and examining the whiteboards that are being prepared. This small
group work allows students to work together on a model-building activity, generate
conceptual models, and practice communicating scientific information in a relatively “low-
stakes” setting.
Large Group Participation: The “Board Meeting”—The practice of having students
first work in small groups and then present their outcomes to a larger group provides
students with multiple opportunities to negotiate the use of conceptual models. The board
meetings involve all students in the class gathering in a circle such that every member can
see every other member and every groups’ boards. During the board meeting, the instructor
assumes the role of disciplinary expert and guides the discourse toward a shared conceptual
model. Facilitating the discussion involves moderating the groups’ whiteboard presentations,
addressing student questions, and helping groups clarify their presentations and
understanding. The instructor’s guidance during the board meetings relies heavily on
providing student groups with formative feedback. The explicit goal of these board meetings
is to reach consensus regarding the conceptual models. In addition to the explicit goals, tacit
goals include establishing the norms of a discourse community and encouraging students to
utilize scientific argumentation strategies (Passmore and Svoboda, 2012). These strategies
include supporting claims with evidence and reasoning based on the shared conceptual
models.
Pairing Large and Small Group Interactions—The combined interaction structure is
designed to elicit target conceptual models. The structure of these interactions also mimics
the structure of science in general and physics in particular as practiced in a research setting.
Students work in small research groups, building up and synthesizing the conceptual model
that is subsequently ‘published’ at the board meeting, much like a scientific meeting. Both
the small and large group settings rely on the pedagogical skill of the instructor. In MI-like
environments (which are less “instructor-centered” than traditional classrooms), the
trajectory of the learning takes varied paths based on the input of the participants. For this
reason, the curriculum and pedagogy of MI are less like a script for an actor to follow, and
more like a set of guidelines for an improvisational comedienne.
Impact on Student Outcomes—The combination of curriculum materials designed to
recursively implement the modeling cycle and a learning environment and pedagogy that are
similarly supportive have been shown to be effective at promoting learning. Like other
transformed curricula in university physics, MI promotes both conceptual understanding and
student success in introductory physics (Brewe et al., 2010b). A survival analysis suggests
Brewe et al. Page 8
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

that the increased success rate in introductory physics is not a result of lowered standards, as
students from MI classes showed equivalent likelihood of success in completing a major in
physics as students from lecture classes (Rodriguez et al., 2016). MI students also report
improved attitudes about learning physics (Brewe et al., 2009, 2013) and these attitudinal
shifts are equitable in terms of ethnicity (Traxler and Brewe, 2015). The group interactions
in a MI class promote more well-developed classroom networks (Brewe et al., 2010a), and
these networks are known to facilitate retention in physics courses (Zwolak et al., 2017).
Positive shifts in self-efficacy associated with participating in MI have been documented,
(Sawtelle et al., 2010) although not consistently (Dou et al., 2016). We are in the process of
studying qualitatively the construction of a conceptual model in MI (Brewe and Sawtelle,
2018) and investigating students’ representational choices in problem solving (McPadden
and Brewe, 2017). These studies are consistent with students constructing and using
conceptual models to solve problems and analyze physical systems. The successes coming
from the MI classroom motivate our current research into the neurobiological mechanisms
of reasoning in physics.
Investigating Mental Model Development Using Neuroimaging
While prior assessments of MI’s impact on students has typically focused on the social
construction of conceptual models (Brewe, 2008, 2011; Sawtelle et al., 2012), here we
consider MI’s potential impact on mental models using brain imaging techniques. This study
aimed to investigate brain activation during a physics reasoning task and changes in brain
activation after MI course instruction relative to before such instruction. Previous
neuroimaging studies have localized brain activity associated with reasoning across various
modalities (e.g., mathematics, formal logic, and fluid reasoning; Prabhakaran et al., 1997;
Arsalidou and Taylor, 2011; Prado et al., 2011), but no investigations have probed for such
brain activity in the field of physics or across physics classroom instruction. Because of this,
no standardized tasks have been adapted for the MRI environment to examine such brain
activation. Therefore, as a first step, we sought to develop a novel neuroimaging paradigm to
probe brain activity during physics reasoning. We focused the development of this task on
mental model use during physics reasoning, as previous research has provided evidence that
students’ use a variety of mental models during conceptual physics reasoning (Nersessian,
1999; Hegarty, 2004). Thus, we adapted items from the well-known
Force Concept
Inventory
(FCI; Hestenes et al., 1992) which is known to engage conceptual physics
reasoning. FCI questions were modified to fit with the parameters of the MRI data
collection, and to investigate physics reasoning, (see section “Physics Reasoning Task” for
further details. Simultaneously, to facilitate formation of neuroanatomical hypotheses
regarding the brain networks we might observe during physics reasoning, we conducted a
neuroimaging meta-analysis (Bartley et al., in press) of fMRI studies that investigated
problem solving across a diversity of representation modalities. Briefly, the primary outcome
of that meta-analysis was that similar reasoning tasks using mathematical, verbal, and
visuospatial stimuli involving attention, working memory, and cognitive control, activated
dorsolateral prefrontal and parietal regions. Participants completed this physics reasoning
task while undergoing functional magnetic resonance imaging (fMRI) scanning, both before
(pre) and after (post) completing a physics course in order to investigate the putative impact
of physics instruction on brain function. Driving this neuroeducation project were two main
Brewe et al. Page 9
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

hypotheses: (1) This novel physics reasoning task would induce increased activity in brain
regions previously associated with attention, working memory, and problem solving (e.g.,
lateral prefrontal and parietal regions), and (2) Activation patterns would differ from pre- to
post-course, indicating that brain activity can be modified as a result of physics instruction.
A few prior studies have demonstrated that short- and long-term course instruction can
impact brain function. Differences in brain function have been observed from pre- to post-
course among students enrolled in a 90-day Law School Admission Test preparation course
(Mackey et al., 2013). Mason and Just (2015) showed that providing information to research
participants about mechanical systems while in the MRI scanner, which they called physics
instruction, led to changes in knowledge representation during successive stages of learning.
