Anda di halaman 1dari 8

The effect of multiple internal representations on context-rich instruction

Nathaniel Lasrya兲
School of Engineering and Applied Sciences, Harvard University, Cambridge, Massachusetts 02138
and Department of Physics, John Abbott College, Montreal, Canada H9X 3L9
Mark W. Aulls
Department of Educational and Counseling Psychology, McGill University, Montreal, Canada H3A 1Y2
共Received 12 May 2006; accepted 16 August 2007兲
We discuss n-coding, a theoretical model of multiple internal mental representations. The n-coding
construct is developed from a review of cognitive and imaging data that demonstrates the
independence of information processed along different modalities such as verbal, visual, kinesthetic,
logico-mathematic, and social modalities. A study testing the effectiveness of the n-coding construct
in classrooms is presented. Four sections differing in the level of n-coding opportunities were
compared. Besides a traditional-instruction section used as a control group, each of the remaining
three sections were given context-rich problems, which differed by the level of n-coding
opportunities designed into their laboratory environment. To measure the effectiveness of the
construct, problem-solving skills were assessed as conceptual learning using the force concept
inventory. We also developed several new measures that take students’ confidence in concepts into
account. Our results show that the n-coding construct is useful in designing context-rich
environments and can be used to increase learning gains in problem solving, conceptual knowledge,
and concept confidence. Specifically, when using props in designing context-rich problems, we
find n-coding to be a useful construct in guiding which additional dimensions need to be attended
to. © 2007 American Association of Physics Teachers.
关DOI: 10.1119/1.2785190兴

I. INTRODUCTION this paper will discuss our findings, their limitations, and
provide recommendations for maximizing physics learning.
In his Nobel lecture, Richard Feynman stated: “Perhaps a
thing is simple if you can describe it fully in several different II. FROM ENCODING TO N-CODING
ways without immediately knowing that you are describing
The mind is currently conceived as a set of specialized
the same thing.”1 The converse may also be true because the
information processors that are spatially independent but
ability to represent an object or phenomenon in multiple
functionally interrelated.7 The classical neuroscience text-
ways may simplify one’s understanding of it. Several studies book by Nobel laureate Eric Kandel and coauthors8 states
have addressed the importance of multiple representations in that the “idea that different regions 关of the brain兴 are special-
learning physics.2,3 However, previous studies have defined ized for different functions is now accepted as one of the
representations primarily as external representations such as cornerstones of modern brain science.” Local brain areas
mathematical, diagrammatic, or graphical representations.2 have different processing functions and although spatially
The present paper shifts the focus to internal mental repre- independent, these processors interact9 with each other, a
sentations to provide insights into optimizing student learn- collaboration of separate entities that Minsky has poetically
ing and potentially explain the effectiveness of external rep- called the Society of Mind.10 Cognitive scientist and Nobel
resentations. Just as multiple external representations can laureate Herbert Simon had argued that this type of modular
enhance problem solving,4 previous cognitive studies have design of the mind is a special case of modular, hierarchical
shown that the construction of multiple mental representa- design of all complex systems.11
tions can also enhance problem-solving abilities.5 Multiple The process by which information is taken from the exter-
mental representations complete each other, resulting in a nal environment and “coded” for the mind is called encod-
more authentic portrayal of a problem than any single source ing. Encoding can take place in several modes. Consider for
of unimodal information.6 instance dual-coding theory. This model12 attempts to give
The first part of our paper surveys the cognitive literature equal weight to verbal and visual processing. The reasoning
to build a coherent model of internal representations that is behind dual-coding theory is that both the visual and the
sufficiently simple to be used in classrooms. The second part auditory system can be activated independently although the
presents a classroom study of the effectiveness of the pro- two systems are interconnected 共Fig. 1兲. In the neuro-
posed model of internal representations. To achieve a com- cognitive literature these independent systems are referred to
prehensive picture of effectiveness, we develop new mea- as the auditory or phonological loop and the visuospatial
sures of learning in physics. By shifting the focus away from sketchpad.13 This independence can be easily demonstrated
a Boolean view of students’ conceptual states 共that is, away by asking subjects to perform two simultaneous tasks. If both
from a perception that “they either get it or they don’t”兲, tasks are auditory 共or both visual兲, an interference occurs
these new measures acknowledge the complexity of stu- prohibiting their simultaneous completion. When one is vi-
dents’ conceptions by taking into account affective factors sual and the other auditory, simultaneous tasks can be per-
such as students’ confidence in concepts. The third part of formed. It has been suggested that the connectedness of both

1030 Am. J. Phys. 75 共11兲, November 2007 http://aapt.org/ajp © 2007 American Association of Physics Teachers 1030
Fig. 2. Auditory processing areas contrasted in darker gray 共blue online兲;
Visual processing areas contrasted in lighter gray 共orange online兲. Image by
Stephen Smith, University of Oxford Functional MRI of the Brain Center,
Copyright 2000–2001. Reproduced with author’s consent.

