Simulations and
Visualizations:
Issues for REC
Paul Horwitz
The Concord Consortium
This paper arose out of a discussion held at the
REC PI meeting on May 16, 2002. The ideas and opinions are derived from those
expressed at that meeting, but have been integrated and edited by the author so
that attribution of them to individuals would be both difficult and
misleading.
Some definitions
This is one of those fields where the descriptive
terms seem to be used differently by different practitioners. Accordingly, let
us start by giving some working definitions of three such terms: visualization,
simulation, and model. To introduce these definitions we remind ourselves that
there are only two things that a computer can do for its users: it can show
them things and it can let them do things (Horwitz, 1999). That's really all
there is to it - every computer application, no matter how complex, is simply a
set of choices of what to show and what to let the user do. Of course there is
a wealth of complexity hidden in that simple statement. It isn't just a
question of what to show - different representations of what is shown
can have very different consequences; similarly, different user
interfaces can dramatically affect how easy it is to do one thing or
another. So, specializing now to software intended to teach something, what is
learned will depend, among other things, on one's choice of representations and
interfaces.
Basically, visualizations are what we choose to
show users, simulations are what we let them do. And models are
what link the two. In the case of educational software, the design of the
visualizations, simulations, and models will depend critically on what is being
taught, for what purpose, and to whom.
An Example
Say we are building a simulator of a nuclear reactor
for the purpose of training people who will later be running the real thing.
Our choice of visualizations will no doubt include a representation of a
control room, complete with all, or an important subset of, the dials,
switches, buttons, digital and graphic readouts, and other displays and
controls present in the real control room. We will make an effort, presumably,
to assure that the simulated displays and controls look the same and are in the
same relative locations as their real-world counterparts. In fact, given the
lack of standardization in the industry (Frogner and Meijer, 1980), if we do
our job well our simulator may strictly apply only to one particular reactor
control room in the whole world!
The controls of our simulated control room will, of
course, be "active" in the sense that they can be manipulated by the learner to
set various parameters of an underlying model of nuclear reactor dynamics. They
constitute the interface to the reactor simulation - they encompass, and
consequently constrain, what the user can do.
Now imagine that our reactor simulator has been built
to teach nuclear physics, say at the undergraduate level. In this case the
visualizations will be considerably less realistic. The control room, in fact,
will probably disappear completely, to be replaced, perhaps, by a visualization
of uranium (or possibly plutonium) nuclei emitting and absorbing neutrons,
fissioning into faster-moving fragments, and sometimes decaying spontaneously.
The time scale of the simulation, which may have been sped up in the previous
example, will here have to be slowed dramatically. The controls will also be
quite unrealistic, offering students the ability to add or delete nuclei,
change reaction cross-sections, or alter the speed of the neutrons - operations
they could not perform in "real life."
The underlying models of our two contrasting examples
will be quite different as well. The simulation intended for reactor operator
training will most likely be run by a top-level model consisting of rules that
govern the macroscopic behavior of the principal components. Thus, activity
levels will depend on core temperature according to the dictates of a table
derived from experimental data; the effects of water flow on reactor cooling
will be similarly determined. The model for the nuclear physics simulation, in
contrast, will pay much less attention to reactor-scale parameters,
concentrating instead on the details of sub-atomic processes hidden to the
real-world observer.
The differences between the models, visualizations,
and simulations in the two applications arise from their different learning
goals. The training of reactor operators is designed to give them a wealth of
procedural knowledge that will guide them in carrying out their functions.
Physics or engineering students, in contrast, seek to acquire a deep
understanding of the physical principals that govern how the reactor works. To
the operator, the nuclear reaction taking place in the reactor core can largely
be treated as a "black box" heat source. The physicist can equally well ignore
the details of how that reaction is monitored and controlled.
