SD 257 441 IR 011 673 Kurland, D. Midian, Ed. Title: Docuxidit Resume

DOCUXIDIT RESUME
SD 257 441 IR 011 673
AUTHOR Kurland, D. Midian, Ed.

TITLE Developmental Studies of Computer Programming Skills.
A Symposium: Annual Meeting of the American
Educational Research Association (New Orleans,
Louisiana, April 23-27, 1984). Technical Report No.
29.
INSTITUTION Bank Street Coll. of Education, New York, NY. Center
for Children and Technology.
PUB DATE Oct 84
NOTE 135p.
PUB TYPE Viewpoints (120) -- Reports - Research/Technical
(143) -- Collected Works - Conference Proceedings
(021)
EDRS PRICE MF01 Plus Postage. PC Not Available from EDRS.

DESCRIPTORS Cognitive Development; *Cognitive Processes;
*Computer Science Education; Computer Software;
Curriculum Development; Elementary Education;
Epistemology; *Learning Processes; *Programers;
*Programing; Secondary Education; *Skill
Development
ABSTRACT
The five papers in this symposium contribute to a
dialog on the aims and methods of computer education, and indicate
directions future research must take if necessary information is to
be available to make informed decisions about the use of computers in
schools. The first two papers address the question of what is
required for a student to become a reasonably proficient programmer.
The first--"Mapping the Cognitive Demands of Learning to Program" (D.
Midian Kurland, Katherine Clement, Ronald Mawby, and Roy D.
Pea)--reports a study of high school programming novices who
participated in an intensive summer programming course. The second
paper--"The Development of Programming Expertise in Adults and
Children" (D. Midian Kurland, Ronald Mawby, and Nancy
Cahir)--examines how expert programmers acquired their skill, with
attention to the amount of time invested and the type of resources
available when they were learning to program. The last three papers
look beyond programming to the issue of transfer. The third-- "Issues
and Problems in Studying Transfer Effects of Programming" (Kate
Ehrlich, Valerie Abbott, William Salter, and Elliot
Soloway)--examines whether learning to program helps students solve
problems in other related intellectual domains. The fourth--"What
Will It Take to Learn Thinking Skills Through Computer Programming?"
(Roy D. Pea)--discusses research on the transfer of high level
thinking skills from programming. The final paper--"Making
Programming Instruction Cognitively Demanding: An Intervention Study"
(John Dalby, Francois. Tourniaire, and Marcia C. Linn)--describes a
study in which a curriculum was designed explicitly to make
programming more cognitively challenging. A concluding commentary by
Jan Hawkins discusses the issues raised in the papers and offers
thoughts on current and future directions for research in this field.
(THC)
r-
aftl"
1%*.
tr.
4
u* DEPARTNINIT OF EDUCATION
NA NAL INSTITUTE OF EDUCATION
EDU TIONAL RESOURCES INFORMATION
CENTER (ERIC,
L. TIM, document has been reproduced
received from the person at °roam:awn
onornaung rt
1 Mena chengss hive been made to improve
reproductron wieldy
_
rooms I we.* at 000n,erie 'UMW P1 the Paco 1
moot do not oacasaailk intraimont official ME
paestion Or pokey
"PERMISSION TO REPRODUCE THIS

MATERIAL IN MICROFICHE ONLY
HAS BEEN GRANTED BY
b_4122.4*-;.4:3 11,A4011N40-..
TO THE EDUCATIONAL RESOURCES

INFORMATION CENTER (ERIC)"
1984 AERA Annual Meeting

April 23-27, New Orleans, LA
Symposium:
Developmental Studies of Computer
Programming Skills
Edited by
D. Midian Kurland
Technical Report No. 29
2
0
ti.
1984 AERA Annual Meeting

April 23-27, New Orleans, LA
Symposium:
Developmental Studies of Computer
Programming Skills
Edited by
D. Midian Kurland
October 1984
3 1
DEVELOPMENTAL STUDIES OF COMPUTER PROGRAMMING SKILLS
D. Midian Kurland 1 INTRODUCTION

D. Midian Kurland, 2 MAPPING THE COGNITIVE DEMANDS OF
Katherine Clement, LEARNING TO PROGRAM
Ronald Mawby, and
Roy D. Pea
D. Midian Kurland, 3 THE DEVELOPMENT OF PROGRAMMING
Ronald Mawby, and EXPERTISE IN ADULTS AND CHILDREN
Nancy Cahir
Kate Ehrlich, 4 ISSUES AND PROBLEMS IN STUDYING
Valerie Abbott, TRANSFER EFFECTS OF PROGRAMMING
William Salter, and
Elliot Soloway
Roy D. Pea 5 WHAT WILL IT TAKE TO LEARN THINKING
SKILLS THROUGH COMPUTER PROGRAMMING?
John Dalby, 6 MAKING PROGRAMMING INSTRUCTION
Francoise Tourniaire COGNITIVELY DEMANDING: AN INTERVENTION
and Marcia C. Linn STUDY
Discussant:
Jan Hawkins 7
4
INTRODUCTION
D. Midian Kurland
Center for Children and Technology
Bank Street College of Education
This collection of papers was originally presented at a symposium entitled

Developmental Studies of Computer Programming at the 1984 American
Educational Research Association annual meeting in New Orleans. The
symposium was a forum for presenting research on the development of
computer programming skills. While the papers were intended for a
research audience, there was considerable interest in this topic by
teachers and educational policy makers as well. Clearly, programming is
a topic of great concern to many in the educational community. While
these particular papers do not specifically address policy issues, many of
the issues they do raise have direct bearing on educational practices.
Thus, they should be of interest to all those concerned with the role of
computer programming in the precollege curriculum.
The systematic study of how children interact with computers is only just
beginning. The fields of software psychology and developmental cognitive
science are still very much in ti,,tir infan...y. Yet schools are under
pressure right now to make dech..../ns about what they should be doing
with programming. Whether because of a belief in the educational value
of programming or fear of being left behind, in the past two years many
schools have opted to introduce some form of programming instruction into
their curriculum. In many schools it has now replaced traditional. CAI as
the primary way in which computers are being used at most grade levels.
The determination to teach programming has occurred despite the almost
total absence of information on how to teach programming, or what
aspects of programming languages children of various grade levels are
most able to understand and use effectively.
The rush to acquaint students of all ages and abilities with the basics of
computer operation and programming has engendered a raft of new
problems which extend far beyond the selection of the best hardware and
software packages. While promoters of computer programming have been
quick to make promises about its educational benefits, they are promises
that have not as yet been fulfilled through practice, nor addressed
adequately through systematic research. Instead, the field is being
driven by largely philosophical debates on the potential educational value
of letting students discover science, math, and linguistic concepts
through the medium of programming. The claims for the benefits of
programming have excited the imagination of thousands of teachers,
parents and researchers. Unfortunately, however, in moving from
educational philosophy to pedagogical practice, we find ourselves on less
fully developed ground. Educators are giving their students time to
freely explore with the computer using languages such as Logo, or are
teaching students to code in BASIC or Pascal on the grounds that
programming is a fundamental skill necessary to participate fully in the
new "information society". Yet because no one knows quite what to
expect from such instruction, nor even how or what to assess in order to
know whether the instruction is successful, programming's foothold in the
schools, particularly at the earlier grades, is tenuous, and the directions
it should be taking unclear.
At the center of the current uneasiness over the role of programming in

schools is concern with its relationship to the the rest of the curriculum.
Teachers are realizing that despite all the rhetoric about how "easy" it is
for kids to program, to learn to program well is a difficult and time
consuming enterprise. Yet without achieving at least certain minimal
levels of competency, many students (as well as teachers) become
frustrated and feel stymied in their efforts to use the computer in ways
they feel are appropriate. However, simply allotting more time to
programming instruction is not an appropriate response in most cases.
Nothing that is new and time consuming can be added to the school
curriculum without it replacing something else. What then should
computer programming replace? For how long? For which students? At
what ages? And once students have mastered the rudiments of
programming, how can it be incorporated into the rest of curriculum so
that it can be used in the service of other educational goals?
Pressure is mounting on policy makers to find answers to these questions.
Yet at the present time there is little systematic research which policy
makers can draw on to help guide their decisions. Aside from a limited
number of anecdotal reports, little is known of how novices, particularly
children, learn programming languages and the pragmatics of debugging
and testing programs. Little too is know, despite all the claims, about
the impact programming may have on the development of a wide range of
high level thinking skills. While computers are rapidly becoming
intricately interwoven into the very fabric of our society, just what
students should know of their inner workings, and how this knowledge
should be taught must for the most part be left to speculation.
The five papers in this symposium examine computer programming from a

variety of perspectives. In doing so they raise a number of important
issues which in turn have serious policy implications. The purpose of
these papers, it should be pointed out, was not to provide answers for
all the issues that they raise. Rather, they contribute to a much needed
dialog on the aims and methods of computer education, and indicate
directions future research must take if we are to have the information
required to make informed decisions about the use of computers in
schools. By providing systematic data on how children learn about
programming a computer, contentions about what children can or can not
do with computers and programming can be more fully constrained by
actual observations of novices in the process of learning. At the same
time, such data moves us closer to achieving our long range goal of
creating a developmental theory of programming. Such a theory must
ultimately serve as the cornerstone of any effective programming
pedagogy.
The papers in this collection fall into two categories. The first two
papers address the question of what is required for a student to become
a reasonably proficient programmer. The last three papers look beyond
programming per se to the difficult issue of transfer. Specifically, they
ask whether programming promotes the development of generalizable
thinking skills or mathematical concepts.
The first paper by Kurland, Clement, Mawby and Pea reports on a study
of high school programming novices who took part in an intensive summer
programming course. This paper examines some of the potential cognitive
prerequsites for being able to program well. The second paper by
Kurland, Mawby and Cahir reports on an interview study with expert
child and adult programmers. This paper examines how expert
programmers acquired their skill, with particular attention to the amount
of time they invested and the type of resources they had available when
they were learning to program.
The third paper by Ehrlich, Abbott, Salter and Soloway directly

addresses the important issues of transfer. They examine whether
learning to program helps students solve problems in other closely related
intellectual domains. The issue of transfer is picked up again in the
fourth paper by Pea who provides a discussion of research on the
transfer of high level thinking skills from programming. The fifth paper
by Dalbey, Tourniaire, and Linn describes a study in which they tried
explicitly to design a curriculum to make programming more cognitively
challenging.
Finally, Jan Hawkins provides a commentary in which she discusses the

