Sci & Educ (2014) 23:897–921

DOI 10.1007/s11191-013-9673-7

Epistemological Issues Concerning Computer

Simulations in Science and Their Implications for Science
Education

Ileana M. Greca • Eugenia Seoane • Irene Arriassecq

Published online: 29 January 2014

 Springer Science+Business Media Dordrecht 2014

Abstract Computers and simulations represent an undeniable aspect of daily scientific

life, the use of simulations being comparable to the introduction of the microscope and the
telescope, in the development of knowledge. In science education, simulations have been
proposed for over three decades as useful tools to improve the conceptual understanding of
students and the development of scientific capabilities. However, various epistemological
aspects that relate to simulations have received little attention. Although the absence of this
discussion is due to various factors, among which the relatively recent interest in the
analysis of longstanding epistemological questions concerning the use of simulations, the
inclusion of this discussion on the research agenda in science education appears relevant, if
we wish to educate scientifically literate students in a vision of the nature of science closer
to the work conducted by researchers today. In this paper we review some contemporary
thoughts emerging from philosophy of science about simulations in science and set out
questions that we consider of relevance for discussion in science education, in particular
related with model-based learning and experimental work.

1 Introduction

Computers represent an undeniable aspect of daily scientific life (Kaufmann and Smarr
1993). They are applied to all areas of traditional natural sciences, whether theoretical or
experimental, and have generated specific scientific disciplines. Not only have they

I. M. Greca
Dpto. de Didácticas Especı́ficas, Universidad de Burgos, Burgos, Spain
e-mail: [email protected]

E. Seoane
Fac. de Cs. Exactas - ECienTec, Universidad Nacional del Centro de la Provincia de Buenos
Aires, Pinto 399 - (7000) Tandil, Buenos Aires, Argentina

I. Arriassecq (&)
Fac. de Cs. Exactas - ECienTec, Universidad Nacional del Centro de la Provincia de Buenos
Aires - CONICET, Pinto 399 - (7000) Tandil, Buenos Aires, Argentina
e-mail: [email protected]

123
898 I. M. Greca et al.

increased the speed and changed the way in which calculations are done, but they have also
changed the way in which the data are inspected (Lenhard 2010), the type of questions that
may be asked—very often, the subject of the investigation is conditional upon whether it is
computationally possible-, and even the way in which the data are presented, with dynamic
and highly visual presentations. Although this connection between computers and scientific
practice began at the start of the 1940s, it has been further strengthened and has reorga-
nized scientific practice as digital computer and its availability has developed, with the
consequent affordability of personal computers, as well as the standardization of hardware
and software (Johnson and Lenhard 2011).
Thus, computational techniques have introduced new tools into science. Within the broad
spectrum of techniques and methods that fall under the loose heading of computational
science, we will focus on computer simulations,1 which form a special and very important
sub-domain of computational science. The use of simulations in the development of
knowledge has been compared to the introduction of the microscope and the telescope,
constituting ‘‘a significant and permanent addition to the methods of science’’ (Humphreys
2004). This has attracted the attention of some epistemologists who have pointed out that
computer simulations not only constitute a new tool, but a new form of scientific production
(Galison 1996; Winsberg 1999). This new form of scientific production would also present
epistemological problems that are also new, such as the modification of the role of differ-
ential equations as the principal tool of physics (Fox Keller 2003; Johnson and Lenhard
2011); the nature of modeling and its relation with existing theories (Winsberg 2010); the
classic division between scientific theory and empirical methods (Humphreys 2004) and the
meaning and the objective of explanations (Johnson and Lenhard 2011).
Since the 1980s simulations have been proposed in the area of science education as a
useful tool for improving the conceptual understanding of students and in general the
development of scientific capabilities.2 Moreover, some researchers support the idea that they
are one of the most powerful applications of the new technologies: as not only can they
simulate real processes in all areas (movement, photosynthesis, atomic configurations, etc.),
but they can also simulate the performance of ‘‘virtual experiments’’, which are too haz-
ardous and costly to perform in school laboratories (Hsu and Thomas 2002; Hennessy 2006).
Besides, if we wish to train students in a vision of the nature of science that is closer to
the research work that is now prevalent, it would appear necessary to include simulations
in science teaching practices because of their centrality to the daily tasks of contemporary
science. However, epistemological aspects that relate to simulations have received little
attention in science education. The absence of this discussion is due to various factors,
among which the relatively recent interest in the analysis of longstanding epistemological
questions concerning the use of simulations, which only began to attract attention in the
philosophy of science towards the end of the 1990s, and the differences between simu-
lations used in the science classroom and those that are specifically scientific (Doerr 1997),
which is an important point that we will address later on.
Nevertheless, the inclusion of some key aspects of the current epistemological debate on
the research agenda in the area of science education, at least for secondary and university
1
As in the literature, we shall also use the following terms interchangeably, throughout the text: simula-
tions, computer simulations, computational models and computational modeling (that is, a model running on
a digital computer, with special characteristics that differentiate it from more traditional modeling, a point
discussed in Sect. 2.3).
2
See de Jong and Njoo (1992); Hsu and Thomas (2002); Huppert and Lazarowitz (2002); Kaput (1995);
Tao and Gunstone (1999); Zacharia (2003); Zacharia and Anderson (2003) and a recent critical review on
the topic by Smetana and Bell (2012).

123
Epistemological Issues Concerning Computer Simulations 899

educational levels, appears relevant. On the one hand, the new epistemological questions
that are raised by simulations affect various points that are considered fundamental for the
understanding of the nature of science (e.g. Osborne et al. 2003; McComas and Olson
1998; Lederman et al. 2002): particularly, data analysis and interpretation, the construction
of hypotheses and predictions, and the diversity of scientific methods. As an example, one
of the research results of Wong and Hodson (2009), in which scientists were interviewed
about various aspects of the scientific enterprise and its nature, showed that recent tech-
nological advances are making it possible, in some areas, to generate knowledge without
the need to generate prior hypotheses.
In addition, there is another significant reason to give serious consideration to episte-
mological aspects of simulations for the training of scientifically literate citizens. Over
recent years, simulations in the area of climate science and the knowledge gained from
them have had a profound impact on public policy (Winsberg 2010). Arguments for and
against the reasons and effects of global warming are usually centered on the results of
climate simulations (Guillemot 2010). However, the layman has no clear conception of
what those simulations are and the validity of their results, especially if we take into
account the difference between the use of the word simulation in everyday language and in
the sciences (Fox Keller 2003). Moreover, it would appear that we run into some diffi-
culties when we try to separate simulation and reality, as simulations appear to be
‘‘obviously true’’ because of their apparent capability to ‘‘imitate’’ reality, in such a way
that some educational researchers (for example, Lunetta et al. 2007) suggest that a new
objective for students in the twenty first century would be to learn to discriminate between
reality and virtual reality.
The main objective of this paper is to review epistemological issues concerning sim-
ulations and to discuss their possible implications for research and teaching in science
education. We have organized it into four sections, as follows: in section two, after a brief
introduction of the principal historical steps in the development of computational mod-
eling, we try to explain, in very general terms, how simulations, despite their great
diversity, are constructed in science, in order to highlight where their specific features may
appear that made them so peculiar. This section ends by addressing certain epistemological
issues, which we think are significant for science teaching and that emerge from their use.
In section three, we revise the different kinds of simulations used in science education and
the main results of the research in this area. This section ends by positing the relevance and
benefit of approaching epistemic issues concerning simulations in two areas of science
education, experimentation and model-based learning, that frequently use them in their
didactic strategies. In the fourth and final section, we set out our concluding remarks.

2 Simulations in the sciences

According to the dictionary of the Real Academia Española [Royal Spanish Academy], to
simulate is to represent something, by pretending or imitating what is not. This definition,
common to several different languages, denotes the negative nature that the term simu-
lation has in colloquial use. However, as Fox Keller (2003, p. 198) has pointed out, the
Oxford Dictionary has, since the Second World War, when computer simulations first
began to appear, incorporated a definition that is not only of a positive character, but
reflects its scientific meaning: ‘‘The technique of imitating the behaviour of some situation
or process… by means of a suitably analogous situation or apparatus, especially for the
purpose of study or personnel training.’’

123
900 I. M. Greca et al.

This is certainly a very wide definition that covers all of the different types of simu-
lations currently used in different areas: Ören (2011a, b) has listed more than 100 defi-
nitions on simulations and about 400 different types of modeling and simulations are
currently in use. As a working definition, for the purposes of this study, simulations are the
representation of the dynamic behavior of a system that moves it from state to state in
accordance with an approximate (mathematical) model that is used to implement it on a
computer. It would be of interest to learn about the origins and evolution of simulations, in
order to arrive at a more precise definition of what today we understand as computer
simulations, which is nevertheless sufficiently broad to include the majority of their dif-
ferent types.

