ArticlePDF Available

Abstract and Figures

Instruction abstracted from specific and concrete examples is frequently criticized for ignoring the context-dependent and perspectival nature of learning (e.g., Bruner, 196227. Bruner , J. S. 1962. On knowing: Essays for the left hand, Cambridge, MA: Harvard University Press. View all references, 196628. Bruner , J. S. 1966. Toward a theory of instruction, Cambridge, MA: Harvard University Press. View all references; Greeno, 199754. Greeno , J. G. 1997. On claims that answer the wrong questions. Educational Researcher, 26: 5–17. View all references). Yet, in the effort to create personally interesting learning contexts, cognitive consequences have often been ignored. To examine what kinds of personalized contexts foster or hinder learning and transfer, three manipulations of perspective and context were employed to teach participants Signal Detection Theory (SDT). In all cases, application of SDT principles was negatively impacted by manipulations that encouraged participants to consider the perspective of the signal detector (the decision maker in SDT situations): by giving participants active detection experience (Experiment 1), biasing them to adopt a first-person rather than third-person perspective (Experiment 2), or framing the task in terms of a well-known celebrity (Experiment 3). These contexts run the risk of introducing goals and information that are specific to the detector's point of view, resulting in sub-optimal understanding of SDT.
Content may be subject to copyright.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
Taylor & Francis Group, LLC
ISSN: 0737-0008 print / 1532-690X online
DOI: 10.1080/07370000802584539
Contextualization in Perspective
Ji Y. Son
University of California, Los Angeles
Robert L. Goldstone
Indiana University5
Instruction abstracted from specific and concrete examples is frequently criticized for ignoring the
context-dependent and perspectival nature of learning (e.g., Bruno, 1962, 1966; Greeno, 1997). Yet,
in the effort to create personally interesting learning contexts, cognitive consequences have often
been ignored. To examine what kinds of personalized contexts foster or hinder learning and transfer,
three manipulations of perspective and context were employed to teach participants Signal Detection 10
Theory (SDT). In all cases, application of SDT principles was negatively impacted by manipulations
that encouraged participants to consider the perspective of the signal detector (the decision maker
in SDT situations): by giving participants active detection experience (Experiment 1), biasing them
to adopt a first-person rather than third-person perspective (Experiment 2), or framing the task in
terms of a well-known celebrity (Experiment 3). These contexts run the risk of introducing goals and 15
information that are specific to the detector’s point of view, resulting in sub-optimal understanding
of SDT.
Cognition can be painted as both context-dependent and context-independent. On the context-
dependent side, problem solving is often easiest when framed by supportive, concrete contexts
(i.e., Baranes, Perry, & Stigler, 1989; Nisbett & Ross, 1980; Wason & Shapiro, 1971). Unfortu-
nately, there have also been many documented failures for such contextualized understandings
to generalize flexibly (e.g., Lave, 1988; Lave & Wenger, 1991; Nunes, Schliemann, & Carraher,
1993). Having wide experience with a range of contexts can help students recognize relevant
information for generalization (for a thorough treatment of this idea, see coordination classes,
diSessa & Wagner, 2005). But when only a limited exposure to contexts is available, strate-
gic decontextualizations, if successfully employed, can allow human learners to function across
contexts and extend prior learning to solve new problems. When symbolic representations such
as graphs, equations, or rules are acquired through time-consuming and effortful study, such
achievement has been shown to lead to flexible transfer to new situations and novel problems
(e.g., Bassok & Holyoak, 1989; Judd, 1908; Novick & Hmelo, 1994). Transfer of learning is
often critically important because the specific training domain is just one example of a much
Correspondence should be addressed to Ji Y. Son, UCLA Psych-Cognitive, BOX 951563, 1285 FH, Los Angeles, CA
90095-1563. E-mail:
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
deeper principle (Morris, Bransford, & Franks, 1977; Bransford & Schwartz, 1999; Thorndike
& Woodworth, 1901). For example, a teacher presents a case study of evolved coloration of
Manchester’s peppered moths not only for students to learn about these moths specifically, but
so that they will understand and apply natural selection, variation, and phenotype concepts when
they arise subsequently in scenarios involving alligators or bacteria.
Successful pedagogy recognizes the potentials and pitfalls of decontextualized abstractions in
the attempts to impart knowledge of general forms, stripped of situational details, actions, and
perspectives. The philosophy of decontextualization has historically been challenged because
abstract formalisms often seem unnaturally difficult for novices to grasp and learn (Bruner,
1962, 1966). Learning decontextualized representations can result in “inert” knowledge because
they are unconnected to particular contexts (Bransford, Franks, Vye, & Sherwood, 1989). For
example, statistical tests are useful abstract formulations for a wide variety of situations, but
students are often at a loss as to when to actually implement them beyond their classroom
examinations (Franks, Bransford, Brailey, & Purden, 1990; Schwartz & Martin, 2004; Schwartz,
Sears, & Chang, 2007). Abstract representations that are too separate from the rest of a student’s
knowledge will not promote the discovery of productive commonalities across diverse situations.
In a recent study exploring the tradeoffs between contextually grounded versus abstract (equation-
based) representations, Koedinger, Alibali, and Nathan (2008) found that for simple problems,
grounded word problems were solved better, but for complex problems, equations were solved
more accurately.
In the effort to bring contexts back into learning, the learning sciences have seen a proliferation
of what can be considered “contextualization, which can broadly be construed as how much
learning is embedded in a specific domain or situation. Many types of contextualizations are
aimed toward the goal of student-centered learning (McCombs & Whistler, 1997), to make what
is learned meaningful to the student. Personalizing learning through direct experience, perspec-
tive, and interests offers straightforward means of couching concepts in meaningful experiences.
In the richest form, this could mean full immersion in real-world problems in communities
of practitioners (i.e., Brown, Collins, & Duguid, 1989; specific instantiations include Gorman,
Plucker, & Callahan, 1998; Barab, Squire, & Deuber, 2000). A caveat is that these rich programs
are complex and difficult to incorporate immediately into traditional instructional contexts. Con-
sidering the potential gains in motivation and knowledge that may come through personalization,
it is worthwhile to consider how to personalize learning in traditional instruction as well. Micro-
manipulations of contextualization can be implemented easily, allowing them to be replicated
many times in the course of a teaching unit.
Empirical results of even minor contextualizations incorporating direct action, perspective, and
personal interest have been fruitful for both a basic understanding of cognition and pedagogical
inquiry. For example, when children manipulated objects and performed actions referred to
in text using physical objects, their reading comprehension improved (Glenberg, Guitierrez,
Levin, Japuntich, & Kaschak, 2004). Contextualizing abstract problems in personally relevant
or interesting situations (such as familiar schemas: Nisbett & Ross, 1980; Wason & Shapiro,
1971; or fantasy situations: Parker & Lepper, 1992) has also produced learning benefits. These
manipulations may have their effect by enhancing intrinsic motivation (Lepper, 1988), but the
implications may be much broader. Consider that human reasoning may be inherently grounded
in modality-specific, action-specific, and perspective-specific interactions between thinkers and
their environments, as embodied psychology (e.g., Barsalou, 1999; Glenberg, 1997) and situated
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
education (Greeno, 1997) suggest. In light of such views, perhaps personalization has more of an
effect than simply increasing interest and personal relevance. Personalized contexts may have an
effect on the content that is actually learned.
Although there may be motivational reasons to personalize, the research reported here explores
the possibility of counteracting cognitive implications. One important implication of personalized
contexts has always been this extreme possibility: if all learning is tied to specific contexts, the
possibility of transfer across domains and phenomena comes into question (e.g., Detterman &
Sternberg, 1993; Lave, 1988). After all, if we define transfer as thinking and reasoning across
contexts (e.g., Barnett & Ceci, 2002), knowledge must be decontextualized and abstracted (from
the Latin, Abstrahere, to pull away) from particulars in order to be transferred (see Reeves &
Weisberg, 1994).
The potential for transfer and generalization over a variety of situations provides compelling
reasons to understand how individuals might benefit from decontextualization. Discovering, un-
derstanding, and using deep principles across domains seem critical for students (see Anderson,
Greeno, Reder, & Simon, 1996 for a defense of this assumption) and have historically been com-
mon goals for educators (Klausmeier, 1961; Resnick, 1987). The benefits of decontextualization
may be most apparent on tests of transfer, but more generally, transfer has also been proposed as
a more sensitive indication of learning than other measures such as memory retrieval (Michael,
Klee, Bransford, & Warren, 1993; Schwartz & Bransford, 1998).
However, there are reasons to be skeptical of transfer as a pedagogical goal and decontextu-
alization as a strategy toward that goal. Evidence of life-to-school transfer failures such as the
mathematical successes of housewives in supermarkets (Lave, 1988) and Brazilian fishermen at
the fish market (Nunes, Schliemann, & Carraher, 1993; Schliemann & Nunes, 1990) coupled with
their inability to exhibit their mathematical knowledge in school settings could be used to argue
that authentic knowledge is based in concrete, real-world situations (Lave, 1988). Decontextu-
alization is called into question considering the failures in the opposite direction, school-to-life
transfer, such as the failure of American children who successfully represent negative numbers
on a school-learned number line to relate that knowledge to money transactions (Mukhopad-
hyay, Peled, & Resnick, 1989). One reaction in education is to abandon efforts to foster transfer
through abstract instruction and instead focus on training students in situations that are directly
pertinent to important and probable future applications. This effort suggests that education should
be contextualized in concrete domains as much as possible.
Also, decontextualization may not be the only way to foster transfer. Proper contextualization
can also result in robust transfer. A situated learning perspective asserts that generalization occurs
because of contextual interactions and commonalities (e.g., Beach, 1997; Lemke, 1997). In other
words, transfer situations that are appropriately contextualized can reveal the influence of past
learning. For example, a student who brings tools from school (e.g., calculators, software) into
the workplace effectively changes their work context to be more like school and consequently
implements school learning in the workplace (Beach, 1995). Situated perspectives also endorse
learning based on problems that are continuous with everyday knowledge (Lampert, 1986).
Changing the type of contextualizations in learning and transfer situations is a non-mutually
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
exclusive alternative to decontextualizing learning in hopes of generalizing to dissimilar transfer
The research reported here is motivated by this broad interest in generalization but specifically
examines the role of decontextualization in learning and transfer. We have focused our efforts on
variations of personalizing contexts (over other types of contextualizations) that often seem thor-
oughly beneficial (or at worst, benign). How could making something more “learner-centered” or
“learner-relevant” be a bad thing? We suspect that any contextualization, implemented without
considering cognitive implications, could have unexpected consequences. Personalization should
be considered in light of the mounting evidence for inherently perspectival, action-specific,
modality-specific representations (see Barsalou, 2003 for a review; Kosslyn & Thompson, 2000;
uller, 1999). If cognition is inherently perspectival, personalization may induce a partic-
ular perspective on the learning content. If that perspective is aligned with the abstract principles
of the system, then such a perspective may be helpful. However, if the induced perspective en-
courages incorrect inferences, counter or orthogonal to the principles to be taught, then we may
see a detrimental effect on learning.
We have focused on manipulations of personalization that are closely controllable to fa-
cilitate the laboratory study of learning and generalization. However, these are also pertinent
to the simple kinds of personalization decisions that arise in everyday lesson planning. Three
experiments explore different types of personalization: (a) action-involving experience, (b) con-
versational narration that places the reader in the story, and (c) familiar case studies with popular
actors/characters. Despite the modest nature of our manipulations of personalization, they are
nonetheless important because these manipulations reflect the types of contextualizations that
are pervasively implemented by teachers, textbook writers, and educational media developers.
Whenever pedagogical texts or materials are developed, design decisions related to contextual-
ization are made. Common design decisions include how much background experience to give
the learner with the domain, what voice to use in positioning the reader in the text, and whether to
take advantage of well-known cases and situations. First we will describe the overall organization
and predictions of the three experiments. Then we will review the potential effects of each type
of personalization in turn as we introduce the corresponding experiment.
To examine the extent to which contextualizations affect the subsequent application of learned
principles, we designed a computer-based tutorial about Signal Detection Theory (SDT), a use-
ful structural description of decision-making based on uncertain evidence. A signal detector
makes decisions regarding whether a signal is present or absent and each decision can have
two outcomes—the signal is actually present or absent. There are four types of events: hits (de-
tector decided “present” and actually present), misses (decided “absent” and actually present),
false alarms (decided “present” and actually absent), and correct rejections (decided “absent”
and actually absent). The contingent relationships between these four categories (e.g., deciding
“present” more often results in more hits but also more false alarms) provide structure that can
be used to effectively describe and predict outcomes in many different situations such as doctors
detecting sick patients, meteorologists predicting storms, analysts picking out market trends, and
farmers classifying ripe fruit.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
In many classrooms, books (i.e. Thomas Wickens, Elementary Signal Detection The-
ory), handouts, (i.e., David Heeger, http://www.cns.
as well as currently available computer-based tutorials (Claremont Graduate Uni-
versity,; California State University, Long Beach, college/hfes/sdt.htm/), SDT is introduced through some example of
an individual who must decide whether some signal is present or absent. A common exam-
ple used in SDT is a doctor trying to determine whether a person’s medical test results in-
dicate sickness or not. Most SDT tutorials describe such cases and then apply a more gen-
eral framework with more abstract descriptions that could be applied to a variety of SDT
In all of the experiments that follow, participants were taught SDT through a click-through
tutorial, which implemented visuospatial representations accompanied by explanatory text since
such simultaneous presentation (of graphics and text) has been empirically shown to be effective
(Mayer, 2003). To contextualize SDT, the tutorial expanded the typical description of a case study.
Instead of describing the principles of SDT in a general abstracted form, they were embedded
in the context of a doctor trying to diagnose patients. By using a common example of a real-life
SDT situation in a more narrative form, we hoped students would gain an advantage of using
a familiar and more compelling problem context. A more detailed description of the tutorial is
presented in the methods section of Experiment 1.