issues raised in the symposium and offers some thoughts on current and
future directions for research in this field.
4
8
MAPPING THE COGNITIVE DEMANDS OF LEARNING TO PROGRAM
D. Midian Kurland, Catherine Clement, Ronald Mawby and Roy D. Pea
Introduction
Vociferous arguments have been offered for incorporating computer
programming into the standard pre-college curriculum (Pale .t, 1980;
Luehrmann, 1981; Snyder, 1984). Many parents and educators believe
that. computer programming is an important skill for all children in our
technological society. In addition to pragmatic considerations, there is
the expectation among many educators and psychologists that learning to
program can help children develop general high level thinking skills
useful in other disciplines such as mathematics and science. However,
there is little evidence that current approaches to teaching programming
bring students to the level of programming competence needed to develop
general problem solving skills, or to develop a model of computer
functioning that would enable them to write useful programs, Evidence of
what children actually do in the early stages of learning to program (Pea
& Kurland, 1984; Rampy, 1984) suggest that in current practices,
programming many not evoke the kinds of systematic, analytic, and
reflective thought that is characteristic of expert adult programmers (cf.
Kurland & Cahir, 1984).
As the teaching of programming is initiated at increasingly early grade

levels, questions concerning the cognitive demands for learning to
program are beginning to surface. Of particular interest to both teachers
and developmental psychologists is whether there are specific cognitive
demands for learning to program that might inform our teaching and tell
us what aspects of programming will be difficult for students at different
stages in the learning process.
The work reporte was supported ThT7ET N a wr'T-3W='--IsfittdeTaT
Education (Contract No. 400-83-0016). The opinions 'expressed do not
necessarily reflect the position or policy of the N:-..tional Institute of
Education and no official endorsement should be inferred. We would like
to thank Karen Sheingold and Linda Caporael for their comments on
earlier drafts of this paper.
In the first part of this paper we explore factors that may determine the
cognitive demands of programming. In the second part we report on a
study of these cognitive demands conducted with high school students
learning Logo. The study was premised on the belief that in order for
programming to help promote the development of certain high level
thinking skills, students must attain a relatively sophisticated
understanding of programming. Therefore we developed two types of
measures: measures to assess programming proficiency, Lnd measures to
assess certain key cognitive abilities which we hypothesized to be
instrumental in allowing students to become proficient programmers. The
relationship between these two sets of measures was then. assessed.
Issues in Determining the Cognitive Demands of Programming
One of the main iosues in conducting research on the cognitive demands
of programming is that the term "programming" is used loosely to refer to
many different activities involving the computer. These activities range
from what a young child seated in front of a computer may do easily
using the immediate command mode in a language such as Logo to what
college students struggle over, even after several years of programming
instruction. Contrary to the popular conceptions that young children
take to programming "naturally" while adults do not, what the child and
the adult novice are actually doing and what is expected of them is
radically different.Clearly the cognitive demands for the activities of
the young child and the college student will also differ. Thus what is
meant by programming must be clarified before a discussion of demands
can be undertaken.
Defining programming and assessing its cognitive demands is problematic

because programming is a complex configuration of activities. These
activities vary according to what is being programmed, the style of
programming, and how rich and supportive the surrounding programming
environment is (Pea & Kurland, 1983; Kurland, Mawby & Cahir, 1984).
One consequence of the fact that programming refers to a configuration of

activities is that different combinations of activities may be involved in
any specific programming project. These activities include, at a general
level, problem definition, design development and organization, code
2 0
writing and debugging (See Pea and Kurland, Different
1983).
combinations of activities will entail different cognitive demands. For
example, a large memory span may facilitate the mental simulations
required in designing and comprehending programs. Or analogical
reasoning skill may be important for recognizing the similarity of different
programming tasks and for transferring programming methods or
procedures from one context to another. An adequate assessment of the
cognitive demands of programming will depend on analyses of the
programming activity and examination of the demands of different
component processes.
Specifyink Levels of Programming Expertise
In assessing the cognitive demands of programming, specifying the
intended level of expertise is essential. Different levels of expertise will
entail different cognitive demands. In many Logo programming
classrooms, we have observed children engaging in what we term brute
force "paragraph" programming, or what Rampy (1984) has termed
product-oriented programming. This style is analogous to so-called
spaghetti programming in BASIC. When programming, students decide on
desired screen effects, then write linear programs, lining up commands
that will cause the screen to show what they want in the order they want
it to happen. Students do not engage in problem decompositie or use
the powerful features of the language to structure a solution to the
programming problem. For example, if a similar shape is required several
times in a program, students will write new code each time the effect is
required, rather than writing one general procedure and calling it
repeatedly. Programs thus consist of long strings of Logo primitives that
are nearly impossible, even for the students who have written them, to
read, modify or debug. Though students may eventually achieve their
goal, or at least end up with a graphics display they are content with,
the only "demands" we can imagine for such a linear approach to
programming are stamina and determination.
Thus, as a first step in determining what cognitive demands there are for
learning or doing programming we need to distinguish between linear and
modular programming (or between learning to program elegantly and
efficiently, in contrast to a style emphaaizing the generation of effects
without any consideration of how they were generated.)
The beginner& linear style of constructing programs, whether in Logo or

BASIC, contrasts with modular programming, a planful process of
structured problem solving. Here, component elements of a task are
isolated, procedures for their execution developed, and the parts
assembled into a program and debugged. This type of programming
requires a relatively high level understanding of the language. Modular
programming in Logo, wl'Iere programs consist of organized reusable
subprocedures, requires that students understand the flow of control of
the language, the powerful control structures such as recursion, and the
passing of values of variables between procedures. The cognitive
demands for doing this kind of programming are different from the
demands for linear programming, as are the potential cognitive benefits
which may result from the two programming styles.
Distinguishing Between Product and Process
In assessing the demands for different levels of expertise, however,
it is important that level of expertise not be equated with the effects the
students' programs produce. We must distinguish product form process
(Werner, 1937). We have seen very elaborate graphics displays created
with entirely brute force programming. One characteristic of highly
interactive programming languages such as Logo and BASIC is that
students can often get the effects they want simply by repeated trial and
error, without ani overall plan, without fully understanding how effects
are created, without the use of sophisticated programming techniques,
and without recognizing that a more planful program could be used as a
building block in future programs.
Furthermore, in school classrooms we have often seen students borrow

code from each other and then integrate the code into their programs
without bothering to understand why the borrowed code does what it
does. Students therefore can often satisfy a prograL,.....ng assignment by
piecing together major chunks imported from other sources. Though such
"code stealing" is an important and efficient technique widely employed by
expert programmers, an overreliance upon other people's code that is
beyond the understanding of the borrower is unlikely to lead to deeper
understandings of programming. Therefore, if we simply correlate
students' products with their performance on particular demands or
4
programming proficiency measures, we are likely to find the correlations
greatly attenuated.
Compensatory Strategies
This point suggests another important factor which complicates the
identification of cognitive demands of programming. Any programming
problem can be solved in many ways. Different programmers can utilize a
different mix of component processes to write a successful program. This
allows for high levels on some abilities to compensate for low levels on
others. For example, a programmer may be deficient in planning skills
needed for good initial program design but may have high levels of skills
needed to easily debug programs once drafted. Thus it will not be
possible to identify a unique set of skills which are necessary for
programming. Instead, different programmers may possess alternative
sets of skills, each of which are sufficient for programming competence.
The Programming Environment
The features of the programming environment may also increase or
decrease the need for particular cognitive abilities important for
programming. We cannot separate the pure demands for using a
programming language from the demands and supports provided by the
instrumental, instructional, and social environments. For example, an
interactive language with good trace routines can decrease the need for
pre-planning by reducing the difficulty of debugging. Similarly,
implementations of particular languages that display both the student's
program and the screen effects of the code side by side in separate
"windows", such asInterlisp-D, can reduce the difficulty in
understanding and following flow of control.
In learning to program, the instructional environment can reduce certain

cognitive demands if it offers relevant structure, or can increase demands
if it is so unstructured that learning depends heavily on what the
students themselves bring to the class. For example, understanding the
operation of branching statements of the IF- THEN -ELSE type requires an
appreciation of conditional logic and an appreciation for the operation of
truth tables. If students have not yet developed such an appreciation,
doing programs which require even simple conditional structure can be
very confusing. However, with appropriate instruction, an understanding
of how to use conditional commands in some limited contexts (such as
conditional stop rules to terminate the execution of a loop) can be easily
picked up by students. Thus in the absence of instruction, conditional
reasoning skill can be a major factor in determining who will learn to
program. However, with instructional intervention, students can pick up
enough functional knowledge about conditional commands to take them
quite far.
Instruction is important in other ways as well. For example, students in

our experience are very poor at choosing appropriate programming
projects on their own that are within their current ability, yet which will
stretch their understanding and force them to think about new types of
problems. They are poor at constructing for themselves what Vygotsky
would describe as the "zone of proximal development" (Rogoff & Wertsch,
1984). Thus, too little guidance on the part of the teacher can lead to
inefficient or highly frustrating programming projects. Yet too much
teacher imposed structure can make the projects seem arbitrary and
uninteresting, and thus less likely to evoke the full attention and
involvement of the student. Finding the right balance between guidance
and discovery will have a major impact on the kinds of cognitive abilities
which the students will have to, or choose to, bring to the programming
task.
Finally, the social context can mediate the demands placed on an

individual for learning to program since programming, particularly in
elementary school classrooms, is often a collaborative process (Hawkins,
1983). The varying skills of the collaborators can compensate for one
another in producing a program. Thus groups of students may be able
to produce programs that any one of them alone could not have produced.
While work in teams is typical of expert programmers, it currently raises
thorny assessment pr,..blems in an educational system that stresses
individual accountability.
In summary, several factors complicate the identification of general
cognitive abilities which will broadly effect a child's ability to learn to
program. In asking about demands we must consider level of expertise,
the impact of supportive and/or compensatory programming environments,
and the role of instructional and social factors which interact with
children's initial abilities for mastering programming.
Analysis of the Cognitive Demands of Modular Programming