2.1 Brief history of simulations

Since the end of the nineteenth century, various analogical models were proposed that
imitated the behaviours of real systems, the 1872’s tide-predicting machine of Lord Kelvin,
or those that resolved specific classes of mathematical problems, such as the differential
analyzer of Vannevar Bush of 1927 (Mindell 2002). However, we had to wait for the
appearance of programs that functioned in digital computers for calculus and the imitation
of systems to acquire a new dimension. The origin of these programs is found in the
development of techniques to determine the reliability of various nuclear weapons, at the
Los Alamos research laboratory, between 1946 and 1952. The assessment of these pro-
posals implied finding the solutions to equations to predict highly non-linear phenomena,
such as neutron diffusion, for example. To do so, various approaches to the computational
procedures of that time were prepared, the most famous of which was the Monte Carlo
method (Galison 1996).
So, computer simulations opened the door to the study of complex systems, which,
because of their characteristics, could not be covered in an exact analytic manner. This was
its first contribution: to provide work plans to find approximated solutions (not the exact
ones of analytical methods) with sufficient precision and speed. However, according to Fox
Keller, although they ‘‘started out as little more than a mechanical extension of conven-
tional methods of numerical analysis, where what was being ‘‘simulated’’ were the pre-
computer, handwritten equations … such methods rapidly grew so effective that they
began to challenge the status of the original, soon threatening to displace the very equa-
tions they were designed to simulate.’’ (op. cit., p. 210).
Following the historical development of simulations outlined by Fox Keller, three stages
may be identified, in which successful developments gradually developed and were
accumulated, generating new effects, and progressively undermining the traditional notions
of theory, experimentation and data. However, the basic concepts and techniques of
simulations that may be highlighted at each stage all emerged in the initial stage: Monte
Carlo methods, finite element methods, cellular automatism, and artificial neuronal net-
works (Lenhard 2010). Moreover, Fox Keller insists that the novelty that epistemologists
now see in simulations was neither found at one instantaneous moment, nor was it a pattern
at that initial stage. It came instead from an accumulative process of small perturbations,
based on what had already been achieved, in which the simulations gradually gained more
ground and were converted into an indispensable tool in all scientific areas. It is worth
mentioning that the uses that these three stages characterize are still valid in research into
various scientific areas, particularly in physics.
The first of the stages identified by Fox Keller is the use of the computer to find
solutions to pre-established mathematical models, which are analytically unsolvable, in

123
Epistemological Issues Concerning Computer Simulations 901

terms of numeric analysis. The consolidation of this way of using the simulations began to
raise questions, fundamentally, over the role of differential equations as a principal tool in
theoretical physics.
A second stage, which started in the mid-1950s in the area of fluid dynamics and plasmas
may be characterized by the emergence of the use of simulations to determine the standard
features required in realistic approximations to physical models of complex systems. A
general theory exists for these systems, but their application to specific cases, in other words,
to the preparation of models is extremely complex. So, first of all, a simplified system replaces
the real system and then the equations that support the theory of this simplified system are
transformed for subsequent treatment by numeric analysis, and for input into the computer;
the resultant simulations are compared to the ‘‘true’’ behavior of the system, in order to
evaluate the simplifications. Rather than the evaluation or resolution of the mathematical
expressions, the priority in this method is the simulation that the computer produces of this
simplified version of the physical system. As a result are obtained models (equations) that are
physically more realistic and computationally treatable. This type of simulation, of great use
in various branches of the experimental sciences, gives rise to the so-called ‘‘computational
experiments’’ or ‘‘virtual experiments’’. They have emerged as an alternative somewhere
between theoretical and laboratory-based experiments, thereby establishing new relations
between the nature of modeling and its relation with theory and experimentation.
Finally, Fox Keller highlights a third use, or stage: the construction of (theoretical and/
or experimental) models of phenomena for which there is neither a theory, nor are there
exact nor approximate equations, but only a rudimentary idea of the underlying dynamic;
for example, in the case of the modeling biological phenomena. The objective in this case
is the simulation of the phenomena in itself, questioning, both the meaning and the
objectives of a scientific explanation. The agent-based models used in biology fit into this
group, as do the works developed in the 1980s in the area of artificial life.
Thus, the original use of simulations as tools for the resolution of unsolvable scientific
equations3 has gone far further, such that they are now used for practical reasons (for example,
experimental costs or impracticable experiments, such as the formation of black holes) and ethics
(for example, the diffusion of a new virus in the population). They are more appropriate than
experiments when it is necessary to optimize any given experiment or when there is no theory that
can directly explain a phenomenon and an effort is made to reproduce it and to understand the
factors that might influence it (such as in the social sciences) (Humphreys 2004). None of these
applications share a theory or have common laws, but instead a set of skills, ‘‘a new mode of
producing scientific knowledge that was rich enough to coordinate highly diverse subject matter.’’
(Galison 1996, p. 119), which is how scientists working with simulations understand them.
Therefore, simulations can not simply be reduced to numerical methods broadly
improved by the use of rapid calculation processes on computers. Fundamentally, this new
mode of producing scientific knowledge has increased the number of phenomenon that can
be modeled and has enormously increased our capability to apply theories to the world.
Although we may think that ‘‘in principle’’, if the equations that describe a phenomenon
exist, they can be solved, it would be ingenuous to suppose that, even if we had an infinite
amount of time, a group of ‘‘human calculators’’4 could, for example, find the solutions to
forecast European weather trends 1 week in advance (Humphreys 2004).

3
Although this, as Humphreys (2004) has highlighted, is no slight matter: a great part of the success of
physics is due to the development of better methods of calculation.
4
We recall that the term ‘‘computer’’ referred to people whose job it was to make calculations, in general
women; it was only later that it came to refer to an electronic device (Galison 1996).

123
902 I. M. Greca et al.

It is worth remembering that when Newton introduced the fundamental equation of

mechanical dynamics, he affirmed its (almost) universal application, even though (almost)
nowhere was it analytically solvable: the analytic solutions that might have been achieved,
would only apply within a range of idealized conditions. In fact, almost all of the many
classes of differential equations are analytically unsolvable, as far as we are currently
aware; the number of models that may be resolved using solely analytical techniques being
very limited. So, our current understanding of all complex and most especially all non-
linear systems that characterize the vast majority of phenomena in the universe, have in
fact been made possible by the use of simulations.

2.2 How to Simulate a Phenomenon

Computer simulations, for Humphreys (2004, p. 110), may be defined as follows:

System S provides a core simulation of an object or a process B just in case S is a concrete
computational device that produces, via a temporal process, solutions to a computational model that
correctly represents B, either dynamically or statically. If in addition the computational model used
by S represents the structure of the real system R, then S provides a core simulation of system R with
respect to B.5

However, this definition alone is not enough to understand the epistemological partic-
ularities introduced by simulations. To do so, it is necessary to gain a better understanding
of how they function. Although, as stressed in the brief history of the development of
simulations presented in the previous section, simulations vary in many forms, from the
closed ended ones that were first developed to the more open as agent-based simulations,
all of them may be characterized as transformations of mathematical models in discrete
algorithms that imitate the behavior of systems, for which different methods exist to
transform the equations into computationally treatable algorithms (among which, Monte
Carlo, finite differences,6 etc.). This would appear to imply a relatively simple process in
areas with well-established theories: given the phenomena and having selected one part of
its behavior for simulation, the physical principles are chosen that are the most appropriate
for its description, the mathematical model (or set of differential equations emerging from
the theory that describe the phenomena) is determined, the parameters and the initial values
of the variables are established, the type of computational method is chosen to transform
the differential equations into algorithms that the computer can solve, and the algorithm is
then fed into the computer to obtain the results.
Seen in this way, it would appear that simulations are nothing more than improved
methods of solving equations. However, the situation is much more complex and from this
complexity emerges great part of the epistemological issues that we will discuss in
this paper and that apply to almost all kind of simulations, although the problems

5
This definition includes the simulation of mathematical objects, because B is not required to be real.
6
In the Monte Carlo method, a formal isomorphism is established between differential equations with
certain equations in probabilistic theory, using the probabilistic relations to resolve the differential equations
and replacing the calculation of all combinatorial possibilities for an entire sequence of events, by an
estimation of the results obtained for a ‘‘sample’’ of attempts. The general idea of the finite-element method
is the division of a continuum by a series of points known as nodes into a set of small interconnected
elements, based on the idea that the equations that govern the behaviour of the continuum will also govern
that of the element. Thus, it is possible to pass from a continuous system (infinite degrees of freedom),
governed by one or by a system of differential equations, to a system with finite degrees of freedom, the
behaviour of which is modelled by a system of either linear or non-linear equations. Visually, it is like
dividing the space into a reticular mesh, seeking the solution at the points that are determined by the mesh.

123
Epistemological Issues Concerning Computer Simulations 903

increase in areas that may be defined as theoretically poorer. In the first place, the trans-
formation of the model with its initial parameters and values into an algorithm that may be
implemented on the computer is neither an obvious nor a simple procedure. In line with
Winsberg (2010, pp. 10–17), there are two aspects in the creation of a viable computer
algorithm. The first aspect is related to the algorithm that results from the direct trans-
formation of the continuous differential equations into discrete differential equations. This
step can result in a very computationally costly algorithm or an algorithm that, as a
consequence of the approximations used to move from continuous to discrete equations, is
unstable, and produces errors and unreliable results. It is therefore necessary to set aside
the algorithm, which would otherwise be the next step on the basis of the mathematical
model, in order to simplify the model, by ignoring or by discarding some factors, by
reducing the model’s degrees of freedom and by adopting what are known as unrealistic
assumptions of symmetry in the computational model.
The other point is the inclusion in the algorithm of mathematical relations to model
factors of the physical model that are fundamental for an understanding of the behavior of
the system, but would, if applied in the computer model with its precise mathematical
formulae, be computationally untreatable. These relations are usually very simple and have
no direct connection with the original differential equations. Their construction is some-
times guided by theory, at other times by physical ‘‘intuition’’, but also in response to the
computational limitations observed by trial and error. They may be considered ‘‘rough-
and-ready, theoretically unprincipled model-building tools’’ (Winsberg, ibid., p. 12) con-
structed to capture some natural important effect that has been left out of the computational
model because of technical limitations. When these model-building tools are combined
with the more theoretical equations, they produce more realistic results than those that
would otherwise have been produced, had those tools not been taken into consideration.
One of these tools is ‘‘eddy viscosity’’, widely used in the simulation of fluids with
turbulent flows, in the dynamics of fluids and meteorology as well as in the study of the
connective properties in giant dwarf stars (Winsberg, ibid.). Another example of these tools
is the ‘‘Arawaka operator’’ used in the simulations of atmospheric dynamics (Küppers and
Lenhard 2005).
A further possibility in this same direction is the substitution of the real physics of a
process that might be highly complex by phenomenological relations. For example, in the
case of red dwarves, to account for surface energy loss, the real physical process was
substituted by the standard formula for the radiation of a black hole, solely applied at the
points at which it was considered that the star would radiate heat efficiently. In short, the
parametric relations that appear in a simulation often have no direct counterpart—in a
strictly realistic sense, from an ingenuously realistic point of view—in a real system.
Once the computer model has been implemented, the algorithm produces a data set
that requires interpretation. A variety of complex visualization techniques are used to
interpret the results, an effort is made to integrate the data cloud with other sources of
knowledge, including observational data, and the credibility and reliability of some of
the features of those data sets are determined. With all of this, a series of static or
dynamic images are finally generated,7 to arrive at the end-result of the simulation,
which Winsberg refers to as ‘‘models of phenomena’’, on which basis scientists can start
to study the emergent patterns.