In addition to these efforts at general contextualization, each experiment also manipulated
an additional aspect of personalization, commonly used to motivate students. In Experiment 1,
participants were given discovery experience through a short hands-on activity of detecting signals
(diagnosing people as sick versus healthy from their cells) and finding out the outcomes of those
decisions. This experience may engage students by allowing them to actually make decisions
under uncertainty; this may also motivate the need for a framework such as SDT. Experiment 2
personalized learning by using a more active and engaging tone, simply by using the term “you”
in the SDT tutorial. Participants were told that they were the active detecting agent with the “you”
pronoun or they were detached from the agent with the pronoun “he. Experiment 3 examined
whether participants benefit from a tutorial that included contextualizing details about a specific
and familiar doctor, a character from a popular medical television drama, compared to the same
tutorial couched in terms of a generic doctor.
These particular types of personalizing contextualizations were chosen because, although they
reflect learner-centered design, their cognitive implications have not been considered as much as
their motivational benefits. Also, these personalizations put learners in a particular perspective
with respect to the material to be learned. The manipulations of Experiment 1 and 2 place the user
in the point of view of the signal detector either directly, through experience, or linguistically.
The perspective fostered by Experiment 3, a well-known character, gives students an opportunity
to take a perspective that has been portrayed in an entertaining way (on a television show).
Although the goal of this tutorial is to teach the larger structure of SDT, our manipulations
of personalization emphasize the signal detector’s perspective. If this perspective is helpful for
understanding the abstract structure of the system, students may enjoy both motivational and
learning benefits of personalization. However, if the perspective of the signal detector limits
students’ understanding of SDT, then, although there may be a motivational benefit, learning or
transfer may be compromised.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
Personalization with experiential, conversational, or familiar information may motivate students
to give extra effort to comprehending the SDT material. However, if the goal is to understand
the general structure of SDT, learning and transfer may benefit from a certain amount of decon-
textualization as well. If decontextualization and detachment from a specific learning situation
are necessary for identifying functionally relevant structure (Goldstone, 2006), then hands-on
experience or a particular point of view or familiar details may anchor learning too much to
irrelevant (or worse, misleading) past experiences or real-world knowledge (e.g., Rumelhart &
Ortony, 1977; Schank & Abelson, 1977). Preconceptions of doctors and their methods of di-
agnosis may hinder students from appreciating the new insights that a SDT perspective might
bring. Additionally, the anchoring influence of personalization may also result in poor transfer
to dissimilar (e.g., non-doctor) situations. By this account, there is a possibility for too much
There is a tension between the concretely experienced and personally relevant on the one
hand and the transportable and general on the other hand that is appreciated by researchers in
both cognition and education. Goldstone (2006) argues that hybrids such as “recontextualiza-
tion, combining contextualization and decontextualization to create new categories for making
predictions and inferences, may be necessary to foster both grounded, connected understanding
and flexible transfer. Related notions in education such as “situated generalization” (Carraher,
Nemirovsky, & Schliemann, 1995), “progressive formalization” (Freudenthal, 1983 ) or “action-
generalization” (Koedinger, 2002; Koedinger & Anderson, 1998) characterize generalization225
behavior through a combination of decontextualizing processes and active, concrete, and specific
situations. Other approaches suggest interfacing between contextualized and decontextualized
learning through coordination classes (diSessa & Wagner, 2005) that emphasize the extraction
of invariant information among a wide variety of contextualized situations. One of the benefits
of a less active or specific perspective on a doctor diagnosis situation is that by not committing
learners to a particularly detailed construal, they can adopt a more general view of the entire
system. Personalization may have the consequence of encouraging students to adopt the signal
detector’s perspective. However, this may interfere with a more structural understanding of SDT
beyond what the detector might know or not know.
Personal motivation is important to contemporary educators who are concerned with pro-
moting inquiry over the acquisition of factual knowledge, reflecting a classic Piagetian view
where true and deep understanding comes from activity at the hands of the learner (Pi-
aget, 1970). Constructivist frameworks endorse rich activities that engage learners rather
than merely studying texts (e.g., Savery & Duffy, 1994). By involving students in ac-
tivities, their learning environment is set up for them to explore and discover princi-
ples. Several studies have documented a positive relationship between hands-on activity
and test scores found in school settings (Bredderman, 1983; Inagaki, 1990; Stohr-Hunt,
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
Unfettered action-oriented experience is often contrasted to direct instruction where a teacher 245
(or text or computer program) guides the activities of a student. Direct instruction may be partic-
ularly effective for demonstrating difficult concepts and abstract regularities because students are
unlikely to discover these through unconstrained experience. Recently, Klahr and Nigam (2004)
found that mastery acquired from directed activity resulted in more sophisticated learning than
mastery by discovery. Experience-based learning can be made more effective with teacher guid-
ance (e.g., Brown & Campione, 1994; Cognition and Technology Group at Vanderbilt (CTGV),
1996). Current debates take into consideration the capacity limits of cognitive architecture and
suggest that direct instruction (Kirschner, Sweller, & Clark, 2006) or well-supported discovery
activity (Schmidt, Loyens, van Gog, & Paas, 2007; Hmelo-Silver, Duncan, & Chinn, 2007) may
aid novices by easing cognitive load (Sweller, 1988).
Studying the interaction between guiding information and active experience is a valuable do-
main for education research given the possible overarching implications for curriculum design
and classroom content. Research shows that although active experience is influential and poten-
tially helpful, not all active experiences are equal. Mayer, Mathias, and Wetzell (Experiment 3,
2002) presents a case of active experience being helpful for learning only when it comes before
explicit instruction but provides no benefit when it comes after. They suggest that becoming
familiarized with a hands-on model first prepares students to understand components that will be
critical when learning about the more abstract causal model. Active experiences can change the
way students organize and interpret future learning.
Unfortunately, these changes may not always facilitate learning and transfer. DeLoache’s
studies (2000) regarding young children’s use of scale models provide empirical evidence that
hands-on activity is not always beneficial. In these experiments, a 3-year-old child observes a
miniature doll hidden in a scale model of a room. The child is told that a larger version of the
doll is in the corresponding location in a larger room and is given an opportunity to search. Some
children were given 10 minutes of free play and manual exploration of the model before the
hiding event and others were simply shown the hiding event. Children who had an opportunity to
play with the model were less effective at using the model as a map of the real room. Furthermore,
children in another experiment who had less opportunity to interact with the model (by placing
it behind a window) were more effective at searching in the real room. DeLoache (1995, 2000)
notes that although experience typically facilitates performance, in this case, experience may
have changed the way children viewed the model room. Although a lower level of experience-
based salience allowed children to view the model room as a representation, a higher level of
interaction with the object prevented this more sophisticated understanding. Hands-on experience
induced the children to see the model as a physical toy rather than as a representation for another
object. Recently, Uttal and colleagues (Uttal, Bostwick, Amaya, & DeLoache, in preparation)
observed that using block letters and numbers to play basketball (throwing numbers into a basket)
or blow bubbles (i.e., using the “P” shape as a bubble wand) did not improve performance on
letter and number comprehension. In actuality, playing irrelevant games may have actually hurt
performance because control children, who did not play with the letter and number toys, showed
better comprehension at the end of the study. The lesson from these results is that hands-on
experiences can limit or distort a learner’s perspective; these perspectives may change the content
and quality of what is learned.
To understand the pedagogical effect of active experience, our experiment takes into consider-
ation the particular perspective induced in our participants through experience. In Experiment 1,
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
participants were either given the SDT tutorial alone (the Control condition), or a short sequence290
of active SDT decisions, in which they effectively play the role of signal-detector, followed by
direct instruction in the form of the tutorial (the Experience condition). Ideally in educational
practice, “experience” and “activity” are much richer than the implementation we have provided
here, and experiences can also be tailored to foster an appropriate perspective on SDT. The min-
imal experience we provide is not designed to be optimal for teaching SDT, but for fostering a
signal detector’s perspective. However, it is worth noting that even this limited contextualization
might still be an improvement because most SDT tutorials do not implement any active expe-
rience at all and merely describe applicable contexts. Another potential benefit of the minimal
signal-detecting activity is that this experience is relevant to learning SDT. Actively making
responses and examining the consequences might give rise to action-specific insights such as
the understanding that hits and false alarms are both possible outcomes of making a “present”
response. If these action-specific insights were intentionally guided towards a more general un-
derstanding of SDT (such as tallying these active responses into the four decision outcomes
of SDT), one might predict that actually experiencing signal detection may be beneficial since
transferable learning can draw on activity contexts (Greeno, Smith, & Moore, 1992). Alterna-
tively, the action-specific insights alone may not lead to a general understanding of SDT and
instead could bias students to adopt perspectives that turn out to be detrimental to learning and
Participants and Design.
Seventy-three undergraduates from Indiana University partic-310
ipated in this experiment for credit. A computer program randomly assigned participants the
Experience (n = 41) or Control condition (n = 32).
To quickly read the text presented in the experiment (including tutorial and quizzes) takes
about 15 minutes. Four additional participants (from the Control condition) who spent less than
15 minutes to complete the experiment were excluded from analysis. When participants were
debriefed at the end of the experiment, they reported how much they previously knew about SDT.
All of our participants said they did not know it at all or had heard of it but did not know what it
was about.
Materials and Procedure.
All participants read through a computer-based SDT tutorial
made up of pictures and explanatory text. The principles of SDT were embedded in the context
of a doctor trying to diagnose patients with leukemia by examining blood cell distortion levels.
Participants were told that patients with leukemia typically have more distorted cells, and healthy
patients have less distorted cells. Although cell distortion is an imperfect indicator of health, the
doctor tries to optimize his decisions based on this imperfect evidence.
After a brief introduction to this scenario, participants in the Experience condition were told:
“Pretend you are a doctor who has gotten the results of a blood test.” They were shown a graphic
of a continuum of cell distortion levels, and were shown that the left and right ends had mostly
healthy and mostly sick patients, respectively. Then participants were shown cells one at a time
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
FIGURE 1 Screenshot of signal detecting categorization experience seen by the Experience condition of Ex-
periment 1.
to diagnose as “sick” or “not sick” as shown in Figure 1. After each diagnosis, the participant
received feedback indicating the actual sickness/health of the patient. The diagnosis activity
included a total of 40 cells that took most participants less than 2 minutes to complete. Then,
they were introduced to a tutorial. Participants in the Control condition were shown the brief
introduction to the scenario and then immediately received the tutorial without any categorization
The tutorial has been used in other experiments (i.e., Son, Doumas, & Goldstone, under
review) and uses a combination of descriptions and diagrams to teach basic SDT concepts. Q7
The tutorial is available online at (the URL is case sensitive).
Pilot experiments teaching students SDT with traditional normal distributions contrasted to other
attempts using frequency bar graphs supported the claim that frequency information is far easier to
understand than probability information (both in general cognition, Gigerenzer & Hoffrage, 1995;
and pedagogy, Bakker & Gravemeijer, 2004). We speculated that the overlapping region of the
traditional distributions (e.g., Figure 3) was particularly crucial for understanding SDT but also
particularly confusing for students. Because we were not interested in teaching graph reasoning
per se, we developed bar graphs that utilized non-overlapping spaces and color codes tailored to
represent critical concepts of SDT (see Figure 2). Non-overlapping regions of the screen (i.e., top
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
FIGURE 2 These are a few screenshots of the tutorial. The distribution of boxes on top of these screens were
outlined in red to indicate the actually sick status of these individual patients whereas the boxes presented on the
lower half of these screens were outlined in green to indicate their actually healthy status. Boxes on the right side
of these screens were labeled with a red “S” in order to show that the doctor had diagnosed these cases as “sick.
Boxes on the left were labeled with a green “H” to indicate that the doctor had diagnosed these as “healthy.
and bottom) were used to represent two different distributions (i.e., actually healthy and actually
sick people). Colored labels provide a perceptual indicator for the categorization the detector has
made (i.e., diagnosed healthy versus diagnosed sick). The tutorial implemented the combination
of these features because SDT requires an understanding of both which cases are actually in
which categories and which have been categorized as such by the detector.
Each case is represented by a distorted cell in a box outline. To show the results of the
doctor’s categorization, each case is labeled with a red “S” or green “H.” To show actual category
membership, the outline of each case is red or green for the two categories of sick and healthy
FIGURE 3 The underlying SDT structures of the tutorial and transfer stories are shown here. The participants
never saw this figure.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
patients, respectively. The cases are spatially ordered, from left to right, by increasing cell
distortion. If the likelihood of a particular level of distortion indicating leukemia is high, there are
more red cases in the column than green. If the likelihood of leukemia is low for a particular level
of distortion, then there are more green cases than red ones. Columns that include both category
types can be seen as analogous to the overlapping regions of traditional SDT distributions because
the same level of distortion could belong to either category. Although in typical SDT distributions
there is an actual physical overlap to indicate both categories being associated with the same cell
distortion level, in our diagrams, occurring in the same column indicates the same meaning as
the overlap. Pilot experiments with non-overlapping representations seemed more effective than
traditional SDT distributions, which frequently led to incorrect interpretations such as, “Cases
in this overlapping area are both sick and healthy. The cases are then separated by category
into two different histograms to clearly show the actual status of these patients. This reflects the
SDT distinction between actual signal and noise distributions, as contrasted with the doctor’s
The tutorial was a 47-screen self-paced slide show covering basic SDT concepts such as the
difference between evidence for a decision, the decision, and the actual status of the case. The
tutorial was designed to help students understand the relationship between changes in decision
thresholds and resulting changing patterns of diagnosis errors, and between changes in the actual
population distribution and patterns of errors. At the end of the tutorial students are expected to
reason about how decisions and distributions affect outcomes. Toward this end, students were
shown how a decision boundary could lead to two ways of making the right decision (hits and
correct rejections) and two ways of being incorrect (misses and false alarms). This was followed
by two examples where the decision boundary was moved in order to show the relationship
between these categories. Additionally, participants were shown what would happen if the signal
distribution (all of the targets) shifted along the evidence continuum.
Due to time constraints, the tutorial was limited to showing basic reasoning with SDT concepts
rather than the equations and other abstract formalisms typically used in SDT. These principles
were fully contextualized in narrative form rather than generic statements about SDT. Meaningful
principles in SDT were couched in terms of a doctor and his patients. For example, the concept
of hits was translated into the story context as “diagnosed sick, actually sick. An example of
moving the decision boundary was presented in the story context as a situation where the doctor
must avoid misses (“diagnosed healthy, actually sick”) because the undiagnosed disease is fatal.
When the doctor moves the decision boundary, more patients, both actually healthy and sick, are
diagnosed as sick. Within the entire tutorial, the category labels (“hits, “false alarms, “correct
rejections,” “misses”) were never explicitly mentioned.
