9JANUARY/FEBRUARY 2003
In this article, the authors first indicate the range of purposes and
the variety of settings in which design experiments have been con-
ducted and then delineate five crosscutting features that collectively
differentiate design experiments from other methodologies. Design
experiments have both a pragmatic bent—“engineering” particular
forms of learning—and a theoretical orientation—developing domain-
specific theories by systematically studying those forms of learning
and the means of supporting them. The authors clarify what is in-
volved in preparing for and carrying out a design experiment, and in
conducting a retrospective analysis of the extensive, longitudinal data
sets generated during an experiment. Logistical issues, issues of mea-
sure, the importance of working through the data systematically, and
the need to be explicit about the criteria for making inferences are
discussed.
In this short article, we draw on our collective experience ofconducting design experiments for a range of purposes in va-riety of settings in order to delineate prototypical characteris-
tics of the methodology and to describe what is involved in
conducting a design experiment. Although the term design exper-
iment is most closely associated with Brown (1992) and Collins
(1992), pedagogical design has informed the development of
theories of instruction for well over a century. Prototypically, de-
sign experiments entail both “engineering” particular forms of
learning and systematically studying those forms of learning within
the context defined by the means of supporting them. This de-
signed context is subject to test and revision, and the successive
iterations that result play a role similar to that of systematic vari-
ation in experiment.
Design experiments are conducted to develop theories, not
merely to empirically tune “what works.” These theories are rel-
atively humble in that they target domain-specific learning
processes. For example, a number of research groups working in
a domain such as geometry or statistics might collectively de-
velop a design theory that is concerned with the students’ learn-
ing of key disciplinary ideas in that domain. A theory of this type
would specify successive patterns in students’ reasoning together
with the substantiated means by which the emergence of those
successive patterns can be supported. This emphasis on theories
reflects the view that the explanations and understandings in-
herent in them are essential if educational improvement is to be
a long-term, generative process. Design experiments ideally re-
sult in greater understanding of a learning ecology—a complex,
Design Experiments in Educational Research
by Paul Cobb, Jere Confrey, Andrea diSessa, Richard Lehrer, and Leona Schauble
Educational Researcher, Vol. 32, No. 1, pp. 9–13
interacting system involving multiple elements of different types
and levels—by designing its elements and by anticipating how
these elements function together to support learning. Design ex-
periments therefore constitute a means of addressing the com-
plexity that is a hallmark of educational settings. Elements of a
learning ecology typically include the tasks or problems that stu-
dents are asked to solve, the kinds of discourse that are encour-
aged, the norms of participation that are established, the tools
and related material means provided, and the practical means by
which classroom teachers can orchestrate relations among these
elements. We use the metaphor of an ecology to emphasize that
designed contexts are conceptualized as interacting systems
rather than as either a collection of activities or a list of separate
factors that influence learning. Beyond just creating designs that
are effective and that can sometimes be affected by “tinkering to
perfection,” a design theory explains why designs work and sug-
gests how they may be adapted to new circumstances. Therefore,
like other methodologies, design experiments are crucibles for
the generation and testing of theory.
Design experiments are pragmatic as well as theoretical in ori-
entation in that the study of function—both of the design and of
the resulting ecology of learning—is at the heart of the method-
ology. This emphasis on function in a realized context holds for
all design experiments even though they are conducted in a di-
verse range of settings that vary in both type and scope:
• One-on-one (teacher-experimenter and student) design ex-
periments in which a research team conducts a series of
teaching sessions with a small number of students. The aim
is to create a small-scale version of a learning ecology so that
it can be studied in depth and detail (Cobb & Steffe, 1983;
Steffe & Thompson, 2000).
• Classroom experiments in which a research team collaborates
with a teacher (who might be a research team member) to as-
sume responsibility for instruction (Cobb, 2000; Confrey &
Lachance, 2000; Gravemeijer, 1994).
• Preservice teacher development experiments in which a re-
search team helps organize and study the education of pro-
spective teachers (Simon, 2000).
• In-service teacher development studies in which researchers
collaborate with teachers to support the development of a
professional community (Lehrer & Schauble, 2000; Stein,
Silver, & Smith, 1998).
• School and school district restructuring experiments in
which a research team collaborates with teachers, school ad-
ministrators, and other stakeholders to support organiza-
tional change (Confrey, Bell, & Carrejo, 2001).