Fig. 1. Dual-coding theory: Information from the sensory system is selec-


tively channeled to verbal or nonverbal processing units. The encoding and disorders.21 However, encoding information about the mind
representation of words 共logogens兲 thus differs from the encoding and rep- of others is done somewhat independently from its related
resentation of images 共imagens兲. Reproduced with consent from TIP data- visual perceptual process. Autistic individuals can recognize
base 共http://tip.psychology.org/paivio.html兲. people without being able to tell their emotional state, some-
thing that autistic professor of animal studies Temple Gran-
din has likened to being as an anthropologist on Mars,23
systems allows individuals to cue from one system to the constantly trying to figure out how different behaviors trans-
other, which facilitates interpreting the environment.14 The late into emotional states. In a nutshell, the possibility to
inference is that because there are two distinct ways of en- n-code is the ability to represent information mentally along
coding and representing information in our mind, the use of multiple dimensions, for example, verbal, visual, logico-
both representations allows parallel processing to occur thus mathematical, kinesthetic, or social.
reducing computational time. From a cognitive perspective, in everyday settings the rep-
Evidence of specialized processing—such as verbal or vi- resentational modes of a problem are not sequential, for ex-
sual processing—is abundant. Through studies of injured pa- ample, auditory then visual. Multimodal inputs are physi-
tients, it has been known for over a century that the process- cally integrated and simultaneously encoded. N-coding can
ing of language is located in what are now called Broca and be related to the everyday characteristic of information in-
Wernicke’s areas in the brain. In the past 15 years, the use of puts as being multimodal, concomitant, and physically inte-
neural imagery has revealed an increasingly clear picture of grated. Lifelike multiple representations are less of a cogni-
localized processing. The impact of current advances in im- tive load by virtue of being physically linked. Multiple
aging has been likened to Ramon y Cajal’s first observation representations may also be a key feature that differentiates
of an individual nerve cell.15 Imaging data of localized pro- experts from novices. In a study of chemistry experts and
cessing shows that visual and auditory words activate 共a novices Kozma and Russel24 showed that experts not only
largely left sided兲 set of areas of the anterior and posterior organize larger chunks of knowledge, which had been well
cortex and the cerebellum, while simple arithmetical tasks described in the literature,25 but that these chunks are orga-
activate left and right occipital and parietal areas.16 There is nized across different representational modes. How can these
now also information on spatial tasks,17 on understanding of findings be applied to instruction?
the minds of others18 and of oneself,19 and even on the pro-
cessing of musical tasks.20 Encoding information from the
environment is a parallel process: One part of the brain does III. FROM THEORY TO PRACTICE
not wait in sequence to start processing if another part is
activated; more than one part can be activated at once. Thus, It is necessary to err on the side of caution when going
current imaging studies suggest that multiple forms of en- from a “descriptive learning theory” to a “prescriptive in-
coding can take place. In a similar way that dual-coding structional theory.”26 It would be ill advised to take imaging
theory urged us to consider two encoding modes, these im- findings and make explicit conclusions and recommenda-
aging findings encourage us to reconsider encoding as tions as to what instructional environments should be like.
n-coding. The term n-coding is coined here to emphasize Current educational trends, such as “brain-based” learning,
that ‘n’ is no longer two 共as in dual-coding兲 but the number have been severely critiqued27 for inferring one-to-one rela-
of identifiable encoding dimensions, which may increase tions between fundamental research findings and instruc-
with further imaging studies 共see Fig. 2兲. tional practices. Before jumping into tentative applications,
N-coding modes need not be strictly perceptual. For in- we are reminded of the wise comments of a founder of
stance, imaging data on understanding the minds of others18 American psychology, William James:28 “You make a great,
supports cognitive “theories of mind” positing the existence a very great mistake if you think that psychology, being a
of an internal “mind-reading system.”21 This modality can be science of the mind’s laws, is something from which you can
seen as rooted both in genetic and environmental settings as deduce definite programmes and schemes and methods of
illustrated by its normal developmental trajectory in instruction for immediate schoolroom use.”
children22 and its dysfunction in children with autistic Although we must be wary of translating neuroscience