The criteria by which to judge the quality of the
models will vary as well. In the training example one would expect the model to
be faithful to the reactor itself - to reproduce, in other words, as accurately
as possible the behavior of its real-world counterpart. In the case of the
physics model, since the goal is to teach students how to interpret the
macroscopic behavior of a large ensemble of unstable nuclei by extrapolating
from the statistically deterministic behavior of a few hundred, accuracy is
less critical than explanatory power. Inclusion of extraneous detail, in
addition to restricting the generality of the model, might actually hinder
understanding. The model will therefore be purposely simplified, glossing over,
perhaps, such practical considerations as the exact geometry of the reactor
core, in order to clarify basic concepts common to all situations.
Issues for educational simulations
The REC panel discussed a number of issues and
questions associated with the creation of simulations for teaching.
Accuracy of models
The nuclear reactor simulation example above
illustrates a key question: how realistic should the model underlying a
simulation be? Although there was general agreement that the simple answer is,
"it all depends," in fact, the consensus of the REC simulation and
visualization panel was that in most instances scientific accuracy of a model
is not as important, for educational purposes, as simplicity and explanatory
power. A chess-playing program designed to teach the game to novices is likely
to be a weaker player than one that is optimized for performance.
Similarly, a simplified model of a complex phenomenon is often better suited to
teach beginners than a more sophisticated model that matches experimental data
more precisely at the cost of being more difficult to understand. The ideal
situation, it would seem, would be to have a model that is more sophisticated
than strictly necessary for beginners, but to simplify the simulations and
visualizations associated with the model so that naïve learners will not
become confused. To do this effectively, however, calls for scripting
the model so that it can alter its mode of presentation to fit changing
circumstances. This possibility is discussed further below.
Revising models
Aside from precision, another criterion for judging a
model is the range of situations to which it can be successfully applied. The
process of continual refinement and generalization of models is central to the
scientific enterprise, and a valuable process for students to observe. Once
again, an example may serve to illustrate the point1.
A special-purpose rule for predicting the results of a
head-on, elastic collision between two equal-mass objects is a far cry from the
universally applicable law of conservation of momentum. A strategy for helping
students appreciate the power of such general laws is to guide them through a
succession of what White and Frederiksen (1990) have termed "upwardly
compatible causal models." Thus, one might start by modeling head-on collisions
between identical particles, one moving and one stationary. Once the students
are familiar with this case, one can first allow both masses to move, then
generalize to glancing collisions. At each stage, students are encouraged to
derive a sequence of ever more general rules, such as "When one ball is at
rest, the moving ball stops dead and gives all its velocity to the stationary
one," or "When both balls are moving before the collision, they exchange their
velocities," to "The total velocity of the system is the same before and after
the collision." Once we allow the masses to be different, the velocity
conservation rule no longer works and must be generalized to the familiar law
of conservation of momentum: "The product of mass times velocity is conserved
in the collision."
As they move from one of these rules to the next,
students are not only coming closer to an accepted scientific law governing
collisions - they are also learning to test their models against experiments -
either real or computer-based - revise them if need be, and generalize them to
cover more and more situations. Finally, the sequence of rules becomes a
genuine scientific model when the students are able to link it back to Newton's
Third Law, and to recognize that momentum conservation is actually a
consequence of the equality of forces between the two colliding masses.
In helping students to follow the path we have laid
out above, we can make use of technology in a variety of ways - data collection
probes can measure forces over very short times, computer simulations make the
experiments easy to perform and noise-free. Finally, visualization techniques
enable us to represent abstract concepts such as forces and velocities - and,
if we choose, to make them, rather than the masses themselves, the logical
objects of student investigation. A straightforward extrapolation of this
latter idea would be to make the physical objects that represent the real-world
situation disappear entirely! This possibility is taken up in the next section.
Hiding information
Visualization enables us to show information that
would not normally be available; it also gives us the ability to hide
information that would normally be present. Though this may seem at first
glance counterproductive, in fact it makes it possible to present challenges to
students that mirror those that real scientists confront. Science, after all,
is largely a process of trying to figure out what is going on, in the absence
of full information. Here is an example, once again based on the situation
described in the previous section.