Crosscutting Features of Design Experiments
We identify five crosscutting features that apply to these di-
verse types of design experiments. First, the purpose of design

experimentation is to develop a class of theories about both the
process of learning and the means that are designed to support that
learning, be it the learning of individual students, of a classroom
community, of a professional teaching community, or of a school
or school district viewed as an organization. We interpret processes
of learning broadly to encompass what is typically thought of as
knowledge, but also the evolution of learning-relevant social prac-
tices and even constructs such as identity and interest. When we
look across these diverse types of design experiments, the means
for supporting learning encompass the affordances and con-
straints of material artifacts, teaching and learning practices, and
policy levers (e.g., performance-based pay), as well as other forms
of mediation that might, for example, include the negotiation of
domain-specific norms—such as what counts as a “good” scien-
tific question in a classroom (Wertsch, 1998). It is apparent from
this broad view of means of support that it is often necessary to
document learning ecologies at multiple levels (Kelly & Lesh,
2000). In the case of an in-service teacher development experi-
ment, for example, the research team might focus simultaneously
on the norms and practices of a professional teaching commu-
nity, the participating teachers’ pedagogical reasoning and in-
structional practices, and their students’ reasoning in a particular
content domain. A challenge that arises in such cases is therefore
that of coordinating multiple levels of analysis.
Although, as a practical matter, a design experiment is con-
ducted in a limited number of settings, it is apparent from the
concern for theory that the intent is not merely to investigate the
process of supporting new forms of learning in those specific set-
tings. Instead, the research team frames selected aspects of the en-
visioned learning and of the means of supporting it as paradigm
cases of a broader class of phenomena. In the case of a one-on-
one design experiment, for example, the broader theoretical goal
might be to develop a psychological model of the process by
which students develop a deep understanding of particular math-
ematical ideas, together with the types of tasks and teacher prac-
tices that can support that learning. In the case of a school district
restructuring experiment, the theoretical goal might be to de-
velop an interpretive framework that explicates the relations be-
tween teachers’ instructional practices and the institutional
settings in which teachers develop and refine their practices. In
these and other types of design experiments, the initial design
formulated when preparing for an experiment and the new form
of learning it is designed to support are viewed as instances of
broader classes of phenomena, thereby opening them to theoret-
ical analysis.
The second crosscutting feature is the highly interventionist
nature of the methodology. Design studies are typically test-beds
for innovation. The intent is to investigate the possibilities for
educational improvement by bringing about new forms of learn-
ing in order to study them. Consequently, there is frequently a
significant discontinuity between typical forms of education (these
could be studied naturalistically) and those that are the focus of
a design experiment. The design developed while preparing for
an experiment draws on prior research and attempts to cash in
the empirical and theoretical results of that research. The process
of engineering the forms of learning being studied provides the
research team with a measure of control when compared with
purely naturalistic investigation. Furthermore, in attempting to
EDUCATIONAL RESEARCHER10
support a specified form of learning, the researcher is more likely
to encounter relevant factors that contribute to the emergence of
that form and to become aware of their interrelations.
By its very nature, the study of phenomena as complex as learn-
ing ecologies precludes complete specification of everything that
happens. It is therefore all the more important to distinguish in
the specification of the design between elements that are the tar-
get of investigation and those that may be ancillary, accidental, or
assumed as background conditions. For example, in a study of
children’s mathematical development, classroom norms of justi-
fication might be assumed as background and the emphasis
placed instead on conceptual development. Alternatively, the de-
velopment of norms might serve as a primary target of investiga-
tion (e.g., Yackel & Cobb, 1996). The use of prior research to
both specify a design and justify the differentiation of central and
ancillary conditions is central to the methodology.
The third crosscutting feature builds on the first two: Design
experiments create the conditions for developing theories yet
must place these theories in harm’s way. Thus, design experiments
always have two faces: prospective and reflective. These two faces
are familiar to all empirical scientists, but the forms they take in
design experiments are somewhat specialized. On the prospec-
tive side, designs are implemented with a hypothesized learning
process and the means of supporting it in mind in order to ex-
pose the details of that process to scrutiny. An equally important
objective is to foster the emergence of other potential pathways
for learning and development by capitalizing on contingencies
that arise as the design unfolds.