After reading through the tutorial, participants answered eight multiple-choice questions about
the tutorial’s doctor situation that could be answered correctly by applying SDT principles. There
was a ninth tutorial question (“Which of these is the worst scenario for the patient?”) that was
used for exploratory assessment, but was not included in the main analyses because it did not
have a correct answer. Difficult quiz questions were purposefully used to ensure that participants
needed to use SDT principles rather than relying on common sense (tutorial quiz questions are a
subset of the questions included in Appendix A).
After the tutorial quiz, participants received an opportunity to analogically transfer what they
had learned to a different story context, although they were not explicitly instructed that the two
scenarios were related. Participants read a few paragraphs (included in Appendix A) presented
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
on three slides describing a small town that wants to export sweet melons and avoid exporting
bitter melons. Sweet melons, laden with juice, tend to be heavier, so this town decides to sort
the melons by weight (even though weight is only an imperfect indicator of sweetness). Heavy
melons are exported and sold whereas light melons are rejected. However, all of the melons are
subject to consumer reports that allow the town to find out which melons are actually sweet/bitter.
A nine-question transfer quiz was subsequently administered. The transfer quiz is available on
the extended Appendix online,
At the end of the experiment, participants were told that these two stories were analogous and
asked to explicitly place elements of the two stories into correspondence with each other in a
six-question multiple choice mapping quiz. Each mapping question presented an element from
one of the stories, such as “sick patient” (signal) and four possibilities from the other story, two of
which are viable answers according to SDT—“sweet melon” (signal) and “bitter melon” (noise).
The mapping of sick patient to bitter melon is not necessarily incorrect because the mathematics
of SDT binary choice is not affected by which category is called “present” and which is called
“absent. The tutorial and transfer situations were designed to reflect the signal-to-signal mapping
because the doctor seeks sick patients for treatment and the town seeks sweet melons for shipping.
One reflection of that intention is found in the spatial organization of the tutorial and transfer
figures, with cell distortion and melon weight increasing from left to right, and signals appearing
on the right side (for a schematic illustration see Figure 3). Because of this spatial alignment, we
will call this “sick patient-sweet melon” mapping the structural answer. However, to make this
match, one must overcome an appealing semantic match between sick patients and bitter (sickly)
melons. Thus the mapping quiz was scored in three ways: a structural score, a semantic score,
and a combined score (structural + semantic).
Tutorial and transfer scores were subjected to a quiz-type × condition, 2 × 2, repeated-measures
ANOVA and the results are shown in Table 1. We found a significant effect of condition (experience
+ tutorial versus tutorial only), F (1, 71) = 8.33, p<.01, and a marginally significant effect of
quiz-type (training versus transfer), F (1, 71) = 3.04, p<.10, but no interaction, F (1, 71) =
The Control condition (tutorial only) scored .48 (SD = .19) correct on the quizzes while the
additional experience dropped performance down to .40 (SD = .27). Although the interaction
was not significant, there is evidence that the effect of condition was stronger for the transfer quiz
than tutorial quiz. Although the Control condition participants significantly outperformed those
Tutorial and Transfer Quiz Results for Experiments 1
Tutorial Transfer
(8 Questions in Tutorial) (9 Questions in Transfer)
Signal Sick patient Sweet melon
Noise Healthy patient Bitter melon
No Experience 0.58 (SD = .26) 0.56 (SD = .22)
Signal-Detecting Experience 0.49 (SD = .18) 0.41 (SD = .12)
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
that received SDT experience on transfer performance, t(72) = 11.84, p<.001, there was no
difference of initial learning on the tutorial, t(72) = 2.61.
One might suppose that transfer differences may have been caused by initial differences in
learning such that students in the Experience condition acquired less (or perhaps less accurate)
knowledge. Although that may be part of the story, differences in learning do not completely
account for the differences on transfer. This is evidenced by an ANCOVA with transfer score as
the dependent variable which found that even though tutorial score was a significant covariate,
F (1, 69) = 15.36, p<.01, condition still exerted a significant influence, F (1, 69) = 8.88,
p< .01, with transfer performance better for participants in the Control condition than the
Experience condition. There was no significant interaction between condition and the tutorial
score, F (1, 69) = 1.49. If tutorial learning was the primary factor for differences in transfer,
there should be no additional influence of condition. This supports the notion that an advantage
of decontextualization goes beyond helping students learn more.
One simple explanation of the Experience condition’s poor performance might be that the
participants in this condition had more work, an extra initial task, so they were simply tired or
not trying as hard as those in the Control condition. We examined this hypothesis by looking
at the total time spent on the experiment. Although overall the Experience condition seemed
to spend a bit more time on the entire experiment (M = 23.25 minutes, SD = 5.81) than the
Control condition (M = 21.85, SD = 5.40), this difference was not significant, t(72) = 1.12. The
categorization task was very short and integrated into the beginning of the tutorial to seem less
like a separate task.
This is a counterintuitive finding in some ways because detecting signals would give partic-
ipants an experiential distinction between being right and being wrong in different ways (e.g.,
correct rejects, false alarms, hits, and misses), an insight that is relevant to SDT. To understand
these results better, we examined the individual questions and found a clue. One of our questions
asked, “Which of the following decision strategies will ensure that the doctor maximizes the
number of actually healthy people he diagnoses as healthy?” In SDT terms, this asks how the
doctor can maximize correct rejections. The correct answer offered is “diagnose everyone as
healthy” (the rest of the answer choices are shown in Appendix A) and while fourteen control
participants gave correct responses, no one in the Experience condition made this choice. This
difference was statistically confirmed by a chi-square analysis on the four answer choices, χ
= 16.61, p<.001. From a doctor’s point of view, a strategy that diagnoses everyone as healthy
is unreasonable and perhaps even preposterous.
If the results of our study were only caused by a few odd questions, the effect of SDT experience
may not be as negative as our initial analyses suggest. Given that the individual questions asked
in these quizzes differed widely in their mean accuracies, a relevant analysis concerns whether
experience exerted an impact across all 17 questions (8 tutorial, 9 transfer). If the effect of
experience was a general one, across many questions, we should see the effect of condition
even in an item-based analysis but no effect of particular items. To answer this question, here
and in the subsequent experiments, we report item analyses to compare conditions. The effect
of item-type (tutorial or transfer) and tutorial condition (Control, Experience) were evaluated
ina2× 2 repeated-measures ANOVA, with item-type as a between-item factor. Even in this
item analysis, there was a significant main effect of condition with the control performing better
than the Experience condition, F (1, 15) = 4.91, p<.05, but no main effect of item-type, F (1,
15) = 3.00, and no interaction between condition and item-type, F (1, 15) = .01. A paired t-test
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
Tutorial and Transfer Mappings from Experiment 1
Structural mapping Semantic mapping
Sick Patient–Sweet Melon Sick Patient–Bitter Melon
Signal–Signal Signal–Noise
No Experience .22 (SD = .27) .57 (SD = .35)
Signal-Detecting Experience .39 (SD = .29) .48 (SD = .32)
revealed that this effect of condition was due to a significant advantage of the Control condition
over the Experience condition (a difference of M = .08, SD = .15), t (16) = 2.29, p<.05. This
result is consistent with the hypothesis that the disadvantage of experience was found throughout
the tests rather than being explained by one or two damaging questions.
Recall that there were two types of mappings, structural or semantic, that could have been
made on every mapping question. The six mapping questions were analyzed relative to these
two mapping types and two conditions in a 2 × 2 repeated-measures ANOVA and the results
are shown in Table 2. Although there was no reliable effect of condition, F (1, 10) = 2.69, there
was a significant effect of mapping type, F (1, 10) = 10.02, p<.01. A paired t -test revealed
that semantic mappings were more frequent than structural ones, mean difference of .26 (SD =
.72), t (72) = 3.05, p<.01. Additionally, there was a significant interaction between condition
and mapping type, F (1, 10) = 30.62, p<.001. Paired t-tests showed that control participants
made significantly more semantic mappings than experience participants, t (5) = 4.00, p<.01.
The Experience condition made significantly more counterintuitive, structural mappings than the
Control condition, t(5) = 4.36, p<.01. The structural mappings are more true to SDT principles
than the semantic mappings because the structural mapping respects that the signal is defined by
what the detector is looking for and the noise is what interferes or is confusable with that signal.
However, this tendency for the Experience condition to map structurally was accompanied by
worse, not better, performance on the tutorial and transfer quizzes.
Why is it that, even across questions, there is a negative effect of a brief diagnosis experience?
Or (perhaps put more optimistically) why is there a benefit of having no experience? Could
it be because of the induced perspective? The situation where a doctor diagnoses patients and
experiences hits, misses, false alarms, and correct rejections is structurally identical to the town
making decisions about the sweetness of melons—but only from a perspective that emphasizes the
overall structure of SDT. To understand these situations from an SDT viewpoint, some of the ideas
about doctors and fruit farming must be ignored. From an SDT perspective, the decision criterion
and its relationship to the underlying signal and noise distributions is more important than doctors
using common sense in their diagnoses or caring deeply about the well-being of their patients.
Part of the difficulty of learning SDT after experiencing the categorization task is that participants
may have attended to the patient-diagnosing aspect of this task, which may recruit knowledge
extraneous to SDT, such as ideas about practicing medicine and how people deal with illness.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
This doctor-based perspective and the resulting patient-based construal may have competed with
an SDT perspective that emphasizes certain aspects of the decision making process: signals,
thresholds, decisions, hits, misses, false alarms, and correct rejections. Sometimes the goals of
a doctor and the solutions offered by through the application of SDT may not match. When
discrepancies between the two construals arise, concrete experiences from the doctor’s point-of-
view may tilt the balance in favor of attending more to the doctor-specific factors rather than the
more formal SDT perspective.
Experience condition participants do seem to develop a notion of signal as “that which is being
looked for. This characterization of targets leads to structural matches, and the successful ignoring
of semantic features. This perspective converges with the structure of SDT that educators generally
want students to learn. Many psychology classes use this kind of basic feedback experience
as demonstrations of experimental paradigms presumably based on the intuition that “putting
students in the experiment participant’s shoes, prepares them to learn the organizing concepts.
The implicit pedagogical expectation might be this: by showing students what an experimental
participant would see and respond to, students might learn more about experimentation in general.
However, our mapping results suggest that experience leads to SDT-compatible interpretations,
but the quiz results also indicate that experience leads to competing doctor-specific interpretations.
So although there are some benefits, they may not be powerful enough to overcome irrelevant
interpretations that come along as baggage.
Experiment 1 examined the effect of a hands-on perspective by giving participants actual
signal-detecting experience. When the learner feels like part of the system (the detector), this
can lead to different learning outcomes than a perspective that fosters a more distant, observer
viewpoint. Providing active experience fosters an active personal perspective for our learners, but
there are other ways of manipulating perspective. Can this perspective be induced by a more minor
manipulation such as a change in textual voice? Experiment 2 used a personalized conversational
narration to further investigate the role of induced perspective on learning.
There are two main reasons to posit benefits of personalized narratives: they play critical roles in
engagement and mental models. The first is that personalized conversation engages learners and
readily connects them to the learning content. Comments that are directed at the learner, such
as strategically placed questions in computer-based tutorials (i.e., “Which do you think. . . ?”),
might engage students to think more critically at crucial junctures (Moreno & Mayer, 2004).
Contextualizing the learning with personal goals (i.e., “your mission, “your journey”) also
facilitates learning (Moreno & Mayer, 2004). Even with subtle grammatical cues such as personal
pronouns (“your lung”) versus a more generic framing (“the lung”), this personalized text aids
learning (Mayer, Fennell, Farmer, & Campbell, 2004). Some researchers recommend the second-
person pronoun for textbooks and computer-based tutorials because, “it connects the reader to
the mathematics because it speaks to the reader directly” (Herbel-Eisenman & Wagner, 2005).
It is important to note that while engagement is often equated with excitement and interest,
this connectedness to the material seems to go beyond that. Although vividness and interest have
been linked to active perspectives (Fernald, 1987; Velasco & Bond, 1998), interest may not be
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
necessary for enhanced learning. For example, Mayer et al. (2004) showed increases in learning550
from the pronoun manipulation but did not show parallel increases in several measures of interest.
This suggests that readers are not simply more interested by conversational style but can be
engaged with the content in particular ways. This leads to the second influence of conversational
narration—textual voice can change how the reader understands the text by changing the mental
models that people build. Theories in discourse processing have claimed that through text compre-
hension, agents and situations are mentally constructed (Chafe, 1994; Graesser, Bowers, Bayen,
& Hu, 2000). Bergen (2006) showed that pronoun-introduced perspectives, such as “you,” which
engage participants in the activity, are influential in mental modeling, a process often implicated
in high-level reasoning tasks (Johnson-Laird, 1983).
More specifically, first-person (“I am a doctor. . . ”) and second-person (“You are a doctor. . . ”)
narration might unite the reader with the actions of the character or narrator. Third-person narration
(“He is a doctor. . . ”) is considered a detached, omniscient observer in formal literary study and
often invisible to na
ıve readers (Duchan, Bruder, & Hewitt, 1995; Graesser et al., 1996). There
may be a particular superiority that comes from relating the material to the self because such
information might be easier to process (Spiro, 1977).
d’Ailly, Simpson, and MacKinnon (1997) applied these ideas about mental modeling and
personalization to grade school children solving word problems. Children were asked to solve
different types of comparison word problems. There were problems where the second-person
language identified participants with the known anchoring quantity for the comparison (i.e., the
you-Known problems, “You have 3 balls. You have 2 more/less balls than Bob. How many balls
does Bob have?”) or identified them with the unknown quantity (i.e., the you-Unknown problems,
“Bob has 3 balls. Bob has 2 more/less balls than you. How many balls do you have?”). When
the pronoun “you” was substituted in the known position and served as the anchor, children
were better able to solve these word problems compared to problems with other names in them.