7
The series of dynamic images constitute the animations.

123
904 I. M. Greca et al.

2.3 Epistemological Problems Relating to Simulations

The methodological novelties introduced by simulations have challenged some relevant

epistemological notions in science,8 some of which we shall discuss in this section.
(a) New role of mathematics in research
The origin of physics coincided with the origin of differential calculus. Since Newton, who
also invented methodological criteria for rationally testing mathematical models,
unmatched even to this day (Harper 2011), this form of calculus has been considered its
principal tool, as it would not otherwise have been possible to describe the continuum and
therefore, to represent reality. In fact, ever since mathematics became a precise tool for
prediction in the eighteenth century, mathematical models described by differential
equations came to be considered as ‘‘true’’ representations, as their predictions were
verified. However, the ‘‘experiments in theory’’, characteristic of the first and second stages
in the development of simulations, started to gain their own life, questioning the primacy of
the most conventional mathematical techniques, in particular differential equations (Fox
Keller 2003; Johnson and Lenhard 2011).
Likewise, the success of simulations increased the legitimacy of the practice of
numerical analysis among scientists, which was not, until the appearance of computing,
anything more than an auxiliary calculus tool. During the second half of the nineteenth
century numerical methods were developed from the need to solve specific problems in
astronomy, physics and engineering, but only during the 1940s did numerical analysis
really became a bona fide mathematical discipline and not just a collection of recipes
(Benzi 2009). This change not only challenged the hegemony of a particular type of
mathematical tool, but also fundamentally challenged the realism associated with the
representations of continuous variables (Fox Keller 2003), even going so far as to call into
doubt the very concept of space as a continuum. So, in the field of cellular automatism, it
was proposed that space could be a discrete temporal spatial network of bits of information
(Vichniac 1984; apud. Fox Keller 2003).9
In relation to the acceptance of this change in the scientific community, the empirical
study conducted in the area of astro-physics by Sundberg (2010b) appeared to show that
both postures coexist: on the one hand, the skeptics express doubts over the elevation of
simulations to the level of analytical methods, and on the other, there are those that propose
the analysis of simulations in themselves, without any need to return to underlying
equations to justify the resulting simulation or to improve ease of computation for a
relevant simulation feature.
(b) Simulations as semi-autonomous models
A further question has to do with the status of simulations as models.10 Over recent years,
the traditional view of models, which gave them a secondary role in relation to theory,

8
It is important to highlight that we are focusing only on the debate surrounding the epistemology of
science. The lively discussion among mathematicians concerning epistemological issues (among others, the
notion of proof) relating to the use of computers is beyond the scope of this study.
9
Similar ideas have been proposed by some theoreticians in the field of quantum information (Wheeler
1990; Brukner and Zeilinger 2005).
10
The status of models in relation to theory and experimentation is not, in fact, an epistemological problem
specific to simulations, but a general problem (Frigg and Reiss 2009); however the reappraisal of models in
relation to theory coincided with the generalization of the use of simulations in all scientific areas, which
may not be coincidental.

123
Epistemological Issues Concerning Computer Simulations 905

insofar as it considered that models were fundamentally nothing more than a representation
that makes sense of mathematical formalism, has run into criticism from semanticists
(Cartwright 1999; Sismondo 1999). In their view, models are something other than only
theory plus data and, as they are partially independent from theory and from the world,
they have an autonomous component that turns them into instruments of exploration in
both domains (Morgan and Morrison 1999), into tools for intervention and for the
manipulation of phenomena (Cartwright 1983; Hacking 1983), increasing the number of
phenomena and processes that may be explained.
Simulations fit perfectly into this vision of models as instruments of mediation. For
example, in the previously discussed case of the construction of simulations for complex
systems in fluid dynamics, the real systems in themselves are not well understood, even if
the theoretical elements for an understanding of their basic dynamics are known; the
construction of the simulations are guided, but not determined by the theory and data from
different sources has also to be used. Thus, emergent simulations can not be reduced to
mere calculations. Moreover, simulations gradually and in an iterative way perfect the
models with which the phenomena are described, as they allow the determination of their
most relevant parameters for their description, producing new results in this way, which are
beyond the reach of theories on the basis of which they were constructed, functioning more
as mediators between theory and experimentation (Galison 1996).
Besides, due to the construction process itself, it is very difficult to be certain about the
causes of a successful simulation. For several authors, the objective of a simulation is the
construction of instrumentally reliable models (Suárez 1999), which are representative of a
physical system, but without the aim of being a realistic representation of the physical
system and its behavior. Therefore, there are authors that affirm that simulations will never
equal the traditional notion of models, remaining a sort of second-order model (Küppers
and Lenhard 2005). Computer simulations, for Winsberg:
involve a complex chain of inferences that serve to transform theoretical structures into specific
concrete knowledge of physical systems… [he argues that] this process of transformation is also a
process of knowledge creation, and that it has its own unique epistemology (Winsberg 1999, p. 275).

It is interesting to highlight that computer models in climate science are criticized by

people who do not consider them ‘‘good science’’, because they are not founded on data
and solid theories (Guillemot 2010) and who also question the possibility of verifying the
projections of the models in relation to the data. The first of these criticisms relates more to
the role of the models in general, and not only the computer models, in relation to theory.
In response to these criticisms, in climate science, Edwards (1996, apud. Guillemot 2010)
pointed out that the relation between models and data, even though interdependent, is
symbiotic rather than circular. Norton and Suppe (2001) argued that that interdependence
also exists between theory and experimentation and that the absence of certainty and the
construction of simplified hypotheses are not inherent to computational models.
(c) Simulations and experimentation
Scientists construct computational models and work on them, even though they are
considered experimentalist or theoretical, understanding work on a model as the action of
exploring the relations between data input and output, in order to produce verifiable
predictions or better models, changing, adding and adapting parameters and repeatedly
running the simulations (Winsberg 2003; Lenhard 2010). This double relation with theory
and experimentation has disconcerted epistemologists of scientific disciplines who seek to
understand ‘‘how simulation can have methodological and epistemological features in

123
906 I. M. Greca et al.

common with experimentation, while still playing the role of a form of scientific
theorizing’’ (Winsberg 2003, p. 106).
The relation between simulation and experimentation is seen by epistemologists in
different ways. The so-called ‘‘numerical experiments’’ that underlie simulations are lik-
ened to laboratory experiments, insofar as they can represent the system under study, with
the possibility of varying parameters and testing theoretical hypotheses, as well as in the
type of results that arise (data sets that have to be organized and interpreted). Indeed, some
people see no difference between experiments and simulation (for ex., Hughes 1999;
Humphreys 2004, Norton and Suppe 2001). ‘‘Simulation modeling is just another form of
experimentation, and simulations are nothing other than models of data’’ (Norton and
Suppe 2001, p. 92). According to these authors, simulations imitate the systems that are of
interest, making it possible to perform experiments on them, in the same way as on any
other experimental objectives. In this case, the physical object experimented upon is the
computer.
Other authors, however, consider that simulations are not comparable to experiments, as
they lack ‘‘materiality’’ (Guala 2005; Morgan 2003). Parker (2009), guided by Hacking’s
definition of an experiment as a research activity into a system to see how the interesting
properties of that system change, considers that the problem is not materiality, given that
what is relevant for the justification of certain inferences on systems is the similarity that
may arguably exist between the system on which the experiment is based and the target
system. In the case of simulations, the intervention is on the computer program the
parameters of which are modified, and about which arguments are advanced in support of
its similarity with the target system.
In fact, simulations and experiments have many points in common, one of which is error
management. As previously discussed, errors in simulations arise from the transformation
of continuous into discrete equations and from the conversion of the mathematical struc-
ture of the model into a computationally feasible structure.11 Researchers working with
simulations have to learn to appraise, in the same way as experimental scientists, the
classes of error that can appear. ‘‘Precision, accuracy, error analysis and calibration are
concepts that we typically associate with experimentation and not with theorizing, but they
are also very much a part of the vocabulary of the simulationist’’ (Winsberg 2010, p. 43).12
In addition, simulations and instruments share similar calibration processes, such as their
use in situations in which the result is known, or, the evaluation of their reliability, by
reproducing the results with other instruments. In the case of simulations this evaluation is
done by trying to achieve similar results from the algorithms constructed in a different
way. The history of simulations for Winsberg (2010) is very similar to the history of
scientific instruments: an evolving set of techniques, practices and circumstances that
mature over time, which are refined when more precise and reliable techniques are needed,
in a process that gives them further credibility that is not exclusively dependent on their
theoretical grounding.
However, there is a fundamental difference: computer experiments are based on sym-
bols and digits; there is no direct contact with the world. Like experimentation they also
11
Durán (2013, pp. 107–108) divides the systematic errors in simulations into three kinds: physical errors
(related to the malfunctioning of any physical component of the computer), logical errors (related to coding
errors or a part of a faulty compiler or a computer language, leading to instabilities in the behavior of the
computer program) and representational errors (the most common ones, located at the level of the mathe-
matical model or the specification, as, for example, a grid too big for precise results, bad approximations,
unacceptable mean square errors, etc.).
12
The preferential vocabulary among those that use simulations is full of experimental metaphors.