In a separate study, they were also able to use machine learning and factor analysis to
identify neural representations of four physics concepts: motion visualization, periodicity,
algebraic forms, and energy flow (Mason and Just, 2016). However, to our knowledge, this
is the first neuroeducational study to consider the impact of a full, semester-long physics
class on the brain.
Brief Primer on Neuroimaging Studies—This manuscript is intended for an
educational research audience, with the expectation that readers have not had extensive
experience with neuroimaging as a research methodology. As such, this section provides a
brief overview of neuroimaging studies, particularly fMRI. In neuroimaging studies,
researchers develop an experimental task to isolate mental operations of interest that
participants perform lying in a MRI scanner while a series of three-dimensional brain
images are acquired. Typically, these brain images are acquired approximately every 2 s and
are composed of small volume elements called voxels, which in this study measured 3.4
mm3. Within each voxel, the blood’s changing oxygen levels (known as the blood-
oxygenation level-dependent [BOLD] signal) are measured. Task-related changes in the
BOLD signal provide an indirect measure of brain activity. In one implementation of fMRI
experimental design, brain images are collected in blocks. During ‘active task’ blocks,
participants are presented a stimulus (e.g., a physics question) engendering cognitive
processes of interest (e.g., physics reasoning) and are instructed to make a response using a
MRI-compatible keypad. During carefully constructed ‘control task’ blocks, participants are
also presented with stimuli and give responses; however, the stimuli presented do not
engender the cognitive processes of interest. Contrasting active blocks with control blocks
presumably isolates task-related brain activity associated with the cognitive processes of
interest and excluding those common to both conditions (e.g., visual processing, word
reading, button pressing).
Following data collection, fMRI data are processed to correct for in-scanner head movement
and fitted to a standardized brain template to enable averaging over a group of participants.
BOLD time series from each voxel are input into a general linear model (GLM) including
distinct regressors for various task events (and other known sources of noise) to characterize
the degree to which variability in the BOLD signal correlates with those task events.
Resulting beta weights from active and control task blocks can then be contrasted and
significant differences are interpreted as differences in brain activity between blocks. This
procedure is repeated for the BOLD time series across all voxels in the entire brain.
Brewe et al. Page 10
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Additional multi-level modeling can be performed on these results, as was done in this
study, to test for changes in brain activity across repeated measures (i.e., from pre- to post-
instruction).
METHODS
Participants
Participants were drawn from MI classes at FIU over the course of 3 years (academic years
2014–2017). We recruited 55 students (33 male, and 22 female) in the age range of 18–25
years old (mean ± SD: 20.1 ± 1.4). All participants were screened to be right-handed, not
using psychotropic medications, and free of psychiatric conditions, cognitive or neurological
impairments, and MRI contraindications. Volunteers invited to participate had not previously
taken a college physics course and met either a GPA (>2.24) or SAT Math (>500) inclusion
criteria. These criteria were implemented to minimize between-participant variability that
could confound brain measurements associated with the experimental conditions. Written
informed consent to a protocol approved by the FIU Institutional Review Board was
obtained from all participants. Imaging data were collected on a General Electric 3-Tesla
Healthcare Discovery 750 W MRI scanner located in the Neuroimaging Suite (NIS) of the
Department of Psychology at the University of Miami (Coral Gables, FL). Each participant
completed a 90-min MRI scanning session at both a pre- and post-instruction time point.
The presession scans were scheduled within the first 4 weeks of the semester and the post-
session scans were completed in the first 2 weeks following the semester. All participants
were compensated for their time participating in the MRI assessment ($50 for pre- and $100
for post-scans).
Physics Reasoning Task
We adapted a set of questions from the Force Concept Inventory (FCI) for presentation in
the MRI scanner (Figure 2A). The FCI was chosen given the substantial amount of extant
data from students in MI at FIU on this measure (Brewe et al., 2010b), established reliability
measures (Lasry et al., 2011), and known time requirements (Lasry et al., 2013). The FCI is
a 30 question, multiple choice conceptual survey of students understanding of Newtonian
mechanics (Hestenes et al., 1992). Each question has five multiple choice options, one
correct and four distractors which were originally generated from student responses to open-
ended versions of the same questions. The questions present “every-day scenarios,” do not
require any mathematical calculations, and are presented as text describing the scenario
accompanied by a representational diagram. To ensure that MRI data collection sessions
were manageable and well-tolerated by participants, we reduced the number of FCI
questions from 30 to nine (FCI 2, 3, 6, 7, 12, 14, 27, and 29). These nine questions were
selected to span a range of difficulty levels that were simultaneously challenging enough to
tax the mental resources of participants, but not necessarily the most difficult items in the
FCI, as determined by item response curves in Morris et al. (2012) (Table 2). Additionally,
because measurement of brain networks via fMRI require the repeated observations across
multiple yet similar experimental trials, we sought to narrow the broad range of physics-
related cognition being probed in this task and selected questions that required students to
determine the trajectories and motion of objects as resulting from different scenarios and
Brewe et al. Page 11
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

combinations of initial velocities and/or force configurations. Given technical constraints
associated with the use of a four-button MRI-compatible keypad, the questions were
modified by removing the least chosen of the five multiple choice options, as indicated by
the item response curves of Morris et al. (2012). In the current neuroimaging task
implementation, each question was parsed into three self-paced presentation phases;
participants were allowed to control the timing of these phases. The first phase of the
question involved presentation of the text describing the phenomena and an accompanying
diagram. The second phase posed the question, and the third phase presented the multi-
choice answer options. FCI responses were assessed for overall and item-specific accuracy.
In addition to FCI questions, participants answered a series of “control questions” (Figure
2B), each of which had similar characteristics to the FCI questions in terms of reading
requirements, visual complexity, and overall design. However, control questions did not
inquire about physics-related content, instead these questions focused on reading
comprehension and shape discrimination. Control questions allowed us to isolate cognitive
processes presumably related to physics reasoning when contrasting FCI (“active task”) vs.
control questions (“control task”).