Let us add to the two colliding balls a third,
invisible one. There are various ways of doing this, but to be specific,
imagine that the three balls are lined up in a row, stationary, with the
invisible one in the middle. The student makes one of the visible balls move
toward the other visible one, but before it gets there it collides with the
invisible ball; that ball in turn will collide, after a short interval, with
the second visible one. By observing the behavior of the visible balls, the
student can easily guess, that there must be an invisible one between them. It
is a little harder to figure out whether the invisible ball is more or less
massive than the visible ones, but it is possible. If one is very clever, one
can determine whether the invisible ball is moving at the end of the experiment
simply by checking to see whether the total momentum of the visible balls is
the same as that originally imparted to the first one. The presence of missing
momentum is an unambiguous indication that the invisible ball is still in
motion. Regardless of how well students perform on this task, it is an
authentic one - the reasoning required mirrors very well how physicists
think2.
Having students create their own models
A recurrent question in the use of computer
simulations and visualizations is whether to allow students to create, as well
as manipulate, the underlying models that link and control them. Simple but
powerful programming languages have been created that make it possible for
students with limited expertise to program their own models of complex systems.
Languages such as NetLogo and StarLogo are especially applicable in domains
that involve interactions among many identical components, whose statistically
deterministic "emergent behavior" mimics that of such real-world analogs as
insect behavior, molecular dynamics, and traffic flow (Wilensky and Resnick,
1999). Since the components usually follow simple rules (e.g., "maintain a
minimum distance between your car and the car ahead of you"), such complex
dynamical models can often be implemented even by novice programmers (Wilensky,
1999).
The power and flexibility that comes with being able
to create one's own model is a distinct advantage of this approach, and the
associated pedagogy complements and empowers a constructivist strategy, but
these advantages can sometimes come at the price of entailing a rather steep
learning curve. It is not necessary for students to recapitulate the work of
Galileo, Kepler, and Newton, for instance, in order to "discover" the models
underlying classical mechanics. Nor do they need to experiment with "triploid"
and "tetraploid" organisms to find that Mendel's diploid model for transmission
genetics offers a better fit to the inheritance data. The research base does
not yet exist that would enable us to determine, in a rigorous way, how much
leeway to give students in the creation of their own models, and under which
circumstances. There is growing evidence, however, that if students are given
ready-made models, rather than being challenged to generate them, one must
ensure that the tasks that accompany the models elicit reasoning and
understanding, rather than requiring only rote learning (Gobert, 2000).
Instructivism
More general than the issue of when to let students
create their own models, is the question of how - and how much - to "scaffold"
students' use of simulations and visualizations. New delivery vehicles, such as
the WISE platform (c.f. Linn and Slotta, 2000), and new software development
environments, such as Pedagogica™ (Horwitz and Christie, 1999), are making
it easier for researchers and educational technology developers to embed
simulations (and by implication their underlying models) in broader educational
contexts. This can serve several purposes (Buckley, Gobert, and Christie,
2002):
- it can situate the simulation in a real-world
context, making it more likely that students' expertise in manipulating a model
will transfer to other situations and contribute to deeper understanding not
only of the scientific domain under investigation, but of others as well;
- it can monitor students' actions and react to them
in the context of a specific challenge or problem;
- by logging and reporting on students' actions and
responses to questions, it offers teachers and researchers a unique and
powerful tool for fine-grained, performance-based assessment.
The embedding of simulations and visualizations within
more traditional curricular materials has created a new kind of interactive
curriculum tool called a "hypermodel" (Horwitz, 1995; Horwitz and Christie,
1999). Hypermodels integrate stored information in the form of multimedia
materials, experimental data, and text, with a manipulable model of the subject
domain. Just as hypertext enables one to navigate through textual materials by
clicking on individual words and phrases, with hypermodels students navigate
through a learning activity by manipulating a computer-based model. The
activity typically presents a more or less open-ended challenge (e.g., "Breed
these organisms as efficiently as possible, trying to get all the offspring to
look like this.") and then leaves the students alone and monitors them (Which
organisms do they choose to breed? How do they react to the outcome?) as they
try to accomplish the goal.
