See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/233661327

Learning by analogy: Discriminating between potential analogs

Article  in  Contemporary Educational Psychology · January 2010

DOI: 10.1016/j.cedpsych.2009.09.001

CITATIONS READS

86 137

2 authors:

Lindsey Engle Richland Ian M Mcdonough

University of California, Irvine University of Alabama
62 PUBLICATIONS   1,456 CITATIONS    59 PUBLICATIONS   647 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Memory Training for Adults with Mild Cognitive Imapirment View project

Analogical Reasoning Development View project

All content following this page was uploaded by Lindsey Engle Richland on 09 November 2017.

The user has requested enhancement of the downloaded file.

 Contemporary Educational Psychology 35 (2010) 28–43

Contents lists available at ScienceDirect

Contemporary Educational Psychology

journal homepage: www.elsevier.com/locate/cedpsych

Learning by analogy: Discriminating between potential analogs

Lindsey E. Richland a,*, Ian M. McDonough b
a
University of California, Irvine, Department of Education, United States
b
University of Chicago, Department of Psychology, United States

a r t i c l e i n f o a b s t r a c t

Article history: The ability to successfully discriminate between multiple potentially relevant source analogs when solv-
Available online 9 September 2009 ing new problems is crucial to proficiency in a mathematics domain. Experimental findings in two differ-
ent mathematical contexts demonstrate that providing cues to support comparative reasoning during an
Keywords: initial instructional analogy, relative to teaching the same analogs and solution strategies without such
Cognition cues, led to increased ability to discriminate between relevant analogs at a later test. Specifically, provid-
Analogy ing comparative gestures and visibly aligned source and target problems during initial learning led to
Comparison
higher rates of positive extension of learning to new contexts, and lower rates of susceptibility to mis-
Problem solving
leading contextual features, both immediately and after a week delay.
Ó 2009 Elsevier Inc. All rights reserved.

1. Introduction learned principles from memory (Bransford, Brown, & Cocking,

2000; National Mathematics Advisory Panel, 2008a, 2008b).
Instructional analogies provide opportunities for teachers to While essential, learners often show difficulties in noticing
clarify similarities and differences among problems, concepts, commonalities between previously learned and new mathematics
and procedures; or among misconceptions and correct strategies across contexts (see Bassok & Holyoak, 1993; National Research
(see Brown & Clement, 1989; Chen & Klahr, 1999; Clement, Council, 2001; Novick & Holyoak, 1991). In the classroom, learners’
1993; Duit, 1991; English, 2004; Gentner, Loewenstein, & Thomp- failure to recognize word problems as comprising known concepts
son, 2003; Kolodner, 1997). Such analogies may compare a math- was recently cited by teachers as one of the most pressing prob-
ematical representation to a non-math entity (e.g., balancing lems in teaching algebra (National Mathematics Advisory Panel,
equations is like balancing a scale), but they may also compare 2008a, 2008b). Word problems and equations may appear different
two math concepts or methods. For example, a teacher might com- at a surface level, even though the underlying mathematical struc-
pare percent and ratio problems, demonstrating their common ture is common.
structure. While experts easily identify such concepts as similar, The literature in analogical reasoning makes a distinction be-
student learners may miss such connections unless they are very tween correspondences based upon surface features, or appear-
explicitly taught. Such comparisons form a backbone of mathemat- ance (e.g., a word problem about trains) and those based on deep
ical thinking and learning, enabling students to build understand- structure (e.g., mathematical structure) (Gentner, 1983). In a math-
ing of new topics based on their prior knowledge. ematical instructional context, therefore, there are at least two lev-
Experimental studies have demonstrated that encouraging stu- els at which mathematics problems can share correspondences
dents to draw such connections facilitates learners’ ability to use with previously instructed problems. New problems may or may
the concepts or methods in a future context (e.g., Rittle-Johnson not share mathematical structure such that they can be solved in
& Star, 2007). However, students must also be able to regulate similar ways. In addition, they may or may not appear similar at
and resist over-extensions of potential analogies. For instance in a surface level independent of mathematical structure. Thus, prob-
the above example, one cannot extrapolate from the ability to lems that are mathematically similar may appear different (e.g.,
add percentages (10% + 5% = 15%) to adding ratios two word problems appear different if one is about pizza and the
(3:5 + 3:5 = 6:10). Accordingly, students must be able to identify other about dividing work hours, even if they are mathematically
and differentiate the conceptual structure of mathematical repre- equivalent), and problems that are mathematically different may
sentations in order to identify relationships with previously appear similar (e.g., two word problems about trains that are math-
ematically dissimilar).
This distinction has proven useful in understanding the chal-
lenges for mathematics students. When contexts are less well
* Corresponding author. Address: University of California, Irvine, 2001 Berkeley
Place, Irvine, CA 92697, United States. understood, novices tend to map correspondences based on sur-
E-mail address: [email protected] (L.E. Richland). face features (e.g., Reed, 1987; Ross & Kennedy, 1990; Schoenfeld

0361-476X/$ - see front matter Ó 2009 Elsevier Inc. All rights reserved.
doi:10.1016/j.cedpsych.2009.09.001
 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43 29

& Herrmann, 1982), while they tend to map better-understood of retention as well as using measures of schematic, conceptual
contexts based on deep structure (e.g., Chi, Feltovich, & Glaser, understanding.
1981; Chi & Ohlsson, 2005, chap. 14; Novick, 1988; Schoenfeld & Despite these many studies finding benefits of analogical instruc-
Herrmann, 1982). Mapping correspondences based on mathemat- tion, providing an instructional analogy far from guarantees sponta-
ical structure leads to more flexible, expert-like mathematical pro- neous use of the analogs on a further test. Several studies have
ficiency. Thus, seeking instructional strategies for improving shown little benefit for various strategies designed to encourage
learners’ ability to spontaneously notice and attend to structural learners to map correspondences during instruction. Reed (1989)
correspondences is a crucial research goal. found no benefits for providing two algebra analogs during instruc-
tion and prompting comparison using a paper–pencil worksheet
2. Cues to draw attention to relational similarity packet. He argued that this failure may have been due to the high do-
main knowledge required in algebra learning. Similarly Gerjets,
The current study experimentally tests an instructional strategy Scheiter, and Catrambone (2006) found no benefits of having learn-
for improving learners’ likelihood of noticing and learning from ers generate self-explanations during analogical learning for molar
instructional analogies. Specifically, the study uses videotaped and modular worked examples (problems taught with focus on
instruction in a laboratory setting to determine whether cues to the overall solution strategy or broken into smaller, more easily pro-
support learning from a presented comparison would impact stu- cessed meaningful components, respectively). In related work, Sche-
dents’ later rates of positive and negative extension on a posttest. iter, Gerjets, and Catrambone (2006) did find that static pictures and
The tested instructional strategy derived from two sources. cues requesting learners to visually imagine the steps of worked
First, the basic literature on analogical reasoning suggests that ex- examples facilitated learning, but this was not the case when dy-
plicit cuing and reducing processing demands should improve namic animations were provided. The authors suggested that higher
learning. Second, a video analysis of typical 8th grade mathematics processing load was necessary to interpret and remember what had
instruction in the US and two higher achieving countries, China been shown in the animation since it was no longer available
(Hong Kong) and Japan, identified classroom-relevant strategies (whereas the crucial steps were available when presented in static
for cuing and reducing processing demands, and revealed that photographs). By contrast, reducing cognitive load by breaking
teachers in both the higher achieving countries provided more sup- problem analogs into subgoals and chunking steps into labeled
port for learning from analogies than did the US teachers. This groupings facilitated learning and transfer (Catrambone, 1998).
background is briefly reviewed below, and then the intervention Many of the existing studies seem to coalesce on this point:
is described more specifically. higher cognitive load reduces learners’ ability to focus on struc-
tural commonalities during instruction, which makes them less
2.1. Basic studies of analogy likely to benefit from the instructional analogy. Instead, learners
tend to focus on surface similarities when under load (e.g., ‘‘Ah, this
One line of well-replicated studies has shown that providing problem is about pizzas. I just have to remember how we solved
cues at the time of final test improves the likelihood that learners that last pizza problem”). Basic studies of analogical reasoning sup-
notice the relations to prior instructed problems (Gick & Holyoak, port this view. Adding a working memory load when undergradu-
1980). These cues may take the form of general reminding state- ates solved picture analogy problems led to greater rates of
ments that encourage participants to consider prior instruction, mapping correspondences based on surface features rather than
or of more specific information about which elements might be on structural correspondences (Tohill & Holyoak, 2000; Waltz,
mapped successfully (e.g., Blessing & Ross, 1996; Novick, 1988; Lau, Grewal, & Holyoak, 2000). Increasing working memory load
Novick & Holyoak, 1991). by making source and target analogs not visible together, as in
A second strategy has involved providing multiple analogs dur- the study of Rittle-Johnson and Star (2007), also made participants
ing instruction and encouraging learners to map the structural cor- more susceptible to distraction from surface features and to the
respondences as a learning tool, rather than simply as a final test. complexity demands of the analogs themselves (Cho, Holyoak, &
The act of mapping relational correspondences between problem Cannon, 2007). The cognitive demands of analog complexity and
analogs, based on analogical reasoning, seems to operate as a distraction from irrelevant surface features are particularly striking
strong learning opportunity, leading to increased rates of sponta- for those with already somewhat limited cognitive resources, such
neous transfer to new problems at the time of a final test (Catram- as young children (Richland, Morrison, & Holyoak, 2006) and aging
bone & Holyoak, 1989; Gick & Holyoak, 1983; Novick & Holyoak, adults (Viskontas, Morrison, Holyoak, Hummel, & Knowlton, 2004).
1991; Rittle-Johnson & Star, 2007; Ross & Kennedy, 1990). Thus instructional analogies may be most effective when taught
In the classic demonstration of this effect with general insight in a way that reduces processing load for learners as much as pos-
problems, Gick and Holyoak (1983) showed learning gains for pro- sible. But a question remains. What exactly does this mean in the
viding two analogous problems (one about a fireman, one about a context of realistic, everyday mathematics classroom instruction?
General attacking a castle) during a learning phase and giving par- Recent practice recommendations (Pashler et al., 2007) have de-
ticipants questions that asked them to map correspondences be- rived from a body of work informed by Cognitive Load Theory
tween elements in the two problems (e.g., ‘‘what is like the (Sweller & Cooper, 1985). Cognitive Load Theory builds on the
General in the fireman story?”). In the mathematics domain, stud- information processing model of memory and in particular, limits
ies by Novick and Holyoak (1991), Ross and Kennedy (1990) re- in working memory capacity, to argue that instructional designers
vealed that giving reminder hints during an initial problem should consider the inherent demands of instructional materials
solving opportunity led to higher performance on a final test for and tasks so as not to overwhelm learners. Importantly, these con-
those who successfully performed the analogical mapping in the straints are particularly important for novices in a domain (e.g.,
first instance. Catrambone, 1998; Paas & Vanmerriënboer, 1994; Renkl & Atkin-
Rittle-Johnson and Star (2007) recently showed benefits of hav- son, 2003; Sweller, Chandler, Tierney, & Cooper, 1990; Zhu & Si-
ing pairs of students compare two solution strategies (one stan- mon, 1987). Optimal learning environments may be quite
dard, one non-standard) to distributive property problems during different for experts versus novices, based on experts’ ability to
learning. At a final test, those participants significantly outper- schematize, or group, information while novices must instead
formed other student pairs who saw the same problems and solu- interact with all relevant problem solving components separately
tions but on different pages. This result was obtained on final tests (Kalyuga, Ayres, Chandler, & Sweller, 2003; Mayer, 2001).
30 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43