FCI and control questions were presented in pseudo-random orders within three task runs.
Each question was followed by 20 s of “rest,” during which participants maintained their
gaze on a fixation cross centrally projected on the screen. These three runs lasted
approximately 6 min each. Participants received instruction and practice on the task in a
carefully managed mock scanner training session to ensure correct performance during the
MRI session. In addition to acquainting participants to the task, the mock scanner also
allows participants to experience what the actual MRI scan will be like.
Data Analysis
Details on fMRI data acquisition parameters can be found in the Supplementary Materials.
Prior to analysis, the data were preprocessed using commonly used neuroimaging analysis
software packages: FSL (FMRIB Software Library, www.fmrib.ox.ac.uk/fsl) and AFNI
(Analysis of Functional NeuroImages, http://afni.nimh.nih.gov/afni). Standard fMRI
preprocessing procedures involved motion correction to remove signal artifacts associated
with head motion, high-pass filtering to remove low frequency trends in the signal associated
with non-brain noise sources (i.e., cardiac or respiratory), and spatial smoothing to increase
signal to noise ratio during analysis. The data were then mapped to a standardized brain atlas
(MNI152) to allow for group-level assessments.
We conducted two primary analyses to identify: (1) brain regions linked with physics
reasoning (task effect) and (2) changes in brain activity associated with physics instruction
(instruction effect). To delineate brain regions linked with physics reasoning at the pre-
instruction time point, each preprocessed fMRI data set was input into a voxel-level General
Linear Model (GLM) including regressors for the FCI and control task conditions (and
various nuisance signals). Contrast images were created for each participant by subtracting
the beta weights associated with the control questions from those for the FCI questions
representing the degree to which each voxel responded more during physics reasoning as
compared to the control condition (FCI > Control). These participant-level contrast images
Brewe et al. Page 12
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

were then input into a group-level, one-sample
t
-test and significant physics reasoning-
related brain activations were defined using a threshold of
Pcorrected
< 0.05 (
Pvoxel
−
level
<
0.001, family-wise error [FWE] cluster correction). To delineate brain regions showing
physics reasoning-related activation changes following a MI course, the participant-level
FCI > Control task contrast images (described above) from the pre- and post-instruction data
collection sessions were input into a group-level, paired samples
t
-test. Both Pre > Post and
Post > Pre contrasts were computed and significant instruction-related brain activity changes
were defined using a
Pcorrected
< 0.05 threshold (
Pvoxel
−
level
< 0.001, FWE cluster
correction). Follow up correlational analyses were also conducted between the BOLD signal
change across instruction (Post > Pre) in the four largest significant clusters (≥1,000 voxels)
identified in the instruction effect analysis described above and accuracy post-instruction on
the FCI using
P
< 0.0125, Bonferroni corrected,. Because the clusters probed showed
significant extent across multiple brain areas, BOLD signal was extracted from spherical
seeds centered at the peaks z-score of each cluster.
RESULTS
Accuracy
Table 2 includes the accuracy results of student responses for the nine questions in the pre
and post-instruction scans along with item difficulties based in classical test theory, Morris
et al. (2006). A paired-samples
t
-test was conducted to compare post-vs. pre-instruction
means. Cohen’s d, was calculated to identify the magnitude of the effect, and 95%
confidence intervals on the effect. The results of the
t
-test [
t
(55) = 6.31,
p
< 0.001] and
Cohen’s d (
d
= 0.84) with a 95% confidence interval of 0.45–1.23 indicated with a high
degree of confidence that response accuracy increased after instruction. These results are
consistent with prior results examining increased FCI accuracy after course instruction
(Brewe et al., 2010b). Furthermore, these accuracy results from participants in the scanner
are in line with the classical test theory item difficulty (outside the scanner performance),
where difficulty is calculated as the average score on a particular item.
Task Effect
MI students exhibited physics reasoning-related brain activity (FCI > Control) at the pre-
instruction time point in four general brain areas, the prefrontal cortex, the parietal cortex,
the temporal lobes, and the right cerebellum (Figure 3, red; Supplemental Table 1). More
specifically, in the prefrontal cortex (PFC), activation peaks were observed in the left
superior frontal gyrus (SFG), dorsomedial PFC (dmPFC), bilateral dorsolateral PFC
(dlPFC), inferior frontal gyri (IFG), and orbitofrontal cortex (OFC). Within the posterior
parietal cortex, brain activity was observed bilaterally in the supramarginal gyri, intraparietal
sulcus (IPS), and angular gryi (AG). Large bilateral clusters of activation during physics
reasoning were also observed in middle temporal (MT) and medial superior temporal (MST)
areas. These same patterns of task-related brain activity from the pre-instruction stage were
also observed when performing a similar assessment at the post-instruction stage (data not
shown).
Brewe et al. Page 13
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Instruction Effect
Significant increases in brain activity following instruction (Post > Pre) were observed
within prefrontal and parietal cortices (Figure 3, blue; Supplemental Table 2). In particular,
three clusters of increased PFC activity were identified in the left dlPFC along the inferior
precentral sulcus, and bilaterally in the frontal poles. Parietal areas demonstrating increased
activation after instruction were located in the posterior cingulate cortex (PCC) extending
into retrosplenial cortex and the precuneus and in the left angular gyrus. No brain regions
showed significantly more task-related activity at the pre-instruction stage as compared to
post-instruction (Pre > Post). Follow up correlation analysis between the left PCC, left
angular gyrus, left orbital frontal pole, and left DLPFC and accuracy on the FCI yielded no
significant correlation (r
pcc
= −0.12,
pcorrected
= 1; rag = −0.07,
pcorrected
= 1; r
ofc
= −0.01,
pcorrected
= 1; r
dlpfc
= 0.02,
pcorrected
= 1).