One of the central motivations for teaching programming to pre-college
students is to provide them with a tool for understanding mathematical
concepts and for developing general problem solving skills. We believe
that achieving this goal requires that students program well (Mawby,
1984). Knowing only basic turtle graphics commands in Logo can be a
useful goal in some instructional coni.exts, for example, if a teacher wants
students to explore angles or estimate distances. However, to use a
language like Logo to develop an understanding of math concepts like
variables and functions, and to learn problem solving techniques such as
planning and modular problem solving, the aspects of Logo which involve
these concepts must be exploited; Logo cannot be used merely as a
substitute for a ruler, protractor, or form blocks. This means that
students must become reasonably good modular programmers. They must
learn to program with variables and procedures, to generate code that
can be reusable, and to understand the control structure of the
language. Given the importance of learning to program at this
proficiency level, what are likely to be some of the cognitive abilities
such programming requires? A rational analysis of the cognitive
requirements of designing and comprehending modular programs suggests
that, among other skills, means-ends procedural reasoning and
decentering may be particularly important.
We expect that procedural reasoning ability is an important skill

underlying the ability to program since programmers must make explicit
the antecedents necessary for different ends, and must follow all possible
consequences of different antecedent conditions. Designing and following
the flow of control of a program requires understanding different kinds of
relations between antecedent and consequent events, and organizing and
interrelating the local means-end relations (modules) leading to the final
end. Procedural reasoning thus includes understanding conditional
relationships, temporal sequencing, hypothetical deduction, and planning.
Decentration also may be important skill in programming since

programmers must distinguish what they know and intend from what the
computer has been instructed to execute. This is important in program
construction and in debugging. The program designer must realize the
level of explicitness required to adequately instruct the computer, and in
debugging must differentiate what the program "should" do from what it
did do. We have found that such decentering is a major hurdle in
program understanding at the secondary school level (Kurland Pea,
&
1984)
On the basis of this rational analysis we designed a study to investigate

the relationship of measures of procedural reasoning and decentering to
the acquisition of programming skill.
Method
To investigate empirically the relationship between these cognitive abilities
and programming competence, we studied novice programmers learning
Logo. Logo was chosen in part because of the high interest it has
generated within the edx,cational community, and in part because the Logo
language has specific features which support certain important thinking
skills. For example, the strategy of problem decomposition is supported
by Logo's modular features. Logo procedures may be created for each
subpart of a task. The procedures may be written, debugged, and saved
as independent, reusable modules and then used in combination for the
solution of the larger problem. Efficient, planful problem decomposition
in Logo results in flexibly reuseable modular procedures with variable
inputs. While the same can be true of languages such as BASIC, the
formal properties of Logo appeared to be more likely to encourage
structured programming on the part of the student.
Participants and Instructional Setting
Participants in the present study were 79 8th-l1th grade female high
school students enrolled in an intensive six-week summer program
designed to improve their math skills and introduce them to programming.
The goal of the program was to improve students' mathematical
understanding while building up their sense of control and lessening their
anxiety about mathematics. (See Confrey, 1984 and Confrey, Rommney &
Mundy, 1984 for details about the affective aspects of learning to
program.) Those admitted to the program r, e students who were
generally doing very well in school and had high career aspirations, but
who were doing relatively poorly in mathematics and, in some cases,
experiencing a great deal of math related anxiety.
Each day, the students attended two 90 minute mathematics classes as well
as lectures and demonstrations on how mathematics is involved in many
aspects of art and science. Each student also spent 90 minutes each day
in a Logo programming course. The teachers hoped that the programming
experience would enable students to explore mathematical principles, and
lead them to new insights irto mathematics. The guiding philosophy of
the program, which influenced both the mathematics and Logo instruction,
was constructivist. This Piagetian inspired philosophy of instruction
holds that a person's knowledge and representation of the world is the
result of his/her own cognitive activity. Learning will not occur if
students simply memorize constructions presented by their teachers in the
form of facts and algorithms. Thus, students were expected to construct
understandings for themselves through their direct interactions and
explorations of the mathematics or programming curriculums.
The Logo instruction was structured around small classes with the
students working primarily in pairs, i.e. two students to a computer.
There was a six to one student-teacher ratio, and ample access to
printers and resource materials. In order to provide structure for the
students' explorations of Logo, the staff of the program created a detailed
curriculum designed to provide systematic learning experiences involving
the basic Logo turtle graphics commands and control structures. While
the curriculum itself was detailed and carefully sequenced, the style of
classroom instruction was influenced by the discovery learning model of
Papert (1980). Thus students were allowed to work at their own pace
and were not directly accountable for mastery of specific concepts or
commands. The instructors saw their primary role as helping the
students develop a positive attitude towards mathematics and
programming. In this respect, the program seemed by our observations
to have been very successful.
The emphasis of the course was on learning to program and doing turtle
geometry. While the teachers repeatedly drew attention to underlying
mathematical principles at work in assignments given, they also tried to
bring students to an adequate level of programming proficiency. Thus
the curriculum was designed around a series of "challenges" (i.e.
worksheets) that students were to work through in a systematic order.
These challenges included creating graphics using basic Logo primitives,
unscrambling programs, predicting program outcomes, coordinating class
projects to produce large-scale programs, and exploring properties of
degrees, radiants, and circles. It was assumed that the students would
find the challenges and the opportunity to work at the computer
enjoyable, and therefore largely self-motivating.
Measures
We were interes*zd in how their level of programming proficiency would
relate to the spet..,fic cognitive abilities which our analysis of the demands
of programming had indicated to be potentially important for mastery. We
therefore developed measures cognitive demands and programming
proficiency to use in this study.
Demands Tasks
Two demands tasks were developed and administered to students at the
beginning of the program. The first, Procedural flow of control task,
was designed to assess students' ability to use procedural reasoning to
follow flow of control determined by conditional relations. Students had
to negotiate a maze in the form of an inverted branching tree (see Figure
1). At the most distant ends of the branches were a set of labeled
goals. To get to any specific goal from the top of the maze students had
to pass through "gates" at each point where a branching occurred. The
- 10 - ld
conditions for passage through the "gates" required satisfying either
simple or complex logical structures (disjunctive or conjunctive). Passage
through gates was controlled by varying sets of geometric "tokens" that
the student was presented with at the beginning of each problem. Each
gate was marked with the type or types of tokens that were required to
gain passage through it. For example, it students were given a circle
token, they could pass through a circular gate, but net a square gate.
If they had both a square and triangle token, then they co'.ild pass
through a joint square-triangle gate but not a joint square-circle gate.
The task had two parts. In the first, students were presented with five
problems in which they had to find paths through the maze that did not
violate the conditions for passage through the gates. Students were
given sets of tokens and asked to discover all possible goals that could
be reached with that set.
In the second part of the task, we designed two problems, based on a

more complex maze, to add further constraints and possibilities for
finding the optimal legal path to the goals. Unlike part 1, at a certain
point in the maze students could choose to trade tokens of one kind for
tokens of another. In addition, as they passed through each gate, they
lost the token that they used to get through it. This feature introduced
additional planning and hypothetical reasoning requirements as the
students had to foresee the sequential implications for choosing one path
over other possible paths. This task allowed for several possible
solutions that met the minimum requirements of the task (i.e. reaching a
specified goal at the end of one branch). However, some possible
solutions were more elegant than others in that they required the use of
fewer tokens. Thus it was of interest to see whether students would
choose to go beyond an adequate solution to find an elegant one.
The task was designed using non-english symbolisms so that verbal ability
and comprehension of the "if-then" connectives would not be a
confounding factor. In natural is nguage, if-then is often ambiguous, and
its interpretation dependent on context. We did not include standard
tests of the if-then connective in propositir ial logic because computing
1J
CQ
ik.
truth values as these tests require is not strictly relevant to following
complex conditional structures in programming.
The procedural flow of control task, therefore, involved a system of

reasonable, though arbitrary and artificial rules to make it analogous to a
programming language and to prohibit subjects' use of prior world
knowledge. The nested conditional structure of the tree and the logical
structures of the nodes were designed to be analogous to logical
structures in computer languages.
The second demands task was designed to assess decentering as well as

procedural and temporal reasoning. In this debugging task students were
required to detect bugs in a set of driving instructions written for
another person to follow. Students were given a set of written
directions, a map, and local driving rules. They were asked to read
over the directions and then by referring to the map catch and correct
bugs in the directions so that the driver could successfully reach the
destination. Students had to consider means-ends relationships and
employ temporal reasoning to follow the given instructions, determine
their accuracy, and correct "buggy" instructions. They had to decenter,
making a distinction between the their and the driver's knowledge, in
order to tell whether instructions were sufficiently explicit and accurate.
The kind of bugs students were asked to find and correct included:
(a) Inaccurate information bug: instructions given were simply

incorrect (e.g., telling the driver to make a right hand turn at a corner
instead of a left).
(b) Ambiguous information bug: instructions were not sufficiently
explicit to enable the driver to correctly make a choice between
alternative routes (e.g., telling the driver to exit off a road without
specifying which of two possible exits to use).
(c) Temporal order bug; one line of instruction was stated at the
wrong time (e.g., telling the driver to pay a token to cross a toll bridge
before saying where to purchase the tokens).
(d) Bugs due to unusual input conditions, and embedded bugs in
which obvious corrections fail because they introduce and/or leave a bug
k
(e.g. , telling the driver to make a detour in response to a rush hour
traffic rule, but failing to note that the obvious detour violates a second
traffic rule).
Programming Proficiency Tasks
We were interested in determining the demands of modular computer
programming. Therefore we nef.-Jed to develop measures that would
provide a detailed assessment of programming proficiency which entailed
understanding of flow of control, program decomposition, and reusability
of code. In designing the test we were less concerned with assessing
students' knowledge of individual commands than with comprehension of
the overall structure of the language and the pragmatics of programming.
The test consisted of three parts: one production task and two
comprehension tasks.
Production task. The production task was a paper and pencil test
designed to assess planning, problem decomposition, and features of
programming style such as the conciseness and generality of procedures.
Students were shown a set of seven geometric figures, represented in
Figure 2.
The students were instructed to select five of the seven figures and write
Logo programs to produce them. The task called for students first to
indicate the five figures they would write programs for, and then to
number them in the order the programs would be written. This
instruction, it was hoped, would encourage the students to plan before
they began to write their programs. Students were free, however, to
alter their choice of figures and/or order once they began to code. For
each of their five programs, they were to write the code and give the
run command needed to make the program produce the figure.
The task sheet included an area labeled "workspace", analogous to the