On the reflective side, design experiments are conjecture-driven
tests, often at several levels of analysis. The initial design is a con-
jecture about the means of supporting a particular form of learn-
ing that is to be tested. During the conduct of the design study,
however, more specialized conjectures are typically framed and
tested. For example, during a classroom design experiment, an
initial conjecture about a prospective interaction between charac-
teristics of tasks as they are realized in the classroom and student
responses may be tested. If this conjecture is refuted, alternative
conjectures can be generated and tested.
Together, the prospective and reflective aspects of design ex-
periments result in a fourth characteristic, iterative design. As
conjectures are generated and perhaps refuted, new conjectures
are developed and subjected to test. The result is an iterative de-
sign process featuring cycles of invention and revision. Of course,
to design iteratively demands systematic attention to evidence
about learning and, as we later describe, this often involves the
parallel development of measures sensitive to the changing ecology
of learning. The intended outcome is an explanatory framework
that specifies expectations that become the focus of investigation
during the next cycle of inquiry.
The fifth feature of design experimentation again reflects its
pragmatic roots: Theories developed during the process of ex-
periment are humble not merely in the sense that they are con-
cerned with domain-specific learning processes, but also because
they are accountable to the activity of design. The theory must
do real work. General philosophical orientations to educational
matters—such as constructivism—are important to educational
practice, but they often fail to provide detailed guidance in or-
ganizing instruction. The critical question that must be asked is

whether the theory informs prospective design and, if so, in pre-
cisely what way? Rather than grand theories of learning that may
be difficult to project into particular circumstances, design ex-
periments tend to emphasize an intermediate theoretical scope
(diSessa, 1991) that is located between a narrow account of a spe-
cific system (e.g., a particular school district, a particular classroom)
and a broad account that does not orient design to particular con-
tingencies. For example, the claim that invented representations
are good for mathematics and science learning probably has some
merit, but it specifies neither the circumstances in which these rep-
resentations might be of value nor the learning processes involved
and the manner in which they are supported. In contrast to most
research methodologies, the theoretical products of design ex-
periments have the potential for rapid pay-off because they are
filtered in advance for instrumental effect. They also speak di-
rectly to the types of problems that practitioners address in the
course of their work.
Preparing for a Design Experiment
As we have emphasized, a crucial issue to be addressed when one
conducts any type of design experiment is that of clarifying its
theoretical intent: What is the point of the study? For illustrative
purposes, we will exemplify this aim for the case of classroom de-
sign experiments, although it applies equally to other kinds of de-
sign experiments, such as those that focus on school districts or
larger educational systems, out-of-school learning contexts, work-
places, and the like.
Most classroom design experiments are conceptualized as cases
of the process of supporting groups of students’ learning in a par-
ticular content domain. The theoretical intent, therefore, is to
identify and account for successive patterns in student thinking
by relating these patterns to the means by which their develop-
ment was supported and organized. However, different classroom
design experiments may set their focus on different constellations
of issues. For example, one might focus on the relation between
classroom norms or standards for mathematical or scientific ar-
gumentation, and student learning. Another study might em-
phasize the ways in which diversity in students’ prior experiences
can be capitalized upon as a resource to ensure that all students
have access to significant disciplinary ideas.
In addition to clarifying the theoretical intent of the experi-
ment, the research team must also specify the significant discipli-
nary ideas and forms of reasoning that constitute the prospective
goals or endpoints for student learning. This usually involves
drawing on and synthesizing the prior research literature to
identify central organizing ideas for a domain (e.g., the notion
of distribution as a central idea for statistical analysis, Lehrer &
Schauble, 2002; McClain, Cobb, & Gravemeijer, 2000). In the
process of specifying instructional goals, a research team fre-
quently proposes an alternative conception of a domain (e.g.,
typicality, center, variation, and relative frequency as character-
istics of the single, overarching idea of distribution rather than as
a set of discrete curriculum topics). Another source of disconti-
nuity in curricular specification is that new resources, such as
computer software, might be developed to support the envi-
sioned form of learning. Yet another is that evolving theories of
knowledge informed by analyses of how knowledge is used in
complex settings may implicate different performances as in-
dicative of deep understanding (diSessa, in press), such as the
ability to innovate procedures in small-group design episodes in
contrast to individual application of a given procedure.