DISCUSSION
This neuroeducational study represents an initial effort to understand how physics reasoning
may translate to the level of brain function assessed by fMRI and how instruction brings
about changes in brain activity. To this end, we have provided fMRI results of brain
activation from two main assessments. First, we observed that the physics reasoning task
(FCI > Control questions) was associated with increased brain activity notably in lateral
prefrontal and parietal regions. Second, we observed that students who completed the MI
course showed increased activation during the physics reasoning task after the course in the
posterior cingulate cortex and frontal pole regions.
Accuracy and Physics Reasoning
Participant responses to the FCI questions in the scanner show accuracy that is in line with
published item difficulties and post course improvement in accuracy are consistent with
Brewe et al. (2010b). This suggests that the MRI version of the task we developed is
prompting physics reasoning that is consistent with that observed out of scanner
environment. Effect sizes from preto post-instruction indicate similar performance on this
task with modified FCI questions as on the full FCI. This improvement is indicative of a
shift in physics reasoning as a result of instruction. We do not interpret these changes as
recall effects for two reasons, the results of the FCI were not discussed with students, and
the task itself was not identified as being derived from the FCI. Further, Henderson (2002)
has shown that recall effects over the duration of a full semester are minimal. While
accuracy is important for characterizing and to some degree validating the task that was
developed for the fMRI environment, we did not expect accuracy to correlate with brain
activity. Instead, physics reasoning, regardless of accuracy, is linked to brain activity.
Task Effect: Brain Activity Linked With Physics Reasoning
Our initial analysis identified brain activity among college students associated with physics
reasoning (FCI > Control) in lateral prefrontal and parietal regions. One interpretation is that
activity in these regions supports cognitive processes critical for answering physics
reasoning problems such as attention, working memory, spatial reasoning, and mathematical
cognition. More specifically, the lateral PFC’s role in executive functions such as working
Brewe et al. Page 14
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

memory and planning are well-characterized (Bressler and Menon, 2010) and these areas are
important in manipulating representations in working memory and reasoning
(AndrewsHanna, 2012; Barbey et al., 2013). Lateral parietal regions are involved in motor
functioning as well as spatial reasoning, mathematical cognition, and attention (Wendelken,
2015). Such an interpretation is reasonable in the context of the current task which likely
involves generating mental simulations and representations in the service of identifying the
correct answer choice. From a large-scale brain network perspective, the brain regions
showing physics reasoning-related activation resemble one commonly observed functional
brain network known as the central executive network (CEN). The CEN, consisting of lateral
prefrontal and parietal regions (Bressler and Menon, 2010), is generally associated with
externally oriented attentional and executive processes (e.g., working memory, response
selection, and inhibition; Cole and Schneider, 2007; Seeley et al., 2007).
The task-related brain regions we observed were generally similar when separately
considering data collected during the pre- and post-instruction scans. While speaking to the
consistency of such brain activity, this analysis is not intended to determine which brain
regions differ as a function of completing a MI course (see below). We suspect that such
task-related brain activity would be similar among students in other instructional
environments.
Instruction Effect: Changes in Brain Activity Post-instruction vs. Pre-instruction
Our second analysis identified increased brain activity among students completing the
physics reasoning task after taking a MI course (Post > Pre) in the posterior cingulate cortex,
frontal poles, dlPFC, and angular gyrus. These brain regions (PCC, angular gyrus) overlap
with regions of another commonly observed large-scale functional brain network known as
the default-mode network (DMN). The DMN, consisting of posterior cingulate cortex
(PCC), angular gyri, medial PFC, and middle temporal gyri (Raichle et al., 2001; Laird et
al., 2009), is generally associated with internally oriented cognitive processes (i.e., self-
reflection, mind wandering, autobiographical memory, planning; Buckner et al., 2008).
However, other lines of evidence also implicate DMN involvement in complex tasks such as
narrative comprehension (Simony et al., 2016), semantic processing (Binder et al., 2009;
Binder and Desai, 2011) or the generation and manipulation of mental images (Andrews-
Hanna, 2012). In the context of the current task, one interpretation is that students may
generate mental images to simulate events and formulate predictions. Additionally, post-
instruction increase in DMN activity was observed during physics reasoning (which we
show is supported by the CEN), and such coupling between the DMN and CEN during
cognition has been hypothesized to arise during controlling attentional focus, thereby aiding
in efficient cognitive function (Leech and Sharp, 2014).
Other brain regions showing greater activation during physics reasoning after the MI course
included the dlPFC and the frontopolar cortex. The frontopolar cortex is a component of a
decision-making network often involved with learning (Koechlin and Hyafil, 2007). The
dlPFC is critically linked with the manipulation of verbal and spatial information in working
memory (Barbey et al., 2013). Given previous links with, for example, mental simulation,
working memory, mathematical calculations, and attention, we speculate that post-
Brewe et al. Page 15
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

instruction increased activity in the PCC, angular gyrus, dlFPC and frontal pole may reflect
enhanced mental operations and/or models involved with physics reasoning and/or
generation of predictions about physical outcomes.
The PCC, left angular gyrus, left frontal pole, and left DLPFC were the four regions of
greatest extent to show increased activity (Post > Pre), however, we did not see correlation
between change in activity within these areas and accuracy on the FCI after instruction. The
FCI is a cognitively demanding task which includes intuitive but wrong answers. Thus, it
may simply be that even wrong answers on the FCI require significant mental effort.
Inaccurate physics reasoning likely still involves many of the same mental operations
successful physics reasoning does (i.e., mental imagery, visualization, prediction generation,
and decision making, to name a few). Measures of accuracy in and of themselves may not
display a simple one-to-one relationship with changes in brain activity across instruction.
Rather, these changes in brain activity may be related to more complex behavioral changes
in how student’s reason through physics questions post-relative to pre-instruction. These
might include shifts in strategy or an increased access to physics knowledge and problem
solving resources.