However, there was no facilitation when the pronoun occurred in the unknown position. d’Ailly
and colleagues hypothesized that children tend to anchor the referent according to the self term
(“you”), and when they need to reverse the anchor (as in the case of the you-Unknown problems),
they had difficulty doing so. Similar studies with adults solving problems about linear ordering
suggest that the self-term easily becomes the point of reference (d’Ailly, Murray, & Corkill,
These results suggest that personalized conversation may lead to better interpretations—
but not in all situations. Perhaps for cases such as “your cloud” and “your plant” (Moreno &
Mayer, 2004) personalization provides the best interpretation. However, the work by d’Ailly
and colleagues (1997) shows interpreting the “you” perspective as the anchor may not always
be optimal for solving the problem. Experiment 1 provides some evidence that anchoring the
participant to the signal detector’s perspective may not be optimal for learning SDT and may in
fact be detrimental. So, to promote a personalized perspective without manipulating experience,
we changed the wording in the tutorial to reflect a second-person perspective (“you”). If learners
tend to anchor their perspective at the self term, this manipulation puts the learner in the role of
the signal-detecting doctor. To provide a control condition that had a perspective more like an
outside observer, we used the third-person perspective (“he”).
Although this may seem like a modest manipulation, if “you” helps, this can be readily applied
in computer-based tutorials. However, if it does not, this may be a warning that we need to be
careful when applying such personalizing contextualizations.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
Method 595
Participants and Design.
Fifty-five undergraduates participated for credit. Participants
were randomly assigned to be in the Third-person condition (n = 27) or Second-person condition
(n = 28). Four participants (2 from the Third-person condition, 2 from the Second-person con-
dition) were excluded from analysis because they spent less than 15 minutes on the experiment.
Participants reported having no previous knowledge of SDT.
Materials and Procedure.
Participants read through a computer-based SDT tutorial similar
to the one used in Experiment 1. Afterward, they took a tutorial quiz worded in the corresponding
manner (second-person or third-person voice) of their tutorial. The transfer scenario description
and quiz followed. We designed additional questions for a total of 12 tutorial and 12 transfer
questions (these have been included in Appendix A). The same mapping quiz used in Experiment
1 was also administered.
For the Second-person condition, all of the tutorial and the corresponding tutorial quiz ques-
tions were modified to indicate that the participants themselves were the doctor. For example,
the introduction in the Second-person tutorial read, “Imagine that you are a doctor who looks at
blood samples to check if your patients have leukemia, a cancer of the bone marrow. Since bone
marrow produces blood cells, you can look for distorted blood cells to diagnose your patients. In
the matching Third-person tutorial, participants were encouraged to “Imagine a doctor who looks
at blood samples to check if his patients have leukemia, a cancer of the bone marrow. Since bone
marrow produces blood cells, the doctor can look for distorted blood cells to help him diagnose
his patients.
When participants were asked to imagine alternative scenarios in the Second-person condition,
they were instructed to imagine placing themselves in the pertinent position. For example,
participants were told: “Changing the decision boundary is something you can do to change
what kinds of mistakes you make. There are some things that are out of your control that also
affect how good your diagnosis is. Consider a situation where it becomes even harder for you to
diagnose your patients because everyone started taking vitamins that make distorted cells caused
by cancerous cells look better. So now sick people have less distorted cells than they used to.
The corresponding Third-person version read, “Changing the decision boundary is something the
doctor can do to change what kinds of mistakes he makes. There are some things that are out of
his control that also affect how good his diagnosis is. Consider a situation where it becomes even
harder for the doctor to diagnose his patients because everyone started taking vitamins that make
distorted cells caused by cancerous cells look better. So now sick people have less distorted cells
than they used to.
Subject analyses showed no significant main effects of test, F (1, 53) = .97, and no interaction
between condition and test, F (1, 53) = .77. However, there was a marginally significant effect
of condition, F (1, 53) = 2.81, p<.10 with second-person perspective tending to produce worse
performance than the third-person perspective. Given this tendency for worse “you” performance
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
for both tutorial and transfer measures, it may have merely been that they learned less overall from
the tutorial. An ANCOVA on transfer performance revealed tutorial performance as a significant
covariate, F (1, 51) = 24.55, p<.001, no significant effect of condition, F (1, 51) = 1.54, and no
interaction, F (1, 51) = .13, suggesting that the learning differences induced by these perspectives
may have led to the subtle differences seen in transfer. Although the “you” versus “he” perspective
difference was a more modest manipulation than the experience-induced perspective, the more
personalized training still results in somewhat reduced learning.
To more finely assess whether the “you” perspective acted in similar ways to the Experience
condition, we examined the individual quiz items. Because the quizzes in Experiment 2 included
all of the questions used in Experiment 1, we were able to analyze the same question about max-
imizing the number of actually healthy people diagnosed as healthy (tutorial question #7). Recall
the correct answer is an unintuitive one from a doctor’s point of view (“diagnose everyone as
healthy”). Eleven out of 27 participants in the “He” condition were able to make this choice versus
7 out of 28 in the “You” condition. The most popular answer for the “You” condition was a caution-
ary and practical one, but not one particularly informed by SDT (“look more carefully at the cell
distortion levels before your diagnosis”). This choice was made by 10 out of 28 “You” participants
compared to the single person that made this choice in the “He” condition. A chi-square analysis on
the four answer choices confirmed these differences between conditions, χ
(3) = 10.90, p<.05.
These results suggest that the “You” condition prompted participants to use general knowledge
about medical diagnosis rather than solely relying on SDT principles. If general medical knowl-
edge were aligned with the inferences of SDT, then perhaps such contextualization would result in
better learning. Here, however, the common sense answer competed with the choice informed by
The slight disadvantage to the “You” condition seen in the subject-analysis may have been
caused by a few difficult questions. So like Experiment 1, we performed an item-analysis over the
two types of items (those in the tutorial and transfer) and the two perspective conditions (“He”
and “You”). This 2 × 2 repeated-measures ANOVA showed a significant main effect of condition,
F (1, 22) = 19.83, p<.001. There was no significant effect of item-type, F (1, 22) = .12, nor was
there a significant interaction, F (1, 22) = 1.59. This suggests that over all the questions, those
in the “You” condition showed limited learning and transfer. Also, this analysis shows that the
tendency for worse “You” than “He” performance suggested by the subject-analysis is significant
when analyzed by item. Collapsing over the two quizzes, the “He” condition scored an average
of .57 (SD = .20) while the “You” condition averaged .49 (SD = .21). The results, organized
according to quiz, are shown in Table 3.
Tutorial and Transfer Quiz Results for Experiments 2
Tutorial (12 Questions in Tutorial) Transfer (12 Questions in Transfer)
Signal Sick patient Sweet melon
Noise Healthy patient Bitter melon
“He” 0.55 (SD = .20) 0.60 (SD = .22)
“You” 0.49 (SD = .20) 0.49 (SD = .26)
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
Tutorial and Transfer Mappings from Experiment 2
Structural Mapping Semantic Mapping
Sick Patient–Sweet Melon Sick Patient–Bitter Melon
Signal–Signal Signal–Noise
“He” 0.23 (SD = .26) 0.62 (SD = .31)
“You” 0.19 (SD = .16) 0.64 (SD = .26)
Mapping quiz performances were analyzed relative to perspective condition as well as type
of mapping (semantic versus structural). The results are shown in Table 4 and although there
were no differences between conditions, F (1, 10) = .02, there was a significant difference in
prevalence between mapping types, F (1, 10) = 29.10, p<.001, such that semantic mappings
were made more often over all six questions than structural ones.
Research from embodied cognition shows that particular text-based perspectives influence mem-
ory (Abelson, 1975; Pichert & Anderson, 1977), comprehension (Bower, Black, & Turner, 1979),
and category judgments (Barsalou & Sewell, 1984). These studies suggest that a participant’s
adopted perspective can influence cognitive performance. Experiment 2 indicates that these find-
ings are important to pedagogy because grammatically induced perspective can also influence
learning and transfer of structural information. Particularly if the particular vantage point offered
to students is non-optimal or limiting, learning opportunities consistent with an active perspective
may have negative consequences. Our results suggest that the detector perspective elicited by the
second-person pronoun can hinder learning and transfer of SDT.
Our negative result combined with other, typically positive, results of personalization found
by other researchers begs the question: When should personalizing texts be used? We suspect
that it largely depends on the content to be learned. In SDT we are interested in an overall
system, and through personalization, the learner becomes positioned as a participant inside of the
system. The goals, structures, and affordances of an individual within a system may be different
than information available from an outside observer’s perspective. Wilensky and Resnick (1999)
have observed that novices to a domain often have trouble using multiple perspectives. Students
imitating agents have been known to experience difficulty abandoning the agent’s perspective
in order to understand systems-level phenomena (Penner, 2001). Applying this tension between
individual and system-level perspective to our situation suggests that our Second-person condition
may have emphasized the individual level to the detriment of the system level. As an example,
information that is present in the system but unavailable to the doctor (e.g., actual sickness/health
of a patient) seems to be underemphasized for students in the second-person perspective.
The participant perspective induced in our experiments was that of a doctor. This seemed to
activate practical knowledge of medical diagnoses. Background knowledge of doctors and patients
can help contextualize abstract SDT patterns, rendering them more comprehensible. Our results
suggest a caveat: Contextualization could be misleading if the structural content is embedded
in the medical context so that it detracts students from using relevant SDT principles. Instead,
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
students who learn from the doctor’s perspective seem to prefer their common sense notions of
making reasonable diagnoses as opposed to SDT’s focus on optimal decisions over many cases.
The SDT framework is often not enough to decide what to do in real life medical situations
because there are other pressures that may affect the decision boundary (i.e., false alarms may be
financially costly or systematic misses could cost lives). However, SDT does provide clear and
useful principles for evaluating and predicting medical situations that sometimes contradict typical
common sense notions. Less contextualized construals or a more detached learning situation may
draw more attention to structural SDT principles, resulting in better learning and transfer.
The overall pattern of results between Experiments 1 and 2 suggests that the active detec-
tor/doctor point of view can be detrimental to learning and transfer of SDT principles. This active
participant point of view can be instantiated in multiple ways. Hands-on activities may be a
particularly strong manipulation and conversational style might be more subtle. There have been
a number of findings that show imagined actions (e.g., Glenberg et al., 2004; Schwartz & Black,
1999), as well as watching someone else perform the action (i.e., Lozano, Hard, & Tversky,
2006), influence thinking in ways similar to first-hand action.
At the very least, it is important not to assume that student’s active engagement with learning
material necessarily translates to their deep understanding of it. Casual observation of people’s
behavior at science museums suggests that an engaging, interactive display may attract partic-
ipants, but does not guarantee that people will take the time to learn the scientific principle
demonstrated by the display. In fact, the display may be so engaging that people choose to engage
in the display instead of learning the principle. Harp and Mayer (1998) have also shown that
irrelevant “seductive details” reduce learning of main ideas and generation of problem-solving
solutions. Similarly but more subtly, the benefit of Experiment 2’s third-person point of view may
have been its relative neutrality and lack of engagement. Experiment 3 provides a more direct
test of this seductive detail hypothesis.
Thus far, we have seen how two common approaches to personalization do not necessarily provide
the best perspective for learning. In Experiment 3, we want to examine another popular approach
to personalization, using elements that are familiar and popular with students. Using personally
relevant details—information from background experiences, friends and teachers as characters,
favorite topics—has been shown in numerous studies to increase interest, understanding, and
reasoning (Anand & Ross, 1987; Lepper & Cordova, 1992; Ku & Sullivan, 2002; L
opez &
Sullivan, 1991, 1992). Even characters and objects that come from an interesting fantasy context
(i.e., pirates, spaceships) show similar boosts to motivation and learning (Cordova & Lepper,
1996; Parker & Lepper, 1992). Beyond increasing interest, personalized details may also have a
significant effect on learning through cognitive impact. Familiar contexts and details can relieve
cognitive load and boost learning because they may be easier to reason with (Ross, 1983; Wason
& Shapiro, 1971). Also it may be easier for learners to interrelate information into a coherent
model given more familiar and compelling information (Ross, 1983; Anand & Ross, 1987).
On the one hand, the addition of interesting details seems like an obvious way to make learning
more engaging and meaningful. On the other hand, these additional details can also seem like an
obvious addition of seductive details, interesting but ultimately distracting from learning. Much
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
of the evidence regarding personalized details is positive, but there are some limitations. Some
researchers question whether the motivating effects of personalized details are due to novelty and
would wear off if used systemically (Ross, Anand, & Morrison, 1988). There are also several
cases where personalization seems to have no significant improvement on actual test scores
(Bates & Wiest, 2004; Ku & Sullivan, 2000; Wright & Wright, 1986). There are also cases
in which personalization can distract or mislead. Renninger, Ewen, and Lasher (2002) showed
that when the domain is personalized to student’s own interests, such as training dogs or ballet
turnout, this can potentially limit strategies and mask a lack of knowledge. Part of learning in any
domain is distinguishing between relevant and irrelevant information, the addition of extraneous
information that might be confused for relevant information can easily increase demand (Muth,
There are multiple ways to implement familiar or interesting details but it seems that the best
conditions for personalization are: (a) when irrelevant details are easily distinguished as separate
from the problem, and (b) when they do not increase demand (at least in terms of amount of
material). In Experiment 3, we focus on how interesting human stories, such as anecdotes, indi-
vidual cases, and richly developed stories can be incorporated into classroom learning through
highly developed characters/environments as opposed to generic characters/situations. A geom-
etry problem about a basketball player’s position relative to the hoop can also be instantiated
with a popular player such as Kobe Bryant or a generic character. A character could also become
more familiar through the learning series. A well-developed example of teaching with character-
specific narratives is the Jasper series for teaching mathematics (CTGV, 1992), where students
learn about Jasper in a detailed way, with his motorboat on Cedar Creek, stopping at Larry’s dock,
and so on. Students engaged in corresponding projects involving problem identification in these
rich stories have shown improvements on achievement tests (Vye et al., 1989). Another example
of using a well-known character to promote interest is the use of Young Sherlock Holmes to
enrich a series of lessons in language arts and social studies (Bransford, Kinzer, Risko, Rowe, &
Vye, 1989).
Experiment 3 incorporated a currently popular television character into the SDT tutorial to
create two versions, one with an interesting, familiar doctor and the other with a generic doctor. If
this personalization through a well-known character has the effect of drawing users to be involved
and take a personal active stance, we may see effects parallel to that of Experiments 1 and 2.
If the familiarity and background experience with this specific doctor organizes the learner’s
understanding, participants may learn SDT with doctor-specific construals rather than the more
general SDT principles.