123
Epistemological Issues Concerning Computer Simulations 907

require manipulation, but not the fact that experiments constitute ‘‘the acid test’’ of the
world over theories (Guillemot 2010). In other words, experiments, when performed, even
when guided by theory, apply in systems over which we neither have any control nor know
how they will function. And, in that sense, they allow us to test theories. However,
simulations, when performed, apply in systems that we have purposely created, selecting
some data or laws to the detriment of others. As Turkle stressed (2009, p. 40): ‘‘An
experiment, in ideal terms, turns to nature ready to be surprised. But if experiments are
done ‘‘in simulation’’, then, by definition, nature is presumed to be ‘‘known in advance’’,
for nature would need to be embedded in the program’’.
In a more simplified way: in the experimental sciences, there is theory and then
experimentation. The experimental results are confronted with theoretical calculations.
Obviously, approximations or ‘‘calculations’’ (Hacking 1983) or models are necessary, in
order to relate them, in the broadest sense discussed earlier (Cartwright 1999), but both
processes are the result of two types of different yet interconnected practices, in two types
of communities with specific objectives and techniques (Galison 1987).
In computer modeling, the hypothesis to be tested (for example, modifying a parameter)
and the numerical experiments are in a continuum: in order to verify a hypothesis, it has to
be transformed into algorithms and inserted in the computational model, which then
performs the simulation. But it is a virtual experiment, which produces no objective facts,
even though it increases the range of explorable domains (Guillemot 2010). Therefore,
some authors point out that experiments have a superior epistemological status with regard
to simulations, because of their greater potential for the validation of their results (Morgan
2003). Winsberg, however, argues that one may not speak of epistemological superiority,
but rather of priority: the experiments have the crucial role of testing theories, hypotheses
and models (Winsberg 2010, p. 71), which is not possible with simulations, as it is nec-
essary to have both theoretical and experimental knowledge of system dynamics to con-
struct them.13 In other words, it is assumed that several characteristics of the system that
one wants to learn are already known, in order to construct the computational models.
Accordingly, various sociologists and historians (e.g. Rohrlich 1991; Kaufmann and
Smarr 1993; Galison 1996; Dowling 1999) have argued that simulations are a completely
new scientific activity. Even though they share the manipulation of equations with theory
and the way that algorithms are manipulated with experimentation, which produce long
data sets that have to be interpreted, ‘‘the resulting bricolage creates a marginalized nether
land that was at once nowhere and everywhere on the usual methodological map’’ (Galison
1996, p. 20).
(d) Simulations and the objective of the explanations
Another relevant epistemological question has been put forward by people that suggest that
the mass use of simulations has implied or is implying a change in the meaning and the
objective of explanations (Johnson and Lenhard 2011).
Since the scientific revolution of the eighteenth century, knowledge of a system came to
be practically equivalent to knowledge of what might happen next, that is predicting
unknown events and objects. When those events and objects were observed, the mathe-
matical model was considered valid and its future predictions were taken seriously
(Johnson and Lenhard 2011). So, many natural philosophers came to consider that the

13
Theoretical knowledge to which we refer does not necessarily imply having a well-established theory of
the phenomenological dynamics of the system of interest, but having some knowledge about its dynamics, as
in the case of simulations in the area of social sciences.

123
908 I. M. Greca et al.

mathematical models were true representations of the mechanisms of physical systems. In

this sense, nobody maintains that simulations have an ontological similarity to the systems
that they imitate (Johnson and Lenhard 2011), given that simulations depend on computer
mechanisms that have no real counterpart, as previously discussed. What computational
models do is to produce a predictive response, not a mimetic model of the causal mech-
anisms of the real phenomenon. In addition, while traditional mathematical models had the
objective of producing transparent causal explanations, computer models are opaque: what
happens inside the computational model is not clear and, in some cases, the reason for the
coherence between the resulting simulation and reality is not wholly explainable.
Johnson and Lenhard (op. cit., p. 194) point out that given the epistemic opacity and the
lack of ontological references, it is not possible to apply the argument of the ‘‘non-miracle’’
of scientific realism—if there were a coincidence between the results of the mathematical
model and the experimental results, it would be miraculous if there were not some sort of
truth in the model—to the predictions of simulations. In the case of simulations, it is
possible to implement a variety of different models in the computer that ‘‘fit’’ the exper-
imental data, as there are multiple possible routes towards coherence. These authors
therefore speak about the emergence of a ‘‘culture of prediction’’ based on the mass use of
simulation in scientific practice, where simulations would privilege recurrent experimen-
tation on computational models and the evaluation of visual results, instead of the slower
and more laborious work on mathematical models more ‘‘in tune’’ with real systems. The
scientific strategy would, in comparison with traditional practice, be more exploratory,
evolutionary, adaptive, provisional and highly visual. Turkle (1995, 2009) noted the dif-
ferent cultures that may be defined among scientists that use computers, which she clas-
sified as a culture of calculation and a culture of simulation. The first is modern for this
author and is characterized by linearity, logic and depth, with the aim of explaining,
unpacking, reducing and clarifying its outcomes. In opposition, the postmodern culture of
simulation is fluid, decentered, and opaque and the search for mechanisms and depth is
seen as futile.
Nevertheless, it is important to discuss why, in many areas of research, a prediction is
sufficient in itself as a scientific goal. Let us remember our previous discussion about the
limited analytical access we have to the vast majority of natural phenomena, although
knowing, in cases, the differential equations that underlie them. In the case of complex
systems, the introduction of finite element simulations provided a vast expansion for
exploring dynamically sensitive regions. However, this access comes at a cost, in so far as
it is only local regions, larger than any finite element of the whole vector field flow, that
can be explored at any one time. We are not certain whether all significant non-linear
features have been caught or, even, whether non-linear features are in fact hidden inside
these finite elements; an uncertainty which forces us to conduct an exhaustive search. As
Hooker (2011, p. 892) stressed, these searches produce immense data sets of multiple
dimensions, often sparsely distributed and expressing subtle interrelations across data
dimensions, space and time.
This situation makes it necessary to use methods to search into the data. Today, there
are many global stochastic simulations capable of revealing important patterns. Many of
these methods, known as model-free pattern extraction methods (among which, neural
networks), do not rely on dynamical approximation or model interpretation at all; they are
in fact methods for data-driven prediction without model- focused explanation. The
interesting point here is that these non-parametric simulations provide more accurate
predictions than others ‘‘more classical’’ simulations provide. And in the case of poor
goodness-of-fit tests for the selection of one among a variety of computational models,

123
Epistemological Issues Concerning Computer Simulations 909

which is quite a common situation in complex systems, prediction is underlined as a

scientific goal that is in itself sufficient.
(e) The validity of knowledge generated by simulations
Among various epistemologists, historians and sociologists investigating the use of
simulations encounter a recurrent theme: the problem of how to validate the knowledge
that they generate. Given the way of constructing the simulations, which are not
exclusively derived from theory, how is it possible to evaluate the veracity of their results?
The problem of validation is even more important if we consider that in relation to climate
change, a theme that ranks high on the international political and diplomatic agenda, the
majority of the results that drive these agendas emerge from climate modeling and from
simulations of future climate change. However, both in climate science as in other areas,
there are few studies on the validation14 and the evaluation of simulations through
comparison with observational data (Guillemot 2010).
Küppers and Lenhard (2005, p. 1) set out the problem of validation based on the
example of Félix Krull, a character from a book of Thomas Mann, who so perfectly
simulates the symptoms of a neuronal illness that he deceives the doctors and is released
from military service. As pointed out earlier, the mathematical models described by dif-
ferential equations are considered ‘‘true’’ representations, when their predictions are ver-
ified. In that sense, simulations are not ‘‘true’’ representations, as they depend on their
capability to ‘‘imitate’’ the operation of a real system. How, therefore, to evaluate the
knowledge produced by a simulation, even though it imitates or even perfectly reproduces
the results of the real world? The problem resides in this difference between representation
and appropriate imitation.
Scientists working with simulations employ different strategies, in order to argue the
reliability of their results (Winsberg 2010): they argue the theoretical foundation of the
model, the robustness of mathematical techniques used to transform the equations of the
model into algorithms, the ‘‘calibration’’ of the simulation compared with what is known of
the phenomenon (directly with experimental results, or indirectly, on the basis of the
analysis or by comparison with other simulations), and they argue that the system responds
as expected when the parameters are modified, or that it can reproduce some basic relations
that are predicted by more phenomenological laws or theories.
In relation to the first two strategies, the simulations, even in the majority of the
practices of the natural sciences, are not exclusively drawn from theory, for which reason
they can not be validated by theoretical arguments (Küppers and Lenhard 2005). On the
other hand, the translation of the mathematical model into algorithms, which can be run on
a computer, often implies introducing artificial, non-realistic effects, in order to overcome
numerical instabilities that emerge from the particular methods that are used. So, the
validity of a simulation can not be assessed in relation to the validity of the mathematical
model on which it is based. In addition, particular numerical methods can fail under
strongly non-linear conditions that are of interest to simulate (Winsberg 2010).The other
strategies are also used in experimental science, however as simulation and experimen-
tation can not be equated, it is not very clear that the procedural aspects of these strategies
are sufficient for their validation.

14
In a recent work, Durán (2013) points that, although neither software nor hardware can be fully verified
nor validated, researchers are developing methods for reducing the possibility of errors in order to increase
the credibility of the model.

123
910 I. M. Greca et al.

The early simulations of atmospheric dynamics represent one example that clearly
shows how a simulation can behave in an appropriate way and be considered reliable by
the scientific community, even though it is not structurally precise (Küppers and Lenhard
2005, pp. 3–5). The first atmospheric simulation was developed by Norman Phillips in
1955, on the basis of six fundamental equations, which reproduced atmospheric circulation
quite well. However, it had a problem of instability: the simulation was stable for only a
few weeks. This problem was overcome in 1966s, by Akio Arawaka, for whom imitation
was more important than a precise calculation of the solution. Arawaka used the same
basic equations, but also a computational trick: he replaced the Jacobi operator, resulting
from the fundamental equations that described the temporal variation, by another that he
had purposely constructed, in order to support effective imitation. In addition, he intro-
duced other assumptions, in order to guarantee the stability of the solution, which con-
tradicted experience and physical laws, such as for example that kinetic energy in the
atmosphere is conserved.
There is still one further question related to validation. As happens with the other
computer programs, the public availability of simulation codes has increased over recent
years, as well as the availability of commercial codes. This availability allows researchers
to use computational models that ‘‘stem from’’, or have the structure of others already
accepted by the scientific community, nevertheless, their validity has neither been cor-
roborated, nor analyzed, nor studied (Turkle 2009; Sundberg 2010a). It is interesting to
note that, even if up until a few years ago the development of computer codes had a certain
status within the academic field, given their current availability, this activity is now, in
general, considered a waste of time. Sundberg (2010b) has studied this question with
astrophysicists and meteorologists that work with simulations. In her case study, it is
standard practice among doctoral students from important research institutes in both dis-
ciplines to ‘‘fine tune’’ an existing code, rather than a careful and detailed examination of it
all.
In summary, there are various factors that can affect the reliability of the results of a
simulation, at least insofar as we have, up until now, understood the validity of a physical
model. It may be possible to imitate a phenomenon when errors of a different sign are
cancelled. At times, it is necessary to incorporate contradictory hypotheses to imitate the
real behavior of the systems; the commercial software convert the code into ‘‘black boxes’’
and at times a software and its simulations are validated in the absence of empirical
verification, by sharing methods with other verified computational models.
Does this mean that the simulations and emergent knowledge of them are unreliable? If
scientific realism is adopted as a point of view, the simulations would have to be true and
correct representations of the phenomenon or system that they simulate and this, as we
have seen, is not possible to do. However, the problem may be overcome by assuming a
pragmatic position in relation to reliability, a position that separates reliability from truth,
reducing the fundamental arguments of scientific realism, that success implies truth. A
simulation can be, in the terms as defined here, highly reliable without even approaching
the truth. On this point, Winsberg (2010, p. 133) has made it clear that a simulation is
reliable when results are obtained that fit in well with the network of knowledge that is held
on the system (theoretical knowledge of the system, previously accepted experimental or
observational data, analytical results with paper and pencil and intuitive physics) and, in
addition, it is able to produce successful predictions. This is what Suárez (1999) refers to as
instrumentally reliable models. Along the same lines, Humphreys talks about ‘‘selective
realism’’—the aim of the simulation is to represent the real system only up to a prede-
termined degree of realism; an increase in that reality is almost always sacrificed, so that it

123
Epistemological Issues Concerning Computer Simulations 911

may be mathematically and computationally manipulated. In fact, scientists tend to think in

this way when working with simulations (Humphreys 2004).