We posit that the observed pre to post-instruction changes in brain activation during physics
reasoning are consistent with what one may expect to observe as students develop refined
mental models during classroom learning. Physics reasoning, regardless of an individual’s
familiarity with the material, is a process continually scaffolded by mental model use
(Nersessian, 1995, 1999, 2002a,b; Giere, 2005; Koponen, 2006), and effective physics
learning is engendered by building and deploying strategies to appropriately implement
mental models during reasoning (Hestenes, 1987). In this study, we framed our exploration
of learning-induced changes in brain activity in the context of the MI classroom because this
pedagogical approach has been shown to effectively encourage the development and flexible
implementation of models during physics reasoning (Brewe, 2008; Brewe et al., 2010b). Our
experimental results do not go as far as to implicate MI as any more or less effective than
other instructional strategies at supporting instructional-related changes in student’s brain
networks. However, if we accept that physics reasoning inherently relies on mental model
use, we can begin to consider a more truly neuroeducational interpretation of physics
learning in which shifts in network engagement across instruction bring about student
conceptual change. Characterizing these neurobiological changes may ultimately help
researchers and educators understand which instructional strategies may best support
successful model development. We hold that the mental models student’s deployed at the
beginning of the semester during reasoning, upheld by a variety of CEN-supported
attentional and executive processes, shifted after instruction, as evidenced by student’s
overall increased accuracy during reasoning. This instruction-induced shift in model use
promoted increased involvement from key DMN and CEN regions within reasoning. This
study represents an initial step in neuroeducational research demonstrating that such shifts,
indicative of learning, are measurable and detectable using noninvasive brain imaging
techniques. Additional work is needed to understand the relationship between external
conceptual models as studied in science education, with mental models and related cognitive
constructs as studied in neuroimaging literature.
Brewe et al. Page 16
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

This project has several limitations. First, we focused on the MI class and did not assess the
brain activity of students from traditional lecture course sections or other active learning
environments. Based on the data presented, we do not make claims that MI is a better or the
only instructional tool capable of inducing brain network alterations. Rather, in the current
study, we used MI as an exemplar case. It remains to be determined if different pedagogies
differentially influence how physics reasoning-related brain networks develop. As noted
above and consistent with recommendations (Freeman et al., 2014), we will explore this in
the future and a future direction could investigate differences among active learning formats.
Second, these analyses addressed brain activation and did not consider correlation with other
behavioral measures, such as mental rotations, science anxiety, or academic performance
measures which could further aid in the interpretation of these fMRI outcomes. Third,
consideration of potential differences between female and male students remains for future
investigations.
Notwithstanding these limitations and future direction, these preliminary outcomes implicate
brain regions linked with physics reasoning and, critically, suggest that brain activity during
physics reasoning is modifiable over the course of a semester of physics instruciton. Further
work should investigate differences between MI and lecture instruction, as well as
addressing differences among different active learning strategies across disciplines. Studying
active learning broadly has the potential to more clearly elaborate how these pedagogies
impact student learning and brain function.
Supplementary Material
Refer to Web version on PubMed Central for supplementary material.
ACKNOWLEDGMENTS
Primary funding for this project was provided by NSF REAL DRL-1420627 (AL, EB, SP, and JB). Contributions
from coauthors were partially provided by NSF 1631325 (AL, MR, and TS), NIH R01 DA041353 (AL, MS, and
MR), NIH U01 DA041156 (AL, MS, MR, KB, and ERB), NSF CNS 1532061 (AL), NIH K01DA037819 (MS),
NIH U54MD012393 (MS), and FIU Graduate School Dissertation Year Fellowships (AN and RO). Additional
thanks to the FIU Instructional and Research Computing Center (IRCC, http://ircc.fiu.edu) for providing HPC and
computing resources that contributed to the research results reported within this paper, and to the Department of
Psychology of the University of Miami for providing access to their MRI scanner. Lastly, special thanks to the FIU
undergraduate students who volunteered, participated, and contributed to this project.
REFERENCES
Andrews-Hanna JR (2012). The brain’s default network and its adaptive role in internal mentation.
Neuroscientist 18, 646–656. doi: 10.1177/1073858411403316
Arsalidou M, and Taylor MJ (2011). Is 2+2=4? Meta-analyses of brain areas needed for numbers and
calculations. NeuroImage 54, 2382–2393. doi: 10.1016/j.neuroimage.2010.10.009 [PubMed:
20946958]
Barbey AK, Koenigs M, and Grafman J (2013). Dorsolateral prefrontal contributions to human
working memory. Cortex 49, 1195–1205. doi: 10.1016/j.cortex.2012.05.022 [PubMed: 22789779]
Bartley JE, Boeving ER, Riedel MC, Bottenhorn KL, Salo T, Eickhoff SB, et al. (in press). Meta-
analytic evidence for a core problem solving network across multiple representational domains.
Neurosci. Biobehav. Rev
Brewe et al. Page 17
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Beichner RJ, and Saul JM (2003). “Introduction to the SCALE-UP (studentcentered activities for large
enrollment undergraduate programs) project.” in Proceedings of the International School of Physics,
(July), 1–17. Available online at: http://www.ncsu.edu/PER/Articles/Varenna_SCALEUP_Paper.pdf
Binder JR, and Desai RH (2011). The neurobiology of semantic memory. Trends Cogn. Sci 15, 527–
536. doi: 10.1016/j.tics.2011.10.001 [PubMed: 22001867]
Binder JR, Desai RH, Graves WW, and Conant LL (2009). Where is the semantic system? A critical
review and meta-analysis of 120 functional neuroimaging studies. Cereb. Cortex 19, 2767–2796.
doi: 10.1093/cercor/bhp055 [PubMed: 19329570]
Bressler SL, and Menon V (2010). Large-scale brain networks in cognition: emerging methods and
principles. Trends Cogn. Sci 14, 277–290. doi: 10.1016/j.tics.2010.04.004 [PubMed: 20493761]
Brewe E (2008). Modeling theory applied: modeling instruction in introductory physics. Am. J. Phys
76, 1155–1160. doi: 10.1119/1.2983148
Brewe E (2011). Energy as a substancelike quantity that flows: theoretical considerations and
pedagogical consequences. Phys. Rev. Spec. Top. Phys. Edu. Res 7:020106 doi: 10.1103/
PhysRevSTPER.7.020106
Brewe E, Kramer LH, and O’Brien GE (2010a). “Changing Participation Through Formation of
Student Learning Communities.” in AIP Conference Proceedings doi: 10.1063/1.3515255
Brewe E, Kramer L, and O’Brien G (2009). Modeling instruction: Positive attitudinal shifts in
introductory physics measured with CLASS. Phys. Rev. Spec. Top. Phys. Edu. Res 5:013102. doi:
10.1103/PhysRevSTPER.5.013102
Brewe E, and Sawtelle V (2018). Modeling instruction for university physics: examining the theory in
practice. Eur. J. Phys doi: 10.1088/1361-6404/aac236
Brewe E, Sawtelle V, Kramer LH O’Brien GE Rodriguez I, and Pamelá P (2010b). Toward equity
through participation in Modeling Instruction in introductory university physics. Phys. Rev. Spec.