Participants and Design.
Sixty-four undergraduates participated for credit. Participants
were randomly assigned to be in the Generic (n = 31) or Specific (n = 33) condition. Nine
additional participants who spent less than 15 minutes to complete the experiment were excluded
from analysis (3 from the Generic condition, 6 from Specific condition). When participants were
debriefed at the end of the experiment, they reported how much they previously knew about SDT.
Two additional participants told us they already knew SDT and were excluded from analysis
(both from the Generic condition). The rest of our participants did not know SDT at all or had
heard of it but did not know what it was about.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
Materials and Procedure.
Participants read through a computer-based SDT tutorial similar
to the ones used in the previous two experiments. Undergraduate students from an upper-level
research class and undergraduate research assistants in two laboratories (not in the subject pool)
were informally polled for popular doctors on television. The Specific condition was constructed
around the most popular of these suggestions, a well-known character on a currently popular
television drama, Dr. Derek Shepherd, a neurosurgeon on “Gray’s Anatomy. The tutorial used
in the previous two experiments was modified such that the doctor is now a neurosurgeon
detecting operable versus inoperable tumors. In the Specific condition, the tutorial starts off with
this paragraph of background information about Dr. Shepherd: “Derek Shepherd is a highly
successful neurosurgeon. He works at a prestigious research hospital called Seattle Grace and
he has performed numerous complicated, risky procedures including a stand-still operation, a
double-barrel brain bypass, and spinal separation surgery for adult conjoined twins. Interspersed
with this paragraph were four pictures of the specific doctor in surgery situations, taken from the
television show.
The Generic tutorial contained the same information only framed for neurosurgeons in gen-
eral: “Neurosurgery is a complex and difficult field. The procedures involved are often risky,
dangerous, and require great skill and training. Some of the more difficult procedures in-
cluding stand-still operations, double-barrel brain bypasses, and the brain/spinal separation
surgery. There were four pictures of unknown neurosurgeons performing operations. Other
than these opening paragraphs and periodic mentions of the name “Dr. Shepherd” in the
Specific condition or “the doctor” in the Generic condition, the tutorials were identical. Both
Generic and Specific conditions received the same tutorial quiz where there was no mention of
“Dr. Shepherd.
The 12-question tutorial quiz from Experiment 2 was modified to reflect the new tutorial
situation about neurosurgery. The transfer situation and 12-question transfer quiz were the same
as those used in Experiment 2. The six-question analogical mapping quiz included modified
tutorial elements. Note that the mapping situation has changed and the tutorial signal is “op-
erable tumor” (which has a known but difficult treatment, surgery) and noise is “inoperable
tumor” (which does not have a known treatment). A signal-to-signal mapping (what the doctor
is looking for and what the farmers are looking for) is an operable tumor to a sweet melon,
both a structural (they are both targets) and semantic mapping (they are both good). Because of
this overlap, this mapping was more intuitive than Experiment 1 and 2’s sick patient to sweet
melon mapping. Thus, each of Experiment 3’s mapping questions has one correct answer, the
structural and semantic match. We created this congruent alignment to confirm that contextu-
alization is not only influential in incongruent alignments (Experiments 1 and 2). Ideally we
would test matching and mismatching versions of the tutorial and transfer stories with each
manipulation of contextualization but this would be prohibitively expensive in terms of time and
participants, so we chose to implement one experiment with an intuitively appealing structural
At the very end of the experiment, participants were asked with a multiple choice question
how familiar they were with this particular television show. Participants who reported watching
“a few” to “a lot” of episodes of the show were placed into the familiar category and those that
reported “not at all familiar” and “not too familiar” with the show were put in the unfamiliar
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
Tutorial and Transfer Quiz Results from Experiment 3
Tutorial (12 Questions in Tutorial) Transfer (12 Questions in Transfer)
Signal Operable tumor Sweet melon
Noise Inoperable tumor Bitter melon
Generic .55 (SD = .19) .63 (SD = .18)
Specific-Celebrity .45 (SD = .22) .61 (SD = .13)
Results 830
Deviating from previous results, subject analyses showed no significant main effect of condition,
F (1, 62) = 1.98, no significant interaction, F (1, 62) = 1.65, yet showed a highly significant effect
of test, F (1, 62) = 13.98, p<.001. With performance on the transfer quiz being significantly
greater (M = .62, SD = .22) than the tutorial quiz (M = .50, SD = .21), t (63) = 3.7, p<.001.
This boost in transfer that we have not seen in the previous two experiments may be due to the
better alignment of the learning and transfer contexts. The match between structural and semantic
mappings also resulted in a high prevalence of signal-to-signal mappings made by both Specific
(M = .78, SD = .23) and Generic conditions (M = .71, SD = .25); these were not significantly
different from each other, t (63) = 1.53. Both groups made very few signal-to-noise mappings
(shown in Table 5) and did not differ from each other, t(63) = 1.58.
Although there was no effect of condition on individual subjects, there is still the possibility
that for each question, one of the training conditions may have found it a bit easier to answer. An
item analysis over the two types of questions (those in the tutorial and transfer quizzes) and the two
tutorial conditions (specific and generic) resulted in a 2 × 2 repeated-measures ANOVA that did
show a significant main effect of condition, F (1, 22) = 6.53, p<.05, but no significant effect of
item-type, F (1, 22) = 2.69. The interaction was marginally significant, F (1, 22) = 2.81, p<.10.
Across all 24 items (12 items from tutorial quiz and 12 from the transfer quiz), participants in
the Generic condition significantly outperformed those in the Specific condition. These results
are shown in Table 6. These results reflect a disadvantage for the specific perspective, consistent
with the other manipulations of contextualized perspective. As with Experiment 2, the results are 850
significant by the item, but not participant, analysis.
Tutorial and Transfer Mappings from Experiment 3
Structural and Semantic Mapping
Mapping Other Mapping
Operable Tumor–Sweet Melon Operable Tumor–Bitter Melon
Signal–Signal Signal–Noise
Generic “A doctor” 0.71 (SD = .25) 0.13 (SD = .20)
“Derek Shepherd”
0.78 (SD = .23) 0.08 (SD = .13)
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
Curiously, the item analysis did not reveal any difference across the two types of test questions.
Further examination revealed that the generic group did not show significant differences in their
tutorial and test quiz performance, t(23) = .96, while the specific group did, t (23) = 4.54,
p<.05. In even closer detail, one of the tutorial questions sets up an ambiguous situation and
asked the participants what the doctor would do if he had to choose between making a diagnosis
according to SDT principles (using the decision boundary) versus “trust[ing] his experience as a
doctor. . . ” (as well as two other foils—available in Appendix A). Half of the participants in the
Generic condition (16/31) chose the SDT solution compared to 5/33 in the Specific condition who
preferred to trust the doctor’s opinion (15/33) more than those in the Generic condition (9/31)
and this difference was reflected in the chi-square analysis, χ
(3) = 11.12, p<.05. A salient
principle character may detract attention away from SDT structure or other cogent parts of the
problem and highlight qualities of the specific person.
We have mainly construed the between-participants manipulation in terms of a specific versus
generic doctor, but another way to think about this is that we included a familiar (famous) doctor
as well. Because this categorization would be most appropriate for participants who are actually
familiar with the particular television show, we categorized participants according to their reports
of familiarity. The scores from the specific doctor condition, shown in Table 5, were analyzed
according to a 2 × 2 repeated-measures ANOVA with respect to item type and show familiarity.
Although there were no main effects of item type or familiarity, F (1, 22) < 2.91, there was a
significant interaction, F (1, 22) = 6.21, p<.05. Paired t-tests showed that the participants in the
Specific condition who were familiar with the show had poorer performance on the tutorial quiz,
scoring 10% (SD = .17) less than unfamiliar participants, but this difference was only marginally
significant, t(11) = 1.98, p = .07. However on the transfer quiz, the difference trended in the
opposite direction with familiar participants doing 6% (SD = .14) better than those who were
unfamiliar with the show, t(11) = 1.52, p = .15. This is an odd pattern of results because better
performance on the tutorial is typically expected to produce better transfer performance. In the
current case, those participants that are familiar with the television show do not initially seem
to show good learning but their transfer performance does not suffer. This result suggests that
educators should examine the role of contextualization in assessment (also recommended by
Bates & Wiest, 2004) as well as during learning.
Familiarity with the TV show did not generally hurt or help performance because when the
scores from the Generic condition were analyzed with regard to item type and show familiarity,
there were no effects of familiarity, F (1, 22) < .17, only an effect of item type, F (1, 22) = 4.61,
p<.05. These results are also shown in Table 5.
Tutorial and Transfer Results of Experiment 3 Broken Down by Familiarity with this Particular
Television Show
Generic Condition
Tutorial Transfer
Unfamiliar with TV show .53 (SD = .18) .61 (SD = .19)
Familiar with TV show .55 (SD = .21) .64 (SD = .19)
Specific Condition
Unfamiliar with TV show .52 (SD = .26) .56 (SD = .14)
Familiar with TV show .42 (SD = .22) .62 (SD = .15)
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
These results reflect a slightly different pattern from the other two experiments so far, where
transfer scores were usually close to tutorial scores and participants in the more richly contextual-
ized condition suffered in transfer. In the current experiment, the transfer scores were appreciably
better than tutorial scores. Although the same SDT pictures and graphs were used across all of
the experiments, the current tutorial had a significant overall change to the story context, from a
doctor examining distorted blood cells in order to diagnose leukemia to a doctor examining tumor
density to diagnose operability. This change may have resulted in a generally increased similarity
to the melon farmers examining melon weight to decide sweetness. Tumors and density may
provide a better match to melons and weight than cells and distortion. These changes to the story
resulted in more consistent mappings between the tutorial and transfer situations. Experiments
1 and 2 had a signal-to-signal match of sick patients to sweet melons—a counterintuitive align-
ment. However, Experiment 3’s change resulted in a signal-to-signal match between operable
tumor and sweet melons—a more intuitive, semantically congruent mapping. Previous research
has suggested that variations in the similarity and alignability between two situations can have a
strong effect on transfer (i.e., Gentner & Toupin, 1986; Gick & Holyoak, 1983; Ross, 1987; Son,
Doumas, & Goldstone, under review) .
Recall the effect of contextualization in Experiments 1 and 2—participants defaulted to
domain-specific knowledge about medical diagnosis and doctors rather than the use of SDT
principles. This may have reduced learning of structural principles that could be used in future
transfer situations (Bransford & Schwartz, 1999; Schwartz, Bransford, & Sears, 2005). However,
in Experiment 3, the familiar character personalization did not result in poorer transfer results
despite poorer performance on the tutorial quiz. This may have been due to the difference between
pretending to be a doctor when the learner is not actually a doctor versus learning about an actual
doctor who presumably knows more about medicine and has more experience than our partici-
pants. Instead of triggering schematic doctor knowledge, this manipulation may have activated
specific doctor knowledge—a perspective that may be detrimental for answering questions that
probe participants’ knowledge of SDT in the context of this doctor, but not for learning SDT
principles that can be applied to different domains.
Contextualizing elements such as activities, personal perspectives, and concrete examples are
prevalent in education. Perhaps this is, in part, a reflection of the intuition that well-grounded
understandings come from concrete experience. However, the literature on the effect of actual
concrete experience on transfer is mixed. There are examples of concrete experiences that foster
transferable knowledge (e.g., Inagaki, 1990; Mayer, Mathias, & Wetzell, 2002) but there are also
examples of knowledge that is fixed to particular aspects of the learning situation (e.g., DeLoache,
1995; Lave, 1988). Our experiments suggest that it is not that concrete experiences, activities,
and demonstrations are generally good or bad for transfer, but rather these manipulations cause
particular construals that affect learning and transfer.
As a summary, the ways we personalized instruction hurt performance on quizzes that hinged
on knowledge of abstract principles. More specifically, perspectives fostered by active experi-
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
ences as well as personalized language were detrimental to learning and subsequent transfer
(Experiments 1 and 2). Experiment 3 tested whether personalization through specific and famil-
iar details produced the same effects. The specific contextualization seemed to hinder learners
from revealing what they learned about SDT in questions about the specific celebrity doctor.
Even though the personalization provided in these experiments was relatively subtle, all three
experiments showed similar detriments for personalization. However, these analyses were only
significant when analyzed according to item; only Experiment 1 showed significant detriment of
personalization with subject analysis.
Our results should not be taken as opposing contextualization, personalization, or learner-
adapted approaches. Instead, our experiments show that the “one size fits all” approach, where
personalized contexts are simply grafted onto contents, could have negative cognitive conse-
quences. Undoubtedly, participating in activities and evoking real world knowledge is influen-
tial and can result in effective activity-specific encoding. For example, warehouse drivers in a
dairy organize information according to warehouse location and pallet size whereas consumers
typically encode by general categories (Scribner, 1985). Their activities and perspectives al-
low them to selectively encode relevant information. However, this context-bound encoding
leads to potential pitfalls. Trying to answer questions about SDT after hands-on activity or
through an active participant perspective seems to lead to construals based on common sense
knowledge about doctors, rather than SDT decision making more generally. A decontextualized
understanding may be beneficial for learning structural principles by leading to a broad, detached
understanding of a situation rather than being guided by a particular perspective.
Even though a high degree of contextualization may be detrimental to transfer of system-
level knowledge, research suggests that being able to adopt or simulate an active view might be
important for other learning situations. For example, Barab et al. (in press) has found that students
learn to apply scientific methodology better using avatars in a fully immersed virtual world
(the Quest Atlantis project: Barab, Thomas, Dodge, Squire, & Newell, 2004; Barab, Thomas,
Dodge, Carteaux, & Tuzun, 2005) than in the context of a third-person story problem clicking
through hyperlinks. Skilled understanding requires actions to transfer across contexts and active
perspectives may be a necessary part of skill acquisition. Certain problem-solving skills have
been shown to benefit from rich problem-based environments (Cognition and Technology Group
at Vanderbilt, 1992). Evidence from cognitive science (Needham & Begg, 1991) and education
(Barron et al., 1998) show that active problem-solving or project-based perspectives are good for
understanding information relevant to these skills. Framing contexts should not just be seen as
scaffolding for static knowledge but as relevant perspectives for particular actions.
Activities and frames that are fine-tuned to the learning situation at hand could be an important
educational contribution because it can be a significant improvement over real-world experience.