Logo workspace, in which students could write procedures to be called by
their programs. The layout of the task sheet, two sample problems, and
explicit instructions made it clear that procedures, once written in the
"workspace," were available to all programs.
:
am
oni i ia
ate
Mena WI I Sung:
laspermsa swag Ng MINN
no mum um
'
111,81116411O8
in
i mu ma
MIR MO
Jilin: a ommimpeasso
me WIMP
me
11110§MIUSOOSIMIEMS
cwmgmfaXMOWOOMMIS WI
IMMO US
MINIM
sallarnallia umasAMMISMOUSOURO genops4 Us
mmommangESUMBUSIMOSOUSSMOICAOMENOW a
WWII ,
ITAIRTIMASPUSErall,:=1 11111111WI
Illiminitsuldiiiiiiiiiii1111111111111111
...
i 1111111111111
EMS .IM 111
Itsimast iumegus.
1111111111111
I
1111=11 ammusammonse 11111111Mlesumwm
111111111:11:1121111111.4111111111 Willi'
IfiligNI11111111111111111111:14g111111111
wassumummusammumesommennumasousal11111111:::
iiiiini1111111111111111M0111111/111
11111111111111/1111111°""""I
WallininniUMUMILIMIIIMMUMMEI
gmemensammummommummommossommannewaramommaluat
mils:::::::::::::::::::::::::::::::::::::::::!:::
Emma
masimmftsmsussammummummunmenum............imismos
1111111111111111111:1: IIIIIIIII
R"..."1"28.11177:allidiiiiiiiirain
"Mina 11101
RUMWOMMUsomensmaSSIONWOO:UMMEMSIMOM:SOmmuuMM
immumatimummisinsummunilliiir
":77:Enegassulle.. I
num min wilignillig
SOMISSMOSimmumlle
mum
MOUUMMUMUMMOMMOU
MOOMMUOURSOUSUOMOBERUMUSUROMROMM
MOSMOMMOSUMMROU
mopommummmn.wOMMMOOSUOMOMS
minummiummatmommesimmummum
mum
rtr
The task was designed to encourage planning for modular procedures
which could be reused across programs. In fact, figures B, C, E, F,
and G could be programmed by writing just three general purpose
procedures. An optimal solution would be to write a procedure with two
variable inputs to produce rectangles, a "move -ver" procedure with one
input, a "move up" procedure with one input, and then to use those
three procedures in programs to produce figures B, C, E, F, and G.
Also, Figures B and G could be most efficiently produced using recursive
programs, though recursion was not necessary.
Figures A and D were included as distractor items. Unlike the other five
figures, they were designed not to be easily decomposed and cannot be
easily produced with code generated for any of the other figures.
The task, then, could be solved by planful use of flexible modules of

code. It could also be solved in many other ways, such as writing low
level, inelegant "linear" code consisting of long sequences of FORWARD,
LEFT and RIGHT commands, thereby never reusing modules of code. We
were particularly interested in this style dimension since a "linear"
solution gives no evidence that the student is using the Logo constructs
which support and embody high level thinking.
Comprehension Tasks. Each of the two comprehension tasks
presented four procedures: one superprocedure and three subprocedures.
The students were asked first to write functional descriptions of each of
the procedures, thus showing their ability to grasp the meaning of
commands within the context of a procedure. Then the students were
asked to draw on graph paper the screen effects of the superprocedure
when executed with a specific input. To draw the screen effects
students had to hand simulate the program's execution, thus providing a
strong test of their ability to follow the precise sequence of instructions
dictated by the program's flow of control.
In the first comprehension tasks the superprocedure was named

TWOFLAGS and the subprocedures were CENTER, FLAG, and BOX.
Figure 3 presents the Logo code for the procedures and a correct
drawing of the screen effect of TWOFLAGS 10.
The second comprehension task included procedures with two inputs and a
recursive procedure with a conditional stop rule. The task was designed
to make the !minter procedure progressively harder to follow. The
superprocedure was named ROBOT and the three subprocedures BOT,
MID, and TOP. Figure 4 presents the Logo code, and correct drawing of
the screen effects of ROBOT 30 25.
Both programming comprehension tasks were designed as paper and pencil

tests that did not require the use of the computer. Students were given
a sheet that listed the programs, a sheet on which to write their
descriptions of what each procedure would do, and graph paper on which
to draw their prediction of what the program would do when executed.
Procedure
The demands measures were administered to the students on the first day
of the program along with a number of mathematics, problem solving and
attitude measures. (See Confrey, 1984 for a discussion of the attitude
measures.) Students were tested together in a large auditorium.
Instructions for each test were read by the experimenters who monitored
the testing and answered all questions. Students were given 17 minutes
for the procedural reasoning task and 12 minutes for the debugging task.
In the final week of the program, the students were administered the
Logo proficiency test. Testing was done in groups of approximately 30
students each. Again the experimenters gave all instructions and were
present throughout the testing to answer students' questions. Students
were given 30 minutes for the production task and 15 minutes each for
the comprehension tasks.
Results
Programming Proficiency Tasks
To use Logo as a tool for high level thinking one must use relatively
sophisticated Logo constructs, such as procedures with variable inputs
and super procedures which call subprocedures. To write and
understand Logo program- using these language constructs one needs to
understand something about the pragmatics of writing programs as well as
developing a good grasp of Logo's control structure, that is, how Logo
- 17 -
25
Demands of Programming
FIGURE 3
First Logo Comprehension Task with Correct Drawing of
the Resulting Screen Effects
LOGO PROCEDURES: DRAWING OF SCREEN EFFECTS OF

TWOFLAG'S
TO CENTER
PENUP
HOME
PENDOWN
END
TO FLAG :X
FORWARD 15
BOX :X
CENTER
END
TO BOX :SIDE
REPEAT 4 [FORWARD :SIDE RT 90]
END
TO TWOFLAGS :X
E
CENTER
FLAG 15
PENUP
RT 90 FORWARD 20 LT 90
PENDOWN
FLAG :X
END
26
-18-
Demands of Programming
FIGURE 4.
First Logo Comprehension Task with Correct Drawing of
the Resulting Screen Effects
LOGO PROCEDURES: DRAWING OF SCREEN EFFECTS OF

ROOT 3D 25
TO BOT :X :Y
FORWARD :X
RT 90
FORWARD :Y
END
TO MID :X :Y 1
BOT :X :Y
RT 90
ROT :X :Y
END
TO TOP :X
IF :X 5 RT 90 BACK 10 STOP
REPEAT 4 [FORWARD :X RT 90]
FORWARD 5 LT 90
TOP :X 10
END
TO ROBOT :X :Y
HT
MID :X :Y
BACK 15 LT 90
BOT :X - 10 :Y - 15
RT 90 PU FORWARD 50 PD
TOP :Y - 10
END
determines the order in which commands are executed. The empirical
question addressed here is whether students develop such an adequate
understanding as the result of five weeks (approximately 45 hours) of
intensive Logo instruction.
Comprehension tasks. The assessments of Logo proficiency given at
the end of the course indicated that mastery of Logo was limited. On the
TWOFLAGS task, 48% of the students correctly drew the first flag, which
required simulating the execution of TWOFLAGS through its call to FLAG
in line 2. Then 21% correctly drew the second flag with 19% of the
students correct on both flags, showing that in almost all cases
performance way cumulative.
A third of the students were partially right on the second flag. Analysis
of errors on the so- .id flag by these students indicated that more of
them had trouble following the flow of control than keeping track of the
values of the variables. An error in place on the second flag suggests
that the student's simulation did not execute all the positioning lines of
code, especially the call to CENTER in the last line of FLAG. This
reveals an error in flow of control. An error in size on the second flag
suggests that the student did not correctly pass the variable from
TWOFLAGS to FLAG to BOX.
On the ROBOT task, 65% of the students correctly drew the body of the
robot, which involved simulating the execution of ROBOT through its call
to MID. Then 37% correctly drew the leg, which involved following the
execution through ROBOT's call to BOT in line 4. TOP is the recursive
procedure which, with inputs to ROBOT of 30 25, would execute three
times. The first time TOP draws the head, the second time it draws the
nose, and the last time it draws the mouth, and then stops. Finally, 16%
of the students correctly drew the head, 13% succeeded with the nose,
and only 2% were able to follow the program execution all the way through
to the mouth. The cumulative percentages are within 3% of these absolute
percentages.
Again analysis of the errors of students who were partially correct

showed that more of these students correctly passed values of variables
zo 28
than followed the flow of control. In partially correct drawings, the
parts of the robot were more often correctly sized than correctly placed.
The students' written descriptions of the procedures in both the

TWOFLAGS and ROBOT tasks showed that many had at least a vague,
general understanding of the procedures. Often students seemed to
understand the code in that they gave adequate glosses of individual
lines. But when tested by the drawing task, many revealed that they
did not understand Logo's control structure well enough to trace the
program's execution. This becomes especially clear when the order of the
lines in a listing of the program differed from the order in which the
lines execute.
Some students failed to grasp the fact that since variable values are local
to the procedure call, values can be passed among procedures under
different names. Even more failed to understand the most basic fact of
flow of control: after a called procedure is executed, control returns to
the next line of the calling procedure. Inadequate models of recursion,
including conflation of recursion with Logo's REPEAT command, were
shown by some students (cf. Kurland & Pea, 1984),
Production task. Results with the production task showed limited
use of efficient programming techniques that required variables and
reusable subprocedures. While most students were able to generate the
figures, many did so following the "linear" programming style. ally 21%
of the students avoided both distractor items. An additional 35% avoided
either A or D singly. Thus 44% of the students wrote programs for both
A and D. Given a low level of programming proficiency choosing the
distractors is reasonable because, by design, linear programs for the
distractors are easier than linear programs for figures B and G (and
comparable to C and F).
Among the possible approaches to the task are "analytic" and "synthetic"
decomposition. By "analytic" decomposition, we mean analyzing a single
figure into compcnent parts, writing procedures for the parts, and
having the program call the procedures. By "synthetic" decomposition we
mean decomposition of the entire problem set into components, writing
procedures for the parts, and then having each of the five programs call
the appropriate modules of code. Note that while the five non-distractor
figures contain only rectangles, the rectangles are of different sizes.
Thus high level "synthetic" decomposition, unlike "analytic"
decomposition, requires a general procedure with variable inputs for
producing the rectangles.
Students were much more likely to use "analytic" than "synthetic"

decomposition. 88% of the students at least once wrote, used and reused
a procedure, thus giving evidence of some "analytic" decomposition.
However, only 20% students gave evidence of "synthetic"
of the
decomposition by using a procedure for more than one program.
Figure 5 provides more detail on the features of Logo students used to

produce the individual figures. In the analysis represented by Figure 5,
we wished to know, for each figure, whether students could write code to
produce it, and whether they could correctly use REPEAT, variables, and
recursion. The REPEAT command is the simplest within-procedure
embodiment of Logo's modular features. Variables go further in
transforming procedures into reusable functions, making the procedures
more general and hence more useful. Recursion is an extremely powerful
Logo construct in which a procedure can call on copies of itself from
within other copies. These features of Logo make modular code possible
and thus support problem decomposition strategies. So in addition to the
product measure of "does it run", we looked at these other indicators of
programming style.
A good summary indicator of style is the number of commands used to
produce the program. For these tasks elegant programs use few
commands. We counted each use of a Logo primitive as one command.
Each procedure call was counted as one command and on the first call to
a procedure the commands within the procedure were counted. On
subsequent calls to that procedure only the call itself was counted.
The graph at the top of Figure 4 shows, for each f: are, the range of
number of commands used, the mean, and the region containing the
middle 50% of the scores. For comparison, we include the number of
_ 22ao
501 Interval
EM3111it niA 61)u:1911RM:I
nein
0 *star of oconaves in on
*tint solution of the 5-fl osge took
*seer
of
figze A figure D fig.rce E figare F figure C figure 8 figure G