Together, these laboratory-based research programs provide a 3. Experimental test of cuing during instructional analogy
theoretical explanation for why learners require additional support
for learning from comparisons between analogs. Laboratory stud- The current study uses a controlled, laboratory format to pro-
ies provide some insight into those supports – opportunities to vide a direct test of the utility of these strategies for improving stu-
study multiple analogs simultaneously with explicit visual or ver- dents’ learning from a mathematics instructional analogy. Two
bal cues to reduce the cognitive processing load for novices. Next experiments were conducted comparing undergraduates’ learning
we discuss a classroom study that used these principles to identify from analogies in which the most common of these strategies were
everyday classroom practices that seemed likely to support stu- used, versus learning from the same content but in which these
dents’ analogical thinking. supports were not used. The instruction in each case used a video-
taped lesson to simulate whole-class instruction and to maximize
2.2. Classroom-based strategies for supporting comparisons comparability between conditions. Highly supported analogies in-
volved a structural comparison in which the teacher used a combi-
Richland, Zur, and Holyoak (2007) examined typical classroom nation of the four most common strategies identified in the coding
analogy practices in 8th grade mathematics classrooms videotaped study (source presented visually, source remains available during
as part of the Trends in International Mathematics and Science target analog instruction, visual alignment between the two ana-
Study (Hiebert et al., 2003). By comparing US teachers’ practices logs, and gestures between the two analogs). Minimally supported
with those of teachers in higher achieving countries, the authors analogies involved instruction using the same two analogs but
sought to gain insight into classroom feasible strategies for sup- without any of the above support cues.
porting high quality learning opportunities. The two experiments were conducted in different mathematical
Richland et al. (2007) studied US, Hong Kong, and Japanese les- contexts to maximize the generalizability of the results. Experi-
sons, sampled from the larger TIMSS-R dataset which was a ran- ment 1 provided instruction in permutation and combination
domized probability sample of all lessons taught in each region problems, and Experiment 2 involved teaching a limit to the line-
over the course of an 8th grade academic year. Teachers’ uses of arity assumption of proportional reasoning. These materials were
instructional analogies were analyzed within ten randomly sam- selected for several reasons. Both are mathematical content areas
pled lessons from each country, each lesson taught by a different in which undergraduates are well known to demonstrate a lack
teacher. Each lesson was analyzed first for presence of instructional of success, but both are included on the GRE and so are at a suitable
comparisons between the mathematical structure of a problem or level for undergraduate students. Also several analogical studies
concept and another problem, concept, or non-math example. have examined the permutation, combination context (e.g., Ross,
Next, each such comparison was analyzed according to a set of 1989; Ross & Kennedy, 1990; Vanderstoep & Seifert, 1993) so we
six quantitative codes that assessed adherence to principles likely can build on prior work in Experiment 1, and expand to a new con-
to reduce learners’ processing load and enhance relational learn- text in Experiment 2. Finally, we sought to generalize across the
ing. These codes derived from an integration of the literature and two most common types of mathematical analogies in US instruc-
observations of the videos. tion (Richland, Holyoak, & Stigler, 2004), between two problem
Analyses revealed that teachers across countries used similar types (Experiment 1) and between two solution strategies (Exper-
total numbers of comparisons between two or more analogs (be- iment 2).
tween 7 and 20 per lesson). However, these patterns varied dra- Posttests were administered immediately (Experiments 1 and
matically when comparisons were assessed for adherence to 2) and after a week delay (Experiment 2). Posttests measured
principles likely to reduce cognitive load. Codes measured: (1) learners’ ability to solve test problems that: (a) appeared similar
learners’ familiarity with the source, (2) whether source analogs to instructed problems, and (b) in which the appearance and math-
were presented visually versus only orally, (3) the source’s visibil- ematical correspondences were cross-mapped. Cross-mapping
ity during instruction of the target, (4) presence or absence of vi- meant here that the story context for one instructed problem
suo-spatial cues to the correspondences between items being was the same as the story context for a posttest problem with dis-
compared, (5) presence or absence of gestures that move back similar mathematical structure. The posttest measures derived
and forth between the items being compared, and (6) the use of from the complexities of adequately measuring analogical learn-
mental imagery or visualizations. These strategies presumably ing. The literature on such measurements is briefly reviewed. Fol-
supported learner’s processing by increasing their relative domain lowing this review, the experiments are reported.
expertise (using familiar sources and mental imagery), reducing
demands on working memory (source presented visually, left visi-
ble during target instruction), and reducing demands on attention 4. Measuring analogical learning
by drawing learners’ eyes to the comparison itself (comparative
gesture and visuo-spatial cues within visual representations). Several strategies have been used in the literature to measure
The findings were quite clear: for all codes, the US teachers analogical learning. Novick and Holyoak (1991) use the term ‘‘ana-
were least likely to use these cuing support strategies. Teachers logical transfer” to describe the retrieval of a previously taught
in Hong Kong and Japan were significantly more likely to use all problem to help generate potential solutions to a test problem.
of these strategies, sometimes using them in double or triple the Cross-mapping surface and structural similarities on test problems
percent of instructional analogies. has also been used to assess the depth of learners’ understanding
Richland et al.’s (2007) data indicate that these cuing support and their focus on structural relations, as well as their resistance
strategies are not yet being used regularly by US teachers, and to irrelevant surface cues (e.g., Gentner & Toupin, 1986; Goldstone
could provide new ways to optimize US teachers’ current uses of & Medin, 1994; Novick, 1988; Ross, 1987, 1989; Ross & Kilbane,
instructional analogies. A limit to this study, however, was that 1997).
no student outcome data were available from the TIMSS studies In a paradigm closely related to the current Experiment 1, Van-
to directly tie instructional practices to student learning (Hiebert derstoep and Seifert (1993) taught participants how to solve two
et al., 2003; Perry, Vanderstoep, & Yu, 1993). Thus while there is mathematically similar (permutation and combination) problems,
theoretical support for these strategies as methods for increasing and then tested participants’ ability to determine which of the in-
students’ ability to learn from relational correspondences, a more structed procedures would be most appropriately applied to test
direct test is necessary. problems. The authors drew on Schwartz and Bransford’s (1998)
 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43 31

notion of contrasting cases, comparing analogs that shared high excluded for failure to complete the study. Participants were re-
structural similarity aside from a key variation. To measure learn- cruited from the university subject pool; generally these were stu-
ing, the authors distinguished between learning how and learning dents taking introductory psychology and education courses. For a
when to use solution strategies, or in Ross’s (1989) terms, retrieval baseline measurement, 28 additional undergraduates (23 females,
access and use. M = 21 years) were given the posttest only.
Vanderstoep and Seifert’s (1993) experiments support the con-
clusion that considerable instructional support is necessary to al- 5.2. Materials
low learners to benefit from a comparison between similar or
contrasting analogs. Written text-based instructional manipula- 5.2.1. Instructional videos
tions tested the benefits of an explicit rule for when to use each Videotaped instruction taught students about two math con-
formula versus providing worked solutions to the two problems cepts tested on the GRE: permutations and combinations. Permu-
separately or together. The authors measured two indicators of tation problems involve determining the maximum possible
learning: (1) accurate problem solving and (2) ability to determine arrangements of a set of items (e.g., the number of alternative or-
which instructed problem solution matched each test problem. ders that four runners can be awarded a gold and silver medal).
Findings showed no differences between conditions in solution Combinations are permutation problems with the additional con-
accuracy, but providing the explicit rule for when to use each for- straint that certain alternative orders of the same objects are con-
mula led to improvements in appropriate formula use. Allowing sidered the same (e.g., the number of different ways that four
participants to induce the rule by providing the two worked prob- students can be awarded two entrance tickets to a show, where
lems together did not improve solution selection. receiving ticket number one and ticket number two is functionally
These experiments point out the importance of experimental the same). The problem solution therefore must exclude equivalent
design and posttest measures that allow for disentangling the ef- orders.
fects of teaching on (1) memory for instructed problems and solu- Videotaped instruction was developed for two conditions: (a)
tions and (2) ability to know when to use these solutions. These high cuing for comparison and (b) minimal cuing for comparison.
types of knowledge seem to be distinct, such that one might recall Separate videos were constructed for the two conditions, allowing
a solution strategy but not spontaneously notice the utility of that for instruction that simulated classroom teaching but was easily
strategy at the time of test. controlled between conditions. Both videos taught the same two
Posttest measures in the current experiments therefore were problems and solution strategies, and were approximately the
designed to provide data on both points through a manipulation same length (11.5 and 13 min, respectively). In both videos, the
of surface and structural similarity. A cross-mapping procedure learner first received instruction about solving permutation prob-
was used to create problems in which the setting of a problem lems embedded within an example of runners in a race. Next, they
taught during instruction either mapped directly onto the setting received instruction about solving combination problems embed-
of a problem solved the same way on a posttest problem (facilitory ded within an example of students vying for tickets to a lecture.
similarity), or mapped onto the setting of a problem solved a differ- The key instructional elements of the two videos are described
ent way (misleading similarity). This measurement decision al- and aligned in Appendix A.
lowed us to begin to differentiate between alternative The instructional manipulation involved the manner in which
characterizations of the impact of cuing on learners’ later problem support was provided for the learners to notice and benefit from
solving. Four alternative explanations were possible for how cuing a comparison between the source and target analogs. The high cu-
might aid learning. Increased cues during analogical instruction ing video included explicit cues to the alignment and correspon-
might lead to: (1) abstracted schematic representations of the dences between the two analogs. Instruction in this video
two analogs, (2) production of a decision rule for when to apply included the four most common cuing and processing support
each taught solution strategy, (3) improved retention of the indi- strategies identified within the cross-cultural video analysis: com-
vidual problems, or (4) more expert-like processing of problems parative gesture, source visually supported, source visible during
(attention to structural versus surface features). These possibilities instruction of target, and visual alignment. The teacher gave the
were explored in the data gathered in the following two experi- same instruction about permutations provided in the minimal cu-
ments. Results from the two mathematical contexts are discussed ing condition, but in the high cuing video, the teacher left the prob-
separately, and then the strength of generalizability across the lem on the board while he described how to solve the combination
two contexts is used in a final discussion to try and determine problem. Learners could still visually reference the permutation
(a) whether providing support in the form of analogy cues im- problem while they were learning about the combination problem
proves learning, and if so, (b) disentangle these four possibilities to check commonalities or differences. The teacher in the high cu-
for how the learning is improved. ing video also used visual supports to promote the students’ com-
parisons between these two problems. He gestured back and forth
between the two problems (comparative gesture) and wrote the
5. Experiment 1
problems on the board in a linear, sequential style that highlighted
the alignment between the two. For example, each line of the prob-
Experiment 1 compared undergraduates’ learning from a highly
lem solution contained an underscore for numbers not yet filled in.
versus minimally cued instructional analogy between a permuta-
This procedure ensured that the structure was preserved for every
tion problem and a combination problem. We hypothesized that
line of the problem (see Fig. 1), and there was a record of the se-
the highly cued instructional analogy would lead to greater ability
quence of the steps visible on the board.
to map instructed problems to test problems on the basis of
By contrast, in the minimal cuing video, the commonalities be-
structure.
tween the two problems were not immediately apparent by look-
ing at the spatial information on the board (see Fig. 2). There
5.1. Method were two main differences. First, the text of the two problems were
each written out more completely than in the high cuing condition,
5.1.1. Participants which did not as explicitly highlight the key problem elements. The
Seventy undergraduates participated for partial course credit solution methods were also written differently, such that there was
(56 females, M = 21 years). Data from six of these participants were not a new sequence of underscore lines written for each step of the
32 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43