1031 Am. J. Phys., Vol. 75, No. 11, November 2007 N. Lasry and M. W. Aulls 1031
findings directly into educational practice, it would be un- Table I. Schema of various treatment sections.
conscionable not to use neuroscience findings as a guide for
empirically driven educational development. Control 共TI兲 nCodLo nCodMed nCodHi
To implement n-coding in the classroom, we must avoid Cooperative-group N Y Y Y
unimodal tasks 共for example, text only兲 and design activities problem solving.
that require students to represent the information along dif- n-coding/multiple N N Y Y
ferent modalities. Designing n-coding tasks goes beyond re- representation required
placing text-based activities by hands-on activities. Tasks for task solution.
should require student representations that are verbal 共putting n-coding in task N N N Y
ideas into words兲, visual 共being able to depict a situation or presentation and task
problem兲, logico-mathematic 共ability to make inferences and solution.
deductions in context兲, kinesthetic 共for example, extracting
information through manipulations or measurements兲, and
social 共collaborate with others and share expertise兲. Having
all of these modalities at once is not always feasible. To give The second group, labelled medium n-coding group
an example of such activities, we present in the following a 共nCodMed兲, did not include emulated environments that re-
classroom experiment where students in different sections quired elaborate setups. Problems were presented in text for-
were assigned to activities based on context-rich problems mat. However, the solution of these problems required stu-
consistent with the cooperative-group problem-solving dents to inquire about the objects of the emulated
共CGPS兲 approach29 but that varied in n-coding opportunities. environment and make some measurements. Thus, the sec-
The CGPS approach was chosen because of its demonstrated ond group was presented neither with an emulated environ-
effectiveness30 and because it constitutes an exemplary inter- ment nor with physical props. To solve the problem students
active engagement approach using context-rich problems. had to enquire about the props. For instance, the actual 9 mm
The classroom experiment attempted to test whether increas- slug was not presented but was available on student request
ing n-coding opportunities in the context of cooperative- and its measurement was essential to the problem solution
group problem solving has even more beneficial effects on 共without the information that it was a 9 mm slug, student
learning. could not know for instance the initial speed of the bullet兲.
Thus, although the same problem was given and similar
measurements had to be done in both groups, only one group
IV. DESCRIPTION OF TREATMENT CONDITIONS had the props presented with the problem while the other had
Students enrolled in an algebra-based mechanics course in to enquire about them. The laboratory problems for the two
a Canadian two-year community college were given homo- first groups were designed to use measurement instruments
morphic CGPS problems. The classroom study consisted of that were familiar to students 共everyday tools such as a bath-
four sections that differed by the level of n-coding intrinsic room scale or a measuring tape, and stop watches兲.
to the environment and the problem structure. To obtain a The third group, labelled low n-coding group 共nCodLo兲,
better picture of the difference between these treatment sec- had conventional CGPS context-rich problems presented in
tions we briefly describe one CGPS activity and its imple- text format without transforming the environment. This
mentation in the different groups. The problem was based on group had fewer opportunities for multiple representations as
a popular television show, CSI: Crime Scene Investigators, none of the props were available, and all data encodable
and put students in the shoes of a detective trying to solve a along visual or tactile dimensions was presented in text form.
murder. The objective of this problem was to help students For instance, the problem stated: “a 9 mm slug was recov-
better acquire and synthesize notions of two-dimensional ki- ered from the scene.” Given the caliber of the bullet, the
nematics. The first section, the high n-coding group students could determine its muzzle speed from a table. As in
共nCodHi兲, incorporated multiple representations in the prob- the other two groups, all the information was not available in
lem presentation and required n-coding in the problem solu- the initial problem description. However, students could not
tion as well. In this section the text problems were accom- measure the missing information from various props.
panied by rich visuo-spatial presentations. Besides diagrams, The control group was comprised of traditional instruction
an actual scene reconstitution included an outline of a victim 共TI兲 laboratories, that is, highly structured “cookbook”
taped on the floor and chalk dust emulating gun powder dis- labs,31 which assessed the same learning outcomes 共for ex-
tribution so as to locate the initial horizontal position of the ample, two-dimensional kinematics兲. All the groups 共treat-
gun. Kinesthetic data was also available such as a block of ments and control兲 had the same instructor 共NL兲 so as to
wood in which a stray bullet and an actual 9 mm slug was minimize interinstructor differences and macrodifferences in
found. 共The slug graciously provided by the police technol- classroom culture. All three treatment groups required stu-
ogy program of the college.兲 Students were then given a dents to collaboratively solve problems in groups of three or
table of muzzle speeds 共that is, the initial speed of the bullet four. Students in the control section were assigned to groups
as it exits the gun兲 for various calibers. Once extracted from of two. Table I summarizes the presence or absence of char-
the block of wood it had supposedly been shot in, the slug’s acteristics for the groups.
angle of entry was to be measured 共a 9 + mm hole was
drilled at 5° from horizontal, and the slug was inserted prior V. MEASURES
to lab兲. Using all the information available, students had to
collaboratively determine the variable they were ultimately Physics understanding is traditionally measured by proce-
seeking. In this case, the initial vertical position of the bullet dural problem solving. In this study, these skills were as-
was to be found so as to find the approximate shoulder height sessed using the local physics department’s comprehensive
of the shooter to later identify him from a line-up. final examination. This exam was constructed by a commit-

1032 Am. J. Phys., Vol. 75, No. 11, November 2007 N. Lasry and M. W. Aulls 1032
tee of physics professors and was approved unanimously by instance, are students significantly more confident of correct
all those teaching the course 共10–12 instructors兲. The exam answers at the end of the semester? Also, are students more
score was also consistent as each instructor marked a single confident in their right than wrong answers before/after in-
exam question for the entire cohort 共not just for his or her struction? Here again, confidence gains can be normalized.
students兲. This procedure insured that no group had either an
exam of differing difficulty or a corrector of different gener-
osity. The correctors of the exam questions were unaware of
which students belonged to which treatment condition. C. Weighted FCI score
Students may know how to solve physics problems with-
Assuming that a five-point Likert scale can be treated as a
out having a complete conceptual understanding of the phys-
continuum 共which is implicitly done when researchers per-
ics involved.32 Therefore, conceptual understanding was also
form t-tests on Likert scale data, for instance兲, we can asso-
measured. Students were assessed the first and last week of
ciate a numerical value to each level of confidence and use
the semester with the force concept inventory 共FCI兲.33,34
this value as a factor in determining a weighted FCI score.
Hake’s normalized gain30 g was used as a gauge of the rela-
We attribute 1 point for a correct answer and −1 point for an
tive effectiveness of different approaches. The normalized
incorrect answer. Levels of confidence are values corre-
gain is defined as the ratio of the increase in correct answers
sponding to the student entry: 0 on the scale indicating
after instruction to the maximal possible increase:

冉 冊
“guessing” and 4 indicating “very confident.” A student en-
PostT − PreT tering a correct answer with maximum confidence obtains 4
g= . 共1兲 points 共1 ⫻ 4兲, and a student entering a wrong answer with
maxT − PreT
maximum confidence receives −4 points 共−1 ⫻ 4兲. Students
Note that the symbol T in Eq. 共1兲 refers to the test score 共in that are not at all sure of an answer 共confidence level 0兲 such
this case the FCI score兲 and maxT to the maximal achievable as students that are guessing, receive 0 points regardless of
grade on the test. Furthermore, PreT and PostT refer to the whether the answer is right or wrong. The 2-point true-false
test scores when administrated at the beginning and end of representation of students conceptions can now be mapped
the semester, respectively. The normalized gain may be cal- on a 9-point pseudo-continuum: from highly confident in a
culated for each student and the average student gain g can misconception 共−4兲 to highly confident in a correct concep-
be found. tion 共+4兲. The resulting total weighted scores for the 30 FCI
Students’ conceptual models are loosely organized35 as
items therefore vary between −120 and 120. The differences
can be seen when instances of a concept are correctly ex-
in weighted FCI score across all groups can then be com-
pressed in one context and not in a somewhat different
pared between both testing occasions. Here again, weighted
context.36 Because students’ conceptions seem not to fit in
FCI gains can be normalized by replacing T in Eq. 共1兲 by the
true-false categories,37 a new measurement is proposed com-
weighted FCI score 共wFCI兲 yielding
bining levels of confidence to FCI questions. In his Peer