When students are simply thrown into real-world experience, the appropriate perspective to take
is not immediately apparent. Additionally, combinations of active and less active perspectives
can be highly effective. For instance, Chi, Roy, and Hausmann (2008) have shown that watching
interactive tutoring is as effective as actually being tutored. Projection into and identification with
an actor can be a useful pedagogical tool for fostering particular perspectives. Students can be
asked to imagine themselves in the place of an observer, participant, or protagonist, fostering a
multi-perspective view of a situation. Just as some situations are easier to imagine than others
(Hegarty, 1992), some perspectives may be easier to adopt than others. If an active perspective
is relatively easy to adopt, then perhaps educators should focus on perspectives that students are
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
less likely to spontaneously adopt. An important skill to acquire might be the ability to choose the
right perspective in order to extract relevant information or to integrate information from several
perspectives. DiSessa and Sherin (1998) argue that “Shifting the means of seeing, a fortiori, is
the core problem of conceptual change. The role of coordinate classes is to unify apparently
disparate situations and construe them in a common manner. For example, in the case of applying
SDT principles, perspectives that focus on entities such as targets and signals are more apt than
those framed in terms of doctors and patients. Framing tutorials to teach these SDT-appropriate
perspectives has benefits over simply assuming that emphasizing elements of the specific doctor
scenario will promote effective learning.
Students need meaningful understanding and perhaps activity-based contextualization or famil-
iar contexts provide meaning. However, particular perspectives lead to certain aspects becoming
highlighted relative to others. If there is a conflict between perspectives, such as seeing individual
trees or a unified forest, making trees salient can detract from understanding the forest. Scientific
and abstract understanding often requires appreciating high-level organization, so in order to em-
phasize system-level meaningfulness, one must have a perspective that can take the entire system
into account. Activities, personal perspectives, and well-known specific characters can sometimes
produce a limited or skewed understanding of the system. Models targeting levels that are relevant
for the problem at hand are more effective for explaining phenomena (Gentner & Stevens, 1983;
Wilensky & Resnick, 1999). If we aim to foster this kind of perceptual reorganization (Goldstone,
2006), we may want strategically to encourage students to create such perspectives.
If cognition is for the purpose of action (e.g., Glenberg, 1997), one’s perspective is vitally
important. However, just because we experience events through first-person, action-oriented,
detailed perspectives does not mean that emphasizing these perspectives is the optimal way to learn
about generalizable structure. Instead, pedagogical efforts to make learning more intrinsically
motivating and personalized to students should consider the cognitive effects of such perspectives.
Cognitive research can also learn an important lesson from educational efforts: The influential
role of perspective and actions in learning should be considered in theories about learning and
transfer. There are technically no context-free, perspective-free learning opportunities, so how
do we acquire and impart seemingly context-independent skills and knowledge? To examine the
intersection between transportable and generalizable structures with real-life, richly encased in
stories, details, and activities, is an important ongoing goal for both cognitive and pedagogical
Anderson, J. R., Greeno, J. G., Reder, L. M., & Simon, H. A. (2000). Perspectives on learning, thinking, and activity.
Educational Researcher, 29, 11–13. Q16
Anand, P. D., & Ross, S. M. (1987). Using computer-assisted instruction to personalize arithmetic materials for elementary
school students. Journal of Educational Psychology, 79, 72–78.
Bakker, A., & Gravemeijer, K. (2004). Learning to reason about distribution. In D. Ben-Zvi & J. B. Garfield (Eds.), The
challenge of developing statistical reasoning, literacy, and thinking (pp. 147–168). Dordrecht, Netherlands: Kluwer.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
Barab, S. A., Scott, B., Ingram-Goble, A., Goldstone, R., Zuiker, S., & Warren, S. (in press) . Contextual embodiment as Q17
a curricular scaffold for transferable understanding. Contemporary Educational Psychology.
Barab, S. A., Squire, K., & Dueber, B. (2000). Supporting authenticity through participatory learning. Educational
Technology Research and Development, 48(2), 37–62.
Barab, S. A., Thomas, M., Dodge, T., Squire, K., & Newell, M. (2004). Critical design ethnography: Designing for
change. Anthropology & Education Quarterly, 35(2), 254–268.
Barab, S. A., Thomas, M., Dodge, T., Carteaux, R., & Tuzun, H. (2005). Making learning fun: Quest Atlantis, a game
without guns. Educational Technology Research and Development, 53(1), 86–108.
Baranes, R., Perry, M., & Stigler, J. W. (1989). Activation of real-world knowledge in the solution of word problems.
Cognition and Instruction, 6, 287–318.
Barnett, S. M., & Ceci, S. J. (2002). When and where do we apply what we learn? A taxonomy for far transfer. Journal
of Experimental Psychology: General, 128, 612–637.
Barsalou, L. W. (1999). Perceptual symbol systems. Behavioral and Brain Sciences, 22, 577–609.
Barsalou, L. W. (2003). Situated simulation in the human conceptual system. Language and Cognitive Processes, 18,
Barsalou, L. W., & Sewell, D. R. (1984). Constructing representations of categories from different points of view. Emory
Cognition Project Report #2, Emory University, Atlanta, GA.
Bassok, M., & Holyoak, K. J. (1989). Interdomain transfer between isomorphic topics in algebra and physics. Journal of
Experimental Psychology: Learning, Memory, & Cognition, 15, 153–166.
Bates, E. T., & Wiest, L. R. (2004). Impact of personalization of mathematical word problems on student performance.
The Mathematics Educator, 14, 17–26.
Beach, K. D. (1995). Activity as a mediator of sociocultural change and individual development: The case of school-work
transition in Nepal. Mind, Culture, and Activity, 2, 285–302.
Beach, K. D. (1997). Socially-organized learning and learning-organized practices [Review of the book Understanding
practice: Perspectives on activity and context]. Journal of Applied Cognitive Psychology, 11, 104–105.
Bergen, B. (November, 2006). Mental simulation, embodiment, and grammar. Presented at Conceptual Structure, Dis-
course, and Language, San Diego, CA.
Bower, G. H., Black, J. B., & Turner, T. J. (1979). Scripts in memory for text. Cognitive Psychology, 11, 177–220.
Bransford, J. D., Franks, J. J., Vye, N. J., & Sherwood, R. D. (1989). New approaches to instruction: Because wisdom
can’t be told. In S. Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 470–497). Cambridge:
Cambridge University Press.
Bransford, J., Kinzer, C., Risko, V., Rowe, D., & Vye, N. (1989). Designing invitations to thinking: Some initial thoughts.
In S. McCormick & J. Zutell (Eds.), Cognitive and social perspectives for literacy research and instruction (pp.
35–54). Chicago, IL: National Reading Conference.
Bransford, J. D., & Schwartz, D. L. (1999). Rethinking transfer: A simple proposal with multiple implications. In A. Iran-
Nejad & P. D. Pearson (Eds.), Review of Research in Education, 24, 61–100. Washington, DC: American Educational
Research Association.
Bredderman, T. (1983). Effects of activity-based elementary science on student outcomes: A quantitative synthesis.
Review of Educational Research, 53, 499–518.
Brown, A. L., & Campione, J. C. (1994). Guided discovery in a community of learners. In K. McGilly (Ed.), Classroom
lessons: Integrating cognitive theory and classroom practice (pp. 229–272). Cambridge, MA: MIT Press.
Brown, J. S., Collins, A., & Duguid, P. (1989). Situated cognition and the culture of learning. Educational Researcher,
18, 32–42.
Bruner, J. S. (1962). On knowing: Essays for the left hand. Cambridge, MA: Harvard University Press.
Bruner, J. S. (1966). Toward a theory o f instruction. Cambridge, MA: Harvard University Press.
Carraher, D. W., Nemirovsky, R., & Schliemann, A. D. (1995). Situated generalization. Proceedings of the 19th annual
meeting of the International Group for the Psychology of Mathematics Education (Vol. 1, p. 234). Universidade
Federal de Pernambus, Recife, Brazil.
Chafe, W. (1994). Discourse, consciousness, and time. Chicago: University of Chicago Press.
Chi, M. T. H., Roy, M., & Hausman, R. G. M. (2008). Observing tutorial dialogues collaboratively: Insights about human
tutoring effectiveness from vicarious learning. Cognitive Science, 32, 301–341.
Cognition and Technology Group at Vanderbilt (1992). Technology and the design of generative learning environments.
In T. M. Duffy & D. Jonassen (Eds.), Constructivism and the technology of instruction: A conversation. Hillsdale,Q18
NJ: Lawrence Erlbaum Associates.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
Cordova, D. I., & Lepper, M. R. (1996). Intrinsic motivation and the process of learning: Beneficial effects of contextu-
alization, personalization, and choice. Journal of Educational Psychology, 88, 715–730.
d’Ailly, H., Murray, H. G., & Corkill, A. (1995). The cognitive effects of self-referencing. Journal of Contemporary
Educational Psychology, 20, 88–113.
d’Ailly, H., Simpson, J., & MacKinnon, G. (1997). Where should you go in a math compare problem? Journal of
Educational Psychology, 89, 562–567.
DeLoache, J. S. (1995). Early understanding and use of symbols: The model. Current Directions in Psychological Science,
4, 109–113.
DeLoache, J. S. (2000). Dual representation and young children’s use of scale models. Child Development, 71, 329–
Detterman, D. K., & Sternberg, R. (1993). Transfer on trial: Intelligence, cognition and instruction. Norwood, NJ: Ablex
DiSessa, A. A., & Sherin, B. L. (1998). What changes in conceptual change? International Journal of Science Education,
20, 1155–1191.
DiSessa, A., & Wagner, J. (2005). What coordination has to say about transfer. In J. P. Mestre (Ed.), Transfer of learning
from a modern multidisciplinary perspective. Greenwich, CT: Information Age Publishing Inc. Q19
Duchan, J. F., Bruder, G. A., & Hewitt, L. E. (1995). Deixis in narrative: A cognitive science perspective. Hillsdale, NJ:
Fernald, L. D. (1987). Of windmills and rope dancing: The instructional value of narrative structures. Teaching Psychology,
14, 214–216.
Franks, J., Bransford, J., Brailey, K., & Purdon, S. (1990). Cognitive psychology: The state of the art. In R. Hoffman &
D. Palermo (Eds.), Cognition, perception, and action: A festschrift for J. J. Jenkins. Hillsdale, NJ: Erlbaum. Q20
Gentner, D., & Stevens, A. L. (Eds.). (1983). Mental models. Hillsdale, NJ: Erlbaum.
Gentner, D., & Toupin, C. (1986). Systematicity and surface similarity in the development of analogy. Cognitive Science,
10, 277–300.
Gick, M., & Holyoak, K. (1983). Schema induction and analogical transfer. Cognitive Psychology, 15, 1–38.
Gigerenzer, G., & Hoffrage, U. (1995). How to improve Bayesian reasoning without instruction: Frequency formats.
Psychological Review, 102, 684–704.
Glenberg, A. M. (1997). What memory is for. Behavioral and Brain Sciences, 2, 1–55.
Glenberg, A. M., Guitierrez, T., Levin, J. R., Japuntich, S., & Kaschak, M. P. (2004). Activity and imagined activity can
enhance young children’s reading comprehension. Journal of Educational Psychology, 96, 424–436.
Goldstone, R. L. (2006). The complex systems see-change in education. Journal of the Learning Sciences, 15, 35–
Gorman, M. E., Plucker, J., & Callahan, C. M. (1998). Turning students into inventors: Active learning modules for
secondary students. Phi Delta Kappan, 79, 530–535.
Graesser, A. C., Bowers, C. A., Bayen, U. J., & Hu, X. (2000). Who said what? Who knows what? Tracking speakers and
knowledge in narrative. In W. van Peer & S. Chatman (Eds.), Narrative perspective: Cognition and emotion. Buffalo, Q21
NY: SUNY Press.
Greeno, J. G. (1997). On claims that answer the wrong questions. Educational Researcher, 26, 5–17.
Greeno, J. G., Smith, D. R., & Moore, J. L. (1992). Transfer of situated learning. In D. Detterman & R. J. Sternberg
(Eds.), Transfer on trial: Intelligence, cognition, and instruction (pp. 99–167). Norwood, NJ: Ablex.
Harp, S. F., & Mayer, R. E. (1998). How seductive details do their damage: A theory of cognitive interest in science
learning. Journal of Educational Psychology, 90, 414–434.
Hegarty, M. (1992). Mental animation: Inferring motion from static displays of mechanical systems. Journal of Experi-
mental Psychology: Learning, Memory & Cognition, 18, 1084–1102.
Herbel-Eisenmann, B., & Wagner, D. (2005). In the middle of nowhere: How a textbook can position the mathematics
learner. In H. L. Chick & J. L. Vincent (Eds.), Proceedings of the 29th Conference of the International Group for the
Psychology of Mathematics Education (Vol. 3, pp. 121–128). Melbourne: PME.
Hmelo-Silver, C., Duncan, R., & Chinn, C. (2007). Scaffolding and achievement in problem-based and inquiry learning:
A response to Kirschner, Sweller, and Clark (2006). Educational Psychologist, 42, 99–108.
Inagaki, K. (1990). The effects of raising animals on children’s biological knowledge. British Journal of Developmental
Psychology, 8, 119–129.
Johnson-Laird, P. N. (1983). Mental models. Cambridge, MA: Harvard University Press.
Judd, C. H. (1908). The relation of special training to general intelligence. Educational Review, 36, 28–42.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
Kirschner, P. A., Sweller, J., & Clark, R. (2006). Why minimal guidance during instruction does not work: An analy-
sis of the failure of constructivist, discovery, problem-based, experiential and inquiry-based teaching. Educational
Psychologist, 41, 75–86.
Klahr, D., & Nigam, M. (2004). The equivalence of learning paths in early science instruction: Effect of direct instruction
and discovery learning. Psychological Science, 15, 661–667.
Klausmeier, H. J. (1961). Learning and human abilities: Educational psychology. New York: Harper.
Koedinger, K. R. (2002). Toward evidence for instructional design principles: Examples from Cognitive Tutor Math 6.