3 Simulations in Science Education

3.1 The Uses of Simulations in Science Education

Simulations that facilitate the learning of sciences in schools are considered one of the
most effective modes of computer-assisted instruction for science subject areas and have
been studied for over three decades (Bayraktar 2002). According to a recent comprehen-
sive and critical review of their usefulness in science education (Smetana and Bell 2012),
simulation appears to be as efficient, if not more so, than other more traditional practices to
promote learning about concepts, conceptual change and the development of procedural
abilities. Like any other educational tool, they are of course dependent on the ways in
which they may be used. In this sense, the importance of the teacher in providing guidance
and support (Smetana and Bell 2012) is continually stressed in the literature. Teachers,
however, do not appear to be sufficiently well prepared, it being helpful to provide them
with opportunities to unpack to unpack how these new approaches and learning tools
would benefit their teaching (Waight et al. 2013).
Many of the simulations that are used in science education are related to experimen-
tation. Thus various authors maintain that simulations can replace real experiments, in
those cases where the latter are very costly, dangerous, rapid and complex (Doerr 1997;
Hsu and Thomas 2002; Hennessy 2006). In these simulations, the students manipulate
variables, observe results and analyze tables, graphs and equations to identify and to
describe the data (e.g. Confrey and Doerr 1994; Thornton 1987). In addition, in the case of
more complex phenomena, simulations allow students to simplify them through the iso-
lation and manipulation of one variable at a time, which helps their understanding of causal
relations (Doerr 1997; de Jong and van Joolingen 1998).
However, the benefits of simulations are increasingly prescribed for the development of
inquirer-based and learner–centered instruction as they appear to assist students in their
understanding of the various phenomena and natural processes through the construction
and evaluation of different hypotheses, obtaining rapid feedback, which involves them in
active problem-solving process (White and Frederiksen 1998; Hargrave and Kenton 2000).
Moreover, simulations make it possible to work with multiple representations, at the same
time, and on the same screen, allowing the integration of various forms of scientific
representation. Related to this representational characteristic, simulations allow the
‘‘visualization’’ of processes at a microscopic level, such as for example in chemistry,
enabling the development of molecular-level thinking and at the same time allowing their
visualization at a microscopic level and the establishment of relations with macroscopic
observations (Özmen et al. 2009; Liu et al. 2008).
We should, however, distinguish between scientific simulations, which we have dis-
cussed in earlier sections, and what is understood by simulations in science education.
Unlike simulations for scientific study, educational simulations may be defined as
‘‘interactive learning environments in which a model simulates characteristics of a system,
depending on actions made by the student’’ (de Vries and Huisman in Kirschner and
Huisman 1998). The fundamental difference is that, whereas scientific simulations seek a
better understanding of complex phenomena and processes based on the construction of
computational models, using well known theories (or theoretical considerations) as well as

123
912 I. M. Greca et al.

other sources of information, educational simulations aim to develop an understanding of

the underlying model and from it, the theoretical principles among the students. Education
simulations therefore require some type of guidance to be effective (de Jong and van
Joolingen 1998). In a comprehensive review of the literature on virtual lab and simulation
software for grades 6–12, Scalise et al. (2011) considered that the simulations used in
science education can be grouped into two main categories: virtual laboratories and sim-
ulations of phenomena, the first used for on-screen simulation of the experiments that are
traditionally performed in real school laboratories as part of the science curriculum and the
second used to model a system or a process.
Two different forms of working may be distinguished in educational simulations (Doerr
1997) related to the agent that designs the simulation. On the one hand, exploration, in
which students explore the consequences of their actions within the boundaries of a
simulation created by the teacher or an expert to represent the knowledge of a specific
content of an ideal model. The model is, therefore, already constructed. Used in this way, it
allows the exploration of questions such as ‘‘what happens if…’’, by modifying parameters
(Doerr 1997). The majority of applications available through Internet are designed for
exploration and this was, in fact, the most common method in the investigations that were
reviewed by Smetana and Bell (2012) and Scalise et al. (2011).
Another form of working with simulations that is closer to the scientific paradigm is
called ‘modeling’ in the literature on science education. To do so, programs are used that
allow students to create their own simulations. In the construction of the model that must
underlie the simulation, students should identify the relevant variables, quantify the rela-
tions between them, and evaluate their validity; exercises which allow them to express
their own concepts and to learn from the processes used to represent them. There are two
types of programs that allow this kind of work. On the one hand, programs such as
STELLA15 (e.g., Mandinach 1989; Steed 1992) enable students to model systems and
automatically generate the mathematical relations from the qualitative relations that the
students establish in graphic form. The modeling system of STELLA is relatively simple
and it has powerful options for visualization and for the creation of graphic interfaces, as it
is principally visual, although this may mean that certain models may become ‘‘tiresome’’.
One of its advantages is the facility with which the students may modify the structure of the
model that is generated and its parameters, after examining its output (Doerr 1996).
Other programs such as Modellus and Easy Java Simulator allow students to model
phenomena, through the use of equations with ordinary derivatives, and to combine the
representation of mathematical objects with analytical, analogic, and graphic representa-
tions (Teodoro 2002). These types of programs require a relatively high level of mathe-
matical knowledge, appropriate only for the final years of secondary education. Programs
such as Interactive Physics and Physlets (Christian and Belloni 2001), which are some-
where between the two types of uses, allow students to generate simulations in which they
can select and measure different variables. These simulations are not completely
straightforward; the student selects the physical situation to be studied—for example, in a
collision between two bodies, the student can choose the shape of the bodies, the values of
the relevant variables, etc.—but once the initial conditions are set, the objects are governed
by Newton’s laws, so the student will neither write down the equations, nor the relations
that govern the movement of the objects.

15
Other programs that allow modelling in K12 science teaching are LOGO (Papert 1980); Model-It
(Jackson et al. 1996); ThinkerTools (White 1984); and BioLogicaTM (Buckley et al. 2004).

123
Epistemological Issues Concerning Computer Simulations 913

In general, research on educational simulations have fundamentally centered on cog-

nitive aspects associated with its usage, while epistemological questions have received
little attention. For example, looking in depth at all the articles analyzed in the review by
Smetana and Bell (2012), only two of them (Hennessy 2006; Marshall and Young 2006)
point, in a roundabout way, to an aspect with epistemological implications relating to the
authority that students project onto the computer, which may lead to misconceptions if they
understand animations and images of abstract concepts in a literal way. Beyond this
review, Kirschner and Huisman (1998) looked at the importance of discussing with stu-
dents the restricted validity of simulations that were generated when working with a
program that used some 500 input variables. Depending on the values assigned to those
same variables, the simulation could, therefore, generate models that did not represent real
processes. A further work that discusses this question is the review by Doerr (1996) on
various investigations using the STELLA program. Doerr pointed out that the program can
run a simulation regardless of the significance of the data inputs, although validation of the
model with experimental data is fundamental. However, in relation to this point, Doerr
recognized that there have been no investigations that examine the systematic evaluation of
the models by students or how students understand that some features are intentionally
excluded from a model or modified in a significant way, while others are artifacts of the
simulation program or are invalid assumptions.
It could be argued that the dearth of investigations on the epistemological aspects of
simulations is related to the aforementioned differences between scientific simulations and
those used for didactic purposes. However, as we shall see next, some relevant episte-
mological points should still be discussed with the simulations designed by experts. If they
have not been discussed, it is probably because up until now, in the field of investigation in
science education, and for some time in epistemology, they have only been considered a
tool, and not a specific scientific method.