Fig. 1. Final screen from Experiment 1 instructional video: high cuing comparison.

problem. Rather, the underscore lines were only written once, and than using a pretest format with the same participants who com-
the steps were all completed using that same row of lines. This was pleted our study, because of the sensitivity of the instructional
more efficient than in the high-cue condition, and is the way one manipulation. A growing body of literature suggests that testing it-
would typically solve the problem independently. However, it did self can provide a learning benefit for test-takers (see Roediger &
not provide the same visual record and required that the viewer re- Karpicke, 2006), even when there is no feedback provided and
call in working memory the sequence of the steps. when all answers to pretest questions are incorrect (Richland, Kor-
nell, & Kao, 2009). We were particularly concerned about this in
5.2.2. Posttest the current experimental case, since we didn’t want our partici-
Two kinds of key test problems were developed for the posttest, pants to spontaneously compare and contrast the permutation
resulting in four total problems used in analyses. Two Facilitory Sim- and combination problems on the pretest. That initial experience
ilarity problems were designed to align with both the mathematical might have directly impacted their learning from the two condi-
and the surface story context features of the instructed problems. tions, perhaps influencing each condition differently.
One problem was a permutation problem about a race, and the sec- Participant responses on all test questions were scored both for
ond was a combination problem about tickets to a lecture. their accuracy and for the type of error made. Accuracy was scored
The second key test problems were Misleading Similarity prob- as percent of problems set up correctly. Correct set up included
lems, in which the word problem context was misaligned, or both use of the appropriate formula and adaptation to the correct
cross-mapped, with the mathematical structure. The misleading values within the target problem context. Calculation errors were
similarity permutation problem was set in the context of procuring not taken into account in accuracy scores, since our goal was to
tickets to a lecture, and the misleading similarity combination measure appropriate decisions about how to classify problems
problem was set in the context of a race. These problems were pre- and map them onto source analogs. Partial credit for retrieving
sented in randomized order. the source analog without ability to draw inferences about the tar-
Each posttest contained a permutation problem and a combina- get problem (adaptation) was not assigned, since the design of the
tion problem with facilitory similarity and one of each with mis- posttest made retrieval without understanding uninteresting. Cor-
leading similarity, yielding a total of four key test problems. Each rect set up was scored as 1; all other answers were scored as 0.
participant also solved five additional word problems of similar Second, instances of cross-mapping were assessed, since these
length. These served as distractor problems and tests of learners’ were diagnostic of failures to appropriately discriminate between
over-extensions of the solution procedures. The five problems analogs on the basis of structure. Errors were coded as cross-map-
were selected in ten balanced configurations from the following ping errors when participants used a combination strategy on a per-
set: three factorial problems (one about races, one about tickets mutation problem and vice versa. Other types of solution errors
to lectures, and one set in an irrelevant context); and four more ad- were not coded as cross-mapping attempts. All participants who at-
vanced combinatoric problems (not directly solvable with the in- tempted to use the strategy that would be correct for the alternative
structed solution), either set in the context of a race or a lecture. type of problem (i.e., using a permutation strategy to solve a combi-
These were interspersed in randomized order between the key, nation problem or vice versa) were given a 1, while those who
analyzed problems, in order to simulate the necessity in classroom scored correctly or made a different type of error were scored 0.
settings for learners to identify relevant analogs amid other dis- The distractor problems were coded by examining whether the
tracting problems. participants made expected errors, which meant using correctly
Baseline performance on posttest problems were assessed on a either the permutation or combination solution to that test prob-
separate group of students sampled from the same population as in lem, though neither was actually a valid way to answer the test
the experimental conditions, but who did not watch any of the question. This was the same as the way the cross-mapping scores
instructional videos. These data were collected to ensure that stu- were calculated, but we scored both whether participants made an
dents would not begin the instruction at ceiling levels of perfor- attempt to use a combination or a permutation solution. As before,
mance. An independent group of participants was tested, rather calculation errors were ignored in this scoring.
 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43 33

Fig. 2. Experiment 1 instructional video: low cuing comparison.

5.2.3. Procedure 5.3.1. Baseline accuracy

In a between-subjects design, participants were randomly as- Participants who did not watch either of the instructional con-
signed to one of the two instruction conditions (minimal cuing dition videos provided evidence that undergraduates at this uni-
condition, N = 30; high cuing condition, N = 34). Participants were versity were not overall mathematically competent on the test
tested individually with a trained experimenter in a laboratory set- problems prior to instruction. The four problems of interest were
ting. They entered and were seated in front of a Macintosh com- analyzed separately. Participants were approximately equally
puter. They were told they would watch a short mathematics accurate on all four problems. For the problems that served as facil-
lesson, and were shown either the high or low cuing video. They itory similarity problems in the experimental conditions, accuracy
were then given a 5-min timed word-finding distractor task. Final- was 10% correct for the permutation problem (SD = 30), and 3%
ly, participants were given a paper packet that included all of the correct for the combination problem (SD = 19). For the problems
posttest problems and distractor problems, with two problems that served as misleading similarity problems in the experimental
per page. This final testing was not timed. Participants were not gi- conditions, accuracy was 10% correct for the permutation problem
ven any explicit cues to use the videotaped instruction when solv- (SD = 30), and 10% correct for the combination problem (SD = 30).
ing the posttest problems. Thus these data show that for the studied population of undergrad-
uates, participants were unlikely to solve any of the posttest prob-
lems correctly prior to instruction, with accuracy rates ranging
5.3. Results
from 3% to 10%.
Analyses were conducted on the two types of dependent vari-
ables: accuracy and presence of a cross-mapping error. Accuracy 5.3.2. Posttest accuracy
data are first described below, and then data from cross-mapping Analyses next examined the impact of the instructional manip-
errors are presented. ulation on posttest performance. Two types of dependent mea-
34 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43

sures were included in a repeated-measures model. Performance The main effect of surface similarity suggested that even the
on facilitory similarity problems was viewed as an index of partic- high-cue condition did not produce fully schematized knowledge.
ipants’ retention and ability to use the instructed problem solu- This interpretation was supported by an analysis comparing only
tions when minimal demands were placed on noticing the the high-cue condition scores on the facilitory and misleading sim-
relevance of the solution strategy. Performance on the misleading ilarity problems. A paired-samples t-test revealed the difference
similarity problems was viewed as an index of participants’ ability remained significant, t(29) = .26, p = .01, d = 2.69, indicating that
to ignore irrelevant surface cues, and instead notice and extend the even following highly cued relational instruction, source analogs
instructed solution strategies to a context that appeared mislead- were retrieved as whole problems rather than simply as solution
ing and dissimilar. schemas or decision rules. The difference between conditions ap-
A repeated-measures ANOVA used accuracy as a dependent var- pears to reflect which elements of a target problem were used as
iable with two within-subjects levels of surface similarity (facilito- retrieval cues for selecting between the two potential source
ry versus misleading similarity) to examine the effects of the analogs.
between-subjects independent variable of instructional condition
(high versus low cuing). As can be seen in Fig. 3, there was a main 5.3.3. Cross-mapping errors
effect of surface similarity F(1, 62) = 29.9, p < .001, g2p ¼ :33, such The cross-mapping error data further support the interpretation
that a higher proportion of facilitory similarity problems (problems that high cuing during instruction impacted participants’ later
that shared surface and mathematical similarity with the in- spontaneous attention to relational structure as opposed to surface
structed problems) were set up correctly (M = .80, SD = .32) than features. The same analyses were repeated using cross-mapping
were problems with misleading surface similarity (M = .51, errors as a dependent variable. These data again showed a main ef-
SD = .38). Thus, as expected, the manipulation of surface similarity fect of surface similarity, F(1, 62) = 36.38, p < .001, g2p ¼ :37, sup-
was effective and accuracy was higher when surface similarity was porting the validity of the surface similarity manipulation. This
correlated with structural similarity than when the surface similar- analysis also revealed an interaction between instructional cuing
ity was misleading. and surface similarity that approached significance, F(1, 62) =
Surface similarity also significantly interacted with instruc- 3.14, p = .08, g2p ¼ :05. As shown in Table 1, the pattern of errors
tional cuing condition, F(1, 62) = 4.05, p < .05, g2p ¼ :06. Impor- supports the interpretation drawn above for the accuracy data.
tantly, this interaction revealed that rates for the proportion of Subjects in both instructional conditions made relatively few of
problems set up correctly were high for both conditions on the these diagnostic errors on the facilitory similarity problems
facilitory similarity problems (M = .80, SD = .31, and M = .81, (t(62) = 1.86, p = .85, d = .05); but on misleading similarity prob-
SD = .32). However, there was a difference in accuracy by condition lems, the low cuing condition resulted in marginally higher error
for the misleading similarity problems. A separate univariate anal- rates than did the high cuing condition (t(62) = 1.9, p = .06,
ysis of the misleading similarity problems revealed that the aver- d = .47). This pattern of errors suggests that participants in the
age proportion correct for the high cuing condition (M = .62, low cuing condition were more likely to retrieve a source analog
SD = .36) was significantly higher than for the low cuing condition on the basis of surface similarity. In contrast, those in the higher
(M = .41, SD = .40), F(1, 62) = 4.57, p < .05, d = .55. This finding sug- cuing condition were presumably more likely to retrieve a source
gests that those participants in the former condition were more analog on the basis of relational structure, and thus were less im-
attentive to structure, as opposed to surface features, and were pacted by the cross-mapping procedure.
better able to discriminate between problems on that basis. It ap-
pears that high cuing led to greater schematization of the source 5.3.4. Over-extension to distractor problems
analogs, and thus lowered participants’ susceptibility to mislead- Distractor problems were analyzed for errors in which partici-
ing surface similarity. pants attempted to map taught solution strategies to the mathe-
Even so, these participants were evidently not reasoning entirely matically dissimilar problems that shared context features. There
on the basis of fully schematized, or rule-based, representations. were no differences across conditions: 30% of participants in both
conditions sought to map a solution strategy based on the surface
1 context, even when the problems were not mathematically compa-
rable. This finding suggests that when they lacked the content
0.9 knowledge to interpret test problems mathematically, all partici-
Low Cuing Condition
High Cuing Condition pants moved to map correspondences on the basis of surface fea-
0.8
tures in some instances. This result also supports the
0.7 interpretation that even participants in the highly-cued compari-
Proportion Accuracy