As part of the process of preparing for a classroom design ex-
periment, the research team also specifies its assumptions about the
intellectual and social starting points for the envisioned forms of
learning. To achieve the instructional agenda, the team identifies
current student capabilities, current practices, and other resources
on which it might be able to build. In relatively well-researched do-
mains, the team can draw on the literature to develop conjectures
about students’ initial interpretations and understandings. How-
ever, in less researched areas, the team typically needs to conduct
pilot work to document these understandings and, thus, the con-
sequences of students’ prior instructional histories. In the course
of this pilot work, the team might also develop new methods for
assessing aspects of student reasoning that need to be documented,
given the purposes of the experiment.
When the conjectured starting points, elements of a trajectory,
and prospective endpoints have been specified, the challenge is
to formulate a design that embodies testable conjectures about
both significant shifts in student reasoning and the specific means
of supporting those shifts. In well-studied domains, the research
team might have a reasonable level of confidence in some of their
conjectures. However, in others, where knowledge is less devel-
oped, the team regards its conjectures as speculative and begins
the experiment with the expectation that many will prove to be
unviable. Even then, the advantage of explicating conjectures at
the outset is that they orient the research team to identify and ac-
count for successive patterns in student thinking.
The means of supporting student learning are usually con-
strued broadly, consistent with an acknowledgement of the com-
plexity of teaching and learning. This relatively encompassing
view of the means of support implies that the research team must
generate multiple forms of data to adequately document the
learning ecology. Because we have focused on classroom learn-
ing, it is important to emphasize that the focus and means of doc-
umentation vary with the institutional setting. For example, in a
science museum, the built environment may constitute an im-
portant means for focusing visitor attention, communicating
how to initiate the activity at hand, and framing reasonable in-
terpretations of the outcome.
Conducting a Design Experiment
As we have indicated, a primary goal for a design experiment is
to improve the initial design by testing and revising conjectures
as informed by ongoing analysis of both the students’ reasoning
and the learning environment. The size of the research team and
the expertise of the members vary depending on the type and
purpose of the experiment. For example, it might be feasible for
a single researcher who conducts the teaching sessions and a
graduate assistant who records the sessions to carry out a one-on-
one design experiment. In the case of a classroom design experi-
ment conducted in collaboration with a teacher in a relatively
well-researched domain, the team might include the teacher, a
researcher, and two graduate assistants. The crucial determinant
in any type of design experiment is that the team collectively has
the expertise to accomplish the functions associated with devel-
oping an initial design, conducting the experiment, and carrying
11JANUARY/FEBRUARY 2003

out a systematic retrospective analysis. Thus, in an experiment
with a relatively broad scope that encompasses multiple class-
rooms and attends to the organizational setting at the school and
district level, two or more researchers might be involved whose
combined expertise includes the design and analysis of classroom
learning environments, professional teaching communities, and
schools as institutions.
Regardless of the type of experiment, strong involvement of
the leaders of the research team is essential. The locus of that par-
ticipation is again defined by the scope and purpose of the ex-
periment. Accordingly, if the scope is district reform, the team
leaders will need to be actively involved in nested levels of activ-
ity, extending from policy forums (such as school board or con-
tent standards meetings) to professional development settings to
classrooms. If the scope is more constrained, for example, to a
single classroom, the team leaders may be present in the class-
room as the design unfolds.
There are at least four important functions that require ongo-
ing direct engagement in the research setting and the associated
planning and interpretive activities. These functions collectively
compose researcher leadership in the conduct of design experi-
ments. First, a clear view of the anticipated learning pathways
and the potential means of support must be maintained and
communicated within the research team, even while responding
to contingency. Maintaining such an overview can be a daunt-
ing challenge, even for an experienced researcher. Second, the ex-
tended nature of most design experiments calls for the cultivation
of ongoing relationships with practitioners. These relationships
are sustained by the negotiation of a shared enterprise, which is
typically developed over the long haul as lead researchers consis-
tently demonstrate their personal commitment. Third, because
of the reciprocal emphasis on learning and the means that sup-
port it, design researchers seek to develop a deep understanding
of the ecology of learning—not simply to facilitate logistics, but
because this understanding is a theoretical target for the research.