3.2 Model-Based Learning and Experimental Work Can be Enriched

with Epistemological Discussions About Simulations

As simulations are today a fundamental part of daily scientific tasks, working with sim-
ulations in science classes at all educational levels is seen to be as important as experi-
mental work and problem solving (Scalise et al. 2011) and the discussion of some
epistemological questions associated with them should certainly be introduced into any
didactic approach In this section, arguments will be advanced to support the contribution of
the epistemological notions, debated in earlier sections, in two areas in which simulations
are already used as a training tool.
(a) Model-Based Learning (MBL)
The central role claimed by models in science education has been widely debated over
recent years, and is probably the area with the greatest discussion of epistemological
questions (among others, Gilbert et al. 2000; Nola 2004; Izquierdo and Adúriz-Bravo 2003;
Halloun 2007). Among the different epistemological approaches, the semantic vision of
models (van Fraassen 1980; Gière 1999) has reached a certain preeminence in science
education, as it would appear to be the best adapted to teaching in this area, because it
highlights the role of models as an active element in the process of knowledge generation
and construction (Sensevy et al. 2008; Koponen 2007). In this area, many investigations
and didactic proposals based on MBL use simulations, whether using tools for the creation
of models (e.g. Stratford et al. 1998; White and Frederiksen 1998; Sins et al. 2009) or

123
914 I. M. Greca et al.

simulations developed by experts for scaffolding scientific understanding (e.g. de Jong

et al. 1999; Monaghan and Clement 1999; Windschitl 2001). When we think of models as
vehicles for the generation of new ideas, the role of simulations, as used by scientists, is
fundamental. However, a close analysis of the research that use simulations in MBL shows
that the characteristics within the modeling that relate to the role of simulations are not
emphasized and simulations are, in general, only used as a tool that can facilitate learning
about modeling (e.g., Windschitl 2001; Sins et al. 2009; Waight et al. 2013) and not as a
particular form of modeling in itself.
Considering simulations as a peculiar form of modeling implies approaching with
students the way in which computational models emerge and their intricate form of relating
theory, experimental data, intuition and tricks, to give the mathematical models a com-
putationally viable form. As indicated earlier, students should manage to understand that
some features are intentionally excluded from a model or modified in a significant way, in
as much as they are either artifacts of the simulation program or non-valid assumptions
(Doerr 1996). This necessarily leads to the problem of the validation of simulations; a
much more complex problem, as we have seen, than the validation of traditional models in
science. This is especially relevant, as it is precisely the notions that students acquire about
this point, which will permit them to evaluate the results of scientific simulations.
Perhaps one way to implement these insights into the classroom could be through the
use of the history of science, for example, related with the origins of simulations and also
with the development of some paradigmatic computational models, such as those used in
climate science, that are well documented. These examples can show students how sci-
entists try to understand complex phenomena and the different ways they use to obtain
computational tractable models and useful simulations that enable good predictions.
Another way may be to introduce the epistemological questions through students’
reflection on what they think the simulation is doing, and how it is executed, and imple-
mented. Certainly, the use of simulations that allow students to design the models them-
selves, to determine the relations between their respective elements, to generate the
dynamic of these relations, and to understand their consequences, is the most appropriate
way of integrating this discussion, as many programs are available at present, even for
students that have mastered neither mathematical techniques nor programming skills.
However, even in the case of closed simulations, prepared by experts, students have to be
able to understand that mode of scientific production. Without this discussion, it may
happen that students eventually consider that when they use simulations, the model is only
for ‘‘playing with values and formulas on the computer model so that they match the
measured values’’ (quoted from a student of Sins et al. 2009, p. 1219).
On another point, the discussion of the veracity or reliability of computational models
inevitably leads us to revise the realist notion that both students and teachers tend to adopt
in science education. As discussed earlier, it would appear necessary, in the case of
simulations, to adopt more restricted stances, such as, for example, the selective realism of
Humphreys (2004) or the notion of instrumental reliability proposed by Suárez (1999).
Koponen (2007, p. 765) also discussed this point, although it was not directly related to
simulations, but to the semantic view of models. He suggested that if we accept this
viewpoint, which implies changing our point of view on models from only a mere rep-
resentation of phenomena to the use of models as matchmaking tools, then realism in
relation to the truth of the theories is challenged. This does not mean abandoning the
positions of a realist, but of its discussion in parallel with other complementary views
(Koponen 2007; Wong and Hodson 2009).

123
Epistemological Issues Concerning Computer Simulations 915

According to the results of the investigation by Wong and Hodson (2009), scientists are
used to playing with both realism and instrumentalism, depending on their immediate
proposals. For this reason, these authors argue that students should be taught to be realistic
critics, capable of evaluating the status of a particular piece of knowledge and using it
either in a realistic or in an instrumental way, in accordance with the demands of each
situation. Doing otherwise, would not give an authentic vision of science. Simulations, in
this sense, have a lot to contribute in a specific way to this discussion.
(b) Experimental work
Much of the research into simulations refers to its use as a complement or substitution of
experimental work. Simulations are used in relation to experimentation in two different
ways: the majority use them as a substitute for real experiments (e.g., Hsu and Thomas
2002; Huppert and Lazarowitz 2002; Kaput 1995; Tao and Gunstone 1999; Zacharia
2003), but also as ‘‘dry laboratories’’ (Kirschner and Huisman 1998), to achieve specific
cognitive skills, such as analysis, synthesis and evaluation (e.g., de Jong and van Joolingen
1998; Plass et al. 2012; Kukkonen et al. 2013).
However, many professors and researchers consider that experimentation with physical
manipulatives is the only real, ‘‘hands-on’’ experimental activity, excluding from this
categorization, virtual laboratories, which simulate experiments (Zacharia et al. 2008;
Klahr et al. 2007). In fact, the absence or the low frequencies of some typical activities of
real laboratories have been noted—peer participation, the analysis of sources of error and
the comprehension of the complexity and the ambiguity of experimental work—among
professors using simulations as experimental activity (Crippen et al. 2012). It is interesting
to note that the America’s Lab Report (National Research Council 2006) concluded that
the lack of available studies on these points left the review committee unable to draw
conclusions on the benefit of using virtual labs. Nevertheless, the results of the investi-
gations into virtual laboratories would appear to contradict this view, because the cognitive
gains of students with either one or the other form of experimentation seem to be equal (for
two well documented reviews on this topic see Triona and Klahr 2003 and de Jong et al.
2013). Some researchers consider that they could even be better than the laboratories with
physical manipulatives, because they are easier and students can work with the data in a
controlled environment and can exercise control over their variables, which is not gen-
erally achieved in standard laboratories (Klahr et al. 2007; Baser 2006).
De Jong et al. (2013, pp. 305–306) stressed that although physical and virtual labora-
tories can achieve similar student-related objectives, related to stimulating their interest in
science, their conceptual understanding, and their inquiry skills, each of these different
kinds of laboratories also have certain specific traits. Physical laboratories allow students to
develop specific laboratory skills, such as practical skills or the ones related with a care-
fully planning of experiments. Virtual laboratories, moreover, are not only less time
consuming, both in the setup as in their output results, but they also allow to adapt reality
in the sense of making unobservable phenomena visible and of removing confusing details.
Along similar lines, Chinn and Malhotra (2002, pp. 207–208), in their discussion of
epistemologically authentic lines of inquiry, suggest that the advantages of a simulated
experiment, in addition to cases in which hands-on activities can not be carried out, is that
it allows the realization of: (a) experiments at the theoretical level of the mechanism, in
other words, the study of theoretical entities that could not otherwise be ‘‘visualized’’(for
example, simulations at a molecular or genetic-molecular level); (b) different types of
experiments on one single case; (c) relatively complex scientific designs. Despite these
advantages, Chinn and Malhotra (2002) highlighted that simulations demystify, in an

123
916 I. M. Greca et al.

artificial way, a large part of the disorder in the natural world and, moreover, students can
not evaluate different models or variables other than those that are programmed, a point
also discussed by Scheckler (2003). Addressing these issues, several authors have began to
propose the benefits of combining both, physical and virtual experiments (for example,
Zacharia et al. 2008; Jaakkola et al. 2011; de Jong et al. 2013).
Thus, the research in science education points to different features of simulations and
experimentation, mainly related with the acquisition or improvement of different skills.
But, as pointed out in the second section, experimentation and simulations are not con-
sidered equivalent in the epistemological discussion that has developed around simula-
tions. As we have seen, experimental work cannot be substituted by simulations, as
experiments continue to be the acid test of all theories relating to the world, having in this
sense an epistemological priority, as they are the only mode of scientific production that
allows us to evaluate hypotheses, models, and theories. Of course, this will never mean that
students should not test hypotheses or models in a virtual lab, a methodology that appears
to be successful at achieving several cognitive goals, or that they should not use simula-
tions in science labs, given its ubiquity in science. But, although this discussion is yet to
arise in science education literature, it appears relevant that students should be aware of the
epistemological differences and similarities between simulations and experimentation.
Perhaps one way of informing students about the epistemological questions relating to
experimentation and simulation might be through the development of research projects that
blend both, in a similar way as scientists do, in which the experimental work serves, on the
one hand, as a database for the generation of computational models and, in turn, as a means
for their validation and as a source of new experiments. This echoes recent discussions on
investigation into modeling in science education as an activity that can not be separated
from experimentation (Sensevy et al. 2008; Koponen 2007). As suggested by Sensevy et al.
(2008, p. 432) ‘‘On the one hand, theory translates Nature itself into semiotic systems
registering the observations (power of the abstract) and, on the other hand, the phenomena
produced by the instruments reach some sort of autonomy that gives feedback on the
theory (power of the concrete)’’. In experimental work, combining hands—on activities
and simulations we could help students achieve not only a more accurate notion of current
scientific practice but also that semi-autonomous vision of models that allow them to
connect with the measurable properties of the phenomena, something which has yet to be
fully discussed in science education.
So, more frequent use of the computer in the science laboratory should take place, in
order to provide a more authentic image of contemporary research in sciences, both for the
purposes of data collection, manipulation, control and presentation as discussed by Wong
and Hodson (2009), and for the analysis of these data through the development or
manipulation of certain sorts of simulations.