冉 冊
Instruction approach, Mazur has shown how students’ confi-
dence levels for in-class concept tests vary at different test Post wFCI − Pre wFCI
wg = . 共2兲
times.38 It is interesting to assess students’ confidence for 120 − Pre wFCI
each FCI item at the beginning and the end of a semester.
The level of confidence expressed allows us to infer how These measures may address some of the concerns raised
strongly a conception is held. Associating a level of confi- by the interpretation of FCI scores.40 For instance, a student
dence on a five-point Likert scale 共0 = guessing, 1 = not sure, guessing a right answer would not attribute high confidence
2 = pretty sure, 3 = confident, 4 = very confident兲 with each an- to an item. Therefore, a portion of false positives 共students
swer gives a better representation of students’ conceptual guessing a right answer兲 would become identifiable. Further-
state than the true-false view. The procedure of assessing more, these measure are more comprehensive as they assess
confidences for FCI items yields three measures. cognitive changes 共conceptual change兲 and affective changes
共confidence change兲 in physics learning thus addressing two
of the three fundamental components of the mental trilogy:
A. Average level of confidence cognition, conation, that is, motivation, and affect.41
This measure represents an individual’s overall confidence Possibly the most important use of wFCI is as a diagnostic
in answering conceptual physics questions. This level of con- tool. Indeed, compiling confidence data across an item 共in-
fidence can be compared for both test times to determine the stead of across a student兲 before instruction, it is possible to
effect of treatment conditions on students’ overall confidence identify strong group misconceptions 共high confidences for
regarding physics concepts. This comparison could reveal wrong answer兲. For instance, if preinstruction results show
interesting information, particularly if an increase in confi- that the class average wFCI for a given item is highly posi-
dence were to be found in some sections more than others. tive 共closer to +4兲, the instructor can decide to briefly over-
On the other hand, students may be less confident, which view the concept and reallocate the time allotted to teaching
may occur if the new knowledge acquired is under construc- it. If the preinstruction results show that the class average
tion and not fully “compiled.”39 Preinstruction and postin- wFCI for a given item is highly negative 共closer to −4兲 in-
struction average confidence levels can also be used to com- structors would be alerted to a misconception strongly held
pute a normalized average confidence gain. by many students and could devote more time to addressing
these misconceptions. Concrete examples of wFCI uses will
B. Confidence level for right/wrong answers be presented elsewhere as they do not fit within the objec-
tives of the current study. Using the proposed measures, the
This measure isolates the reported confidence for right or central research question is “Does instruction using n-coding
wrong answers and can be contrasted at both test times. For opportunities provide better learning gains?”

1033 Am. J. Phys., Vol. 75, No. 11, November 2007 N. Lasry and M. W. Aulls 1033
Table II. Group differences in dependent learning variables. Learning gains increase as n-coding increases.

nCodHi 共n=14兲 nCodMed 共n = 14兲 nCodLo 共n = 18兲 Control 共n=15兲

FCI g 0.46± 0.06 0.42± 0.06 0.32± 0.05 0.18± 0.07


wFCI g 0.33± 0.08 0.32± 0.07 0.20± 0.04 0.14± 0.05
RConf g 0.21± 0.1 0.22± 0.15 0.09± 0.11 0.03± 0.12
Exam 83.6± 2.5 77.1± 3.1 63.3± 4.1 69.9± 2.8
共Avg±
StdErr兲

