Beruflich Dokumente
Kultur Dokumente
This paper reports on an exploratory study investigating (mathematical) problem solving processes
of fifth graders (ages 10 to 12) from German secondary schools. The videos of 110 processes of 45
different pupils working on 5 tasks were coded using an adapted version of the protocol analysis
framework from Schoenfeld (1985). The data shows a significant correlation between the pupils'
problem solving behavior (characterized by Schoenfeld's episodes) and their success (or failure).
Please note that this article is a major update to Rott (2011) – so there might be some
repetitions regarding the “Design of the Study” and “Methodology” paragraphs (but not the
“Results” section).
BACKGROUND
Problem solving is important in every day life, when one is confronted with situations where
the solution path is not immediately obvious (cf. OECD 2003), as well as in mathematics,
because "what mathematics really consists of is problems and solutions." (Halmos 1980, p.
519) It is widely accepted, that solving problems is important for the learning of mathematics,
thus it is part of many school curricula, e.g. in the United States and in Germany (cf. NCTM
2000; KMK 2003).
Though the importance of problem solving is widely aknowledged, the term “problem
solving” has different meanings ranging from working on routine tasks to solving puzzling or
difficult situations (cf. Schoenfeld 1992, p. 10 ff.). I refer to the latter interpretation,
accounting a “problem” as a task for which no obvious solution method is available (cf.
Mayer & Wittrock 2006, p. 287). Problem solving is cognitive and, that is important to note,
personal (cf. ibid.). One person's problem can be a routine task for another person who knows
an algorithm or a solution method to resolve the situation. Therefore, research on problem
solving should examine processes instead of only tasks and/or their solutions.
Most frameworks regarding the problem solving process “attribute some relationship to
Pólya's problem-solving stages” (Fernandez et al. 1994, p. 196): (1) understanding the
problem, (2) devising a plan, (3) carrying out the plan, and (4) looking back (Pólya 1945).
More modern frameworks – like those of Fernandez et al. (1994), Mason et al. (1982), and
Schoenfeld (1985 ch. 4) – break with the linear nature of Pólya's steps, allowing cycles and
abcde
Rott
steps backwards, and they highlight the importance of metacognitive and self-regulatory
behavior; apart from that, they stay close to the original. In addition to his partition of the
problem solving process into phases, Pólya (1945) proposes assistance for problem solvers by
presenting questions and guidelines regarding heuristic and self-regulatory content.
At present, there is a lack of research regarding the problem solving abilities and processes of
younger children (cf. Heinze 2007, p. 15), thus we started the project presented below to
address (among others) the following research questions:
• How can problem solving and routine processes be distinguished from each other?
• Is there a correlation between pupil's behavior (their decision making) in problem
solving processes and their success in solving the problem?
name of the task mathematical area number of pupils working this task
sum 110
1
Mathematik AG an der Leibniz Universität means to Mathematics Working Group at Leibniz University.
Overall, we used more than 30 different tasks, selected to represent a wide range of
mathematical areas and to allow the use of different heuristics. We chose five of those tasks
(and all related processes) for the analyses that are presented here (see Table 1 and Appendix
1 for the task formulations).
METHODOLOGY
Product Coding: To determine the pupils' success in problem solving, their work results
were graded in four categories: (1) No access, when the pupils did not work on the task
meaningful. (2) Basic access, when they solved (parts of) the problem but the solution had
notable flaws. (3) Advanced access, when they solved the problem for the most part. And (4)
full access, when the pupils solved the task properly and presented appropriate reasons.
This grading system was customized for each task with examples for each category (see
Appendix 2 for examples). Then, all the products were rated independently by the author and
a research assistant. We agreed in almost all cases and discussed the few products with
different ratings, reaching consensus every time. Those discussions led to better defined
categories. It is important to note, that the two members of a pair could – and sometimes did –
gain diverse ratings, when their written results differed.
Process Coding: The pupils' behavior was coded using the framework for the analysis of
videotaped problem solving sessions presented by Schoenfeld (1985, ch. 9).2 He intended to
“identify major turning points in a solution. This is done by parsing a protocol into
macroscopic chunks called episodes […].” (ibid., p. 314) An episode is “a period of time
during which an individual or a problem-solving group is engaged in one large task [...] or a
closely related body of tasks in the service of the same goal [...].” (ibid., p. 292) Schoenfeld
(1992, p. 189) continues: “We found […] that the episodes fell rather naturally into one of six
categories:”
(1) Reading or rereading the problem.
(2) Analyzing the problem (in a coherent and structured way).
(3) Exploring aspects of the problem (in a much less structured way than in Analysis).
(4) Planning all or part of a solution.
(5) Implementing a plan.
(6) Verifying a solution.