son condition were not reasoning on the basis of fully schematized

0.6 knowledge, but rather maintained contextual information in their
knowledge representations.
0.5

0.4 5.4. Discussion

0.3 Experiment 1 was conducted with two main aims: to determine

whether cues to support analogical thinking would increase stu-
0.2

0.1 Table 1
Frequency of cross-mapped strategy use across problem type in Experiment 1.
0
Instructional Facilitory similarity Misleading similarity
Facilitory Similarity Misleading Similarity condition problems problems
Posttest Problem Type High cuing 7 (16)a 28 (31)
Low cuing 6 (17) 46 (41)
Fig. 3. Impact of support for analogical instruction on posttest accuracy (Experi-
a
ment 1). Standard deviations listed in parentheses.
 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43 35

dents’ flexible learning for instructed concepts, and if so, to better over a more substantial delay. As in Experiment 1, posttests mea-
understand how the cues would change learners’ representations sured learners’ ability to solve test problems that (a) appeared sim-
of the instruction. In answer to the first, the data support the inter- ilar to instructed problems, and (b) in which the appearance and
pretation that cuing does provide a benefit for encouraging flexible mathematical correspondences were cross-mapped.
learning. Providing more analogical cues did not affect learners’
retention of instructed solution strategies, participants were able
6. Experiment 2
to solve facilitory similarity problems after both videotapes, but
it did impact their ability to recognize and extend the solution
Experiment 2 compared undergraduates’ learning from a highly
strategies to contexts with misleading surface appearance. Partici- versus minimally cued instructional analogy between two strate-
pants with higher cuing were better able to recognize problem
gies for solving a proportion problem. This experiment also in-
structure and use the appropriate solution strategy on the mislead- cluded a control condition in which participants were actively
ing similarity problems. Participants who received minimal cuing
engaged in the instruction but there was not an analogy presented.
were most likely to make cross-mapping errors on misleading sim- We hypothesized that the highly cued instructional analogy would
ilarity problems. This suggests that they were more reliant on sur-
lead to the greatest ability to map instructed problems to test
face similarity in determining when and where to use a learned problems on the basis of structure.
solution strategy.
These data reiterate the importance of posttest measures that
6.1. Method
allow for disentangling the effects of teaching on memory for in-
structed solutions and ability to know when to use these solutions.
6.1.1. Participants
Further, these data allow for beginning to understand why the cues
Participants were 80 undergraduates who participated for par-
were successful. Four alternative explanations were originally pos-
tial course credit (62 = females, M = 21 years). Four participants
ited for how cuing might aid learning. Increased cues during ana-
were excluded for failing to return on the second day of the study.
logical instruction might lead to: (1) abstracted schematic
Data for the remaining 76 participants were included in analyses.
representations of the two analogs, (2) production of a decision
rule for when to apply each taught solution strategy, (3) improved
retention of the individual problems, or (4) more expert-like pro- 6.2. Materials
cessing of problems (attention to structural versus surface
features). 6.2.1. Pretest
Based on Experiment 1 data, there seems to be little support for A pretest was included in the current study to determine the
the position that participants developed either fully abstracted rate and pattern of participants’ use of the linearity assumption.
representations of the two analogs (a), or production of a decision Because the experimental manipulation involved a comparison be-
rule for when to apply each solution strategy (b). Facilitory similar- tween two solution strategies, we determined that most partici-
ity problems were solved at a higher rate than misleading similar- pants would only solve the pretest problem once, and thus were
ity problems in both conditions, which indicates that even the high less likely than in Experiment 1 to independently make their
cuing participants had stored problem contextual features as well own comparison. Doing so could have impacted the affect of the
as the mathematical structure. At the same time, the interaction instructional manipulation.
between the effects of surface similarity and instructional condi- Pretest packets were constructed to measure both entry math-
tion seems to rule out the third possibility. Despite the effect of ematics knowledge and demographic information. The packet con-
misleading surface similarity on the high cuing participants, the ef- tained a math problem that assessed learners’ ability to design an
fect is relatively less than on the minimal cuing participants. If high appropriate proportion between two ratios: Bob needs 6 h to paint a
cuing led to better retention for the test problems, as indicated in square wall with a side of 10 m. How many hours would he need to
possibility three above, one might expect that the effect of surface paint a square wall with a side of 5 m? The problem also assessed
similarity would be even greater for the high cuing condition. But, the frequency of the linearity misconception: the likelihood of
this is not the case. We do concede, however, that greater retention making an assumption of linearity between the two proportions
for the test problems could allow participants to more easily com- (i.e., setting up the proportion 6/10 = x/5). This is appropriate in
pare both surface and mathematical structure between instructed some contexts, but in the case of an area, linearity does not hold.
and posttest problems. Finally, these data do seem to support the Rather, one must first calculate area and use that quantity in the
fourth possibility, that the higher cuing teaches students to reason proportion. The correct proportion should be 6/100 = x/25.
more like experts and improves their ability to analyze the struc-
tural properties of a new problem. 6.2.2. Experimental videos
Experiment 2 extends these data in several ways, allowing for a Videotaped instruction was developed to alert learners to the
better treatment of these interpretations in the final discussion. A structural similarities and differences between making the linear-
new mathematical content area was used to explore the generaliz- ity assumption and performing relevant calculations (area) before
ability of Experiment 1 findings, particularly because permutation setting up a proportion. Three videos were designed. Two were
and combination problems have been used extensively in the anal- experimental videos that provided an instructional comparison be-
ogy learning literature. Experiment 2 involved teaching a limit to tween these structurally related but different solution strategies. In
the linearity assumption of proportional reasoning, which is an- both versions the teacher showed the use of linearity and then the
other mathematical content area in which undergraduates are well correct solution strategy; however, these videos differed in their
known to demonstrate a lack of success, but is included on the GRE use of cues to promote analogical encoding. One version was a
and so is at a suitable level for undergraduate students. Second, we Low Cuing video, in which the linearity strategy was taught and re-
sought to generalize across the two most common types of math- corded on the board, labeled incorrect for this problem, erased, and
ematical analogies in US instruction (Richland et al., 2004). Exper- then the second strategy was taught and recorded on the board.
iment 1 studied the effects of cuing on a comparison between two The second video, High Cuing, taught the same two strategies to
problem types, and Experiment 2 examined the effects of a com- the same problem, but supplied several educationally realistic lev-
parison between two solution strategies. Finally, Experiment 2 in- els of cuing supplied to ensure learners’ attention to the relational
cluded a delayed posttest in order to measure the impact of cuing structure. The linearity strategy was taught and recorded on the
36 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43