In D. S. Mewborn, P. Sztajn, D. Y. White, H. G. Wiegel, R. L. Bryant, & K. Nooney (Eds.), Proceedings of twenty-
fourth annual meeting of the North American Chapter of the International Group for the Psychology of Mathematics
Education, (Vol. 1, pp. 21–49). Columbus, OH: ERIC Clearinghouse for Science, Mathematics, and Environmental
Koedinger, K. R., Alibali, M. W., & Nathan, M. J. (2008). Trade-offs between grounded and abstract representations:
Evidence from algebra problem solving. Cognitive Science, 32, 366–397.
Koedinger, K. R., & Anderson, J. R. (1998). Illustrating principled design: The early evolution of a cognitive tutor for
algebra symbolization. Interactive Learning Environments, 5, 161–180.
Kosslyn, S. M., & Thompson, W. L. (2000). Shared mechanisms in visual imagery and visual perception: Insights from
cognitive neuroscience. In M. S. Gazzaniga (Ed.), The new cognitive neurosciences (pp. 975–985). Cambridge, MA:
MIT Press.
Ku, H. Y., & Sullivan, H. J. (2002). Student performance and attitudes using personalized mathematics instruction.
Educational Technology Research and Development, 50, 21–34.
Lampert, M. (1986). Knowing, doing, and teaching multiplication. Cognition and Instruction, 3, 305–342.
Lave, J. (1988). Cognition in practice: Mind, mathematics and culture in everyday life. New York: Cambridge University
Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation. Cambridge, MA: Cambridge
University Press.
Lemke, J. (1997). Cognition, context, and learning: A social semiotic perspective. In D. Kirshner & J. A. Whitson (Eds.),
Situated cognition: Social, semiotic, and psychological perspectives (pp. 37–56). Mahwah, NJ: Erlbaum.
Lepper, M. R., & Cordova, D. I. (1992). A desire to be taught: Instructional consequences of intrinsic motivation.
Motivation and Emotion, 16, 187–208.
opez, C. L., & Sullivan, H. J. (1991). Effects of personalized math instruction for Hispanic students. Contemporary
Educational Psychology, 16, 95–100.
opez, C. L., & Sullivan, H. J. (1992). Effect of personalization of instructional context on the achievement and attitudes
of Hispanic students. Educational Technology Research & Development, 40(4), 5–13.
Lozano, S. C., Hard, B. M., & Tversky, B. (2006). Perspective taking promotes action understanding and learning. Journal
of Experimental Psychology: Human Perception and Performance, 32, 1405–1421.
Mayer, R. E. (2003). Elements of a science of e-learning. Journal of Educational Computing Research, 29, 297–
Mayer, R. E., Fennell, S., Farmer, L., & Campbell, J. (2004). A personalization effect in multimedia learning: Students
learn better when words are in conversational style rather than formal style. Journal of Educational Psychology, 96,
Mayer, R. E., Mathias, A., & Wetzell, K. (2002). Fostering understanding of multimedia messages through pre-training:
Evidence for a two-stage theory of mental model construction. Journal of Experimental Psychology: Applied, 8,
McCombs, B. L., & Whistler, J. S. (1997). The learner-centered classroom. San Francisco: Jossey-Bass.
Michael, A. L., Klee, T., Bransford, J. D., & Warren, S. (1993). The transition from theory to therapy: Test of two
instructional methods. Applied Cognitive Psychology, 7, 139–154.
Moreno, R., & Mayer, R. E. (2004). Personalized messages that promote science learning in virtual environments. Journal
of Educational Psychology, 96, 165–173.
Morris, C. D., Bransford, J. D., & Franks, J. J. (1977). Levels of processing versus transfer appropriate processing.
Journal of Verbal Learning and Verbal Behavior, 16, 519–533.
Mukhopadhyay, S., Peled, I., & Resnick, L. (1989). Formal and informal sources of mental models of negative numbers. In
G. Vergnaud, J. Rogalski, & M. Artique (Eds.), Proceedings of the 13th International Conference for the Psychology
of Mathematics Education (pp. 106–110). Paris: Universite de Paris.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
Muth, K. D. (1984). Solving arithmetic word problems: Role of reading and computational skills. Journal of Educational
Psychology, 76, 205–210.
Needham, D. R., & Begg, I. M. (1991). Problem-oriented training promotes spontaneous analogical transfer: Memory-
oriented training promotes memory for training. Memory & Cognition, 19, 543–557.
Nisbett, R. E., & Ross, L. (1980). Human inference: Strategies and shortcomings of social judgment. Englewood Cliffs,
NJ: Prentice-Hall.
Novick, L., & Hmelo, C. E. (1994). Transferring symbolic representations across non-isomorphic problems. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 20, 1296–1321.
Nunes, T., Schliemann, A. D., & Carraher, D. W. (1993). Mathematics in the streets and schools. Cambridge, England:
Cambridge University Press.
Parker, L. E., & Lepper, M. R. (1992). Effects of fantasy contexts on children’s learning and motivation: Making learning
more fun. Journal of Personality and Social Psychology, 62, 625–633.
Penner, D. E. (2001). Complexity, emergence, and synthetic models in science education. In K. Crowley, C. D. Schunn,
& T. Okada (Eds.), Designing for science (pp. 177–208). Hillsdale, NJ: Erlbaum.
Piaget, J. (1970). Piaget’s theory. In P. Mussen (Ed.), Carmichael’s manual of child psychology (Vol. 1, pp. 703–772).
New York: John Wiley & Sons.
Pichert, J. W., & Anderson, R. C. (1977). Taking different on a story. Journal of Educational Psychology, 69, 309–
uller, F. (1999) Words in the brain’s language. Behavioral and Brain Sciences, 22, 253–336.
Reeves, L. M., & Weisberg, R. W. (1994). The role of content and abstract information in analogical transfer. Psychological
Bulletin, 115, 381–400.
Renninger, K. A., Ewen, L., & Lasher, A. K. (2002). Individual interest as context in expository text and mathematical
word problems. Learning and Instruction, 12, 467–491.
Resnick, L. B. (1987). Education and learning to think. Washington, DC: National Academy Press.
Ross, B. H. (1987). This is like that: The use of earlier problems and the separation of similarity effects. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 13, 629–639.
Ross, S. M. (1983). Increasing the meaningfulness of quantitative material by adapting context to student background.
Journal of Educational Psychology, 75, 519–529.
Ross, S. M., Anand, P. G., & Morrison, G. R. (1988). Personalizing math problems: A modern technology approach to
an old idea. Educational Technology, 28, 20–25.
Rumelhart, D. E., & Ortony, A. (1977). The representation of knowledge in memory. In R. C. Anderson, R. J.
Spiro, & W. E. Montague (Eds.), Schooling and the acquisition of knowledge. Hillsdale, NJ: Lawrence Erlbaum Q22
Savery, J. R., & Duffy, T. M. (1994). Problem based learning: An instructional model and its constructivist framework.
In B. Wilson (Ed.), Constructivist learning environments: Case studies in instructional design. Englewood Cliffs, NJ: Q23
Educational Technology Publications.
Schank, R. C., & Abelson, R. P. (1977). Scripts, plans, goals, and understanding. Hillsdale, NJ: Lawrence Erlbaum
Schliemann, A. D., & Nunes, T. (1990). A situated schema of proportionality. British Journal of Developmental Psychol-
ogy , 8, 259–268.
Schmidt, H., Loyens, S., van Gog, T., & Paas, F. (2007). Problem-based learning is compatible with human cog-
nitive architecture: Commentary on Kirschner, Sweller, and Clark (2006). Educational Psychologist, 42, 91–
Schwartz, D. L., & Black, T. (1999). Inferences through imagined actions: Knowing by simulated doing. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 25. Q24
Schwartz, D. L., & Bransford, J. D. (1998). A time for telling. Cognition and Instruction, 16, 475–522.
Schwartz, D. L., Bransford, J. D., & Sears, D. A. (2005). Efficiency and innovation in transfer. In J. Mestre (Ed.),
Transfer of learning from a modern multidisciplinary perspective (pp. 1–52). Greenwich, CT: Information Age
Schwartz, D. L., & Martin, T. (2004). Inventing to prepare for learning: The hidden efficiency of original student
production in statistics instruction. Cognition & Instruction, 22, 129–184.
Schwartz, D. L., Sears, D., & Chang, J. (2007). Reconsidering prior knowledge. In M. Lovett & P. Shah (Eds.), Thinking
with data (pp. 319–344).Mahwah, NJ: Erlbaum.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
Scribner, S. (1985). Knowledge at work. Anthropology and Education Quarterly, 16, 199–206.
Son, J. Y., Doumas, A. A., & Goldstone, R. L. (under review). When do words promote analogical transfer?Q25
Spiro, R. J. (1977). Remembering information from text: The “State of Schema” approach. In R. C. Anderson, R. J.
Spiro, & W. E. Montague (Eds.), Schooling and the acquisition of knowledge (pp. 137–165). Hillsdale, NJ: Lawrence
Erlbaum Associates Inc.
Stohr-Hunt, P. M. (1996). An analysis of frequency of hands-on experience and science achievement. Journal of Research
in Science Teaching, 33, 101–109.
Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12, 257–285.
Thorndike, E. L., & Woodworth, R. S. (1901). The influence of improvement in one mental function upon the efficiency
of other functions. Psychological Review, 8, 247–261.
Uttal, D. H., Bostwick, M., Amaya, M. M., & DeLoache, J. S. (in preparation). Concreteness and symbolic development:Q26
The effect of manipulatives on children’s early literacy skills.
Velasco, C., & Bond, A. (1998). Personal relevance is an important dimension for visceral reactivity in emotional imagery.
Cognition and Emotion, 12, 231–242.
Vye, N., Bransford, J., Furman, L., Barron, B., Montavon, E., Young, M., Van Haneghan, J., & Barron, L. (1989, April).
An analysis of students’ mathematical problem solving in real world settings. Paper presented at the annual meeting
of the American Educational Research Association, San Francisco, CA.
Wason, P. C., & Shapiro, D. (1971). Natural and contrived experience in a reasoning problem. Quarterly Journal of
Psychology, 23, 63–71.
Wilensky, U., & Resnick, M. (1999). Thinking in levels: A dynamic systems perspective to making sense of the world.
Journal of Science Education and Technology, 8, 3–18.
Wright, J. P., & Wright, C. D. (1986). Personalized verbal problems: An application of the language experience approach.
Journal of Educational Research, 79, 358–362.
Full tutorial, transfer, mapping quizzes, as well as the tutorials used in Experiments 1–3 are
available online ( The quiz questions from Experiment 2 are
provided here as examples (Experiment 1 is a subset of these questions and Experiment 3 has a
slightly modified cover story).
Tutorial Quiz Questions for Experiment 2 (“You” Condition)
1. Why do you make mistakes?
a. The cell distortion evidence overlaps between sick and healthy patients.
b. When I move my decision boundary, I tend to make more mistakes.
c. I do not diagnose enough people as sick.
d. If I tried harder, I could reduce many of my errors.
2. The number of actually healthy and sick people are the same two months in a row. However,
in the second month, you are diagnosing more patients as sick when they are actually sick
and more people as sick when they are actually healthy. What must have changed in the
second month?
a. You must be diagnosing people with purer cells as sick.
b. You must be diagnosing people with more distorted cells as sick.
c. You must be diagnosing more people who are actually sick as healthy.
d. You must have become better at diagnosing sick people.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
3. For a particular level of cell distortion, you know from your experience this month that there
is a 50% chance that this level of distortion indicates cancer. What does this mean?
a. 50% of people with leukemia have this kind of cell.
b. 50% of all the patients you have seen this month have leukemia.
c. You have seen equal numbers of people with leukemia and people with distorted cells
this month.
d. You have seen equal numbers of sick people with this level of distortion and healthy
people with this level of distortion.
4. You are looking into a new blood test for finding distorted cells. How can you find out
whether this new test is better than the old one?
a. You change your decision boundary and diagnose more sick people as sick.
b. You change your decision boundary and diagnose more pure cells as healthy.
c. You do not change your decision boundary and diagnose more healthy people as
d. You do not change your decision boundary and diagnose more distorted cells as sick.
5. If you move your decision boundary all the way to include even extremely pure cells as
evidence for sickness, it means:
a. you are generally more accurate because you are able to make less errors.
b. you never diagnose people as sick when they are actually healthy.
c. you always diagnose people as healthy when they are actually sick.
d. you always diagnose people as sick when they are actually sick.
6. This month, each sick person’s cells get a little more distorted while healthy people’s cell
distortions do not get better or worse. You do not know this information. If you do not
change your decision boundary, how does this change in the population help you?
a. you increase the number of actually healthy people you diagnose as healthy.
b. you decrease the number of actually sick people you diagnose as healthy.
c. you increase the number of actually healthy people you diagnose as sick.
d. sick people become more common so you get more experience diagnosing them.
7. Which of the following decision strategies will ensure that you maximize the number of
actually healthy people you diagnose as healthy?
a. diagnose everyone as healthy.
b. look more carefully at the cell distortion levels before your diagnosis.
c. examine the ratio of sick patients with distorted cells to sick patients with pure cells
before your diagnosis.
d. examine the ratio of patients with distorted cells to patients with pure cells before your
8. Which is most likely to lead to inaccuracy in your diagnoses?
a. Sick people that develop extremely distorted cells.
b. Sick people and healthy people have similar distortion levels.
c. The people diagnosed as sick have similar distortion levels.
d. Distorted cells are more common among sick people.
9. Very distorted cells are often caused by protein bundles. Knowing this, your accuracy can:
a. improve at detecting who is actually healthy and sick.
b. improve at detecting who is actually sick.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
c. improve at detecting who is actually healthy.
d. not improve based on this information.
10. Which decisions should you try to maximize to make sure you are accurately diagnosing
both sick and healthy populations?
a. You should focus on increasing the number of sick people diagnosed as sick and
reducing the number of sick people diagnosed as healthy.
b. You should focus on increasing the number of healthy people diagnosed as healthy and
reducing the number of healthy people diagnosed as sick.
c. You should focus on increasing the number of sick people diagnosed as sick and
reducing the number of healthy people diagnosed as sick.
d. You should focus on increasing the number of sick people diagnosed as sick.
11. If you set a very high decision boundary where you only diagnose very distorted cells as
a. You are likely to make many errors in general.
b. You are likely to increase the error of diagnosing sick people as healthy.
c. You are likely to increase the error of diagnosing healthy people as sick.
d. You are likely to increase the number of sick people diagnosed as sick.