We adopted this framework for our study with the following modifications: Firstly, we added
new categories of episodes, because our fifth graders – unlike the university students
Schoenfeld observed – showed plenty of non-task related behavior. For example, there were
2
Though this framework is older than 25 years, it was the most promising one of several frameworks I tried to apply to
our data. It also helps answering the question, whether fifth graders use Pólya-like steps in their problem solving
processes and whether they profit from it (cf. Rott 2011).
periods of time during which our pupils talked about TV series or recreational activities that
were coded as acts of digression. But these new types of episodes are not important for the
results presented in this paper, because I'll focus on the task-related episodes, which are (2) –
(6) of Schoenfeld's list.
Secondly, we operationalized Schoenfeld's episode descriptions, because in the beginning we
had some difficulties in coding reliably; especially figuring out the differences between
Analysis and Exploration was hard (as predicted in Schoenfeld 1992, p. 194). We noticed that
Schoenfeld’s (1985, ch. 9) empirical framework refers to his theoretical model of the problem
solving process (ibid., ch. 4) which is based on Pólya’s (1945) list of questions and
guidelines. By assuming an analogy between Schoenfeld's framework and Pólya's work, we
were able to sharpen the descriptions in our coding manual. Applying Pólya’s questions to the
problem solving processes helped us deciding whether to code Analysis or Exploration (see
Figure 1 for a summary and an excerpt of Pólya’s questions and advices).
The coding of the videotapes was done independently by research assistants and the author.
We applied the “percentage of agreement” approach to compute the interrater-reliability as
described in the TIMSS 1999 video study (cf. Jacobs et al. 2003, p. 99 ff.) for randomly
chosen videos, gaining more than PA = 0.7 for the parsing into episodes and more than PA =
0.85 for the characterizing of the episode types. But, more importantly, every process was
coded by at least two raters. Whenever those codes did not coincide (which they did most of
the time), we attained agreement by recoding together (cf. Schoenfeld 1992, p. 194).
This coding was done separately for the children, even though they worked in pairs, because
it was not uncommon for them to work on different aspects of the tasks sitting next to each
other, thus gaining different episode-codes. To avoid an imbalanced weighting of the data, all
processes were counted independently, even if the two pupils of a pair worked together all the
time and got the same codes.
RESULTS
As stated above, I'm concentrating on the five task-related episode types Analysis,
Exploration, Planning, Implementation, and Verification. The pupils' behavior will be
characterized by the presence or absence of those types of episodes. Due to space limitations
further approaches have to be left out of this paper (see the “Conclusions”).
Routine vs. Problem Solving Processes
As stated above, it is impossible to attribute “routine” or “problem”3 to a task formulation,
because this characteristic depends on the solver and the related process. While extreme cases
might be easy to distinguish, it is not always easy to separate “routine” from “problem”
solving processes. Theoretical considerations, using the Schoenfeld coding as a tool, lead to
the following results: The coding of Exploration should be a sure sign of a “problem” solving
activity, because of heuristic usage and the absence of an algorithm to solve the task. In
contrast, processes with only Implementation or Planning-Implementation episodes are good
candidates for “routine” processes (cf. Schoenfeld 1985, p. 301). It becomes difficult, when
(planning-) implementing behavior is preceded by an Analysis and/or succeeded by a
Verification, because sometimes it is required to understand a task before one can remember a
suitable solution scheme and even solutions obtained by an algorithm should be checked and
verified. On the other hand, a very structured approach to solve a task with no known
algorithm could also lead to a coding of Analysis and (Planning-) Implementation without
any other episode types in the process. Therefore, there is no well-defined theoretical
assignment of problem solving types regarding the presence and absence of Schoenfeld
episodes to the process attributes “routine” or “problem”.
In the context of validating the process codes, we also discussed the performances of the
pupils. Did they show signs of knowing an appropriate algorithm or did they even mention to
know one? These interpretations of the empirical data confirm the theoretical thoughts
insofar, that every process with at least one Exploration episode was considered to be a
“problem” process. We assigned the “Super Car” task (see Appendix 1) with the reasonable
assumption, that every fifth grader knows a scheme to solve it. And indeed, all but one of the
12 pupils working this task showed (planning-) implementing behavior with no approach to
an Exploration and no need for an Analysis; two of those finished their work on this task with
a Verification. The interpretations of the other tasks (especially the “Number Series”
problem) unfortunately confirm, that there are a few videos with Analysis and
Planning-Implementation but without Exploration, that we graded as “problem” processes.
Thus, the empirical result after more than 100 coded processes is: Processes with an
Exploration can be considered as “problem” while processes with only Implementation or
Planning-Implementation can be regarded as “routine” without need to further look at the
regarded videos thoroughly. But processes with all other possible combinations of episodes
3
In the literature, the terms task and problem are often poorly distinguished. In this paragraph, to avoid confusion, the task
and process attributes “problem” and “routine” are labelled by quotation marks.