4 Final Remarks

Certainly, simulations have not impacted on science in a ‘‘conceptual’’ form to the same
extent as certain theories—consider the impact of relativity or quantum mechanics -, which
is one of the reasons noted by epistemologists for the scant attention paid to them.
However, they have impacted and in a strident manner, on the scientific practice and its
application in all scientific areas. They have allowed scientists to greatly expand the
knowledge about the world in such a way that our current understandings of all complex

123
Epistemological Issues Concerning Computer Simulations 917

and above all non-linear systems that characterize the vast majority of phenomena in the
universe have in fact been made possible only by the use of simulations.
However, simulations, as recent epistemological studies have shown, have proven to be
much more than a fantastic tool for calculus, but a new form of scientific production.
Simulations therefore stir up classical epistemological notions, such as the ones presented
in this paper: the modification of the role of differential equations as the principal tool of
physics; the nature of modeling and its relation with existing theories; the classic division
between scientific theory and empirical methods; the prediction as a self-sufficient goal in
some areas of science to the detriment of explanation; and the need to assume a more
pragmatic position in relation to reliability. Although we have discussed these issues in
very general terms, the areas that are denominated complex systems, which spread to every
branch of science, constitute a privileged arena in which to study these special features
provided by simulations in scientific method.
In this paper we have tried to review these issues, arising from a large, diverse and quite
recent literature, highlighting them for research in science education, where we have quite
a good body of knowledge on the use of simulations for cognitive and motivational goals
(although not yet used as widely in teaching as would be desirable), but that have not yet
addressed them from an epistemological point of view. And, although there is today a wide
ranging literature in science education on experimentation and models, which urges
researchers and teachers to address their most relevant epistemological features, as they
support the training of scientifically literate citizens, the same has not yet happened with
simulations. If we are to take the recommendations of the various curricular reforms
seriously that urge us to provide students with an opportunity for authentic inquiry, which
‘‘refers to the research that scientist actually carry out’’ (Chinn and Malhotra 2002, p. 177),
we must include simulations in science education, but not, as seems to happen, uncritically
and only as a tool.
Nevertheless, the way to introduce the epistemological problems discussed throughout
this paper in the secondary and university education and in teacher training is an open
question. We have only addressed, very generally, some possibilities in two research areas,
experimental work and model based learning, that we think are key points to approach
them. As stated above, research into these issues is at a primitive stage and more research is
needed to offer specific suggestions. In fact, these questions form part of an ongoing
research project in which one of the authors of this paper is currently engaged.
Recently, it has been argued that science education should not center exclusively on the
teaching scientific concepts, but also on metaconcepts. Snir et al. (2003) defend the idea
that the notion of model should be applied in this way, so that students know what it is and
how it is used in science, as by ‘‘doing so we are letting students take part in the process of
scientific developments the way scientist do, even though it is in a structured and limited
environment designed by us specifically for these purposes’’ (Snir et al. 2003, p. 803). We
consider that the same should apply to the case of simulations. So, in the same way as
scientific theories and skills associated with scientific development are proposed as
indispensable elements for the general training of students, we should include simulations
not only as a tool to motivate students and to facilitate their learning, but with a similar
status to the inclusion of experiments or modeling in the content of natural sciences. The
training of scientifically literate citizens today requires them to know about the potential
and the limitations of simulations, because simulations are connected to a large part of our
emergent knowledge of the world—i.e. the practical application of theories to the world
that is, as citizens, what interests us most.

123
918 I. M. Greca et al.

Acknowledgments The authors would like to thank the anonymous reviewers for their valuable comments
and suggestions to improve the quality of the paper.

References

Baser, M. (2006). Effects of conceptual change and traditional confirmatory simulations on pre-service
teachers’ understanding of direct current circuits. Journal of Science Education and Technology,
15(5–6), 367–381.
Bayraktar, S. (2002). A meta-analysis of the effectiveness of computer-assisted instruction in science
education. Journal of Research on Technology in Education, 34(2), 173–188.
Benzi, M. (2009). The history of numerical linear algebra. SIAM conference on applied liner algebra.
Brukner, C., & Zeilinger, A. (2005). Quantum physics as a science of information. In A. Elitzur, S. Dolev, &
N. Kolenda (Eds.), Quo vadis quantum mechanics?. Berlin: Springer.
Buckley, B. C., Gobert, J. D., Kindfield, A. C. H., Horwitz, P., Tinker, R. F., Gerlits, B., et al. (2004).
Model-based teaching and learning with BioLogicaTM: What do they learn? How do they learn? How
do we know? Journal of Science Education and Technology, 13, 23–41.
Cartwright, N. (1983). How the laws of physics lie. Oxford, New York: Clarendon Press.
Cartwright, N. (1999). Models and the limits of theory: quantum Hamiltonians and the BCS model of
superconductivity. In M. S. Morgan & M. Morrison (Eds.), Models as mediators: perspective on
natural and social science. Cambridge: Cambridge University Press.
Chinn, C. A., & Malhotra, B. A. (2002). Epistemologically authentic inquiry in schools: A theoretical
framework for evaluating inquiry tasks. Science Education, 86, 175–218.
Christian, W., & Belloni, M. (2001). Physlets: Teaching physics with interactive curricular material. Upper
Saddle River, NJ: Prentice Hall.
Confrey, J., & Doerr, H. (1994). Student modelers. Interactive Learning Environments, 4(3), 199–217.
Crippen, K. J., Archambault, L. M. & Kern, C. L. (2012). The nature of laboratory learning experiences in
secondary science online. Research in Science Education. Online First doi:10.1007/s11165-012-9301-6.
de Jong, T., Linn, M. C., & Zacharias, C. Z. (2013). Physical and virtual laboratories in science and
engineering education. Science, 340(6130), 305–308.
de Jong, T., Martin, E., Zamarro, J.-M., Esquembre, F., Swaak, J., & van Joolingen, W. R. (1999). The
integration of computer simulation and learning support: an example from the physics domain of
collisions. Journal of Research in Science Teaching, 36, 597–615.
de Jong, T., & Njoo, M. (1992). Learning and instruction with computer simulation: Learning processes
involved. In E. de Corte, M. C. Linn, H. Mandl, & L. Verschaffel (Eds.), Computer-based learning
environments and problem solving (pp. 411–427). Berlin: Springer.
de Jong, T., & van Joolingen, W. R. (1998). Scientific discovery learning with computer simulations of
conceptual domains. Review of Educational Research, 68(2), 179–201.
Doerr, H. (1996). STELLA: Ten years later. A review of the literature. International Journal of Computers
for Mathematical Learning, 1(2), 201–224.
Doerr, H. (1997). Experiment, simulation and analysis: An integrated instructional approach to the concept
of force. International Journal of Science Education, 19(3), 265–282.
Dowling, D. (1999). Experimenting on theories. Science in Context, 12, 261–273.
Durán, J. M. (2013). Explaining simulated phenomena. A defense of the epistemic power of computer
simulations. PhD Thesis. Stuttgart University.
Fox Keller, E. (2003). Models, simulations and ‘‘computer experiments’’. In H. Radder (Ed.), The philos-
ophy of scientific experimentation. Pittsburgh: Pittsburgh University Press.
Frigg, R., & Reiss, J. (2009). The philosophy of simulation: Hot issues or same old stew? Synthese, 169,
593–613.
Galison, P. (1987). How experiments end. Chicago: University of Chicago Press.
Galison, P. (1996). Computer simulation and the trading zone. In P. Galison & D. J. Stump (Eds.), The
disunity of science: boundaries, contexts, and power. Stanford: Stanford University Press.
Gière, R. (1999). Science without laws. Chicago: University of Chicago Press.
Gilbert, J. K., Boulter, C. J. & Elmer, R. (2000). Positioning models in science education and in design and
technology education. In: J. K. Gilbert, & C. J. Boulter (Eds.), Developing models in science education
(pp. 3–17). Dordrecht: Kluwer.
Guala, F. (2005). The methodology of experimental economics. Cambridge: Cambridge University Press.

123
Epistemological Issues Concerning Computer Simulations 919

Guillemot, H. (2010). Connections between simulations and observation in climate computer modeling.
Scientist0 s practices and ‘‘bottom-up epistemology’’ lessons. Studies in History and Philosophy of
Modern Physics, 41, 242–252.
Hacking, I. (1983). Representing and intervening. Cambridge: Cambridge University Press.
Halloun, I. (2007). Mediated modeling in science education. Science & Education, 16(7), 653–697.
Hargrave, C. P., & Kenton, J. M. (2000). Pre instructional simulations: Implications for science classroom
teaching. Journal of Computers in Mathematics and Science Teaching, 19(1), 47–58.
Harper, W. L. (2011). Isaac Newton’s scientific method: Turning data into evidence about gravity and
cosmology. Oxford: Oxford University Press.
Hennessy, S. (2006). Integrating technology into teaching and learning of school science: A situated per-
spective on pedagogical issues in research. Studies in Science Education, 42, 1–50.
Hooker, C. A. (2011). Introduction to philosophy of complex systems. Part B. In C. A. Hooker (Ed.),
Philosophy of complex systems, Vol. 10 (Handbook of the Philosophy of Science) Oxford: Elsevier.
Hsu, Y.-S., & Thomas, R. A. (2002). The impacts of a web-aided instructional simulation on science
learning. International Journal of Science Education, 24, 955–979.
Hughes, R. I. G. (1999). The Ising model, computer simulation, and universal physics. In M. S. Morgan &
M. Morrison (Eds.), Models as mediators. Cambridge: Cambridge University Press.
Humphreys, P. (2004). Extending ourselves: Computational science, empiricism, and scientific method.
New York: Oxford University Press.
Huppert, J., & Lazarowitz, R. (2002). Computer simulations in the high school: Students’ cognitive stages,
science process skills and academic achievement in microbiology. International Journal of Science
Education, 24, 803–821.
Izquierdo, M., & Adúriz-Bravo, A. (2003). Epistemological foundations of school science. Science &
Education, 12(1), 27–43.
Jaakkola, T., Nurmi, S., & Veermans, K. (2011). A comparison of students’ conceptual understanding of
electric circuits in simulation only and simulation-laboratory contexts. Journal of Research in Science
Teaching, 48(1), 71–93.
Jackson, S., Stratford, S. J., Krajcik, J. S., & Soloway, E. (1996). Making system dynamics modeling
accessible to pre-college science students. Interactive Learning Environments, 4(3), 233–257.
Johnson, A., & Lenhard, J. (2011). Toward a new culture of prediction. Computational modeling in the era
of desktop computing. In A. Nordmann, et al. (Eds.), Science transformed? Debating claims of an
epochal break. Pittsburgh: University of Pittsburgh Press.
Kaput, J. J. (1995). Creating cyberetic and psychological ramps from the concrete to the abstract: Examples
from multiplicate structure. In D. N. Perkins & J. L. Schwartz (Eds.), Software goes to school:
Teaching for understanding with new technologies (pp. 130–154). London: Oxford University Press.
Kaufmann, W. J., & Smarr, L. L. (1993). Supercomputing and the transformation of science. New York:
Scientific American Library.
Kirschner, P., & Huisman, W. (1998). Dry laboratories in science education; computer-based practical work.
International Journal of Science Education, 20, 665–682.
Klahr, D., Triona, L. M., & Williams, C. (2007). Hands on what? The relative effectiveness of physical
versus virtual materials in an engineering design project by middle school children. Journal of
Research in Science Teaching, 44, 183–203.
Koponen, I. T. (2007). Models and modelling in physics education: A critical re-analysis of philosophical
underpinnings and suggestions for revisions. Science & Education, 16, 751–773.
Kukkonen, J., Kärkkäinen, S. Dillon, P., & Keinonen, T. (2013). The effects of scaffolded simulation-based
inquiry learning on fifth-graders’ representations of the greenhouse effect. International Journal of
Science Education. First on-line, 2013. doi:10.1080/09500693.2013.782452.
Küppers, G. & Lenhard, J. (2005). Validation of simulation: Patterns in the social and natural sciences.
Journal of Artificial Societies and Social Simulation, 8(4)3. http://jasss.soc.surrey.ac.uk/8/4/3.html.
Lederman, N. G., Abd-El-Khalick, F., Bell, R. L., & Schwartz, R. (2002). Views of nature of science
questionnaire (VNOS): Toward valid and meaningful assessment of learners’ conceptions of nature of
science. Journal of Research in Science Teaching, 39(6), 497–521.
Lenhard, J. (2010). Computation and simulation. In R. Frodeman, J. T. Klein, & C. Mitcham (Eds.), The
oxford handbook on interdisciplinarity (pp. 246–258). Oxford: Oxford University Press.
Liu, H.-C., Andre, T., & Greenbowe, T. (2008). The impact of learner’s prior knowledge on their use of
chemistry computer simulations: A case study. Journal of Science Education and Technology, 17,
466–482.
Lunetta, V. N., Hofstein, A., & Clough, M. (2007). Learning and teaching in the school science laboratory:
An analysis of research, theory, and practice. In N. Lederman & S. Abel (Eds.), Handbook of research
on science education (pp. 393–441). Mahwah, NJ: Lawrence Erlbaum.