12. Two patients come into your office today. Patient A seems very weak but has a cell distortion
level of 3. The other one, patient B, seems normal but has a distortion level of 6. Over many
years, you’ve found that setting the decision boundary to a cell distortion level of 4 or more
(range from 1–7) minimizes errors. What should you do?
a. You should use your decision boundary; then diagnose patient A as healthy and
diagnose patient B as sick.
b. You want to make sure patient A gets treatment so change your decision boundary to
include level 3 only for patient A; then diagnose patient A as sick and patient B as
c. You want to make sure patient B does not get exposed to harmful radiation for no reason
diagnose both patients as healthy without changing your decision boundary.
d. You should trust your experience as a doctor and diagnose patient A as sick and patient1340
B as healthy because cell distortion should not be the only evidence that you base your
decision on.
Tutorial Quiz Questions for Experiment 2 (Both Conditions)
1. (Use provided graph.) Remember that the farmers of Chanterais do not know which melons
are sweet and which are bitter before they export them. They only know how much the
melons weigh. The farmers of Chanterais shipped all melons that met the minimum weight
of 1500 grams. How many 1750-gram melons did they export?
a. 200
b. 400
c. 600
d. 1200
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
2. (Use provided graph.) Approximately what percentage of all 1000 gram melons (1 kg) are
a. 10%
b. 25%
c. 33%
d. 50%
3. (Use provided graph.) There was a very bitter shipment of melons last year so the towns
people wanted to be extremely careful this year. They set a 1750 gram minimum weight but
they do not know which are sweet or bitter. How many melons that weighed 1500 grams
were rejected?
a. 300
b. 450
c. 500
d. 750
e. 950
4. (Use provided graph.) With the minimum weight for the pluma melon set at 1750 grams,
how many bitter pluma melons are rejected?
a. 400
b. 950
c. 1550
d. 2000
e. 2600
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
5. Some of the farmers in Chanterais debate over using a high-tech digital scale in place of
their old-fashioned analog scale. What would be evidence that the high-tech scale is a better
a. Chanterais changes their required weight and exports more sweet fruit.
b. Chanterais changes their required weight and rejects more bitter fruit.
c. Chanterais does not change their required weight and rejects more bitter fruit.
d. Chanterais does not change their required weight and rejects more light-weight fruit.
6. For 1750 gram melons, Chanterais knows from last month that there is a 25% chance that
these melons are sweet. What does this mean?
a. 25% of sweet melons will weigh 1750 grams.
b. 25% of the melons will be sweet.
c. 75% of the melons will be bitter.
d. 25% of 1750 gram melons will be sweet.
7. If Chanterais lowers their minimum weight, which of the following would happen?
a. They will export more sweet melons and fewer bitter melons.
b. They will never export bitter melons.
c. They will export less sweet melons and more bitter melons.
d. They will export more sweet melons and more bitter melons.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
8. In a particular year, there is plenty of rainfall and all the melons get about 250 grams heavier.
The prior year Chanterais exported melons that weighed 1500 grams or more. If they do not
change their policy:
a. Chanterais will ship more heavy melons that are sweet and f melons that are bitter.
b. Chanterais will reject more light melons that are sweet.
c. Chanterais will ship more melons.
d. Chanterais will reject more melons.
9. How does this graph support the idea that melon weight is a good predictor of sweet melons?
a. There are fewer heavy melons that are bitter than are sweet.
b. There are fewer light melons that are bitter than are sweet.
c. There are fewer light melons than heavy melons.
d. There are more sweet melons than bitter melons.
e. There are more heavy melons than light melons.
10. Which decisions should the farmers of Chanterais try to maximize to make sure their melons
are of high quality?
a. They should try to export more sweet melons and reject fewer sweet melons.
b. They should try to reject more bitter melons and export fewer bitter melons.
c. They should try to export more sweet melons and export fewer bitter melons.
d. They should try to export more sweet melons.
11. If the town only sells extremely heavy melons:
a. They will make fewer errors in general.
b. They will reject more sweet melons.
c. They will export more bitter melons.
d. They will export more sweet melons.
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
12. Why does Chanterais export bitter melons?
a. Because they previously had no quality control procedure.
b. They do not reject enough melons.
c. Whenever they change their minimum melon weight, they tend to make more errors.
d. Sweet and bitter melons sometimes have the same melon weight.
Analogy Questions for Experiment 2 (Both Conditions)
Italics indicate a structural mapping; ** indicate a semantic mapping.
a. A patient diagnosed as sick but is actually healthy is like what?
b. A bitter melon that is rejected.
c. A bitter melon that is accepted.
d. A sweet melon that is rejected.**
e. A sweet melon that is accepted.
What in the doctor story is most analogous to a heavy melon?
a. A patient who is sick.
b. A patient who is healthy.
c. A patient with distorted cells.
d. A patient with pure cells.**
A melon that is sweet but was rejected is analogous to:
a. a sick patient who had been diagnosed as healthy.
b. a sick patient who had been diagnosed as sick.
c. a healthy patient who had been diagnosed as sick.**
d. a healthy patient who had been diagnosed as healthy.
1. What in the melon export story is most analogous to the sick patient in the doctor scenario?
a. A sweet melon.
b. A bitter melon.**
c. An exported melon.
d. A rejected melon.
2. The patient with leukemia who has been diagnosed as sick is most like:
a. A melon that is rejected and sweet.
b. A melon that is rejected and bitter.**
c. A melon that is exported and bitter.
d. A melon that is exported and sweet.
3. An exported melon is like:
a. a patient who has been given low distortion test results.
b. a patient who has been given high distortion test results.
c. a patient who has been given a sick diagnosis.
d. a patient who has been given a healthy diagnosis.**
P1: ...
HCGI_A_358621 708.cls December 4, 2008 13:34
%Correct Resultsfor Each Quiz Question
Accuracy on Each Question Broken Down by Experiment and by Condition Within
Experiment 1 Experiment 2 Experiment 3
Experience Control “You”
Control Perspective (“He” Perspective) Perspective Generic Specific
Tutorial Questions
1 .93 1.0 .54 .73
2 .61 .50 .54 .48 .65 .48
3 .61 .69 .64 .48 .52 .48
4 .41 .38 .32 .22 .13 .12
5 .71 .56 .79 .52* .71 .45*
6 .37 .31 .36 .48 .52 .67
7 .34 0** .39 .26 .29 .15
8 .95 .88 .89 .89 .87 .76
9 .63 .63 .50 .44 .58 .39
10 .29 .26 .55 .52
11 .68 .63 .58 .48
12 .29 .19 .52 .15**
Transfer Questions
1-Exp. 1 .37 0**
1-Exp. 2, 3 .79 .70 .84 .73
2 .44 .31 .61 .37 .68 .64
3 .39 .13* .61 .37 .61 .48
4 .51 .31 .57 .59 .65 .64
5 .34 .38 .32 .26 .35 .33
6 .54 .69 .79 .63 .74 .67
7 .10 .06 .61 .48 .58 .58
8 .34 .25 .57 .52 .58 .64
9 .59 .75 .75 .52 .81 .61
10 .53 .56 .55 .61
11 .32 .22 .29 .48
12 .79 .67 .84 .85
Note. *Significant difference between the conditions, p<.05; ** p<.01.
... Individuals are then bound to struggle, and sometimes even fail. This is especially problematic since mathematics education does not usually control for content effects Lee, DeWolf, Bassok, & Holyoak, 2016), which is partly due to mathematics being primarily considered the realm of abstraction (Son & Goldstone, 2009;Goldstone & Sakamoto, 2003;Day, Motz & Goldstone, 2015). Similarly to how concreteness fading is proposed as a way to improve transfer by resorting to increasingly abstract examples (Fyfe, McNeil, Son, & Goldstone, 2014), it may be a promising route to develop a semantic congruence fading process using increasingly incongruent examples. ...
Full-text available
With its context-independent rules valid in any setting, mathematics is considered to be the champion of abstraction, and for a long time human mathematical reasoning was thought to follow nothing but the laws of logic. However, the idea that mathematics is grounded in nature has gained traction over the past decades, and the context-independency of mathematical reasoning has come to be questioned. The thesis we defend concerns the role played by general, non-mathematical knowledge on individuals' understanding of numerical situations. We propose that what we count has a crucial impact on how we count, in the sense that human's representation of numerical information is dependent on the semantic context in which it is embedded. More specifically, we argue that general, non-mathematical knowledge about the entities described in a mathematical word problem can shape its interpretation and foster one of two representations: either a cardinal encoding, or an ordinal encoding. After introducing a new framework of arithmetic word problem solving accounting for the interactions between mathematical knowledge and world knowledge in the encoding, recoding and solving of arithmetic word problems, we present a series of 16 experiments assessing how world knowledge about specific quantities can promote one of two problem representations. Using isomorphic arithmetic word problems involving either cardinal quantities (weights, prices, collections) or ordinal quantities (durations, heights, number of floors), we investigate the pervasiveness of the cardinal-ordinal distinction in a wide range of activities, including problem categorization, problem comparison, algorithm selection, problem solvability assessment, problem recall, sentence recognition, drawing production and transfer of strategies. We gather data using behavioral measures (success rates, algorithm use, response times) as well as eye tracking (fixation times, saccades, pupil dilation), to show that the difference between problems meant to foster either a cardinal or an ordinal encoding has a far-reaching influence on participants from diverse populations (N = 2180), ranging from 2nd graders and 5th graders to lay adults, expert mathematicians and math teachers. We discuss the general educational implications of these effects of semantic (in)congruence, and we propose new directions for future research on this crucial issue. We conclude that these findings illustrate the extent to which human reasoning is constrained by the content on which it operates, even in domains where abstraction is praised and trained.
... Sense-making results in some abstract view of the concrete data, i.e., a new "presumptive understanding through progressive approximations" [69, p.412]. Moreover, the process iterates, often in a trial-and-error manner, between a highly contextualized interpretation of a message to a more abstract, and necessarily decontextualized, view of the message [50,61,74]. The result of sense-making, which we refer to as 'conceptualization', may take the form of a list of concepts (themes, constructs, abstractions) and possibly the relationships between the concepts, as well as, classification criteria (rules). ...
Full-text available
The need for advanced automation and artificial intelligence (AI) in various fields, including text classification, has dramatically increased in the last decade, leaving us critically dependent on their performance and reliability. Yet, as we increasingly rely more on AI applications, their algorithms are becoming more nuanced, more complex, and less understandable precisely at a time we need to understand them better and trust them to perform as expected. Text classification in the medical and cybersecurity domains is a good example of a task where we may wish to keep the human in the loop. Human experts lack the capacity to deal with the high volume and velocity of data that needs to be classified, and ML techniques are often unexplainable and lack the ability to capture the required context needed to make the right decision and take action. We propose a new abstract configuration of Human-Machine Learning (HML) that focuses on reciprocal learning, where the human and the AI are collaborating partners. We employ design-science research (DSR) to learn and design an application of the HML configuration, which incorporates software to support combining human and artificial intelligences. We define the HML configuration by its conceptual components and their function. We then describe the development of a system called Fusion that supports human-machine reciprocal learning. Using two case studies of text classification from the cyber domain, we evaluate Fusion and the proposed HML approach, demonstrating benefits and challenges. Our results show a clear ability of domain experts to improve the ML classification performance over time, while both human and machine, collaboratively, develop their conceptualization, i.e., their knowledge of classification. We generalize our insights from the DSR process as actionable principles for researchers and designers of 'human in the learning loop' systems. We conclude the paper by discussing HML configurations and the challenge of capturing and representing knowledge gained jointly by human and machine, an area we feel has great potential.
... Notably, there is no significant relation between personalization and presentation mode (e.g., on-screen text, narration) (Dunsworth, 2005). However, Son and Goldstone (2009) found that the personalized group learned the medicine content less than the non-personalized group with the computer-based material. Yeung et al. (2009) reported no significant difference between personalized and non-personalized groups' performances in their chemistry-based study in e-learning environments. ...
Full-text available
Academia Letters, Article 1090.
... Although arithmetic word problems are a central part of mathematics education and teachers are usually encouraged to provide realworld examples to illustrate the concepts being taught (e.g. Richland, Stigler, & Holyoak, 2012;Rivet & Krajcik, 2008), the use of concrete examples to teach new concepts has also been shown to have a detrimental effect on transfer (Day, Motz, & Goldstone, 2015;Goldstone & Sakamoto, 2003;Son & Goldstone, 2009). In line with current efforts to develop new teaching methods aimed at overcoming the deleterious influence of content effects (e.g. ...
We argue that what we count has a crucial impact on how we count, to the extent that even adults may have difficulty using elementary mathematical notions in concrete situations. Specifically, we investigate how the use of certain types of quantities (durations, heights, number of floors) may emphasize the ordinality of the numbers featured in a problem, whereas other quantities (collections, weights, prices) may emphasize the cardinality of the depicted numerical situations. We suggest that this distinction leads to the construction of one of two possible encodings, either a cardinal or an ordinal representation. This difference should, in turn, constrain the way we approach problems, influencing our mathematical reasoning in multiple activities. This hypothesis is tested in six experiments (N = 916), using different versions of multiple-strategy arithmetic word problems. We show that the distinction between cardinal and ordinal quantities predicts problem sorting (Experiment 1), perception of similarity between problems (Experiment 2), direct problem comparison (Experiment 3), choice of a solving algorithm (Experiment 4), problem solvability estimation (Experiment 5) and solution validity assessment (Experiment 6). The results provide converging clues shedding light into the fundamental importance of the cardinal versus ordinal distinction on adults' reasoning about numerical situations. Overall, we report multiple evidence that general, non-mathematical knowledge associated with the use of different quantities shapes adults' encoding, recoding and solving of mathematical word problems. The implications regarding mathematical cognition and theories of arithmetic problem solving are discussed.
... Even though students in Bring-Your-Own-Device classrooms receive individualized learning plans and can choose their own learning pathways, their learning achievements are assessed through standardised frameworks, national and international tests. As a result, personalised education is limited to initial stages of students' learning and its benefits reduced to students' motivation, something that is not necessarily always linked to learning gains (Son & Goldstone, 2009). As such, the role of personalisation is socially shaped and deradicalised in the same way that the computer language LOGO was in the 1980s and 1990s (Agalianos, Whitty and Noss, 2006). ...