O
% bhp did it: 73 .51 .91 .91 .96 .60 .31

variable paogram: .86 .90 .91 .80 .85 .47 ,aa
varielales used: .05 .02 .14 .12 .10 .43 .40
repeat used: .08 .02 .49 .84 .65 .49 .68
recurslai used: .00 .00 .00 .00 .00 .04 .08
31
commands used in an optimal solution of the task as a whole. This
particular optimal solution "synthetically" decomposes the five rectangular
figures with three subprocedures, and produces the programs in the
order E, F, C, B, G.
The figures fall into three groups: the distractors A and D; C, E, and
-; and B and G. As noted, nearly half of the students chose figures A
and D, and 90% of the students who chose these figures were able to
write a Logo program to produce them. As expected from the design of
the figures, less than 10% of these programs used variables or REPEAT.
Most of the code was low level "brute force" style which could not be
reused in other programs. Thus, while the students wrote programs to
produce the figurt, their programming style gave no indication that they
were engaged in the high level thinking which Logo can support.
The group of figures C, E, and F was chosen by over 90% of the

students, and nearly 90% of these students wrote workable programs for
them. More than half of the students correctly used REPEAT, Logo's
simpler, within-procedure modular construct. Less than 15% of these
prctgrari.3 correctly used variables. This more elegant across-program
cor truct was largely ignored. As a result, most students needed more
than the optimal number of commands to write programs for Figures F
and C.
Fewer students chose figures B and G (60% and 31%, respectively), and
only half of these students wrote wor cable programs to produce them.
These programs used REPEAT and variables more often (REPEAT: 49% in
B, 68% in G; variables: 43% in B, 40% in G). It seemed that the more
skilled students chose these figures and did them quite well. Of the
others who chose these figures, about half the students did not attempt
to use variables, and half used variables incorrectly. Again, because few
students did "synthetic" decomposition, most programs had more than an
optimal number of commands.
No one tried to write a recursive program for any of the figures except B
and G, where recursion is appropriate. But fewer than 10% of the
- 24 -
32
students who chose figure B or G wrote correct recursive programs.
This powerful Logo construct was conspicuous in its absence.
What factors may have kept these students from using the powerful and
elegant features of Logo? It is unlikely that students did not notice the
geometrical sirnilarit:r among, for instance, Figures C, E, and F. But to
do a "synthetic" decomposition of the task one needs to write procedures
with variables. Moreover, coordinating subprocedures in a
superproceeure requires a good understanding of Logo flow of control.
Performance on the comprehension tasks showed that students had a fair
understanding of individual lines of Logo code, but had difficulty in fact
in following program flow of control. Thus to produce elegant Logo
programs one needs to comprehend Logo variables and control structure.
Cognitive Demands Tasks
There was a fairly broad range of performance on the demands tasks
(Clement, 1984). Many students showed moderate or high levels of
reasoning skills as assessed by these tasks, and a few found the tasks
fairly difficult.
Procedural flow of control task. The two parts of this task were
examined individually. The first part included a series of problems for
students to solve. Each problem posed a different set of constraints
and/cr goals for going through the maze. Some difficult problems
required more exhaustive testing of conditions than others (i.e., the
given tokens satisfied many nodes early on). Some had benefits for
using alternate strategies, such as searching from the bottom up rather
than top down. Performance was relatively low on the more difficult
problems (30-40% correct as opposed to 55-70% correct on the less r-)mplex
problems). This indicated that when many possibill.ies had to be
considered, and there were no easy shortcuts to reduce the number of
possibilities, students had difficulty testing all conditions. Doing so
required careful attention to the structure of the maze and the sequential
relationships between the gates.
In the second part, there were three levels of efficiency among correct
routes corresponding to the number of tokens required to successfully
reach the goal. Only 14% of the students on the first problem and 21% of
the students on the second problem found the most efficient route, while
43 of the students on the first problem and 79% of the students on the
second were unable to reach the goal at all. Thus few subjects tested
the hypotheses needed to discover the most efficient route.
Debugging task. Table 1 shows the percent of students detecting
and correcting each of the four types of bugs in the task. As shown,
inaccurate information and temporal bugs were easiest to detect and
correct (72$ to 91% success). It was more difficult to successfully correct
the ambiguous instructions. Only 48$ of the students were able to write
instructions that were explicit enough for a driver to choose correctly
among alternate routes. For the lines with embedded bugs only 21$ h of
the subjects fully corrected the instructions; 40$ caught and corrected
one bug but not the other.
Results indicate that subjects had little difficulty detecting first order
bugs and correcting these when the corrections were simple, for example,
changing a number or direction to turn. However, when subjects had to
be very explicit and exhaustively check for ambiguity and for additional
bugs they were less successful.
Relationship of the Demands Measures to Programming Proficiency
Analysis of the relationship between these demands tasks and the
assessments of programming proficiency yielded an interesting set of
results. As can be seen in Table 2, the demands measures correlated
moderately with composite scores on both tests of programming
proficiency.
Examination of correlations with subscores on the programming production

task showed that students' ability to write an adequate, runnable program
was less highly correlated with demands measures than their appropriate
use of variables, use of subprocedures within programs, or use of a
minimum number of commands to write programs (one indication of
program elegance).
Other subcomponents of the production task that we assumed would

correlate highly with the demands measures, in particular whether
students reused procedures across several programs or used recursion,
TABLE 1
DEBUGGING TASK
(S of students; Nu7?)
NO CHANGE CATCH NO FIX CATCH SOME FIX. CATCH. & FIX

DUG TYPE
WRONG .03 .06 ma" .91

INSTRUCTION
AMBIGUOUS .11 .41 .nal .48

INSTRUCTION
TEMPORAL .16 .11 ma* .73

ORDER BUG
EMBEDDED .29 .10 .40 .21

BUGS
* This category is not applicable

TABLE 2
Correlations of 'Demands' Measures with

Measures of Programming Proficiency (N=79)
Procedural
Reasoning Part 1 A
Procedural
Reasoning Part 2 B .34** --
Debugging
Task C .38*** .27**
Math
Level D .51*** .304** .42***
Production
Proficiency E .45*** .19* .394* .38***
Comprehension
Proficiency F .54*** .50*** .45*** .59*** .26*
Teacher
Rating G .30** .20* .22* .37*** .26** .54*** --
P4.05
** p4.01
* *IP P4.001
were not highly correlated. However, few students engaged in either of
these forms of programming. Thus, because of the floor effect it is
difficult to know how to interpret this lack of a significant correlation.
Interestingly, though few used the more advanced programming
techniques, many students manifested high levels of reasoning skills on
the demands measures. Their demonstrated logical abilities, however, were
not sufficient to enable employment of sophisticated programming
techniques. Other knowledge specific to the programming domain would
appear to be required in addition to the underlying cognitive capacity to
reason in the ways we assessed.
In general the correlations of the demands measure were higher with

programming comprehension than with programming production. The
design of the production task may have contributed to these findings.
Students could write linear programs and still E...icceed on the task, and
most did so. This was true even for those who at times in their class
projects had utilized more advanced programming techniques from time to
time. In contrast, the comprehension task required students to display
their understanding of sophisticated programming constructs. Thus the
comprehension task wl.s better able to test the limits of programming
novices' understanding of the language. However, a production task
such as we employed may prove the better indicator of programming
proficiency for students once they attain a more advanced level of ability.
We examined the relation between math achievement level (assigned on the

basis of grade point average, courses taken in school, and scores on
math tests administered on the first day of the program) and Logo
proficiency. Math level was as good a predictor of programming
proficiency as the specific demands measures taken individually.
However, when math level was partialled out of +1 correlations in
Table 2, they all remained significant at the .01 level or better, with the
exception of the correlation between part two of the procedural reasoning
task and program production proficiency. Thus our demands measures
appear to tap abilities that are independent of those directly tied to
mathematics achievement.
When both mathematics achievement and performance on our demands
measures were entered into a multiple regression analysis, with Logo
proficiency as the dependent variable, the multiple correlations were .71
and .52 for programming comprehension and production respectively.
Thus a quarter to a half of the variability in tested programming
proficiency was accounted for by mathematical understanding and specific
cognitive abilities bearing a rational relationship to programming.
Discussion
The present study was aimed at identifying the cognitive demands for
reaching a relatively sophisticated level of programming proficiency. We
examined students learning Logo in an instructional environment that
stressed self discovery within a sequence of structured activities, but
with no testing or grading. Given this setting and amount of instruction,
we found students for the most part managed to master only the basic
turtle graphics commands and simpler aspects of the program control
structure. While they gained some understanding of such programming
concepts as procedures and variables, most did not develop enough
understanding of Logo to go beyond the skill level of "effects
generation". Thus, for example, though they used variables within
procedures, they seldom passed variables between procedures, used
recursion, or reused procedures across programs. Those aspects of
programming requiring a more sophisticated understanding of flow of
control and the structure of the language were apparently not mastered.
Without this understanding students cannot use the powerful Logo
constructs which engage and presumably encourage the development of
high level thinking skills.
Nonetheless, we did find moderate relationships between the ability to

reason in ways we hypothesized were critical for advanced programming,
and performance on our measures of programming proficiency. The
magnitude of the correlations indicated that the students who developed
most in programming were also those who tended to perform better on
tests of logical reasoning. However, our observations of the students
during the course of their instruction and their performance on the Logo
proficiency measures suggests that the actual writing of programs for
3 '6
- 30 -
many students does not require that they ase formal or systematic
approaches in their work. Programming can .nvoke high level thinking
skills, but clearly such skills are not necessary for students to get by in
the early stages o: writing programs to generate desired screen effects.
Conclusions
The field of computer education is in a period of transition. New
languages and more powerful implementations of old ones are rapidly being
developed, and more suitable programming environments engineered for
both the new and established languages.
We can best assess cognitive demands of programming when we are clear