VI. SAMPLE these comparisons follows from our theoretical model. We


hypothesized that increasing n-coding should entail in-
Participants consisted of a cohort of 84 students following creased learning gains. Because the high and medium
first semester algebra-based physics, pseudo-randomly as- n-coding groups are CGPS groups that were modified by
signed by the registrar to four different laboratory sections additional n-coding opportunities, the first comparison seeks
following distinct instructional formats 共nCodHi, n = 20; differences between these n-coding transformed CGPS
nCodMed, n = 20; N-CodLo, n = 24; Cont: n = 20兲. Instruction groups and conventional CGPS groups. The additional ad-
for all four sections took place in the same conventional vantage of combining both the high and medium n-coding
physics laboratory with seating for 24 students. Of the initial groups is to increase 共double兲 the small number of students
84 students pretested, 61 were also posttested 共nCodHi, n in the sample and provide more ability to find differences
= 14; nCodMed, n = 14; nCodLo, n = 18; Cont, n = 15兲. This between the groups should they exist. We compared the high
attrition is due in part to the loss of first semester students versus medium n-coding groups to see whether adding
due to program changes, as well as decreases in attendance n-coding to the presentation increases learning gains signifi-
in the week prior to final examinations. Because all groups cantly. Finally, we grouped all three CGPS active engage-
were of similar size and each lost five to six students be- ment groups to see whether the differences expected in con-
tween the first and last administration of the FCI, the absence ceptual learning also yield confidence gains and confidence
of 共usually weaker兲 students the last week of the semester weighted FCI gains.
should not work to the advantage of any particular group.
The pre- and post-FCI scores, related confidence levels, and
final exam scores were collected for these 61 participants. VIII. RESULTS
The average value per group for each of the four measured
VII. ANALYSIS dependent variables is presented in Table II.
These results show that the high n-coding group outper-
Using a multivariate analysis of variance 共MANOVA兲, the forms every other group; the exception being a marginal dif-
treatment condition was chosen as the independent variable ference between the medium n-coding and high n-coding
and the FCI, wFCI, right answer confidence 共RConf兲 and the groups on right answer confidence gains. These results also
final exam grade 共Exam兲 were taken as the multiple depen- show that medium n-coding outperformed the conventional
dent variables. These four variables were chosen because CGPS and Control sections for each variable. The conven-
each gives some information the others do not. For instance, tional CGPS group did better than the Control section on all
problem-solving aptitude is a measure of learning. However, measures aside from the final exam grade. Thus it seems that
students may learn how to solve problems algorithmically the greater the n-coding, the greater the learning gains. The
without full conceptual understanding.32 The FCI therefore remaining question is “Are these differences statistically sig-
adds a contribution by measuring conceptual learning. Be- nificant or are they due to chance?”
cause answers on the FCI are either right or wrong and stu- In comparing the difference in these dependent variables
dents’ conceptions seem not to fit in Boolean categories,37 a between groups using a MANOVA, the hypothesis of “no
confidence weighted FCI was developed to create a richer overall treatment effect” between groups was significantly
scale. Finally, because a robust use of a conceptual model rejected 共p = 0.009兲. Because the MANOVA shows that the
entails a good confidence in correct concepts, the right an- groups differ significantly in outcome, the question remains
swer confidence gain was also sought to complete the pic- which groups differ? The differences in the MANOVA for
ture. A MANOVA is an appropriate statistical measure here the three planned comparisons between groups are presented
because it seeks differences between groups by adding the in Table III.
contributions from each dependent variable 共dependent vari-
ables must be linearly independent of one another兲. Thus,
differences under one specific variable such as the FCI may Table III. MANOVA planned comparison. Higher n-coding CGPS groups
not be significant between groups—particularly with small differ statistically from classical CGPS. All CGPS together statistically dif-
sample sizes such as those in this study—but the combined fer from control.
effect of the different variables may result in significant dif-
p
ferences.
Three planned group comparisons seek differences be- nCodMed vs nCodHi 0.740
tween n-coding CGPS 共nCodHi&Med兲 and conventional nCodHi&Med vs nCodLo 0.003
CGPS 共nCodLo兲; high versus medium n-coding groups; and nCodHi&Med&Lo vs Cont 0.0497
all CGPS groups versus control group. The rationale behind

1034 Am. J. Phys., Vol. 75, No. 11, November 2007 N. Lasry and M. W. Aulls 1034
Table IV. IE versus TI groups under each dependent variable. Learning methods in fostering learning, and that this enhancement
differences previously found between IE and TI are found to extend to goes beyond changes in Boolean conceptual states as mea-
student concept-confidence gains. Also, traditional problem solving 共exam兲 sured by FCI scores.
is not affected by IE method.
No significant difference was found between the high and
nCodHi&Med&Lo medium n-coding groups. Recall that the difference between
vs Cont the two n-coding CGPS groups 共nCodMed and nCodHi兲 was
that although both required activation of multimodal input