4
Because of the nominal character of the process categories, no (Pearson or Spearman) correlation could be calculated.
Process / product categories no & basic access advanced & full access sum
sum 43 45 88
The analyses of individual tasks and their processes also show interesting results (see Rott
2011), but only the combination of multiple tasks leads to statistically reliable results. Please
note, that the “routine processes” of the “Super Car” and “Squares on a Chessboard” tasks
(see above) were excluded from this computation.
References
Bruder, Regina (2003). Learning Methods and Techniques on How to Solve Problems. URL
(25.08.2011): http://www.math-learning.com/files/learnmethod.pdf
Fernandez, Maria L.; Hadaway, Nelda & Wilson, James W. (1994). Problem Solving:
Managing It All. In The Mathematics Teacher, Vol. 87, No. 3, p. 195 – 199.
Halmos, Paul R. (1980). The Heart Of Mathematics. In The American Mathematical
Monthly, Vol. 87, No. 7 (Aug. - Sep., 1980), p. 519 – 524.
Heinze, Aiso (2007). Problemlösen im mathematischen und außermathematischen Kontext.
Modelle und Unterrichtskonzepte aus kognitionstheoretischer Perspektive. In Journal für
Mathematikdidaktik (JMD), 28(1), p. 3 – 30.
Jacobs, Jennifer; Garnier, Helen; Gallimore, Ronald; Hollingsworth, Hilary; Givvin, Karen
B.; Rust, Keith et al. (2003). Third International Mathematics and Science Study 1999
Video Study Technical Report. Volume 1: Mathematics. Washington: National Center for
Education Statistics. Institute of Education Statistics, U. S. Department of Education.
KMK (2003). Kultusministerkonferenz (Ed.). Bildungsstandards im Fach Mathematik für
den mittleren Schulabschluss. Bonn: KMK.
Mason, John; Burton, Leone; & Stacey, Kaye (1982 / 2010). Thinking Mathematically.
Dorchester: Pearson Education Limited. Second Edition.
Mayer, Richard E. & Wittrock, Merlin C. (2006). Problem Solving. In Alexander, P. A. &
Winne, P. H. (Ed.): Handbook of Educational Psychology. Routledge, p. 287 – 304.
NCTM (2000). National Council of Teachers of Mathematics (Ed.). Principles and Standards
for School Mathematics. Reston: NCTM.
OECD (2003). Organisation for Economic Co-Operation and Development (Ed.). The PISA
2003 Framework – Mathematics, Reading, Science and Problem Solving Knowledge and
Skills. URL (12.12.2010): http://www.oecd.org/dataoecd/46/14/33694881.pdf
Pólya, George (1945). How to Solve it. Princeton, NJ: Princeton University Press.
Rott, Benjamin (2011). Problem Solving Processes of Fifth Graders: an Analysis. In Ubuz, B.
(Hrsg.): Proceedings of the 35th Conference of the International Group for the Psychology
of Mathematics Education, Vol. 4. Ankara, Turkey, p. 65 – 72.
Schoenfeld, Alan H. (1985). Mathematical Problem Solving. Orlando, Florida: Academic
Press, Inc.
Schoenfeld, Alan H. (1992). On Paradigms and Methods: What do you do when the ones you
know don't do what you want them to? Issues in the Analysis of data in the form of
videotapes. In The Journal of the learning of sciences, 2(2), p. 179–214
For instance, if there are the numbers 10, 6, 3 in three consecutive caskets, the 6 adds up to a
square number with its left (10+6=16) and its right neighbor (6+3=9).
How could Marco fill-up his 15 caskets?
Source: Fürther Mathematikolympiade, 2005/06, 1. round (www.fuemo.de)
Squares on a Chessboard
Peter loves playing chess. He likes playing chess so much that he keeps thinking
about it even when he isn’t playing. Recently he asked himself how many
squares there are on a chessboard. Try to answer Peter’s question!
Idea: Mason, Burton, & Stacey (2010, p. 17)
Seven Gates
A man picks up apples. On his way into town he has to go through seven gates. There is a
guardian at each gate who claims half of his apples and one apple extra. In the end the man
has just one apple left. How many apples did he have first?
Source: Bruder (2003, p. 12)
Squares on a Chessboard:
(1) No access – no (meaningful) work on the task; e.g. just counting 64 (small) squares.
(2) Basic access – being aware that there are larger squares than 1x1; e.g. counting 85 squares
(the number of all squares that cover the whole chessboard without overlapping).
(3) Advanced access – presenting all sizes of possible squares, from 1x1 to 8x8, but
miscounting their number.
(4) Full access – presenting the correct number of squares: 204.