Top. Phys. Edu. Res 6:010106. doi: 10.1103/PhysRevSTPER.6.010106
Brewe E, Traxler A, de La Garza J, and Kramer LH (2013). Extending positive CLASS results across
multiple instructors and multiple classes of modeling instruction. Phys. Rev. Spec. Top. Phys. Edu.
Res 9:20116. doi: 10.1103/PhysRevSTPER.9.020116
Buckner RL, Andrews-Hanna JR, and Schacter DL (2008). The brain’s default network: anatomy,
function, and relevance to disease. Ann. N. Y. Acad. Sci 1124, 1–38. doi: 10.1196/annals.1440.011
[PubMed: 18400922]
Cole MW, and Schneider W (2007). The cognitive control network: integrated cortical regions with
dissociable functions. Neuroimage 37, 343–360. doi: 10.1016/j.neuroimage.2007.03.071
[PubMed: 17553704]
Craik K (1943). The Nature of Explanation. London: Cambridge University Press.
Crouch CH, and Mazur E (2001). Peer Instruction: ten years of experience and results. Am. J. Phys,
69, 970–977. doi: 10.1119/1.1374249
Dou R, Brewe E, Zwolak JP, Potvin G, Williams EA, and Kramer LH (2016). Beyond performance
metrics : examining a decrease in students’ physics self-efficacy through a social networks lens.
Phys. Rev. Phys. Edu. Res 12:20124. doi: 10.1103/PhysRevPhysEducRes.12.020124
Etkina EE, Murthy S, and Zou X (2006). Using introductory labs to engage students in experimental
design. Am. J. Phys 74:979. doi: 10.1119/1.2238885
Etkina E, and Van Heuvelen A (2007). Investigative Science Learning Environment – A Science
Process Approach to Learning Physics. PER-Based Reforms in Calculus-Based Physics. Available
online at: http://www.compadre.org/PER/per_reviews/media/volume1/ISLE-2007.pdf
Freeman S, Eddy SL, McDonough M, Smith MK, Okoroafor N, Jordt H, et al. (2014). Active learning
increases student performance in science, engineering, and mathematics. Proc. Natl. Acad. Sci.
U.S.A 111, 8410–8415. doi: 10.1073/pnas.1319030111 [PubMed: 24821756]
Frigg R, and Hartmann S (2006). “Models in science,” in Stanford Encyclopedia of Philosophy, Vol. 1,
ed Zalta EN (Stanford, CA: Metaphysics Research Lab, Stanford University).
Giere RN (2005). How models are used to represent reality. Philos. Sci 71, 742–752. doi:
10.1086/425063
Brewe et al. Page 18
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Gouvea J, and Passmore C (2017). “Models of” versus “Models for”toward an agent-based conception
of modeling in the science classroom. science and education, 26, 49–63. doi: 10.1007/
s11191-017-9884-4
Greca IMA, and Moreira MA (2001). Mental, physical, and mathematical models in the teaching and
learning of physics. Sci. Edu 86, 106–121. doi: 10.1002/sce.10013
Greca IM, and Moreira MA (2000). Mental models, conceptual models, and modelling. Int. J. Sci.
Educ 22, 1–11. doi: 10.1080/095006900289976
Halloun IA (1996a). Schematic modeling for meaningful learning of physics. J. Res. Sci. Teach 33,
1019–1041. doi: 10.1002/(SICI)10982736(199611)33:9<1019::AID-TEA4>3.0.CO;2-I
Halloun IA (1996b). “Views About Science and physics achievement: The VASS story,” in AIP
Conference Proceedings-Physics Education Research Conference, Vol. 399, (College Park, MD),
605–614.
Halloun IA (2004). Modeling Theory in Science Education. Dordrecht: Springer.
Hegarty M (2004). Mechanical reasoning by mental simulation. Trends Cogn. Sci 8, 280–285. doi:
10.1016/j.tics.2004.04.001 [PubMed: 15165554]
Henderson C (2002). Common concerns about the force concept inventory. Phys. Teacher 40, 542–
547. doi: 10.1119/1.1534822
Hestenes D (1987). Toward a modeling theory of physics instruction. Am. J. Phys 55, 440–454. doi:
10.1119/1.15129
Hestenes D (1992). Modeling games in the Newtonian world. Am. J. Phys 60:732. doi:
10.1119/1.17080
Hestenes D, Wells M, and Swackhamer G (1992). Force concept inventory. Phys. Teacher 30, 141–
158. doi: 10.1119/1.2343497
Johnson-Laird PN (1983). Mental Models: Towards a Cognitive Science of Language, Inference, and
Consciousness. Cambridge, MA: Harvard University Press.
Justi R, and Gilbert J (2000). History and philosophy of science through models: some challenges in
the case of’the atom’. Int. J. Sci. Educ 22, 993–1009. doi: 10.1080/095006900416875
Kober N (2014). Reaching Students What Research Says about Effective Instruction in Undergraduate
Science and Engineering. Washington, DC: National Academies Press.