At various points throughout an activity the computer
may "pop up" and offer help, either as metacognitive prompts (e.g., "Why did
you do that?" "Did it accomplish what you expected?" "What is your immediate
goal?"), or in the form of hints ("What happens when you breed two
recessives?") The trick in programming such interventions is not to overwhelm
the student with "helpful" comments (the infamous Microsoft Word paperclip
comes to mind!), but at the same time to avoid having students become
discouraged or "turned off."
Thus, one of the central questions that the Modeling
Center at the Concord Consortium is studying is how much structure to build
into a hypermodel activity. The two extremes are clearly suboptimal: too much
structure (think of the "drill and kill" examples from computer-aided
instruction) and an activity becomes boring, too little ("just give them some
open-ended software and leave them alone") and most students are likely to give
up in despair. Moreover, students' responses to questions asked in situ
constitute powerful probes of their thinking. They can be useful both for
research purposes and in configuring an activity in real time to meet
individual students' needs.
There is also the problem of transfer. If an activity
is not explicitly tied to specific teaching goals, it runs the risk of becoming
a meaningless videogame - the students may become quite expert at winning the
game without developing the understanding of the domain that we seek - and that
the standards call for. The eternal quest for the happy medium between linear
instruction and radical constructivism is what we have dubbed "instructivist,"
to convey that it seeks a Hegelian synthesis of two contrasting educational
philosophies. Once again, the research literature does not (yet!) provide a
clear guide.
Communication and collaboration
The use of models as communication devices was also
brought up during the REC panel discussion that was the impetus for this paper.
Awareness of the importance of communication and the social context of learning
has been growing for decades (Rogoff and Lave, 1999). The use of models as a
means of communication, however, is relatively new (Black, 1972; Gilbert,
1980). A number of recent projects have demonstrated the extraordinary value of
having students share their models, comment on them, and view each other's
comments (cf. Gobert et al, 2002). Not only does this appear to have a
beneficial effect on students' learning and retention, it also teaches social
and team-building skills, and places students in situations that resemble the
world of real scientists.
But without simulations that enable one to observe the
behavior of a model, and visualizations that convey their effects, it is
difficult to share a model with someone else. Static language and mathematical
equations cannot capture the dynamic aspects of a time-evolving process. The
implications of recent advances, not only in the technology for creating and
viewing models but also in means of sharing them with others, offer a fertile
field for research in social cognition and learning, and are likely to have
significant implications for the use of simulations and visualizations in
classrooms, and for teacher preparation programs.
Teacher professional development
The panel also discussed the question, "What is the
proper role of the teacher in a classroom that uses simulations and
visualizations? The answer, and its implications for teacher professional
development, will surely be influenced by the progress of the instructivism
debate referred to above. No computer activity, no matter how carefully
constructed, is likely to "stand alone" - now, or in the foreseeable future. If
for no other reason than the obvious fact that the curriculum cannot - and
should not! - be delivered entirely on a computer, the teacher will always be a
role model and mentor, and will play an essential role in coordinating
student's activities, managing their interactions and discussions, keeping them
on task, and motivating them to learn. Nonetheless, in a class that uses
simulations there will be times when the students are working on the computers,
either singly or in pairs or small groups. In this situation, the teacher's
role is to support them in that activity. The situation is not unique in
teachers' prior experience - other group activities such as laboratories are
similar in some respects. However, it is harder to determine what students are
doing on a computer than in a "wet lab," and consequently it is more difficult
to detect those "teachable moments" when a small intervention can help a
student make sense of something or discover a connection that might otherwise
remain obscure. Improvements in the technology can help, but clearly teachers
are going to have to become thoroughly familiar with each computer activity
their students will undertake, the common problems they are likely to face, and
strategies for helping what will in fact be a more idiosyncratic and
student-centered learning process than they are accustomed to dealing with.