about our goals for teaching programming, and how much we expect
students to learn. However, to understand the cognitive demands for
achieving a particular level of expertise, we must consider the
characteristics of a specific language (such as its recursive control
structure), the quality of its implementation, the sophistication of the
surrounding programming environment (the tools, utilities and editors
available), and the characteristics of the instructional environment in
which it is being presented and learned.
Our results indicate that certain reasoning abilities are linked to higher
levels of achievement in learning to program, but that most students often
opt for a programming style which negates the need for engaging in high
level thinking or planful, systematic programming. Thus, the demands
issue remains clouded by inherent characteristics of interactive
programming languages, which promote the e of a trial and error
approach to program prL auction, and the particular characteristics of the
instructional environment in which learning ,ccurs.
In conclusion, we have argued that asking what are the cognitive

demands of programming is far from a simple or easily answered question.
On the one hand, programming ability of one form or another is
undoubtedly obtainable regardless of levels of particular cognitive skills.
However, if by "learning to program" we mean developing a level of
proficiency which enables programming to serve as a tool for reflecting on
the thinking and problem solving process, then the demands are most
certainly complex and will interact with particular programming activities
and instructional approaches.
Programming has the potential to serve as a fertile domain in which to

foster the growth and development of a wide range of high level thinking
skills. Studies are needed, however, on two fronts, if this potential is
to be realized.
One the one hand, much more work is needed to discover what kinds of
instructional environments and direction are best suited for achieving
each of the many goals educators may have for teaching programing to
children of different ages. We are only beginning to understand how to
teach programming. In fact it still comes as a surprise to many parents
and educators who read Mindstorms (Papert, 1980) too literally that
programming has to be taught at all. But the unguided, free exploration
approach, while possibly effective for some purposes, does not lead many
students to a deeper understanding of the structure and operation of a
programming language, and thus does not lead them to use or develop
high level thinking skills such as problem decomposition, planning, or
systematic elimination of errors.
New instructional approaches may dramatically facilitate the learning of

programming and attainment of the proficiency levels at which more
abstract and intensive thinking is required. Such approaches must be
investigated. Also, in what ways such instruction will lessen or heighten
the demands for learning to program to the levels specified by the new
curricula will be interesting to monitor.
On the other hand, our ability to design more effective instruction will
depend in part on further experimental work to tease apart the role
various cognitive abilities play in influencing students' ability to master
particular programming commands, constructs and styles. Our knowledge
of the cognitive demands of operating with a language should help focus
our instruction and to identify those aspects of programming which will be
difficult for students of different age and ability levels. While the
relation found here between conditional and procedural reasoning ability
and programming suggests some important skills, our conjecture is that at
a more fundamental level, these tasks correlated with programming
proficiency because they required the ability to reason in terms of formal,
systematic, rule governed systems, and to operate within the limitations
imposed by them. This, we feel, may be the major factor in determining
whether students will obtain expert levels of proficiency. What remains
to be determined is whether programming at proficiency level below that
of the expert require and/or help develop high level cognitive skills and
abilities.
REFERENCES
Confrey, J. (1984, April). An examination of the conceptions of
mathematics of youngs women in high school. 'Paper presentedJ7117
annual meeting of the American Educational Research Association,
New Orleans, LA.
Confrey, J.; Rommney, P.; & Mundy, J. (1084, April). Mathematics
anxiety: A person-context-adaptation Model. Paper presented at ta.
annual meeting of the American tducaignal Research Association,
New Orleans, LA.
Hawkins, J. (1983). Learnin Loo to ether: The social context. (Tech.
Rep. No. 13). ew or : Ban treet o ege of Education, Center
for Children and Technology.
Kurland, D.M. & Cahir, N. (1983). The development of computer
ro rammin ex ertise: An interview stud of ex ert adult
programmers. i npu s e manuscript, ank Street lege o
education, Center for Children and Technology.
Kurland, D.M., Mawby, R., & Cahir, N. (1984, April). The development
of programming. expertise. Paper presented at the annual meeting of
the American Educational Research Association, New Orleans, LA.
Kurland, D.M. , & Pea, R.D. (in press). Children's mental models of
recursive Logo programs. Journal of 'Educational Computing
Research.
Luehrmann, A. (1981) Computer literacy: What should it be? Mathematics
Teacher, 74.
Mawby, R. (1984, April). Determining student's understanding of
programming concepts. Paper presented at the annual meeting of
Mawby, R., Clement, C., Pea, R.D. , & Hawkins, 3. (1984). Structured
interviews on children's conceptions of computers. (Tech. Rep.
\7=9). New fork: Bank Street College of Education, Center for
Children and Technology.
Pea, R. D. & Kurland D.M. (1983). On the cognitive prerequisites of
learning computer _programming. (Tech. Rep. No. 18). New York:
Bank Street College of Education, Center for Children and
Technology.
Pea, R. D & Kurland D. M. (1984) . Logo programming and the develop-
ment of plannin& skills. (Tech. Rep. No. 16) . New York: Bank
Street College of -Education, Center for Children and Technology.
Papert, S. (1980). Mindstorms. New York: Basic Books.
Rampy, L.M. (1984, April). The problem solving st le of fifth graders,
using Lo o. Paper presented at the annual meet ng o the ' merman
ucat on Research Association, New Orleans, LA.
Rogoff, B. & Wertsch, J. V. (Eds. ). (1984). Children's learning in the
"zone of proximal development". New Directions for Child
Development, Number 23. San Francisco, Ca: Jossey-Bass.
Snyder, T. (1984, June). Tom Snyder: Interview. inCider, pp. 42-48.
Werner, H. (1937). Process and achievement. Harvard Educational
Review, 7, 353-368.
THE DEVELOPMENT OF PROGRAMMING EXPERTISE IN ADULTS AND CHILDREN
D. Midian Kurland, Ronald Mawby, & Nancy Cahir
The teaching of high level, general purpose languages such as BASIC,

Pascal, or Logo is the main focus of most precollege computer literacy
courses. Although there has emerged, largely in the absence of
reasonable alternatives, broad consensus that programming is a worthwhile
activity for students, there is much less agreement on why this is so. Is
programming to be taught as a skill worth knowing in its own right, or
should it be used primarily as a vehicle for teaching problem solving and
thinking skills? The problems associated with teaching programming are
most acute at the elementary levels where educators are least equipped
(or convinced of the need) to teach programming simply as a skill like
reading or writing, yet at the same time are unsure of what using
programming as a vehicle for teaching thinking skills actually entails.
The situation is complicated by the misconception being perpetrated

through the mass media, and by educational visionaries of many stripes,
that learning to program is easy for children. Consequently, schools are
adding a -programming requirement to their curriculum without a clear
sense of what to expect other than that students of all ages will quickly
master the pragmatics of programming and be able to write interesting
and useful programs. In addition, particularly at the earlier grade
levels, it is hoped that "doing" programming will enhance students' high
level thinking skills, such as their ability to plan and solve problems by
applying generalizable heuristics.
The work reported here was supported by the National Institute of

Education (Contract No. 400-83-0016). The opinions expressed do not
necessarily reflect the position or policy of the National Institute of
Education and no official endorsement should be inferred. We would like
to thank Roy Pea, Karen Sheingold and Catherine Clement for their
comments on earlier drafts of this paper.
There is a growing awareness that things are not so simple. Learning to
program a computer well is a complex process. In a recent paper, Pea &
Kurland (in press) proposed that desired cognitive outcomes, such as
improved problem solving ability or enhanced planning skills, would be
linked to the level of programming proficiency obtained by students.
This argument has been taken further by Mawby (1984) and Clement,
Kurland, Mawby and Pea (1984). The basic hypothesis is that it is not
enough to simply have children "do programming" since it is how one is
programming, not that one is programming, that makes the difference in
what is being learned. If programming is to be an arena in which high
level thinking skills develop, then students must go beyond the typical
novice level in which their programs are the simple one-to-one mapping of
commands to their corresponding invariant screen effects? If we expect
students to develop high level thinking skills from doing programming,
then they must be writing programs in which high level thinking skills,
such as problem decomposition and planning, will be required. It is
important that students reach certain minimum levels of competency and
understanding in their programming work before we look to programming
as a domain that can foster the development of generalizable thinking
skills. Thus, before addressing the question of how programming affects
a student's thinking abilities, it is important to ask how programming
ability itself develops.
What does it take for students to become reasonably competent

programmers? Recent studies of novice programmers (c.f., Dalbey &
Linn, 1984; Kurland, Clement, Mawby & Pea, 1984; Soloway,, Ehrlich,
Bonar, & Greenspan, 1982) have shown that for many students learning
to program is a difficult undertaking, and many students fail to achieve
even a modest understanding after one or two programming courses.
However, studies of novices do not directly address the question of what
is required to reach a reasonably high level of programming expertise.
The problem with only studying novices in typical classroom programming
courses is that so much of the variance in learning is carried by the
instructional and social environments that surround the activity of
programming (Krisler et al, this volume; Watt, 1984). We may observe
students failing to develop an adequate understanding of programming
fundamentals, not because programming itself is difficult, but because the
way it is presented or contexted is confusing, inappropriate, or
misleading. Therefore, to address more directly the question of what is
required to become a proficient programmer, we elected to take a closer
look at people who were already expert programmers.
Our decision to study expert programmers was prompted by the need to

place limits on our expectations of what levels of proficiency one should
expect novice programmers to reach in their first programming course.
Programming is such a new field, particularly where children are
involved, that there are no established norms for what one can reasonably
expect children of different ages and with different amounts of experience
to accomplish. Thus by probing the developmental histories of expert
programmers, we felt we could gain some insight into what to expect,
and, equally impostantly, and what not to expect of students in
programming classes at the precollege level. In addition, knowledge of
what expert programmers had found effective in helping to learn to
program could be useful to schools and curriculum developers. Such
knowledge could guide decisions about what resources to provide for
students in the programming classroom, and how much time to allocate to
teaching and practice if the intention of the school program is for
students to become proficient programmers.
The study was conducted in two parts. In the first we conducted a

series of in depth interviews with a group of adult programmers. In the
second, we interviewed and tested a group of programming "whiz kids"
under the age of 15.
Expert Adult Programmers