Koechlin E, and Hyafil A (2007). Anterior prefrontal function and the limits of human decision-
making. Science 318, 594–598. doi: 10.1126/science.1142995 [PubMed: 17962551]
Koponen IT (2006). Models and modelling in physics education: a critical re-analysis of philosophical
underpinnings and suggestions for revisions. Sci. Educ 16, 751–773. doi: 10.1007/
s11191-006-9000-7
Krieger MH (1987). The physicist’s toolkit. Am. J. Phys 55, 1033–1038. doi: 10.1119/1.14929
Kuhn TS (1970). The Structure of Scientific Revolutions. Chicago, IL: University of Chicago Press.
Laird AR, Eickhoff SB, Li K, Robin DA, Glahn DC, and Fox PT (2009). Investigating the functional
heterogeneity of the default mode network using coordinate-based meta-analytic modeling. J.
Neurosci 29, 14496–14505.doi: 10.1523/JNEUROSCI.4004-09.2009 [PubMed: 19923283]
Lakoff G (1987). Women, Fire, and Dangerous Things: What Categories Reveal about the Mind, Vol.
64 Chicago, IL: University of Chicago Press.
Lasry N, Rosenfield S, Dedic H, Dahan A, and Reshef O (2011). The puzzling reliability of the force
concept inventory. Am. J. Phys 79:909. doi: 10.1119/1.3602073
Lasry N, Watkins J, Mazur E, and Ibrahim A (2013). Response times to conceptual questions. Am. J.
Phys 81:703. doi: 10.1119/1.4812583
Leech R, and Sharp DJ (2014). The role of the posterior cingulate cortex in cognition and disease.
Brain 137, 12–32. doi: 10.1093/brain/awt162 [PubMed: 23869106]
Lehrer R, and Schauble L (2006). “Cultivating model-based reasoning in science education,” in The
Cambridge handbook of: The learning sciences, ed Sawyer RK (New York, NY,: Cambridge
University Press), 371–387.
Mackey AP, Miller Singley AT, and Bunge SA (2013). Intensive reasoning training alters patterns of
brain connectivity at rest. J. Neurosci 33, 4796–4803. doi: 10.1523/JNEUROSCI.4141-12.2013
[PubMed: 23486950]
Brewe et al. Page 19
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Mason RA, and Just MA (2015). Physics instruction induces changes in neural knowledge
representation during successive stages of learning. Neuroimage 111, 36–48. doi: 10.1016/
j.neuroimage.2014.12.086 [PubMed: 25665967]
Mason RA, and Just MA (2016). Neural representations of physics concepts. Psychol. Sci 27, 904–
913. doi: 10.1177/0956797616641941 [PubMed: 27113732]
Mathieu JE, Heffner TS, Goodwin GF, Salas E, and Cannon-Bowers JA (2000). The influence of
shared mental models on team process and performance. J. Appl. Psychol. 85, 273–283. doi:
10.1037/0021-9010.85.2.273 [PubMed: 10783543]
Matthews MR (2007). Models in science and in science education: an introduction. Sci. Educ 16, 647–
652. doi: 10.1007/s11191-007-9089-3
McDermott LC, and Shaffer PS (2001). Tutorials in Introductory Physics. New York, NY: Pearson.
McPadden D, and Brewe E (2017). Impact of the second semester university modeling instruction
course on students’ representation choices. Phys. Rev. Phys. Educ. Res 13:020129 doi: 10.1103/
PhysRevPhysEducRes.13.020129
Morgan MS, and Morrison M (1999). Models as Mediators: Perspectives on Natural and Social
Science. New York, NY: Cambridge University Press.
Morris GA, Branum-Martin L, Harshman N, Baker SD, Mazur E, Dutta S, et al. (2006). Testing the
test: Item response curves and test quality. Am. J. Phys 74, 449–453. doi: 10.1119/1.2174053
Morris GA, Harshman N, Branum-Martin L, Mazur E, Mzoughi T, and Baker SD (2012). An item
response curves analysis of the force concept inventory. Am. J. Phys 80:825. doi:
10.1119/1.4731618
Nersessian NJ (1995). Should physicists preach what they practice? Sci. Educ 4, 203–226. doi:
10.1007/BF00486621
Nersessian NJ (1999). “Model-based reasoning in conceptual change,” in Model-Based Reasoning in
Scientific Discovery, eds Magnani L, Nersessian NJ, and Thagard P, (New York, NY: Kluwer
Academic/Plenum Publishers), 5–22.
Nersessian NJ (2002a). Abstraction via generic modeling in concept formation in science. Mind Soc.
3, 129–154. doi: 10.1007/BF02511871
Nersessian NJ (2002b). The Cognitive Basis of Model-Based Reasoning in Science. Cambridge:
Cambridge University Press.
Odenbaugh J (2005). Idealized, inaccurate but successful: a pragmatic approach to evaluating models
in theoretical ecology. Biol. Philos 20, 231–255. doi: 10.1007/s10539-004-0478-6
Passmore CM, and Svoboda J (2012). Exploring opportunities for argumentation in modelling
classrooms. Int. J. Sci. Educ 34, 1535–1554. doi: 10.1080/09500693.2011.577842
Prabhakaran V, Smith JA, Desmond JE, Glover GH, and Gabrieli JD (1997). Neural substrates of fluid
reasoning: an fMRI study of neocortical activation during performance of the Raven’s Progressive
Matrices Test. Cogn. Psychol 33, 43–63. doi: 10.1006/cogp.1997.0659 [PubMed: 9212721]
Prado J, Chadha A, and Booth JR (2011). The brain network for deductive reasoning: a quantitative
meta-analysis of 28 neuroimaging studies. J. Cogn. Neurosci 23, 3483–3497. doi: 10.1162/
jocn_a_00063 [PubMed: 21568632]
Raichle ME, MacLeod AM, Snyder AZ, Powers WJ, Gusnard DA, and Shulman GL (2001). A default
mode of brain function. Proc. Natl. Acad. Sci. U.S.A 98, 676–682. doi: 10.1073/pnas.98.2.676
[PubMed: 11209064]
National Research Council (2012). “Discipline-Based Education Research: Understanding and
Improving Learning in Undergraduate Science and Engineering,” in Committee on the Status,
Contributions, and Future Directions of Discipline-Based Education Research; Board on Science
Education; Division of Behavioral and Social Sciences and Education, eds Singer SR, Nielsen NR,
and Schweingruber HA (National Research Council; National Academies Press). Available online
at: http://download.nap.edu/cart/download.cgi?andrecord_id=13362andfree=1
Rodriguez I, Potvin G, and Kramer LH (2016). How gender and reformed introductory physics
impacts student success in advanced physics courses and continuation in the physics major. Phys.