board, but participants were not told whether it was correct or but we predicted that learners who were less effective at process-
incorrect. It was left on the board while the correct strategy was ing the structure of the target analog would over-extend the
demonstrated on the right side of the board. The two strategies instruction to this case.
were recorded on the board in a spatially aligned format (i.e., the Scoring procedures were the same as in Experiment 1. Problem
problem was written in the same way and the proportions were solutions were scored both for their accuracy and the type of error
written at the same spatial level). made. As in Experiment 1, calculation errors were not taken into
In a procedure modeled after that introduced by Gick and Holy- account in the accuracy ratings. Second, errors were coded as other
oak (1983) and that has been widely used since, participants were errors or as cross-mapping errors, since the latter were considered
then given the opportunity to map correspondences between the diagnostic of failures to appropriately discriminate between ana-
two strategies and make a judgment about which strategy was cor- logs on the basis of structure. Errors were coded as cross-mapping
rect. The video stopped and they were handed a worksheet with errors when participants made the linearity assumption when not
four questions. The first three prompted them to compare the appropriate, or applied the instructed strategy when linearity was
two strategies on the basis of structure and map objects between appropriate. All participants who made such an error were given a
the two proportions. The last question asked them to state which 1, while those who scored correctly or made a different type of er-
strategy they believed was accurate. When the experimenter ror were scored 0.
determined that the participant was finished, the video resumed
and the recorded instructor stated the right solution was correct, 6.2.4. Procedure
and used comparative gesture to highlight the relevant structural In a between-subjects design, participants were randomly as-
difference between the strategies (the calculation of area before signed to one of the three video conditions, High Cuing (N = 26),
setting up the proportion). Minimal Cuing (N = 23), and active participation control (N = 27).
A third condition, active participation control, was designed to An experimenter tested each participant individually at a com-
ensure that any performance differences between the participants puter. In all conditions, the experimenter administered the paper
in the two cuing conditions were not due only to subjects’ active and pencil pretest, which was untimed. Afterwards, the experi-
participation in the instruction. Since laboratory-based studies menter removed the pretest and asked the participant to wear
are plagued with participants’ low motivation levels, and because headphones that were provided. The experimenter then started
active participation in instruction is known to facilitate learning, the videotaped lesson on the computer. As noted above, in two
we developed a videotaped instruction that did not use a structural of the conditions the experimenter administered a worksheet be-
comparison, but did teach the correct solution strategy and en- tween two parts of the video. After the instructional video was
gaged learners’ in participating. In this videotape the instructor complete, in all conditions participants were given one of two
wrote the problem on the board and demonstrated how to set up counterbalanced posttests. Testing was untimed. One week later,
the correct proportion. Since most participants used the linearity participants returned and completed the alternative posttest, again
strategy on the pretest problem it was possible that this instruc- individually. The delayed posttest was also untimed.
tion invoked a mental comparison between these strategies, but
such comparison was not supported by the instruction. After set- 6.3. Results
ting up the proportion, the video stopped and the experimenter
handed participants a worksheet. As in the high cuing condition, As in Experiment 1, accuracy data are described first, followed
the worksheet included four questions. These asked the participant by data from cross-mapping errors.
questions about the instruction and asked them to perform the cal-
culations necessary to solve the proportion. Crucially, these active 6.3.1. Pretest
participation prompts did not cue the learners to attend to any Pretest data are reported first to assess the base rates of perfor-
structural differences between the linearity and accurate solution mance across participants, and to verify that participants would
strategies. After the experimenter determined that participants over-extend the linearity assumption, as predicted by the larger
had finished, the video resumed and the recorded instructor per- body of research on undergraduates’ mathematics knowledge.
formed the calculations and gave the correct answer. The details Additionally, the pretest allowed for ensuring random assignment
of the videotape construction are available in Appendix B. of participants across the experimental conditions. Results from
the pretest suggest that the majority of participants in all condi-
6.2.3. Posttest tions did reveal this misconception. Twenty-three percent of par-
Two counterbalanced posttests were created so that each in- ticipants (n = 18) overall correctly solved the problem on the
cluded two target problems alternating with three distractor prob- pretest, whereas 64% made the expected linearity error (n = 48).
lems. For each participant, one of the posttests was used as an The error rates were distributed across all three conditions. There
immediate posttest, and the other was used as a week-delayed were no differences between conditions on the pretest accuracy
posttest. The posttests given at the two time periods were counter- rates, F(2, 73) = 1.1, p = .34, g2p ¼ :03, nor of making the linearity
balanced versions of the same test, so half of the participants got assumption error, F(2, 73) = 1.0, p = .36, g2p ¼ :028. Participants
one version immediately and the second after a delay, and vice ver- were not excluded from further analyses if they gave a correct an-
sa. Thus the same problems were distributed for all participants swer, since we found that their performance on the final and de-
across the two time periods to ensure maximum comparability be- layed posttests revealed the use of the misconception at
tween the two tests. comparable rates to those who used it on the pretests, and they
As in Experiment 1, posttest questions orthogonally manipu- were all similarly impacted by surface similarity.
lated surface similarity and structural similarity. On each test,
one posttest problem appeared similar to the instructed problem 6.3.2. Posttest accuracy
and could be solved using the taught solution strategy that avoids Posttest data were next analyzed to assess the impact of the
the linearity assumption (facilitory similarity). The second problem experimental manipulation on retention of the instructed strategy
was set in the same surface context (painting) and thus appeared (facilitory similarity problems) and extension to problems that
similar to the instructed problem, but the taught instruction was were unlike the instructed problems (misleading similarity prob-
not relevant and the linearity assumption was actually accurate lems). Immediate and delayed posttest data were both included
(misleading similarity). In fact this was an easier test problem, in an omnibus ANOVA to determine whether the performance pat-
 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43 37

terns changed over time with forgetting. The independent variable, a. Immediate Test
instructional condition, had three levels: high cuing, minimal cu-
1
ing, active participation control. The model first tested whether Active Participation Control Condition
the manipulation of surface similarity within the task materials 0.9 Low Cuing Condition
was successful. As in Experiment 1, there was a main effect of
0.8 High Cuing Condition
problem type, F(1, 73) = 69.7, p < .001, g2p ¼ :48, meaning that vary-
ing the surface similarity of problems did impact the cognitive de-

Proportion Accuracy
0.7
mands of the facilitory versus misleading problems. Accuracy was
0.6
proportionally higher overall for facilitory similarity problems
(immediate test: M = .95, SD = .28, delayed test: M = .82, SD = .39) 0.5
than for misleading similarity problems (immediate test: M = .32,
0.4
SD = .47, delayed test: M = .46, SD = .50). However, there was not
a significant interaction between problem similarity type and con- 0.3
dition, F(2, 73) = 1.14, p = . 33, g2p ¼ :03. Unlike in Experiment 1,
0.2
condition effects were visible on both facilitory and misleading
similarity problems (main effect result below), so these data are 0.1
collapsed for the subsequent analyses of variations in learning by 0
condition. Facilitory Similarity Problem Misleading Similarity Problem
Interestingly, there were also few effects of the posttest delay.
The main effect of delay of test was not reliable, F(1, 73) = .04, b. Delayed Test
p = .84, g2p ¼ :00, as there was not a large overall difference be- 1
tween performance on the immediate and delayed posttest (see
Fig. 4a versus b). There was also no interaction between test delay 0.9
and instructional condition, F(2, 73) = 1.14, p = .33, g2p ¼ :03, indi- 0.8
cating that any effects of the instructional manipulations are not
Proportion Accuracy
0.7
easily interpreted as differences in retention or retrieval access.
There was an interaction between problem similarity type and test 0.6
delay F(1, 73) = 11.7, p < .01, g2p ¼ :04, but the pattern is reflective
0.5
of the stimuli. The solution strategy that was correct for the in-
structed problem showed some forgetting (evident in the decrease 0.4
over time in accuracy on facilitory similarity problems), whereas 0.3
the linearity misconception that had been shown to be incorrect
on the instructed problem underwent some reinstatement after a 0.2
delay (evident in the small increase in performance on misleading 0.1
similarity problems). We interpret this interaction as evidence that
as participants increasingly forgot the instruction, they reinstated 0
Facilitory Similarity Problem Misleading Similarity Problem
the misconception error many had revealed on the pretest. The
three-way interaction between test delay, problem similarity type, Fig. 4. Impact of support for active student participation in analogical instruction
and instructional condition was not reliable, F(2, 73) = .59, p = .56, on posttest accuracy immediately and at a delay (Experiment 2).
g2p ¼ :02.
The omnibus model revealed a main effect of condition
F(2, 73) = 8.71, p < .001, g2p ¼ :19. Mean scores for the high cuing
condition were higher than either of the other instructional condi- a difference in cuing between facilitory and misleading similarity
tions for both problem types immediately and at a delay. These problems. This comparison provided an important test of whether
apparent differences between conditions were further explored these participants solved the target problem based on a purely
using a planned contrast analysis. Since neither time nor problem schematic/conceptual understanding or by retrieving the source
similarity type was significantly related to instructional condition, representation. The analyses revealed a difference both immedi-
the following pair-wise comparisons collapsed across those factors. ately, t(25) = 4.2, p < .001, d = 1.7, and at a delay t(25) = 2.5,
These analyses revealed an overall difference between the condi- p < .05, d = 1.0, such that scores were higher on facilitory similarity
tions, F(1, 73) = 8.71, p < .001, g2p ¼ :19, with specific variations be- problems than on misleading similarity problems (immediate
tween the high and minimal cuing conditions (mean means = 1.00 versus .58, and at a delay, means = .88 and .58). This
difference = .23, p < .001), and the high cuing and active participa- indicates that participants even in the high cuing condition were
tion control conditions (mean difference = .19, p < .01), but not be- retrieving the instruction as a source analog while solving target
tween the minimal cuing and active participation control problems.
conditions (mean difference = .04, p = .77).
The high cuing condition thus led to higher performance than 6.3.3. Cross-mapping errors
either the low cuing or control conditions, but there was very little An omnibus ANOVA was next conducted using cross-mapping
difference between the low cuing and control conditions. This find- errors as the dependent variable to examine whether the patterns
ing suggests that without high cuing during training, the compar- observed in the accuracy data were explained by participants’ fail-
ison of two analogs was not more effective than an engaging lesson ure to accurately map known analogs to new posttest problems. As
on one analog only. Further, these data suggest that directing nov- with accuracy, all three conditions were first examined together to
ices’ attention to relational structure during an instructional com- see the overall relationship between level of comparative instruc-
parison between confusable analogs increased later accuracy and tion and performance on facilitory versus misleading similarity
appropriate use of a relevant source analog. problems. A 2 (problem similarity type)  2 (test delay) ANOVA
Finally, paired t-tests were conducted only on the high cuing was performed with instructional condition (three levels) as a be-
participants to determine whether, as in Experiment 1, there was tween-subject factor. As was found for accuracy, there was a signif-
38 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43