As part of the process of refining conjectures, subtle and often
unanticipated cues need to be recognized and drawn into a larger
perspective. Fourth and finally, regular debriefing sessions are the
forum in which past events are interpreted and prospective events
are planned for. These sessions are the sites where the intelligence
of the study is generated and communicated.
One of the distinctive characteristics of the design experiment
methodology is that the research team deepens its understanding
of the phenomenon under investigation while the experiment is
in progress. It is therefore important that the team generates a
comprehensive record of the ongoing design process. It is stan-
dard procedure in most engineering disciplines to keep records
to support the retrospective analysis of the experiment (Edelson,
2002). Accordingly, the research team may employ audio records
of meetings and logs to document the evolving conjectures, to-
gether with the observations that are viewed as either supporting
or questioning a conjecture.
In addition to self-consciously building and documenting the
design and its rationale, the team members, like all researchers,
have a responsibility for communicating what they learn in ways
that are open to public scrutiny. This implies a commitment to
generate data that support the systematic analysis of the phe-
nomenon under investigation. At a minimum, this entails the
generation of data on both learning and the means by which that
learning was generated and supported. In practice, achieving
these aims frequently requires the collection and coordination of
a complex array of data sources—for example, products of learn-
ing (such as student work); classroom discourse; body posture
and gesture; tasks and activity structures; patterns of social in-
teraction; inscriptions, notations, and other tools; and responses
to interviews, tests, or other forms of assessment. Because the
team often intends to use these data sources to track changes over
time, the task is further complicated by the need to collect ex-
tended records of each type. Technological support for the gen-
eration of these forms of data (e.g., video cameras, sophisticated
audio-recording systems, mass electronic storage devices) enables
these efforts but also imposes its own challenges (e.g., the devel-
opment of tools and procedures for managing and analyzing
large quantities of data).
The team draws on a variety of data sources that may bear on
the broader phenomena framing any particular design experi-
ment. Consider, for example, an experiment in which the team
has framed the process of cultivating students’ interests in disci-
plinary ideas as an explicit focus of investigation. In this case, team
members might document the nature of students’ engagement
not only in the target classroom but also in out-of-school activi-
ties. Multiple sources of data ensure that retrospective analyses
conducted when the experiment has been completed will result in
rigorous, empirically grounded claims and assertions. Of course,
no data collection can be complete, and the revision of the data
collection procedures may be a part of the iterative process. As
with traditional experimental and quasi-experimental designs, the
viability of the conclusions drawn from data depends on the
soundness of the process that generated the data.
Attending to the process by which data are generated means
attending to the problem of measure. Much of the cleverness of
excellent design experiments resides in how the team handles is-
sues of measurement. An obvious point, although one that is
often overlooked, is that all measurements (even observations)
are indexes to constructs of interest, not the constructs themselves.
For example, consider all the decisions that must be made when
using video as data, even though the surface impression is one of
non-problematic capture (Hall, 2000). Measures are created, not
found, and decisions about the creation of measures are among
the most important made. An otherwise impeccable design will
produce no useful information about the phenomena of interest
if problems of construct validity are not successfully resolved.
Measures that are feasible to administer, and that provide precise
and reliable scores, may or may not adequately capture the phe-
nomenon of interest. Because design experiments need to gener-
ate results that do work with respect to subsequent cycles of design,
they focus on problems of construct validity.
Conducting Retrospective Analysis
An educational accomplishment is characterized by contingency
in which earlier events open up, enable, and also constrain the
events that follow. Accounting for this process requires an his-
torical or retrospective explanation, one that provides a trust-
worthy account of the process whereby a series of events—each
of which is local and contingent—can be seen as part of an emer-
gent and potentially reproducible pattern. For example, consider
EDUCATIONAL RESEARCHER12

13JANUARY/FEBRUARY 2003
a third-grade class working together to explore conjectures about
whether the volume of a plant’s canopy grows proportionally
over the plant’s life cycle. One might want to understand how
such a capability came to be. Producing an explanation of this
kind requires showing how the students’ earlier histories of learn-
ing (e.g., about geometric similarity, rates, and plants) bear on
the events under consideration. Doing so requires justifying both
selection (among all events) and rational reconstruction that fo-
cuses on issues of cause and relative importance of events in the
class’s unfolding history. For this reason, it is methodologically
advantageous to cultivate diverse points of view from members
of the research team. Diversity of expertise and backgrounds
among members of the research team can be an important re-
source for developing alternative interpretations, as can asking
different team members to assume primary responsibility for rep-
resenting particular perspectives during the analysis.