Recommendations for REC Research
The description above has identified several areas in
which more research is needed. Accordingly, the following recommendations for
REC flow directly from the discussions of the simulations and visualizations
panel, and all of them, in fact, were raised at some point by the panel.
Importance of visualization research
The connection between the two terms used in
describing the panel - simulation and visualization - and the third term,
modeling, needs additional teasing out. For example, there has been extensive
research work on information visualization, but most of it deals with
visualization of data, not models3.
Concept maps (Novak, 1980) constitute an interesting and fruitful attempt to
visualize and represent relationships between concepts, but they are static,
propositional representations that are purely verbal and inherently
paper-based. The ability to create manipulable models in the form of
simulations and visualizations is uniquely a product of the computer age. We
must learn how to adopt this new tool to literally "use vision to think" - to
externalize that critical aspect of the mental process that so augments human
reasoning and memory. The educational and learning sciences can and should
contribute to that work.
Instructivist research
The issue brought up above under the heading
"Instructivism" harks back to the tension between teaching content and
teaching process, and is probably as old as education itself. In science
education, specifically, the tension often arises in the form of arguments over
whether to teach science as a series of stories about the workings of the
natural world, or to concentrate instead on teaching the scientific reasoning
behind the discovery, validation, and refinement of those stories. In reality,
of course, either extreme is counterproductive - science is clearly much more
than a collection of facts, but it is pointless to teach students how to reason
without giving them anything substantive to reason about.
In the context of educational software, the
"product-process" controversy can be characterized as a tension between
computer-aided instruction (CAI) and the "microworlds" most closely identified
with Logo. More research is needed to fully explore the continuum between the
two extremes, the area that we have termed "instructivism," and to bridge the
gap between them. Many questions remain to be answered. Do different students,
as experience and anecdotal data would suggest, require different "treatments"
along the instructivist continuum? Do students pass through that continuum in
stages, perhaps requiring less and less support as their understanding deepens
and their inquiry skills improve? Can we build software that responds
differently to different students, and to the same students at different
points, altering the amount and type of scaffolding as circumstances indicate?
Technology-assisted assessment
One of the main differences between the use of
simulations and visualizations to teach science and more traditional methods is
in the use of, and reliance on, language4.
Prior to the advent of the computer, science teaching perforce relied heavily
on language to describe abstract ideas. This mode of learning works well with
the subset of the population that possesses the requisite linguistic skills,
but very poorly with those students who do not use language and other abstract
representations effectively. More work needs to be done to explore the extent
to which simulations can help such "language deficient" students5 achieve and demonstrate understanding commensurate
with their ability. Can the computer help them by bridging the gap between the
simulation and the linguistic or mathematical representations of knowledge?
Currently, assessment is almost entirely a
language-intensive activity. Much more research is required to enable us to use
simulations as assessment tools, but the initial experience is promising
(Buckley, Gobert, and Christie, 2002). If a student cannot read or write very
well, can't manipulate algebraic symbols with facility, but can solve a
challenging problem using a manipulable computer model, doesn't that imply that
the student knows something? And shouldn't we be able to assess such students
by giving them a simulation-situated problem to solve? As computers become more
and more ubiquitous in schools, the time is fast approaching when it will be
cost effective to deliver the SAT and other high-stakes tests via computer,
rather than shipping stacks of sealed (and ferociously guarded!) paper
assessments around the country. We need to prepare for that time by learning
how to make the radical improvements in assessment that truly interactive
technology potentially affords.
Research on pedagogy
At the moment, most of our information on how to use
simulations and visualizations in the classroom is based on anecdotal evidence.
In particular, we face fundamental questions about how to evaluate students'
work with simulations. In an age of increasing "accountability" and high-stakes
testing, how can we reconcile an approach that encourages collaborative problem
solving with the requirement to assign individual grades? How can the use of
interactive simulations be adapted to individual learning styles and skill
levels? What depth of content knowledge is required to enable teachers to feel
comfortable in assigning open-ended exploratory activities? To answer questions
such as these on a scientific basis will require much broader projects than the
current norm. It is possible that improvements in technology may make such
projects more affordable in the near future.