Rev. Phys. Educ. Res 12:9. doi: 10.1103/PhysRevPhysEducRes.12.020118
Rogoff B (1990).”Shared thinking and guided participation: conclusions and speculation BT -
Apprenticeship in thinking: cognitive development in social context,” in Apprenticeship in
Brewe et al. Page 20
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Thinking: Cognitive Development in Social Context (New York, NY: Oxford University Press),
189–210.
Sawtelle V, Brewe E, Goertzen RM, and Kramer LH (2012). Identifying events that impact self-
efficacy in physics learning. Phys. Rev. Spec. Top. Phys. Educ. Res 8:20111. doi: 10.1103/
PhysRevSTPER.8.020111
Sawtelle V, Brewe E, and Kramer LH (2010). “Positive impacts of modeling instruction on self-
efficacy,” in PERC Conference Proceedings, Vol. 1289, eds Singh C, Rebello NS, and Sabella M
(Portland, OR: American Institute of Physics), 289.
Schwartz DL, and Bransford JD (1998). A time for telling. Cogn. Instrum 16, 475–5223. doi: 10.1207/
s1532690xci1604_4
Schwarz CV, Reiser BJ, Davis EA, Kenyon L, Achér A, Fortus D, et al. Krajcik J (2009). Developing a
learning progression for scientific modeling: making scientific modeling accessible and
meaningful for learners. J. Res. Sci. Teach 46, 632–654. doi: 10.1002/tea.20311
Seeley WW, Menon V, Schatzberg AF, Keller J, Glover GH, Kenna H, et al. Greicius MD (2007).
Dissociable intrinsic connectivity networks for salience processing and executive control. J.
Neurosci 27, 2349–2356. doi: 10.1523/JNEUROSCI.5587-06.2007 [PubMed: 17329432]
Simony E, Honey CJ, Chen J, Lositsky O, Yeshurun Y, Wiesel A, et al. (2016). Dynamic
reconfiguration of the default mode network during narrative comprehension. Nat. Commun
7:12141. doi: 10.1038/ncomms12141 [PubMed: 27424918]
Svoboda J, and Passmore C (2011). The Strategies of Modeling in Biology Education. Science and
Education. Available online at: http://www.springerlink.com/index/10.1007/s11191-011-9425-5
Traxler A, and Brewe E (2015). Equity investigation of attitudinal shifts in introductory physics. Phys.
Rev. Spec. Top. Phys. Educ. Res 11, 1–7. doi: 10.1103/PhysRevSTPER.11.020132
Wells M, Hestenes D, and Swackhamer G (1995). A modeling method for high school physics
instruction. Am. J. Phys 63:606. doi: 10.1119/1.17849
Wendelken C (2015). Meta-analysis: how does posterior parietal cortex contribute to reasoning? Front.
Hum. Neurosci 8:1042. doi: 10.3389/fnhum.2014.01042 [PubMed: 25653604]
Windschitl M, Thompson J, and Braaten M (2008). Beyond the scientific method: model-based inquiry
as a new paradigm of preference for school science investigations. Sci. Educ 92, 941–967. doi:
10.1002/sce.20259
Zwolak JP, Dou R, Williams EA, and Brewe E (2017). Students’ network integration as a predictor of
persistence in introductory physics courses. Phys. Rev. Phys. Educ. Res 13, 1–14. doi: 10.1103/
PhysRevPhysEducRes.13.010113
Brewe et al. Page 21
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

FIGURE 1 |.
Schematic of the relationship between conceptual and mental models in physics curriculum.
Brewe et al. Page 22
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

FIGURE 2 |.
Example items from the physics reasoning fMRI task. (A) FCI questions described a
physical scenario using pictures and words and then asked a physics question followed by
four potential answers. (B) Control question shared basic visual and linguistic features with
FCI questions, however control questions did not ask students to engage in physics
reasoning.
Brewe et al. Page 23
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

FIGURE 3 |.
Group-level fMRI results. (Red) Task effect: Brain regions showing increased activity during
the physics reasoning task (FCI > Control) at the pre-instruction stage. (Blue) Instruction
effect: Brain regions showing increased activity at the post- relative to pre-instruction (Post
> Pre) scan during the physics reasoning task.
Brewe et al. Page 24
Front ICT
. Author manuscript; available in PMC 2019 May 15.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Brewe et al. Page 25
TABLE 1 |
Conceptual and Categorical Hierarchies.
HIERARCHY
Conceptual Level Categorical
Theory Superordinate Animal
Model Basic Dog
Concept Subordinate Retriever
Front ICT
. Author manuscript; available in PMC 2019 May 15.

Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Brewe et al. Page 26
TABLE 2 |
Overall and individual item accuracy for pre and post instruction FCI questions in the scanner.
Pre Post Change Item Difficulty
FCI Question % % (Post–Pre) (%) Morris et al., 2006 (%)
2 29.5 39.3 +9.8 34.6
3 42.6 58.9 +16.3 51.5
6 78.7 78.6 −0.1 73.6
7 54.1 71.4 +17.3 66.4
8 39.3 46.4 +7.1 50.4
12 45.9 69.6 +23.7 65.2
14 24.6 41.1 +16.4 39.5
27 44.3 46.4 +2.1 59.4
29 42.6 85.7 +43.1 50.8
Total 44.6 59.7 15.1
Item difficulty measures from Morris et al. (2006) are included for comparison.
Front ICT
. Author manuscript; available in PMC 2019 May 15.