ANOVA: FCI g p = 0.004 共verbal, visual, logico-mathematical, kinesthetic, and social兲
ANOVA: wFCI g p = 0.001 for problem solution, only one 共nCodHi兲 also had all these
ANOVA: Exam p = 0.217 multimodal stimuli in its presentation 共see Table I兲. It was
ANOVA: RConf g p = 0.0618 expected that as the n-coding opportunities increase, the out-
MANOVA p = 0.0497 come gains should also increase. The remaining question is
why did the two n-coding groups not differ from one an-
other?
To better understand why no difference exists, the previ-
These data shows that the two n-coding CGPS groups ous question can be reformulated as: “Why doesn’t the intro-
共nCodMed and nCodHi兲 differed statistically from the con- duction of multiple representations in the problem presenta-
ventional CGPS approach 共p = 0.003兲. However, the two tion contribute significantly?” Constructivism may hold the
n-coding CGPS groups 共nCodMed and nCodHi兲 did not dif- answer to this question. Essentially, constructivism holds that
fer statistically from each other 共p = 0.74兲, although both in students construct new knowledge from existent knowledge.
combination did significantly differ 共p = 0.003兲 from the con- As well put by Resnick and Hall42 constructivism “confirms
ventional CGPS 共nCodLo兲 group. All the CGPS groups were Piaget’s claim that people must construct their understand-
significantly different from the TI control group 共p ing; they do not simply register what the world shows or tells
= 0.0497兲. The differences between the interactive engage- them, as a camera or a tape recorder does. To “know” some-
ment 共IE兲 and TI groups under the four dependent measures thing, indeed even to memorize effectively, people must
are presented in Table IV. build a mental representation that imposes order and coher-
These results show that besides the expected differences in ence on experience and information. Learning is interpretive
FCI normalized gains 共p = 0.004兲, significant differences and inferential; it involves active processes of reasoning and
were also found in the weighted FCI scores 共p = 0.001兲, and a a kind of ‘talking back’ to the world—not just taking it as it
comes.”
tendency in right answer confidence 共p = 0.0618兲 is also ob-
Having multimodal information necessary for task
served. Furthermore, IE groups performed as well 共or more completion allows students to construct multiple representa-
precisely, nonsignificantly better兲 than TI groups in tradi- tions along modalities that are not usually activated, and use
tional problem solving. Therefore, the previously reported these representations to “talk back to the world.”42 However,
difference between IE and TI did not reside in FCI gains giving multiple representations to students in the problem
alone as previously reported,30 but extended to our new con- presentation may contribute only minimally in constructing
fidence related measures as well. Furthermore, although non- their own representation. The effectiveness of “handing out”
traditional pedagogies may be thought to hinder traditional representations in the problem presentation runs counter to
problem solving skills, IE students performed 共nonsignifi- the notion that students ought to construct their own repre-
cantly兲 better than TI students in the final examination. sentation.
The advantage of this negative result is that the implemen-
IX. DISCUSSION tation of n-coding strategies need not be as laborious as pre-
viously thought. Implementing emulated lifelike environ-
We set out to measure the effect of adding multiple repre- ments is not straight forward. Entire classroom settings need
sentations in a cooperative-group problem-solving environ- to be reorganized and physical props need to be placed in
ment. Our results show that increases in representational mo- meaningful arrangements. The results of this study show that
dalities resulted in increases in conceptual learning and the processing of information 共n-coding tasks兲 is more im-
confidence measures, a finding we attribute to the increase in portant than simulating a lifelike environment. From a prag-
n-coding required from students. We therefore find that matic instructional perspective, this reduces to being con-
n-coding is an effective construct in understanding how to scious of a variety of props and settings that make abstract
enhance conceptual change. problems lifelike and meaningful. Many teachers do use
The second significant difference found was between all props. This study puts an emphasis on the nature of lifelike
CGPS groups and the traditional control section 共p learning by specifying which additional dimensions 共for ex-
= 0.0497兲. These findings replicate meta-analytic differences ample, verbal, visual, logico-mathematical, kinesthetic, so-
found between TI and IE.30 However, the meta-analytic find- cial兲 need to be attended to when using props in designing
ing reported primarily normalized FCI gain differences. ill-structured instructional problems.
Here, this difference extends to confidence weighted FCI
gains and right answer confidence. This result suggests that X. LIMITATIONS
affective gains in confidence accompany the cognitive gains
made by students in nontraditional instruction. Although The results of this study and their generalizabiltiy are af-
more emphasis was given to concepts, students in all CGPS fected by several limitations. Because the sample sizes of
groups did not differ from the control group 共p = 0.217兲 in these groups were relatively small, replications of this study
traditional problem solving. Taken together, these findings would be needed to further validate the results. As an aside,
confirm that IE methods are significantly better than TI small sample sizes also have notable advantages. Small