Support for large scale, scientific education
research
With the internet becoming ubiquitous in the Nation's
schools (Cuban, 2001), it is now feasible to introduce technology and acquire
data over the World Wide Web. The WISE project (Linn and Slotta, 2000) provides
an authoring and curriculum development environment that enables teachers to
write and disseminate their own curriculum material, and to obtain data for
evaluation of educational effectiveness. The Concord Consortium's
Pedagogica™ (Horwitz and Christie, 1999; Horwitz and Tinker, 2001) is a
suite of tools with similar capabilities. Other research groups are working on
the same problem. With respect, specifically, to research into the use of
simulations and visualizations, it will soon be possible for thousands of
schools, and hundreds of thousands of students, to use the same simulations for
the same purposes6. As the students work
through the activities, they will generate a wealth of data that will be made
available instantly to teachers and researchers. Such an infrastructure will
make possible large-scale, quantitative or qualitative research with enough
statistical power to enable us to make scientifically valid statements about
the relative strengths and weaknesses of different pedagogical approaches.
To pursue the thought a step further: if one were
trying to determine the differences between different types of scaffolding, one
could develop multiple versions of the same activity and randomly assign
different versions to different students within the same class.
This would avoid most of the complications inherent with the use of comparison
"control groups" that are never as close to the "treatment group" as one would
ideally like. Just such a research protocol, using Pedagogica, is being
employed by the Concord Consortium, on an IERI grant (Concord Consortium,
2001).
Need for economies of scale
At present, research using simulations and
visualizations is still very expensive. Simulation technology, particularly if
it involves complex models and sophisticated pedagogy, is expensive to produce,
and the obstacles to deploying such technology in schools, and receiving data
back from it reliably, are formidable indeed7.
The problem is particularly daunting for younger researchers who cannot command
the level of resources required to make a contribution to this promising field.
The suggestion has therefore been made (Concord Consortium, 2002) that the NSF
consider funding a national "Educational Accelerator," similar to the National
Accelerator Laboratory (aka "FermiLab") in Illinois, that would take on many
common technology-intensive tasks for "customers" in the education research and
K-12 school communities. Such a national laboratory would:
- create new technology for educational researchers
and help them adapt it to their particular purposes;
- recruit schools and teachers who wish to
participate in research projects, link them to researchers, help them to
install and maintain the technology, and offer professional development
services to help them use it effectively;
- handle the acquisition of data from the schools and
transmit it to the researchers under appropriate guarantees of privacy and
security;
- support the analysis of school-based data by
researchers and communicate the results of such analyses back to teachers,
students, and parents, as desired;
- offer internships and summer courses for Ph.D.
students in education research;
- provide technology to schools of education for
pre-service and in-service training, as well as courses in instructional design
and technology research; and
- host national conferences and symposia on issues
related to educational technology.
Natural scientists have learned take advantage of the
vast economies of scale associated with technology-intensive research. It makes
sense for the education research community to do the same. The formation of a
National Educational Accelerator would turn the notion of "scalability" on its
head - making it possible for many research groups to take advantage of a
common technological infrastructure. Such an initiative could catalyze a
revolution in science education, and bring the elusive national goal of
universal science literacy within our grasp.
Acknowledgements
It is a pleasure to thank Dr. Barbara Buckley and Dr.
Janice Gobert for their insightful comments on earlier versions of this paper,
and their suggestions for its improvement.
Footnotes
1 This example is
taken from a physical sciences curriculum under development at the Concord
Consortium.
2 A more complicated
variant of the "missing momentum" technique led the physicist Wolfgang Pauli to
hypothesize the existence of an unseen, zero-charge, zero-mass particle
(subsequently named the "neutrino" by Fermi) that is emitted in the process
called beta decay. Recently, more delicate experiments have demonstrated that
neutrinos actually have a very small mass.
3 The addition of
animation to one's arsenal of visualization tools does not necessarily address
this problem. An animated weather map does not convey any insight into why that
thunderstorm developed where and when it did, or what is likely to happen to it
next.