123
920 I. M. Greca et al.

Mandinach, E. (1989). Model-building and the use of computer simulation of dynamic systems. Journal of
Educational Computing Research, 5(2), 221–243.
Marshall, J. A., & Young, E. S. (2006). Preservice teachers’ theory development in physical and simulate
environments. Journal of Research in Science Teaching, 43(9), 907–937.
McComas, W. F., & Olson, J. K. (1998). The nature of science in international standards documents. In W.
F. McComas (Ed.), The nature of science in science education: Rationales and strategies (pp. 41–52).
Dordrecht, The Netherlands: Kluwer.
Mindell, D. (2002). Between human and machine: Feedback, control, and computing before cybernetics.
Baltimore, MD: Johns Hopkins University Press.
Monaghan, J. M., & Clement, J. J. (1999). Use of a computer simulation to develop mental simulations for
learning relative motion concepts. International Journal of Science Education, 21(9), 921–944.
Morgan, M. (2003). Experiments without material intervention: Model experiments, virtual experiments and
virtually experiments. In Hans. Radder (Ed.), The philosophy of scientific experimentation
(pp. 216–235). Pittsburgh: University of Pittsburgh Press.
Morgan, M. S., & Morrison, M. (Eds.). (1999). Models as mediators: Perspective on natural and social
science. Cambridge: Cambridge University Press.
National Research Council. (2006). America’s lab report: Investigations in high school science. Wash-
ington, DC: National Academy Press.
Nola, R. (2004). Pendula, models, constructivism and reality. Science & Education, 13, 349–377.
Norton, S., & Suppe, F. (2001). Why atmospheric modeling is good science. In C. Miller & P. Edwards
(Eds.), Changing the atmosphere: Expert knowledge and environmental governance. Cambridge: MIT
Press.
Ören, T. I. (2011a). The many facets of simulation through a collection of about 100 definitions. SCS M&S
Magazine, 2(2), 82–92.
Ören, T. I. (2011b). A critical review of definitions and about 400 types of modeling and simulation. SCS
M&S Magazine, 2(3), 142–151.
Osborne, J. F., Ratcliffe, M., Collins, S., Millar, R., & Duschl, R. (2003). What ‘ideas-about-science’ should
be taught in school science? A Delphi study of the ‘expert’ community. Journal of Research in Science
Teaching, 40(7), 692–720.
Özmen, H., Demircioğlu, G., & Coll, R. (2009). A comparative study of the effects of a concept mapping
enhanced laboratory experience on Turkish high school students’ understanding of acid-base chem-
istry. International Journal of Science and Mathematics Education, 7, 1–24.
Papert, S. (1980). Mindstorms: Children, computers and powerful ideas. New York: Basic Books Inc.
Parker, W. S. (2009). Does matter really matter? Computer simulations, experiments, and materiality.
Synthese, 169(3), 483–496.
Plass, J. L., Milne, C., Homer, B. D., Schwartz, R. N., Hayward, E. O., Jordan, T., et al. (2012). Investigating
the effectiveness of computer simulations for chemistry learning. Journal of Research in Science
Teaching, 49, 394–419.
Rohrlich, F. (1991). Computer simulations in the physical sciences. In A. Fine, M. Forbes, & L. Wessels
(Eds.), Proceedings of the 1990 philosophy of science association biennial meetings (pp. 507–518).
East Lansing, MI: Philosophy of Science Association.
Scalise, K., Timms, M., Moorjani, A., Clark, L., Holtermann, K., & Irvin, P. S. ( 2011). Student learning in
science simulations: Design features that promote learning gains. Journal of Research in Science
Teaching, 48(9), 1050–1078.
Scheckler, R. K. (2003). Virtual labs: A substitute for traditional labs? International Journal of Develop-
mental Biology, 47, 231–236.
Sensevy, G., Tiberghien, A., Santini, J., Laube, S., & Griggs, P. (2008). An epistemological approach to
modeling: Cases studies and implications for science teaching. Science Education, 92(3), 424–446.
Sins, P. H. M., Savelsbergh, E. R., van Joolingen, W., & van Hout-Wolters, B. (2009). The relation between
students’ epistemological understanding of computer models and their cognitive processing in a
modeling task. International Journal of Science Education, 31(9), 1205–1229.
Sismondo, S. (1999). Models, simulations and their objects. Science in Context, 12(2), 247–260.
Smetana, L. K., & Bell, R. L. (2012). Computer simulations to support science instruction and learning: A
critical review of the literature. International Journal of Science Education, 34(9), 1337–1370.
Snir, J., Smith, C. L., & Raz, G. (2003). Linking phenomena with competing underlying models: A software
tool for introducing students to the particulate model. Science Education, 87, 794–830.
Steed, M. (1992). STELLA, a simulation construction kit: Cognitive process and educational implications.
Journal of Computers in Mathematics and Science Teaching, 11, 39–52.

123
Epistemological Issues Concerning Computer Simulations 921

Stratford, S. J., Krajcik, J., & Soloway, E. (1998). Secondary students’ dynamic modeling processes:
Analyzing, reasoning about, synthesizing, and testing models of stream ecosystems. Journal of Science
Education and Technology, 7(3), 215–234.
Suárez, M. (1999). The role of models in application of scientific theories: Epistemological implications. In
M. S. Morgan & M. Morrison (Eds.), Models as mediators. Cambridge: Cambridge University Press.
Sundberg, M. (2010a). Organizing simulation code collectives. Science Studies, 23, 37–57.
Sundberg, M. (2010b). Cultures of simulations vs. cultures of calculations? The development of simulation
practices in meteorology and astrophysics. Studies in History and Philosophy of Modern Physics, 41,
273–281.
Tao, P., & Gunstone, R. (1999). The process of conceptual change in force and motion during computer
supported physics instruction. Journal of Research in Science Teaching, 36, 859–882.
Teodoro, V. D. (2002). Modellus: Learning physics with mathematical modeling. PhD Thesis.
Thornton, R. K. (1987). Tools for scientific thinking-microcomputer-based laboratories for physics teaching.
Physics Education, 22, 230–238.
Triona, L., & Klahr, D. (2003). Point and click or grab and heft: Comparing the influence of physical and
virtual instructional materials on elementary school students’ ability to design experiments. Cognition
and Instruction, 21, 149–173.
Turkle, S. (1995). Life on the screen. Identity in the age of the internet. New York: Touchstone.
Turkle, S. (2009). Simulation and its discontents. Cambridge: MIT Press.
van Fraassen, B. C. (1980). The scientific image. Oxford, UK: Oxford University Press.
Waight, N., Liu, X., Gregorius, R., Smith, E. & Park, M. (2013). Teacher conceptions and approaches
associated with an immersive instructional implementation of computer-based models and assessment
in a secondary chemistry classroom. International Journal of Science Education, On line First.
doi:09500693.2013.787506
Wheeler, J. A. (1990). Information, physics, quantum: The search for links. In W. Zurek (Ed.), Complexity,
entropy, and the physics of information. Redwood City, CA: Addison-Wesley.
White, B. Y. (1984). Designing computer games to help physics students understand Newton’s laws of
motion. Cognition and Instruction, 1(1), 69–108.
White, B., & Frederiksen, J. (1998). Inquiry, modeling, and metacognition: Making science accessible to all
students. Cognition and Instruction, 16(1), 3–118.
Windschitl, M. (2001). Using simulations in the middle school: Does assertiveness of dyad partners
influence conceptual change? International Journal of Science Education, 23(1), 17–32.
Winsberg, E. (1999). Sanctioning models: The epistemology of simulation. Science in Context, 12, 2.
Winsberg, E. (2003). Simulated Experiments: Methodology for a virtual world. Philosophy of Science, 70,
105–125.
Winsberg, E. (2010). Science in the age of computer simulation. Chicago: The University of Chicago Press.
Wong, S. L., & Hodson, D. (2009). From the horse’s mouth: What scientists say about scientific investi-
gation and scientific knowledge. Science Education, 93, 109–130.
Zacharia, Z. C. (2003). Beliefs, attitudes, and intentions of science teachers regarding the educational use of
computer simulations and inquiry-based experiments in physics. Journal of Research in Science
Teaching, 40, 792–823.
Zacharia, Z. C., & Anderson, O. R. (2003). The effects of an interactive computer-based simulations prior to
performing a laboratory inquiry-based experiments on students’ conceptual understanding of physics.
American Journal of Physics, 71, 618–629.
Zacharia, Z. C., Olympiou, G., & Papaevripidou, M. (2008). Effects of experimenting with physical and
virtual manipulatives on students’ conceptual understanding in heat and temperature. Journal of
Research in Science Teaching, 45(9), 1021–1035.

123