1035 Am. J. Phys., Vol. 75, No. 11, November 2007 N. Lasry and M. W. Aulls 1035
sample sizes lead to low statistical power, that is, they are conceptual gains, we also found affective differences to be
unlikely to detect effects because the between-group differ- associated with the change of confidence students had in
ences sought are drowned in the within-group fluctuations concepts. The implication is twofold. First, a richer non-
共like a too low signal-to-noise ratio兲. Yet, if a statistically Boolean picture of students’ conceptions can be derived us-
significant result 共p ⬍ 0.05兲 is found with small sample sizes, ing confidence weighted FCI scores and other confidence
we must infer that the effect size 共or signal兲 is very large. measures such as the change in average confidence or the
Conversely, a major limitation of very large sample-size change in confidence for specifically right or wrong answers.
studies is that most variables turn out to be statistically sig- Second, using these measures, our model of internal mental
nificant even when their effect-sizes are minute. representations, n-coding, has a positive effect on learning
A second limitation of this study resides in its design com- both at the cognitive and affective levels. Teachers wanting
prised of three treatment sections and one control group. All to enhance the learning of their students may opt for using
groups were taught by the same instructor 共NL兲. This con- ill-structured context-rich problems29 where the information
figuration was chosen in order to minimize interinstructor in a laboratory environment can be represented in multiple
differences. However, acting as one’s own control can also ways and where the data measured from various everyday
be seen as a limitation. Much time and effort was invested in objects serve to solve the problem.
creating a range of meaningful activities. The instructor and
primary investigator’s enthusiasm during the treatment sec- ACKNOWLEDGMENTS
tions probably cannot be compared to his involvement in the
“cookbook” control section. On the other hand, involving a The authors thank Michael Witmann, Richard Hake, and
different instructor would have been a limitation at least as an unknown reviewer for the constructive feedback that has
severe. Variables in educational research contexts are ex- helped to improve the quality of this manuscript.
tremely difficult to isolate, and this concern is but one ex-
ample of why research in education is “not rocket science, a兲
Author to whom correspondence should be addressed. Electronic mail:
it’s much harder.”43 A mitigating factor, however, is the dis- lasry@seas.harvard.edu
tance instructors have in CGPS approaches. As a student- 1
R. Feynman, “The development of the space-time view of quantum elec-
centered activity, it can be argued that the real involvement trodynamics,” 具nobelprize.org/physics/laureates/1965/feynman-lecture.
in the CGPS sections was that of a “guide on the side,” html典.
2
leaving learners to construct knowledge and negotiate mean- D. E. Meltzer, “Relation between students’ problem-solving performance
and representational format,” Am. J. Phys. 73, 463–478 共2005兲.
ing between themselves rather than that of a “sage on the 3
A. Van Heuvelen and X. L. Zou, “Multiple representations of work-
stage” who directs and controls each step of the instructional energy processes,” Am. J. Phys. 69共2兲, 184–194 共2001兲; R. J. Dufresne,
process. W. J. Gerace, and W. J. Leonard, “Solving physics problems with mul-
A third limitation of this study is that it takes as a starting tiple representations,” Phys. Teach. 35, 270–275 共1997兲; J. Larkin, “Un-
assumption that guided inquiry is a culturally and socially derstanding and problem solving in physics,” in Research in Science
acceptable instructional strategy. In cultures where instruc- Education: New Questions, New Directions, edited by J. Robinson 共Cen-
ter for Educational Research and Evaluation, Louisville, CO, 1982兲, pp.
tion is expected to be formal and didactic to be effective,
115–130.
such as in Singapore,44 such active engagement approaches 4
P. Kohl and N. D. Finkelstein, “Representational format, student choice,
may not succeed. Cultural expectation of formal and didactic and problem solving in physics,” in PERC Proceedings 2004 共AIP Pub-
learning is also inherent in western college settings, particu- lishing, MD, 2004兲. Online at http://www.colorado.edu/physics/
larly as students advance in their program. Indeed, the suc- EducationIssues/paper/PK_PERC.pdf.
5
cess of this study is not unrelated to the fact that the partici- R. J. Spiro and J. Jehng, “Cognitive flexibility and hypertext: Theory and
pating students were mostly first semester students unaware technology for the nonlinear and multidimensional traversal of complex
subject matter,” in Cognition, Education, and Multimedia: Exploring
of standard college instructional formats. Had these students Ideas in High Technology, edited by R. Spiro and D. Nix 共Lawrence
been finishing students who had been accustomed to formal Erlbaum Associates, Hillsdale, NJ, 1990兲, pp. 163–204.
and didactic college teaching, they might have been less 6
S. Ainsworth, “The functions of multiple representations,” Comput.
likely to embrace the CGPS formats because they place dif- Educ. 33共2–3兲, 131–152 共1999兲.
7
ferent constraints on students. Because student involvement Steven Pinker, How the Mind Works 共W. W. Norton, New York, 1997兲.
8
is a central part of any interactive engagement method, this Eric R. Kandel, James H. Schwartz, and Thomas M. Jessell, Principles of
Neural Science, 4th ed. 共McGraw-Hill, New York, 2000兲.
issue may limit the application of these findings to introduc- 9
F. Crick and C. Koch, “Are we aware of neural activity in primary visual-
tory courses. cortex,” Nature 共London兲 375, 121–123 共1995兲; A. R. Haig, E. Gordon,
V. De Pascalis, R. A. Meares, H. Bahramali, and A. Harris, “Gamma
XI. CONCLUSIONS activity in schizophrenia: Evidence of impaired network binding?” Clin.
Neurophysiol. 111, 1461–1468 共2000兲.
10
Information presented to problem solvers in the everyday Marvin Lee Minsky, The Society of Mind 共Simon and Schuster, NY,
world comes in different forms. To solve an everyday prob- 1986兲.
11
Herbert Alexander Simon, The Sciences of the Artificial 共MIT Press,
lem we usually obtain information by looking, listening, Cambridge, 1969兲.
touching, and interacting with people. It is reasonable to be- 12
J. M. Paivio and A. Clark, “Dual coding theory and education,” Educ.
lieve that problem solving can be optimized by allowing Psychol. Rev. 71, 64–73 共1991兲; Allan Paivio, Mental Representations: A
multiple modes of information to be represented and pro- Dual Coding Approach 共Oxford U. P., New York, 1986兲.
13
cessed in parallel. We have shown that adding multiple in- John R. Anderson, Cognitive Psychology and its Implications, 6th ed.
ternal representations, labeled n-coding, is an effective strat- 共Worth Publishers, New York, 2005兲; Allan D. Baddeley, Human
Memory: Theory and Practice 共Allyn and Bacon, Boston, MA, 1998兲,
egy to enhance learning. To assess this effectiveness, we rev. ed.
developed new measures and made use of students’ confi- 14
L. P. Rieber, Computers, Graphics and Learning 共WCB Brown & Bench-
dence in concepts to better portray their knowledge states. mark, Madison, WI, 1994兲; T. J. Simpson, “Message into medium: An
Thus, beyond finding cognitive differences measured by FCI extension of the dual coding hypothesis,” in Imagery and Visual Literacy:

1036 Am. J. Phys., Vol. 75, No. 11, November 2007 N. Lasry and M. W. Aulls 1036
Selected Readings from the Annual Conference of the International Visual 共1999兲; P. Heller and M. Hollabaugh, “Teaching problem-solving through
Literacy Association (26th, Tempe, Arizona, October 12–16, 1994) cooperative grouping. II. Designing problems and structuring groups,”
共1995兲, pp. 2–10. Am. J. Phys. 60共7兲, 637–644 共1992兲; P. Heller, R. Keith, and S. Ander-
15
Santiago R. y Cajal, E. Horne Craigie, and Juan Cano, Recollections of son, “Teaching problem-solving through cooperative grouping. I. Group
My Life 共The American Philosophical Society, Philadelphia, 1937兲. versus individual problem-solving,” Am. J. Phys. 60共7兲, 627–636 共1992兲.
16 30
M. I. Posner, “Neural systems and individual differences: A commentary R. R. Hake, “Interactive-engagement versus traditional methods: A six-
on Frames of Mind: The Theory Of Multiple Intelligences,” Presented at thousand-student survey of mechanics test data for introductory physics
the 2003 AERA conference, 具www.tc.umn.edu/~athe0007/BNEsig/ courses,” Am. J. Phys. 66共1兲, 64–74 共1998兲.
papers/PosnerArticleMI.pdf典. 31
17
Wilbert James McKeachie and Barbara Hofer, McKeachie’s Teaching
M. Corbetta and G. L. Shulman, “Control of goal-directed and stimulus- Tips: Strategies, Research, and Theory for College and University Teach-
driven attention in the brain,” Nature Reviews Neuroscience 3共3兲, 201– ers, 11th ed. 共Houghton Mifflin Co., Boston, 2002兲.
215 共2002兲. 32
E. Kim and S. J. Pak, “Students do not overcome conceptual difficulties
18
U. Frith and C. Frith, “The biological basis of social interaction,” Curr. after solving 1000 traditional problems,” Am. J. Phys. 70共7兲, 759–765
Dir. Psychol. Sci. 10, 151–155 共2001兲. 共2002兲.
19
D. Gusnard, E. Akbudak, G. Shulman, and M. E. Raichle, “Role of me- 33
I. A. Halloun, R. R. Hake, E. P. Mosca, and D. Hestenes, “Force Concept
dial prefrontal cortex in a default mode of brain function,” Neuroimage Inventory,” in Peer Instruction: A User’s Manual, edited by Eric Mazur
13共6兲, S414–S414 共2001兲. 共Prentice Hall, Upper Saddle River, NJ, 1995兲; D. Hestenes, M. Wells,
20
R. J. Zatorre, “Brain imaging studies of musical perception and musical and G. Swackhammer, “Force concept inventory,” Phys. Teach. 30, 141–
imagery,” J. New Music Res. 28共3兲, 229–236 共1999兲. 158 共1992兲.
21
S. Barron-Cohen, “How to build a baby that can read minds: Cognitive 34
L. C. McDermott and E. F. Redish, “Resource letter: PER-1: Physics
mechanisms in mindreading,” Cah. Psychologie Cognitive 13, 513–552
education research,” Am. J. Phys. 67共9兲, 755–767 共1999兲.
共1994兲. 35
22 I. A. Halloun and D. Hestenes, “Common-sense concepts about motion,”
Alison Gopnik, Andrew N. Meltzoff, and Patricia K. Kuhl, The Scientist
Am. J. Phys. 53共11兲, 1056–1065 共1985兲.
in the Crib: Minds, Brains, and How Children Learn 共William Morrow, 36
D. Hestenes and I. A. Halloun, “Interpreting the force concept inventory:
New York, 1999兲.
23 A response to March 1995 critique by Huffman and Heller,” Phys. Teach.
Oliver W. Sacks, An Anthropologist on Mars: Seven Paradoxical Tales
33, 504–506 共1995兲.
共Vintage Books, New York, 1996兲. 37
24 L. Bao and E. F. Redish, “Model analysis: Assessing the dynamics of
R. B. Kozma and J. Russell, “Multimedia and understanding: Expert and
novice responses to different representations of chemical phenomena,” J. student learning,” 具arxiv.org/abs/physics/0207069典.
38
Res. Sci. Teach. 34共9兲, 949–968 共1997兲. Eric Mazur, Peer instruction: A User’s Manual 共Prentice Hall, Upper
25
R. Glaser, “Expertise and learning: How do we think about instructional Saddle River, NJ, 1997兲.
39
processes now that we have discovered knowledge structures,” in Com- E. F. Redish, R. E. Scherr, and J. Tuminaro, “Reverse-engineering the
plex Information Processing: The Impact of Herbert A. Simon, edited by solution of a ‘simple’ physics problem: Why learning physics is harder
D. Klahr and K. Kotovsky 共Erlbaum, Hillsdale, NJ, 1989兲, pp. 269–282; than it looks,” Phys. Teach. 44, 293–300 共2006兲.
40
R. Glaser and M. Chi, Overview in The Nature of Expertise, edited by M. C. Henderson, “Common concerns about the force concept inventory,”
Chi, R. Glaser, and M. Farr 共Erlbaum, Hillsdale, NJ, 1988兲, pp. xv–xxviii. Phys. Teach. 40, 542–547 共2002兲; D. Huffman and P. Heller, “What does
26
J. Bruner, “Some theorems on instruction illustrated with reference to the force concept inventory actually measure?” Phys. Teach. 33, 138–
mathematics,” in The Sixty Third Yearbook of The National Society for 143 共1995兲; R. N. Steinberg and M. S. Sabella, “Performance on
The Study of Education, edited by E. Hilgard 共National Society For The multiple-choice diagnostics and complementary exam problems,” Phys.
Study Of Education, Chicago, 1964兲. Teach. 35, 150–155 共1997兲.
41
27
J. T. Bruer, “Education and the brain: A bridge too far,” Educ. Researcher Ulric Neisser, Cognitive Psychology 共Appleton-Century-Crofts, New
26共8兲, 4–16 共1997兲; J. T. Bruer, “Brain science, brain fiction,” Educ. York, 1967兲.
42
Leadership 56共3兲, 14–18 共1998兲; John T. Bruer, The Myth of the First L. B. Resnick and M. W. Hall, “Learning organizations for sustainable
Three Years: A New Understanding of Early Brain Development and education reform,” Daedalus 127, 89–118 共1998兲.
43
Lifelong Learning 共Free Press, New York, 1999兲. E. F. Redish, “Millikan Lecture 1998: Building a science of teaching
28
William James, Talks to Teachers on Psychology and to Students on Some physics,” Am. J. Phys. 67共7兲, 562–573 共1999兲.
44
of Life’s Ideals 共Harvard University Press, Cambridge, MA, 1983兲. N. Geoghegan and D. Geoghegan, “Etic and emic views of early child-
29
P. Heller and K. Heller, “Cooperative group problem solving in physics,” hood education in Singapore,” J. Aust. Res. Early Childhood Educ. 11,
具groups.physics.umn.edu/physed/Research/CGPS/GreenBook.html典 1–12 共2004兲.

1037 Am. J. Phys., Vol. 75, No. 11, November 2007 N. Lasry and M. W. Aulls 1037

Anda mungkin juga menyukai