4 Our use of the word
"language" is quite general here, and is intended here to include mathematical
symbolism as well as text.
5 Native speakers of
languages other than English form an important subset of these students.
6 Concord Consortium's
Modeling Across the Curriculum will achieve this goal in the near future.
7 In order to avoid
the problems inherent with students' use of the school computers, many schools
have developed elaborate "firewalls" designed to inhibit indiscriminate
downloading of software, and in many cases, the saving of files in general.
This makes it difficult both to install software in such schools, and to find a
way to save students' work.
References
Black, M. (1972) Models and Metaphors, Cornell
University Press.
Buckley, B.C., Gobert, J., and Christie, M.T (2002)
Model-based Teaching and Learning with Hypermodels: What do they learn? How do
they learn? How do we know? Paper presented at the Annual Meeting of the
American Education Research Association, New Orleans, LA.
Concord Consortium (2002) Center for Educational
Research with Technology, proposal submitted to the Centers for Learning and
Teaching Program, National Science Foundation.
Concord Consortium (2001) Modeling Across the
Curriculum, proposal submitted to the Interagency Education Research
Initiative, National Science Foundation.
Cuban, L., (2001) Oversold and Underused: Computers
in Classrooms, Harvard University Press
Frogner, B. and Meijer, J. (1980) On-Line Power Plant
Alarm and Disturbance Analysis Systems, EPRI Final Report NP-1379.
Gilbert, J.K. (1980) The Use of Models in Science
Teaching, European Journal of Science Education, Vol. 2(1), 1 -
13.
Gobert, J. (2000). A typology of models for plate
tectonics: Inferential power and barriers to understanding. International
Journal of Science Education, 22(9), 937-977.
Gobert, J., Snyder, J., & Houghton, C. (2002,
April 1-5). The influence of students' understanding of models on
model-based reasoning. Paper presented at the Annual Meeting of the
American Educational Research Association, New Orleans, LA.
Horwitz, P. (1995) Linking Models to Data:
Hypermodels for Science Education, The High School Journal, Vol. 79,
No. 2, pp. 148 - 156.
Horwitz, P. (1999) Designing Computer Models that
Teach, in Computer Modeling and Simulation in Pre-College Science
Education, pp. 179 - 196, Nancy Roberts and Wallace Feurzeig, eds.,
Springer Verlag.
Horwitz, P. and Christie M. T. (1999) Hypermodels:
Embedding Curriculum and Assessment in Computer-Based Manipulatives,
Journal of Education, Volume 181, Number 2, pp. 1 - 23 (with Mary Ann
Christie)
Horwitz, P. and Tinker, R. (2001) Pedagogica to the
rescue: A short history of hypermodels, @ Concord, the Concord Consortium
Newsletter, Vol. 5(1), pp. 1-4.
Linn, M.C. & Slotta, J.D. (2000). WISE science.
Educational Leadership.
Alexandria: Association for Supervision and Curriculum Development.
Novak, J. D. (1990). Concept maps and Vee diagrams:
Two metacognitive tools for science and mathematics education.
Instructional Science, 19, 29-52.
Rogoff, B., and Lave, J., Editors (1999) Everyday
Cognition: Its Development in Social Context, iUniverse.com Publishers.
White, B., & Frederiksen, J. (1990) Causal Model
Progressions as a Foundation for Intelligent Learning Environments. Artificial Intelligence,
24, pp. 99-157.
Wilensky, U. (1999). GasLab--an Extensible Modeling
Toolkit for Exploring Micro- and Macro- Views of Gases. In Roberts, N. ,
Feurzeig, W. & Hunter, B. (Eds.) Computer Modeling and Simulation in
Science Education. Berlin: Springer Verlag.
Wilensky, U. & Resnick, M. (1999). Thinking in
Levels: A Dynamic Systems Perspective to Making Sense of the World. Journal
of Science Education and Technology. Vol. 8 No. 1. pp. 3 - 18.
|