icant difference between problems, F(1, 73) = 37.3, p < .001, high supportive cues for an instructional comparison had direct
g2p ¼ :34, such that the rate of cross-mapping errors was lower implications for participants’ future ability to discriminate be-
on facilitory similarity problems overall (immediate test: M = .05, tween previously unseen potential analogs on the basis of struc-
SD = .22, delayed test: M = .16, SD = .37) than for misleading simi- tural correspondences rather than surface features. Importantly,
larity problems (immediate test: M = .47, SD = .50, delayed test: the role of support was crucial to whether or not the instruction
M = .46, SD = .52). was effective at promoting this type of flexible problem solving.
Again, there was not a main effect of test delay, F(1, 73) = .04, Participants in all conditions scored fairly well on problems that
p = .84, g2p ¼ :00, since there was not a large overall difference be- appeared similar to instructed problems immediately after
tween performance on the immediate and delayed posttest (see instruction (over 90% accuracy) and there was not a significant ef-
Table 2 for means). The interaction between problem similarity fect of forgetting after a delay. Thus all instruction was fairly po-
type and test delay was not reliable, F(1, 73) = 1.8, p = .19, tent. However, the crucial finding was that performance on
g2p ¼ :02, nor was the three-way interaction between test delay, misleading similarity problems differed significantly between
problem similarity type, and instructional condition, conditions depending on the level of instructional cuing provided
F(2, 73) = .00, p = .99, g2p ¼ :00. during training. In the General Discussion we consider alternative
In addition, there was no reliable interaction between time and explanations for this effect.
condition, F(2, 73) = .15, p = .86, g2p ¼ :00, indicating that any ef- The mere inclusion of a comparison in instruction was not en-
fects of the instructional manipulations are not easily interpreted ough to promote expert-like reasoning, as evidenced by the results
as differences in retention or retrieval access, since performance from participants in the low support condition. Participants in that
was fairly comparable immediately after training and after a week condition saw the same correct instruction as did participants in
delay. Also, like in the accuracy data, there was not a significant the high support condition and were explicitly shown why the lin-
interaction between problem similarity type and condition, earity assumption did not work in this case before being taught the
F(2, 73) = 1.0, p = . 37, g2p ¼ :03. However, the main effect of condi- correct strategy; yet when compared to the high support condition,
tion was reliable, F(2, 73) = 4.89, p = .01, g2p ¼ :12. As shown in Ta- these participants were significantly less well able to differentiate
ble 2, mean rates of cross-mapping errors for the high cuing when to use each strategy. This gap was observed even immedi-
condition were lower than for either of the other instructional con- ately after training, and the differences remained after a substan-
ditions, for both problem types immediately and at a delay. These tial delay.
apparent differences between conditions were further explored Moreover, active participation in the instruction was not en-
using a planned contrast analysis. Since neither test delay nor ough to ensure participants’ ability to identify and discriminate be-
problem similarity type were significantly related to instructional tween analogs on the posttest. Participants in the control condition
condition, the following pair-wise comparisons collapsed across actively participated in the construction of the correct solution, but
those factors. These analyses revealed an overall difference be- were less able to identify the key structural properties of potential
tween the conditions, F(1, 73) = 4.89, p = .01, g2p ¼ :12, with specific analogs either immediately or at a delay than participants in the
variations between the high and low cuing conditions (mean dif- high cuing condition.
ference = .15, p < .05), and the high cuing and active participation Altogether, these data indicate that dynamic, ecologically-valid
control conditions (mean difference = .16, p < .01), but not between visual and spatial cuing during training impacted learners’ interac-
the low cuing and active participation control conditions (mean tions with posttest problems encountered subsequent to instruc-
difference = .006, p = .92). The high cuing condition thus led to few- tion. Specifically, high cuing led to more successful
er cross-mapping errors than either the low cuing or control con- discrimination between potential source solutions analogs. The
ditions, but there was very little difference between the low impact of cuing at time of test is well known, but these data re-
cuing and control conditions. This finding bolsters the pattern re- vealed that cuing during instruction directly impacted learners’ la-
vealed in the accuracy data, suggesting that without high cuing ter processing of target and retrieved source representations. The
during training, the low cued comparison during training was cues did not seem to impact retention or retrieval fluency for
not more effective at promoting schematization than an engaging source analogs, as participants in all conditions were readily able
lesson using one analog only. Further, these data suggest that to retrieve the relevant analog for solving the facilitory similarity
directing novices’ attention to relational structure during an problems, and performed higher on the facilitory than misleading
instructional comparison between analogs reduced future cross- similarity problems.
mapping errors, and in particular, made learners less distracted
by cross-mapped featural similarity.
7. General discussion
6.4. Discussion
7.1. Summary
Experiment 2 conceptually replicated and extended findings
from Experiment 1. Overall, results in Experiment 2 show that The results of Experiments 1 and 2 both revealed that providing
cues during an instructional comparison led to more flexible and
schematized representations of taught concepts, as measured by
increases in problem solving accuracy and decreases in cross-map-
Table 2 ping errors, than did the same instruction with minimal cuing. This
Frequency of cross-mapped strategy use across problem type in Experiment 2.
basic finding was demonstrated in two mathematical contexts in
Instructional condition Facilitory similarity Misleading similarity which the instructional analogy compared analogs that were con-
Immediate Delayed Immediate Delayed fusable – representations that were structurally identical except
test test test test for a crucial difference that caused one representation to require
High cuing condition 0 (0a) 12 (33) 31 (47) 31 (47) an additional transformation. Confusable analogs are related to
Low cuing condition 4 (21) 17 (39) 57 (51) 57 (59) the concept of a ‘‘near miss” – cases in which all features and struc-
Active participation 11 (32) 19 (40) 56 (51) 52 (51) tures are identical except for one crucial difference. However, the
control
‘‘near miss” concept has been used to emphasize crucial differ-
a
Standard deviations listed in parentheses. ences in object correspondences (see Ross & Kilbane, 1997; Nokes
 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43 39

& Ross, 2007), whereas in the present study the confusable analogs single analog control conditions did not show patterns of forget-
differed on a key structural relation. ting that were suggestive of minimally encoded, short-term reten-
Cuing during training impacted learning from comparisons be- tion for the problem solution analogs. It is possible, of course, that
tween two contrasting problem analogs (Experiment 1), as well as at longer delay differential effects of retention might be observed.
comparisons between two alternative solutions to a single problem While the present data exclude schema-only, decision rule, or
(Experiment 2). These results held both immediately and (tested in retention explanations, participants’ performance in the high cuing
Experiment 2 only) after a week delay. Without such cues, the conditions appears congruous with what one would expect from a
instructional analogy produced less flexible, expert-like reasoning, more expert-like reasoner. Participants in the high cuing condi-
and was not more effective than a lesson in which participants tions may have not only gained more conceptual, structural knowl-
were actively engaged in learning with one analog only (Experi- edge representations of the instructed analogs, but also something
ment 2). more intangible – how to attend to relational structure when pro-
cessing a new problem. The teacher used gesture, structural align-
7.2. Possible mechanisms for flexible learning from analogies ment, and visual cues, which in essence served to demonstrate the
act of identifying and mapping structural correspondences be-
The data from the present experiments provided insight into tween two analogs. Thus learners may have been trained in what
the possible mechanisms by which a highly cued instructional it means to attend to relational structure in that particular mathe-
analogy could yield increases in participants’ ability to differentiate matical context. If true, this interpretation generates two impor-
between confusable analogs. Specifically, we sought to differenti- tant predictions for future research. First, experts’ ability to
ate between four alternative characterizations of the impact of attend to relational correspondences may be distinguishable from
such cuing on novices’ problem solving: (1) abstracted schematic domain knowledge within analogical thinking, since both high and
representation of the two analogs, (2) production of a decision rule low cuing conditions showed high retention of content and ability
for when to apply each taught solution strategy, (3) improved to solve facilitory similarity problems. Second, attention to rela-
retention of the individual problems, or (4) more expert-like pro- tional structure may be trainable.
cessing of previously unseen problems (attention to structural ver-
sus surface features). 7.3. Broader theoretical implications
Data from the cross-mapping manipulation are especially infor-
mative. First, the main effect of surface similarity in both experi- These results have theoretical implications for several bodies of
ments eliminates the possibility that all, or even highly cued, educational and psychological literature on learning by analogy
instructional analogies automatically led to fully schematized and classroom mathematics instruction. Drawing connections
knowledge representations of taught concepts. In both experi- and comparing representations is core to mathematical thinking
ments, differences in accuracy across conditions were obtained be- and generalizable learning (see National Mathematics Panel,
tween the facilitory and misleading similarity problems. If 2008a, 2008b; Gallistel & Gelman, 2005; Hilbert, 1900; Polya,
participants in the high cuing conditions were reasoning on the ba- 1954; Skemp, 1976), but it is seriously underutilized in US class-
sis of purely schematic representations of the analogs, they would room teaching (Hiebert et al., 2003; Richland et al., 2007). While
have shown comparable performance on the two types of prob- the National Council of Teachers of Mathematics and disciplinary
lems. Instead, like participants in the other conditions, participants panels have long recommended using mathematical connections
in the high cuing condition were misled by the irrelevant featural to deepen students’ conceptual understanding, strategies for inte-
similarity. Thus, this evidence suggests that participants had at grating these practices into teachers’ normative routines have
least partially stored the problem analogs intact, and were not not been largely successful (Hiebert et al., 2003).
using a fully abstract schematic representation. This paper used a novel approach to this problem by bridging
It is possible that while stored source analogs were not fully cognitive science models of analogical reasoning (Gentner, 1983;
schematized after high cuing, they were more schematized than Holyoak & Thagard, 1989/2002) with study of practices identified
following a more minimally cued instructional analogy. Perhaps within everyday mathematics lessons taught in the United States
if we had included prompts to evaluate participants’ schema qual- and Internationally (Richland et al., 2007). These practices are fea-
ity (as in Novick & Holyoak, 1991), we would have detected im- sible and require low resource investment by the teacher. Further,
proved ability to represent the relational schemas in the high as evidenced by the experiments demonstrated here, benefits can
cuing conditions in spite of their use of the intact source analogs be realized even without a greater time commitment. Thus the
when mapping to target problems. strategies identified here may provide a window into techniques
In addition, these same accuracy patterns demonstrate that par- that could help scaffold teachers’ existing practices to become
ticipants did not appear to be using a discrimination rule to differ- increasingly effective. Many studies reveal that simply invoking
entiate between the two types of problems. Vanderstoep and an instructional analogy, or another opportunity for comparing
Seifert (1993) found that providing such a discrimination rule problem solutions, is not reliably effective (e.g., Gerjets et al.,
could help participants differentiate between permutation and 2006; Gick & Holyoak, 1980, 1983; Rittle-Johnson & Star, 2007).
combination problems; however, such a rule does not seem to be Thus, the current tested strategies may be an important tool for
an inevitable consequence of comparative instruction, as we would ensuring that students benefit from the analogies as intended by
have expected such a decision rule to be unaffected by irrelevant the teacher.
surface featural similarity. Using a framework of analogical reasoning to consider mathe-
The third possible explanation was that the instructional com- matics instruction and the teaching of problem solving also pro-
parison led to better retention for the instructed problems, perhaps vides insights into the distinction between surface and structural
due to more enriched encoding. But as noted above, if retention features of mathematics problems or representations. Much basic
were the differentiating factor one would expect a significant rela- work in analogy indicates that children have great difficulty inhib-
tionship between forgetting and condition, which was not the case. iting responses to appearance and surface features that appear
Furthermore, the relatively high performance on the facilitory sim- similar, in spite of being structurally irrelevant or misleading
ilarity problems even at a week delay suggests that retention could (e.g., Cho et al., 2007; Richland et al., 2006). This is crucial to math-
not be the causal explanation. Participants in the low cuing and ematical thinking, which relies upon the ability to manipulate and
40 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43