Participants. Twenty adults participated in the first part of the study.
Seven were graduate students (five men and two women, between the
ages of 20 and 30), five were professional software game designers from
an independent software company (four men and one woman between the
ages of 20 and 30), and eight were commercial programmers who served
as systems analysts for a large bank (seven men and one woman between
the ages of 20 and 40). The seven students were doctoral candidates in
computer science, and had completed at least one year of their program.
Three students had obtained an M.A. in computer science. The academic
backgrounds of the game designers ranged from a high school diploma to
a masters degree in Russian studies. The game designers averaged just
over eight computer science courses; the range was from zero to twenty.
The commercial programmers were less varied L. their academic
backgrounds; most had a college degree in engineering, computer science,
or business, though --like the game programrzersone had taken no
computer science courses.
All the participants began programming in their late teens or early

twenties, except for one programmer who began at age 27 and two who
began in high school. Seventeen of 20 reported that their first exposure
to computers had been through programming itself. Of those remaining,
two first used computers in conjunction with college laboratory work, and
one began as a game player. Twelve of the 20 programmers learned
FORTRAN as their first language,four started with BASIC, two with APL,
and one each with Pascal and Assembler.
Procedure. All 20 participants were asked to complete a 65-item
questionnaire. The questions requested information on their work and
educational backgrounds, as well as childhood hobbies and interests,
early programming history, programming related work experience,
programming style, and their thoughts about a range of
programming-related topics (e.g., what abilities they believed were
important for a programmer to possess, how programming had influenced
their social life, how programming fit into their long range career plans).
The questionnaire included multiple choice, fill-in-the-blank and
open-ended questions.
Upon receipt of the completed questionnaires, follow-up interviews were

conducted with the seven graduate students and five game designers.
These taped, one-hour structured interviews further explored the process
by which the programmers actually wrote a program. Interview questions
focused on such topics as the stages they went through when writing a
program, the tools they used to help themselves program, where they got
4
their ideas, what they did when they get stuck, and what role other
programmers and/or supervisors played in their work.
Results
A detailed report of the results from these questionnaires and interviews
is available elsewhere (Kurland & Cahir, 1984). Here we would like to
comment on three of the more striking themes which emerged from the
interviews and, where appropriate, relate these observations of expert
programmers to previous findings about novices.
First, there was little consensus among the programmers over what
specific characteristics or abilities were important for learning to
program. Most indicated that being logical, systematic, and curious about
how a formal system operated were the most important traits required to
excel at programming.However, the programmers also mentioned a wide
variety of other traits or abilities that they believed contributed to
becoming a successful programmer. This list included being creative,
flexible, smart, personable, dedicated, planful, disciplined, organized and
patient. Several emphasized that these characteristics were entering
requirements, not outcomes to be expected from learning to program.
Most indicated that they were logical and disciplined thinkers before they
ever began programming and simply found programming compatible with
their style of thinking. This view of the relationship of programming to
thinking contrasts sharply with one of the major tenets of educational
computing, namely that learning to program (like latin and geometry in
past generations) is good mental exercise for developing logical thinking
processes. The fact that these programmers claimed to have been logical
thinkers prior to learning to program does not preclude the possibility
that learning to program may for others help promote this style of
thinking. However, it does suggest that educators must look more closely
at the cognitive styles of novice programmers, and tailor instruction to
take advantage of, or compensate for, students' preferred thinking
styles.
A related finding concerned the relationship of prior technical ability in

science or mathematics to programming. While the majority of
programmers reported having enjoyed mathematics in school, most felt that
being good at mathematics per se was not important for becoming a good
programmer. As one put it:
...I think the way you look at problems mathematically requires the
same skills [as programming], i.e., reducing the English language to
some sort of algorithmic language or your mathematical notation.
A similar finding has been noted by Fisher (cited in Johns, 1894) who,
on the basis of a survey of programming manages, reported that the 10
most important attributes for success as a programming trainee were:
°willingness to accept responsibility
°thoroughness
°persistence against obstacles
°ability to be a self starter
°ability to communicate
°resourcefulness
°responsibility to fulfill promises and commitments
°enjoyment of the work
°self-confidence and assuredness
°high standards
She concluded that beyond certain minimum levels of intelligence and
background knowledge, personal characteristics, attitude, and
communication skills count more heavily than does strictly technical
knowledge or aptitude in specific content areas.
A second common theme across the interviews was ro.spondents

consistently noted that learning to program well required substantial time
and energy, even for those who greatly enjoyed programming. All those
interviewed claimed to have put enormous amounts of time into their
training. Estimates of the average time spent at a terminal while they
were learning fell in the range of 20 to 35 hours per week. However,
almost all of the programmers also reported that at some point when they
were learning they went through a prolonged period of total immersion.
Non-stop sessions of over 30 hours as well as 60-100 hour weeks at the
computer were typical. For example, one software game designer
described learning to program at college in these terms:
"One semester in school I think I averaged about four hours of
sleep a night for three months. My health was damaged, my
brain was damaged."
A computer science student talked about learning to program in similar
fashion: -$.
"Actually Iwork an incredible number of hours, as most

computer scientists do and I work very weird hours. The thing
about computer science is that it is a very consuming
subject...I regularly spend ninety hours a week at the
computer. It would be very convenient for me to to have a
computer at home, but I'm not sure I'd do anything
else... Unfortunately computer science can be so consuming that
you have to make a very definite effort to have other
interests."
A software game designer noted that the time required to learn to
program seemed to be in excess of what was needed to gain mastery in
other intellectual domains:
"I was always struck at how much time I put into school
compared to everybody else I knew... (those) who were taking
ordinary computer science curriculum compared to English, it
was at least double the amount of time. People were just
putting in amazing amounts of time."
The amount of time expert computer programmers dedicated to learning

contrasts sharply with the way computer time is allocated in schools. As
recent surveys of classroom computer use indicate (Becker, 1983), more
and more schools are including programming in their curriculum.
However, studies of computer use in elementary schools suggest that, at
most, children spend 50 hours a year with the computer, and that this
time is broken into many separate episodes of short duration ( Pea &
Kurland, 1983) . Even in high school programming courses where
students may attend five 40 minute periods per week, the total number of
hours students are likely to spend programming per year is still only as
much as some programmers claim to have spent in a single week when
they were learning . This extreme time differential raises serious
questions about what the got is of programming instruction at the
precollege level should be, given the amount of time schools are typically
willing or able to devote to any single subject.
In addition to spending thousands of hours learning to program, the
programmers also strongly favored uninterrupted marathon sessions when
they did their work. All the programmers reported spending a great deal
more time per session working on a programming problem than students
are able to in schools. While all professionals devote more time to their
discipline than do students working in the same field, the emphasis
expert programmers placed on having long, uninterrupted blocks of time
in which to work was nonetheless striking. They attributed this need to
the fact that to program effectively requires keeping track of the meaning
of current values of many variables and procedures, while also monitoring
the way in which control passes conditionally between lines of code and
program modules. Thus, the programmers felt that once they started on
a problem it paid to keep working for as long as possible so that they
would not lose track of where they were or what they were doing.
In a related study of the work habits of "super" programmers, Molzberger

(1983) also reported being struck by the way expert programmers
worked, and the importance for them of large blocks of uninterrupted
time. For example, one of the programmers Molzberger interviewed stated
that:
51
"Interruptions can be unfortunate. It can take hours after a
telephone call! When I am interrupted at an unfavorable time
where many threads run together and I am not completely
finished yet, I have to start all over."
The third striking characteristic of these expert programmers was the

steps they went through in developing a program, particularly the amount
of planning they reported engaging in prior to the actual coding of a
program. They reported that they put a great deal of thought into how
to design a program before they began the coding phase. They tended
to first map out in detail a structure for the program to serve as a
coding guide. None reported using flow charts to do this. They each
had developed their own system based on personal preferences, given the
particular type and size of program they were writing. For example, one
programmer explained how he prepared to code in this way:
"First I'd do a literature review, see how they [other
programmers] did it. Then I'd sit down to write the algorithm
in English language terms, in very broad structured
programming. Then I'd break it down until I got it to a
language I'm familiar with. Then I'd start coding."
Another programmer said that she tended to plan her assembly language
programs and work out the algorithms she would need in the high level
language PL/I, then use her PL/I program as a g, for the actual
assembly language coding.
We have observed that many novices do little planning of this type (Pea &
Kurland, 1984), and have little idea about what kinds of problems they
and spend almost all of their programming time writing code, In
contrast, these expert programmers viewed coding as much less central to
the programming process. They reported that coding, the actual writing
of programs, took only 20-25 percent of the total time they spent
programming. The rest of the time was spent writing specifications for
programs, planning and designing procedures. and algorithms,
systematically debugging and testing code they had generated, and in
some cases documenting their code to help other programmers understand
it. Just as knowledge of grammar and spelling is not the focus of good
writers, knowledge of the rules for a computer language clearly is t the
focus of good programmers.
While planning and debugging skills were emphasized by the programmers

we interviewed, they were nonetheless all highly proficient coders. Each
knew at least three different languages, with one claiming to know 16.
Many commented that knowing many languages was helpful, and that they
used their knowledge of the strengths and weaknesses of particular
languages in decisions about how to design their programs. Several
programmers reported that they often would design a program in their
favorite language, and then translate it into the language required by the
particular application on which they were working, This technique was
particularly popular with game designers who were forced to produce
programs for small microcomputers which could not support the rich
programming tools and powerful languages that the programmers prefered.
Thus, these expert programmers appeared quite different from novices we
had worked with in previous studies. First they had spent a tremendous
amount of time learning to program. In addition, they now allocated their
programming time in a manner that was very different from that of most
novices. Rather than focusing on code generation, expert programmers
spend much more time planning their programs, designing and testing
algorithms, and testing their routines to be sure they worked properly
under all conditions.
While expert programmers approach programming as a planful,
systematic undertaking, (Pea I. Kurland, 1984). Whether this is due to a
difference in learning style, or is a function of experience is impossible
to tell from these interviews. It seems likely, though, that if the goal of
school-based programming courses is to encourage students to develop a
more planful and systematic problem solving style, then the current heavy
emphasis on coding should be rethought. At some point it becomes
important for students to be able to code in some fashion in order to
program effectively. However, what was clear from the interviews was
that coding skill alone is not sufficient for becoming a programmer.
A clear implication that can be drawn from these interviews is that to

learn to program well takes lots of focused time. Schools could elect to
devote substantially more time than they presently do to the teaching of
programming, but more time could only be added at the expense of other
school activities, and this seems neither likely nor particularly desirable.
Rather, these interviews suggest that acquiring sufficient skill in current
general purpose programming languages takes more time and effort than
schools, particularly at the elementary level, can justify. If this is true,
the question film becomes, how can we design an effective curriculum
that focuses on the planning and problem solving aspects of program
design and development, without requiring that students also spend
excessive amounts of time in order to master coding.
Expert Child Programmers