A central challenge in conducting retrospective analyses is to
work systematically through the extensive, longitudinal data sets
generated in the course of a design experiment so that the result-
ing claims are trustworthy. As part of this process, it is important
to be explicit about the criteria and types of evidence used when
making particular types of inferences so that other researchers can
understand, monitor, and critique the analysis. A primary aim
when conducting a retrospective analysis is to place the design ex-
periment in a broader theoretical context, thereby framing it as a
paradigm case of the more encompassing phenomena specified at
the outset. In this regard, retrospective analyses can be contrasted
with the analyses conducted while the experiment is in progress in
that the latter are typically oriented toward the goal of supporting
the learning of the participants. For example, in a classroom ex-
periment, the research team may, under the pressure of time, in-
tuitively and successfully modify aspects of its instructional design.
Retrospective analysis attempts to generate a coherent framework
that accounts for these effects, thus making it possible to anticipate
outcomes in future designs. In sum, retrospective analyses results
in situated accounts of learning that relate learning to the means
by which it can be supported and organized.
The situated nature of retrospective analyses is a strength of
the methodology, given the overall goal of engineering new forms
of learning and the tendency of “high” theory to pass over what
may be important details in an effort to paint phenomena in uni-
form terms. In particular, because the resulting accounts of learn-
ing are tied to the means by which it was generated, the design
team is always in a position to develop testable conjectures about
how those means of support and, thus, the instructional design
might be improved. “What works” is underpinned by a concern
for “how, when, and why” it works, and by a detailed specifica-
tion of what, exactly, “it” is. This intimate relationship between
the development of theory and the improvement of instructional
design for bringing about new forms of learning is a hallmark of
the design experiment methodology.
In summary, design experiments are extended (iterative), in-
terventionist (innovative and design-based), and theory-oriented
enterprises whose “theories” do real work in practical educational
contexts. Although design experiments share many individual
characteristics with other ways of conducting science in the ser-
vice of education, the constellation of crosscutting themes we
have identified distinguishes a genre of science with high promise
but also with a host of characteristic difficulties that researchers
need to manage effectively to achieve that promise.
NOTE
The authors contributed equally to the manuscript and are listed in al-
phabetical order.
AUTHORS
PAUL COBB is a professor of mathematics education at Vanderbilt Uni-
versity, Peabody College, Box 330, Nashville, TN 37203; paul.cobb@
vanderbilt.edu. His research interests include classroom instructional
design and analysis, the development of professional teaching commu-
nities, the institutional setting of teaching, and issues of diversity and
equity as they play out in the mathematics classroom.
JERE CONFREY is a professor at University of Texas, Austin, Depart-
ment of Curriculum and Instruction, SZB 518, Austin, TX 78712;
jere@mail.utexas.edu. Her research interests include cognition and mul-
tiplicative relations, functions and trigonometry, technology design,
and systemic reform.
ANDREA DISESSA is Chancellor’s Professor at University of California,
Berkeley, Graduate School of Education, 4647 Tolman Hall, Berkeley,
CA 94720; disessa@soe.berkeley.edu. His research interests include
conceptual and experiential knowledge in physics, and the design and
use of flexible, comprehensible computer systems for learning.
RICHARD LEHRER is a professor at Vanderbilt University, Depart-
ment of Teaching and Learning, Peabody College, Box 330, 166 Wyatt
Center, Nashville, TN 37203; rich.lehrer@vanderbilt.edu. His research
interests include the design of learning environments for developing an
understanding of mathematics and science.
LEONA SCHAUBLE is a professor at Vanderbilt University, Depart-
ment of Teaching and Learning, Peabody College, 1930 South Drive,
Nashville, TN 37203; leona.schauble@vanderbilt.edu. Her research in-
terests include cognitive development, especially the development of sci-
entific thinking and model-based reasoning.
Manuscript received August 23, 2002
Revisions received November 6, 2002
Accepted November 7, 2002