engage with abstract concepts rather than focusing on surface classroom feasible and would not require a full re-organization of
appearance or problem context (Bransford et al., 2000; National currently normative teaching practices. They also would not re-
Mathematics Advisory Panel, 2008a, 2008b). Thus drawing on in- quire immense class-time commitments since US teachers already
sights from the literature on analogical reasoning allows for con- regularly invoke mathematical analogies. Even so, teachers would
sidering new strategies for drawing learners’ attention to the key probably need to gain a nuanced understanding of their students’
structural, mathematical correspondences in a learning context comparative thinking and the difficulties inherent in learning by
rather than surface features. analogy to be most effective.
We also caution against the interpretation that including these
7.4. Implications for instructional practice cuing strategies means simply using more direct instruction. Rather,
we argue that these pedagogical tools free learners’ resources and fo-
Finally, this interpretation has direct implications for US cus their attention onto key mathematical structure. By doing so, the
teaching practices. US teachers regularly use analogies in teacher supports and facilitates the learners’ own constructive, ana-
instruction but are less likely to use the high cuing techniques logical thinking. This active sense-making process thus is made pos-
than their Asian peers (at least at the eighth-grade level). At sible by the teacher’s increased support, but it was the learners’ own
the same time, they regularly cite their students’ difficulty notic- reasoning that impacted their more flexible representations and
ing mathematical commonalities between problems that appear more expert-like treatment of new problems. As this is a key feature
different at a surface level (e.g., National Mathematics Advisory of children’s struggles with mathematical proficiency (Kilpatrick,
Panel, 2008a, 2008b). Findings from the current experiments Swafford, & Findell, 2001), the present findings may have important
suggest that modifying teachers’ current practices of using anal- implications for fostering mathematics learning.
ogy to include theoretically grounded strategies for cuing stu-
dents’ comparative thinking might at least partly address this Acknowledgments
problem. While the controlled laboratory context does not allow
for fully generalizable interpretations of these results (they We thank Keith Holyoak for helpful discussions and Yevgen-
would need to be replicated in K-12 classrooms), the findings iya Bukshpun for participation in study design and running par-
do provide important insights into strategies for optimizing ticipants. This material is partially based upon work supported
teachers’ current practices of analogy. by the Office of Naval Research under Grant N000140810186
Of course, we do not minimize the challenges in incorporating and the National Science Foundation under Grant No. IIS-
such strategies into current practices. Any attempt to transform 0757646. Any opinions, findings, and conclusions or recommen-
teaching is likely to fail unless teachers engage in deeply consid- dations expressed in this material are those of the author(s) and
ered conceptual change, though the suggested modifications have do not necessarily reflect the views of the National Science Foun-
several advantages. The identified cuing strategies align with dation. Writing was supported by a National Academy of Educa-
experimental and theoretical principles but derive from classroom tion/Spencer Foundation Postdoctoral Fellowship to the first
teaching practices in the US, Hong Kong, and Japan. Thus they are author.

Appendix A

Minimal comparison condition High comparison condition

Introduction Begin both videos by introducing the concept of permutations. Do not mention combinations yet in either
condition. Define permutations as: ‘a math problem that allows you to find the number of possible
arrangements of items in a particular order’
Source analog ‘‘Suppose there are five people running in a race. The winner of the race will get a gold medal, the person who comes
in second will get a silver medal, and the person who comes in third will get a bronze medal. How many different
orders of gold, silver, and bronze can there be?”
Source solution  Teacher asks how many runners could have come in first, and answers – 5. This number was written down first
 Teacher explains that assuming someone has taken gold, how many runners could possibly come in second,
and answers – 4
 Assuming the gold and silver are taken, the teacher questions how many runners could possibly have come in
third, and answers – 3
 The three numbers are multiplied to reach the total number of possible orders
Source solution representation (see  Three lines, with ‘‘gold,” ‘‘silver,” and ‘‘bronze” writ-  Three lines, with ‘‘gold,” ‘‘silver,” and ‘‘bronze” writ-
Fig. 1a and b for screenshots of ten underneath ten underneath.
the source analogs on the videos)  Each number is filled out in turn – first gold (five  Each number is filled out in turn – first gold (five pos-
possible), then silver (four possible) and bronze sible), then silver (four possible) and bronze (three
(three possible) possible)
 Once one slot is filled, a second row of three lines are
drawn with the one filled in to leave a record of the
thinking process. The last line has all three slots filled
Camera capture Camera moves to capture only the free second half of Camera moves to capture the whole board, including
the board, simulating the teacher erasing or changing the first and second analog
an overhead slide to remove the source analog
 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43 41

Appendix A (continued)
Minimal comparison condition High comparison condition
Transition Teacher introduces combination problems, stating Teacher introduces combination problems, stating that
that he will explain them using the following scenario they are similar to permutations but have one main
difference. He will explain them using the following
scenario
Target analog A professor is choosing students to attend a special seminar. She has 11 students to choose from, but she only has four
extra tickets available. How many different ways are there to make up the four students chosen to go to the seminar?
Target solution  Teacher asks how many students could have received the first ticket, and answers – 11. This number was writ-
ten down
 Teacher explains that assuming someone has taken the first ticket, how many students could possibly take the
second ticket, and answers – 10
 Teacher explains that using the same logic, only nine students could have taken the 3rd ticket
 Finally, only eight students could have taken the 4th ticket
 The four numbers are multiplied to reach the total number of possible orders
 The total number of possible orders are divided by the
Target solution representation (see  Four lines, with ‘‘ticket 1,” ‘‘ticket 2,” and ‘‘ticket 3”  Four lines, with ‘‘ticket 1,” ‘‘ticket 2,” and ‘‘ticket 3”
Fig. 2 for screen shots of the and ‘‘ticket 4” written underneath. and ‘‘ticket 4” written underneath.
source analogs in the videos)  Each number is filled out in turn – 11, 10, 9, 8  Each number is filled out in turn – 11, 10, 9, 8
 Once one slot is filled, a second row of three lines are
drawn with the one filled in to leave a record of the
thinking process. The last line has all three slots filled
Cues supporting comparison Source and target analog presented in immediate Source and target analog presented in immediate
sequence. The solution strategy and visual sequence. The solution strategy and visual
representation was the same for the two analogs save representation was the same for the two analogs save
for the last target analog step for the last target analog step. Teacher referenced the
source analog at each step of the target analog, using
comparative gesture to the relevant element of the
source analog. The teacher also noted when the final
step in the target solution did not map to the source
solution

Appendix B

Comparison: high cuing condition Comparison: low cuing Active participation control
condition condition
Introduction Begin all videos with the same introduction, stating that this would be a problem-solving lesson, using one particular
problem as an example
Problem Teacher writes on board and reads out loud: Bob needs 6 h to paint a square wall with a side of 10 m. How many hours
would he need to paint a square wall with a side of 5 m?)
Source analog solution Teacher describes an incorrect but common solution –
strategy strategy:
 Teacher states that one way to set up the proportion is to
put hours in the numerator and meters in the denominator
using the numbers given in the problem; Bob needs 6 h to
paint a square wall with a side of 10 m. So we get:
6h
10 m
 Teacher restates the question – ‘‘How many hours will he
need to paint a square wall that is 5 m?
6h h
10 m
¼ 5? m
 Teacher narrates and writes the cross-multiply strategy on
the board, calculating the solution as 3 h. Teacher summa-
rizes the reasoning behind this proportion: this makes
sense because the side of a square wall of 5 m is half of
the side of a square wall of 10 m”

Comparative  Teacher states that this is

participation not the correct way to
(Not active) solve the problem
 Describes the issue – Bob
is painting the area of the
whole wall, not just the
side
Camera capture: Source analog is left visible on the left side of the board Board is cleaned and only
the problem statement is
visible
Transition ‘‘That’s one way to approach the problem. Now lets look at
this again

(continued on next page)

42 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43

Appendix B (continued)
Comparison: high cuing condition Comparison: low cuing Active participation control
condition condition
Target analog solution  Teacher states that visualizing the problem should help students, and draws the following diagram: ‘‘Each side of the
strategy square wall is 10 m”

10m
Area = L  W.

10 m
 Next to the diagram, teacher reminds them of the formula for area of a square, and writes the equation:
10 m  10 m = 100 m2
 Reminds them that they are trying to figure out how long it would take for Bob to paint a wall that has a side of 5 m, so
they would need to now calculate again how much space he is painting. So we need to calculate the area of the second
wall. Demonstrates this with a diagram:

5m

5m
 Verbalizes and writes the calculation for area of this square:
5 m  5 m = 25 m2
 Walks the students through setting up the ratio: hours in the numerator and meters squared in the denominator:
6 h
100 m2
 Teacher restates the main question with emphasis: How many hours will he need to paint a 25 m2 wall?
6 h
100 m2
¼ 25 ?m2
Non-comparative active Active participation:
participation teacher asks students to
write and solve the
problem themselves on
their own piece of paper,
showing their work. (video
pauses)
Completion of target  Teacher narrates and writes the cross-multiply strategy on the board, calculating the solution as 1.5 h.
solution strategy  Summarizing, she states that they calculated using this ratio that Bob needs 1.5 h to paint a wall that’s 25 m squared.
Or in another words, it takes Bob 1.5 h to paint a wall with a side of 5 m
 On board: 1.5 h to paint 5 m
Comparative (Video paused) Participants given a sheet of paper with
participation (active) prompts to map correspondences between the two
solution strategies, and asked to decide which one is
correct
(video resumes). Teacher states that the strategy on the left
is not correct because Bob is painting the whole area, not
the edge. Gestures between source and target strategies
Conclusion In strategy one, you have hours per meter, which is the So, we find that in order to
time it takes to paint a side of a square wall. In strategy calculate the correct
two, you have hours per meter squared, which is the time number of hours, we need
it takes to paint an area of the square wall. The first to first calculate the area of
solution strategy is not realistic [Cross out]. We paint the the wall, and then the time
whole wall (or in other words, the area) of the wall and not it takes to paint that area
one side of a wall
Cues supporting Two solutions presented in serial, both to the same Two solutions presented in –
comparison problem, both using a ratio serial, both to the same
problem, both using a ratio.
Solutions visible on the
board simultaneously,
teacher gestures between
strategies when discussing
their accuracy
Now you’ll be doing a set of problems on your own. Good luck and please show all your work

References Brown, D. E., & Clement, J. (1989). Overcoming misconceptions via analogical
reasoning: Abstract transfer versus explanatory model construction.
Instructional Science, 18, 237–261.
Bassok, M., & Holyoak, K. J. (1993). Pragmatic knowledge and conceptual structure:
Catrambone, R. (1998). The subgoal learning model: Creating better examples so
Determinants of transfer between quantitative domains. In D. K. Detterman & R.
that students can solve novel problems. Journal of Experimental Psychology:
J. Sternberg (Eds.), Transfer on trial: Intelligence, cognition, and instruction
General, 127(4), 355–376.
(pp. 68–98). Norwood, NJ: Ablex Publishing.
Catrambone, R., & Holyoak, K. J. (1989). Overcoming contextual limitations on
Blessing, S. B., & Ross, B. H. (1996). The effect of problem content on categorization
problem-solving transfer. Journal of Experimental Psychology: Learning, Memory,
and problem solving. Journal of Experimental Psychology: Learning, Memory, and
and Cognition, 15, 1147–1156.
Cognition, 22, 792–810.
Chen, Z., & Klahr, D. (1999). All other things being equal: Children’s acqui-
Bransford, J. D., Brown, A., & Cocking, R. (Eds.). (2000). How people learn: Mind, brain,
sition of the control of variables strategy. Child Development, 70(5), 1098–
experience and school, expanded edition. Washington, DC: National Academy
1120.
Press.
 L.E. Richland, I.M. McDonough / Contemporary Educational Psychology 35 (2010) 28–43 43

Chi, M., Feltovich, P., & Glaser, R. (1981). Categorization and representation of Pashler, H., Bain, P., Bottge, B., Graesser, A., Koedinger, K., McDaniel, M., et al. (2007).
physics problems by experts and novices. Cognitive Science, 5, 121–152. Organizing instruction and study to improve student learning (NCER 2007-2004).
Chi, M. T. H., & Ohlsson, S. (2005). Complex declarative learning. In K. J. Holyoak & R. Washington, DC: National Center for Education Research, Institute of Education
G. Morrison (Eds.), Cambridge handbook of thinking and reasoning (pp. 371–399). Sciences, US Department of Education. Retrieved <http://ncer.ed.gov>.
New York, NY: Cambridge University Press. Perry, M., Vanderstoep, S. W., & Yu, S. L. (1993). Asking questions in first-grade
Cho, S., Holyoak, K. J., & Cannon, T. D. (2007). Analogical reasoning in working mathematics classes: Potential influences on mathematical thought. Journal of
memory: Resources shared among relational integration, interference Educational Psychology, 85, 40–41.
resolution, and maintenance. Memory & Cognition, 35(6), 1445–1455. Polya, G. (1954). Mathematics and plausible reasoning. I. Induction and analogy in
Clement, J. (1993). Using bridging analogies and anchoring intuitions to deal with mathematics. II. Patterns of plausible inference. Princeton, NJ: Princeton
students’ preconceptions in physics. Journal of Research in Science Teaching, University Press.
30(10), 1241–1257. Reed, S. K. (1987). A structure–mapping model for word problems. Journal of
Duit, R. (1991). On the role of analogies and metaphors in learning science. Science Experimental Psychology: Learning, Memory, and Cognition, 13, 124–139.
Education, 75, 649–672. Reed, S. K. (1989). Constraints on the abstraction of solutions. Journal of Educational
English, L. (Ed.). (2004). Mathematical reasoning: Analogies, metaphors, and images. Psychology, 81, 532–540.
Hillsdale, NJ: Erlbaum. Renkl, A., & Atkinson, R. K. (2003). Structuring the transition from example study to
Gallistel, C. R., & Gelman, R. (2005). Mathematical cognition. In K. J. Holyoak & R. G. problem solving in cognitive skill acquisition: A cognitive load perspective.
Morrison (Eds.), Cambridge handbook of thinking and reasoning (pp. 559–588). Educational Psychologist, 38, 15–22.
New York, NY: Cambridge University Press. Richland, L. E., Kornell, N., & Kao, S. L. (2009). Leave no test behind: Failed retrieval
Gentner, D. (1983). Structure-mapping – A theoretical framework for analogy. on a pretest can enhance future learning. Journal of Experimental Psychology:
Cognitive Science, 7, 155–170. Applied, 15, 243–257.
Gentner, D., Loewenstein, J., & Thompson, L. (2003). Learning and transfer: A general role Richland, L. E., Holyoak, K. J., & Stigler, J. W. (2004). Analogy generation in eighth
for analogical encoding. Journal of Educational Psychology, 95, 393–408. grade mathematics classrooms. Cognition and Instruction, 22, 37–60.
Gentner, D., & Toupin, C. (1986). Systematicity and surface similarity in the Richland, L. E., Morrison, R. G., & Holyoak, K. J. (2006). Children’s development of
development of analogy. Cognitive Science, 10, 277–300. analogical reasoning: Insights from scene analogy problems. Journal of
Gerjets, P., Scheiter, K., & Catrambone, R. (2006). Can learning from molar and Experimental Child Psychology, 94, 249–271.
modular worked examples be enhanced by providing instructional Richland, L. E., Zur, O., & Holyoak, K. J. (2007). Cognitive supports for analogy in the
explanations and prompting self-explanations. Learning and Instruction, 16(2), mathematics classroom. Science, 316, 1128–1129.
104–121. Rittle-Johnson, B., & Star, J. R. (2007). Does comparing solution methods facilitate
Gick, M. L., & Holyoak, K. L. (1980). Analogical problem solving. Cognitive Psychology, conceptual and procedural knowledge? An experimental study on learning to
15, 306–355. solve equations. Journal of Educational Psychology, 99, 561–574.
Gick, M. L., & Holyoak, K. L. (1983). Schema induction and analogical transfer. Roediger, H. L., III, & Karpicke, J. D. (2006). The power of testing memory: Basic
Cognitive Psychology, 15, 1–38. research and implications for educational practice. Perspectives on Psychological
Goldstone, R. L., & Medin, D. L. (1994). The time course of comparison. Journal of Science, 1, 181–210.
Experimental Psychology: Learning, Memory, and Cognition, 20, 29–50. Ross, B. H. (1987). This is like that: The use of earlier problems and the separation of
Hiebert, J., Gallimore, R., Garnier, H., Givvin, K. B., Hollingsworth, H., Jacobs, J., et al. similarity effects. Journal of Experimental Psychology: Learning, Memory, and
. Teaching mathematics in seven countries: Results from the TIMSS 1999 video Cognition, 13, 629–639.
study. (NCES 2003-013). Washington, DC: US Department of Education. Ross, B. H. (1989). Distinguishing types of superficial similarities: Different effects
Hilbert, D. (1900). Mathematical problems. Lecture delivered before the International on the access and use of earlier problems. Journal of Experimental Psychology:
Congress of Mathematicians, Paris, France. Learning, Memory, and Cognition, 15, 456–468.
Holyoak, K. J., & Thagard, P. (1989). Analogical mapping by constraint satisfaction. Ross, B. H., & Kennedy, P. T. (1990). Generalizing from the use of earlier examples in
Cognitive Science: A Multidisciplinary Journal, 13, 295–355. problem solving. Journal of Experimental Psychology: Learning, Memory, and
Holyoak, K. J., & Thagard, P. (2002). Analogical mapping by constraint satisfaction. In Cognition, 16(1), 42–55.
T. A. Polk & C. M. Seifert (Eds.), Cognitive modeling (pp. 849–909). Cambridge, Ross, B. H., & Kilbane, M. C. (1997). Effects of principle explanation and superficial
MA: MIT Press. similarity on analogical problem solving. Journal of Experimental Psychology:
Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). The expertise reversal effect. Learning, Memory, & Cognition, 23, 427–440.
Educational Psychologist, 38(1), 23–31. Scheiter, K., Gerjets, P., & Catrambone, R. (2006). Making the abstract concrete:
Kilpatrick, J., Swafford, J., & Findell, B. (2001). Adding it up: Helping children learn Visualizing mathematical solution procedures. Computers in Human Behavior,
mathematics. Washington, DC: National Academy Press. 22, 9–25.
Kolodner, J. L. (1997). Educational implications of analogy: A view from case-based Schoenfeld, A. H., & Herrmann, D. J. (1982). Problem perception and knowledge
reasoning. American Psychologist, 52(1), 57–66. structure in expert and novice mathematical problem solvers. Journal of
Mayer, R. E. (2001). Multimedia learning. New York: Cambridge University Press. Experimental Psychology: Learning, Memory, and Cognition, 5, 484–494.
National Mathematics Advisory Panel (2008a). Foundations for success: The final Schwartz, D. L., & Bransford, J. D. (1998). A time for telling. Cognition and Instruction,
report of the national mathematics advisory panel. Washington, DC: US 16, 475–522.
Department of Education. Skemp, R. (1976). Relational understanding and instrumental understanding.
National Mathematics Advisory Panel, (2008b). Reports of the task groups and Mathematics Teaching, 77, 20–26.
subcommittees. Washington, DC. Sweller, J., Chandler, P., Tierney, P., & Cooper, M. (1990). Cognitive load and selective
National Research Council (2001). Adding it up: Helping children learn attention as factors in the structuring of technical material. Journal of
mathematics. In J. Kilpatrick, J. Swafford, & B. Findell (Eds.), Mathematics Experimental Psychology: General, 119, 176–192.
learning study committee, center for education, division of behavioral and social Sweller, J., & Cooper, G. A. (1985). The use of worked examples as a substitute for
sciences, and education. Washington, DC: National Academies Press. problem solving in learning algebra. Cognition and Instruction, 2(1), 59–89.
Nokes, T. J., & Ross, B. H. (2007). Facilitating conceptual learning through analogy Tohill, J. M., & Holyoak, K. J. (2000). The impact of anxiety on analogical reasoning.
and explanation. In L. Hsu, C. Henderson, & L. McCullough (Eds.), 2007 Physics Thinking & Reasoning, 6, 27–40.
education research conference (pp. 7–10). American Institute of Physics Vanderstoep, S. W., & Seifert, C. M. (1993). Knowing ‘how’ versus knowing ‘when’:
Conference Proceedings. Improving transfer of problem-solving principles. Journal of the Learning
Novick, L. R. (1988). Analogical transfer, problem similarity, and expertise. Journal of Sciences, 3, 93–111.
Experimental Psychology: Learning, Memory, and Cognition, 14, 510–520. Viskontas, I. V., Morrison, R. G., Holyoak, K. J., Hummel, J. E., & Knowlton, B. J. (2004).
Novick, L. R., & Holyoak, K. J. (1991). Mathematical problem solving by analogy. Relational integration, inhibition and analogical reasoning in older adults.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 17, Psychology & Aging, 19, 581–591.
398–415. Waltz, J. A., Lau, A., Grewal, S. K., & Holyoak, K. J. (2000). The role of working
Paas, F. G. W. C., & Vanmerriënboer, J. J. G. (1994). Instructional-control of cognitive memory in analogical mapping. Memory & Cognition, 28, 1205–1212.
load in the training of complex cognitive tasks. Educational Psychology Review, 6, Zhu, X., & Simon, H. A. (1987). Learning mathematics from examples and by doing.
351–371. Cognition and Instruction, 4, 137–166.

View publication stats