The programmers in the first part of the study were adults, most of
whom had learned to program in their late teens or early twenties. How
much can we generalize from their experiences learning to program to the
experience of much younger students? To find out, we next interviewed
a group of six young programmers, each of whom had begun programming
seriously between the ages of six and twelve.
Participants
To find students for this phase of the study, we contacted programming
teachers, computer user groups, and local programming experts in the
New York City area to identify children under the age of 14 who had
clearly demonstrated expert programming ability.
Finding such children was not easy. Children who knew some
programming or who spent thousands of hours playing sophisticated
computer games were quickly identified, but children who were deeply
involved with programming were much harder to find. We defined a
programming expert as any child who had (1) written a commercially
published program: or (2) had produced programs or utilities that others
(e.g., their school or friends) were using; or (3) who had taught
programming courses; or (4) in some equivalent way was clearly capable
of producing software usable by others. We attempted to screen out
children, whose programming consisted solely of short programming
exercises produced for their own amusement or interest.
We ultimately identified four children who appeared to meet our strict

criteria for inclusion in the study, plus five others who appeared to be
good programmers, though perhaps not quite meeting our criteria of
expertise, In this initial sample were eight boys and one girl between
the ages of 9 and 14. Each visited our Center for a day during which
they were interviewed and given a battery of cognitive aptitude and
programming proficiency tests. They also were asked to bring an
example of their best work for us to examine.
We analyzed the examples of their work and asked them to write several
programs for us in order to verify that they were knowledgeable
programmers. During this phase of the study three of the original nine
participants were dropped from the study because they appeared to lack
sufficient expertise. For example, one child showed us a program he had
written that looked impressive when run, but examination of the code
showed that it had been written in an inelegant, brute force manner, with
the exception of one routine which, it turned out, the child's tutor ha t
provided. We also dropped a child who had experience doing a wide
- 12 - 5 3
variety of activities with computers, such as using Visicalc to keep his
comic book collection organized. However, on closer scrutiny it was
apparent that his father did most of the organizing and conceptual work
while he served as helper. The third child we dropped was the one
female programmer in the group. She used the computer primarily as an
artistic medium for producing pictures. Her programming was restricted
to fairly elementary routines in the Pilot language to generate graphic
images. Thus, though these three children had some knowledge of
programming and were interesting in other ways, they were clearly not
expert programmers of the type required for the purposes of the present
study, and so were not included in our final sample.
Results
The final group of clearly knowledgeable programmers consisted of six
very bright boys between the ages of 9 and 14 (see Table 1 below) from
middle to upper income homes. Though these six programmers were much
more knowledgeable about programming than children in any of the school
samples we had previously studied, they still displayed a very wide range
of programming ability and understanding. However, all appeared to
have a reasonably deep understanding of computer languages that went
beyond simply knowledge about individual commands.
As may be seen in Table 1, all started with BASIC, primarily because

that is what came with their home computer. Several also knew
Assembler, Pascal and Pilot. In addition, all six had learned some Logo,
either in school or at a computer camp. However, none claimed to like
Logo for doing their personal projects. The main reason they gave for
preferring to work in BASiC or Assembler was the greatly increased
speed they felt these languages offered for the fast action, graphics
intensive video games they wanted to program. The boys° personal
judgement of the relative merits of one language versus another tended to
be at the level of what it could do, not how elegantly it did it.
Table 1
Characteristics of Young Programming Experts
Age when Total hours Typical First Number of

Child Age started programming project language languages
(estimated)
1 14 11 1100 utilities BASIC 8

2 13 8 3850 utilities BASIC 3
3 14 12 1000 games BASIC 3
4 12 10 550 games BASIC 4
Note. 'Number of languages' refers to the number of computer languages

children reported knowing at least well enough for them to write a simple
program.
To get a clearer idea of just how thoroughly the boys understood

programming, we had them do several short t...As. One of the tasks was
to describe in detail what a particular program we had prepared would do
when executed. This short program was written in the most unstructured
manner possible with embedded calls back and forth throughout. Novices
whom we have studied usually have terrible difficulty with this kind of
task since knowledge of individual commands is not sufficient for
determining what the program as a whole will do. Novices often read the
program from top to bottom without regard to flow of control and thus
have no idea what the program will do. They also frequently misinterpret
conditional statements which prevent them from accurately predicting what
a program will do when run. In contrast, the boys in this group read
the program following its control structure and thus were able to
understand and accurately predict what the program would do. However,
they varied in how quickly they could comprehend the program. The
better programmers comprehendeci the program almost immediately, while
the others took quite a lot of time. All six also volunteered suggestions
about how the program could be improved (e.g., structure the code,
avoid unnecessary GOTO's, insert error traps).
- 14 -
On a second task, the boys were asked to write a program to determine
how many ways there are to make change for a dollar using nickels,
dimes and quarters. This problem requires the ability to develop an
algorithm (i.e., for testing all possible ways to figure change) plus the
ability to use conditionals and stop rules. Five of the six boys handled
the problem well. Three did it with no help, though again there were
large differences in speed. Two needed one or two small hints on how to
formulate the rule for testing whether a particular combination of coins
satisfied the goal (i.e. , 5*nickels + 10*dimes + 25*quarters = 100). The
last child developed an interesting approach to the problem based on
randomly selecting collections of coins to test, but his program failed to
meet the stated goal of printing out all possible combinations. While his
p...ogram would eventually, through random search, find all the
combinations, he had no provision for testing whether a combination was a
new one or one that had been previously found.
It was interesting to note that all three boys who succeeded without help
tried to go beyond the simple requirements of the task and attempted to
find an optimal solution. One found a tight algorithm that would get the
answer by making the fewest number of comparisons possible. Another
said that he would improve his BASIC program by writing it in assembler
so that h would run faster. The third was distressed at the "slowness"
of his algorithm but could not discover a faster one. This search for the
elegant solution was something we have rarely encountered in novices,
but was commented upon frequently in our interviews with the expert
adult programmers.
Having demonstrated to our satisfaction that we had a group of unusually

talented young programmers (though clearly not as proficient as the
adults interviewed in the first part of the study), we then proceeded to
question them about their work habits and how they learned to program.
The interview results in many ways paralleled those with the expert adult
programmers. First, it was clear that they devoted substantial amounts
of time to learning to program. The boys averaged roughly 500±200
hours of programming a year. For the most experienced boy past five
years he had been programming. Again, this contrasts sharply with the
35 to 50 hours per year that most school programming courses can offer.
While the boys all did some programming in school (and one helped teach
a course in BASIC), they all claimed that they learned nothing about
programming in school since no one in school knew as much as they did.
Their schools did not have the equipment, books, manuals or
knowledgeable people required to help them with their programming
problems. Thus they relied on the programmer's underground and/or
expert programmers (sometimes peers) outside of school for guidance and
assistance.
Not surprisingly, all the boys owned one or more computers where they
did the majority of their programming. In each case, there was a
significant older person--parent, high school tutor, adult friend--who had
provided substantial early encouragement and in some cases, instruction
on a one-to-one basis. The three best programmers in the group (who
were also the oldest) had spent the most time programming, and had the
most access to knowledgeable experts. One boy, for example, spent
hours every day in the computer center of the university where his
mother worked. While there he programmed on the university mainframe
with the help of the people at the center. Another was tutored by an
adult friend with a background in electronics. Together they built
electronic devices and then worked together on programming projects.
Thus, in contrast to the situation in many classrooms, these boys began
in a rich programming environment with ample support materials and, most
'importantly, knowledgeable experts to help guide them through their early
learning.
The appeal of programming for these boys appeared to stem from several
sources. One was the sense of power it provided. For example, one boy
stated that he liked programming because: "Well, the feeling of power,
definitely. Going into the software underground. So that if I write
something good it will probably be copied and recopied for lots of people
in America." Others commented that they liked the challenge and the
feeling experienced when they succeeded in making a complex program
59
- 16
work. However, the domir Int motivator appeared to be social.
Programming played an important role in the social lives of all the boys.
All six reported that many or most of their friends were involved with
computer activities of one sort or another. Programming provided a
common bond with their peers and with adults with whom they shared
mutual interests. In addition, two of the boys had become involved with
local computer bulletin board systems and communicating with other
programmers through the their modems. One had programmed his own
bulletin board system system for a VIC 20 computer and ran it several
hours per day out of his bedroom. He claimed to have met most of his
friends, including a current girl friend, over the modem. Thus, the
boys not only found programming cognitive gratifying, it also gained them
entry into a social network whose purpose went beyond simply helping
each other with technical problems.
Conclusion
What conclusions can be drawn from these interviews with expert
programmers? Three factors of relevance to schools have already been
considered. First, learning to program well, like becoming expert in any
other domain, takes a tremendous amount of time and dedication.
Hundreds of hours per year were spent by both the adult and young
programmers. For the young programmers, even this much time had not
made them all fluent programmers. Several struggled over our assigned
programming tasks and took a lot of time developing solutions. It seems
clear that unless radically better methods for teaching programming are
discovered, not everyone can or should become proficient in the general
purpose programming languages available today.
However, just because becoming a highly proficient programmer requires a

tremendous expenditure of time and effort does not mean that instruction
in programming fundamentals should only be retained for an interested
elite. On the basis of these interviews, a programming curriculum whose
aim is to turn students into proficient programmers in the classroom time
allotted would seem to be unrealistic. Yet we believe that there are ways
to teach fundamental programming and computer science concepts in the
normal school classroom that could serve as a solid foundation for
students' future interactions with computers, whether they continue to
use computers for programming, word processing, or other information
management purposes. The challenge facing educators today is to find a
role for computer programming within the standard curriculum so that
over the course of a student's school career, he or she will gain adequate
exposure to the central computational concepts embodied in programming
languages and increasingly called upon by the new generation of powerful
application programs. Without such appreciation and understanding , it is
difficult to see how student will become flexible users of these powerful
new tools for managing information and learning .
Second, there are clearly significant, qualitative differences in the

programming environments created for students in classrooms and those in
which expert programmers learned and do their work. If students are to
understand more about programming than simply what individual commands
do, then we must find ways to get more expertise into the classroom. In
addition, there need to be more support materials that can help students
gain a larger sense of what programming is, how languages and computers
work, and how particular classes of problems can be solved in alternate
or more elegant ways.
Third, although all the adults and children were good at mathematics,
they seemed to view their abilities in mathematics and programming as
stemming from a more fundamental interest in logical, formal systems.
Similarly, we were intrigued by the observation that many of the
programmers in both age groups were accomplished musicians. We
speculate that there is a particular cognitive style that makes
understanding and controlling an elegant rule governed system such as
music, mathematics, chess, physics or programming highly appealing for
some people. While this is just speculation, it would be interesting to
pursue this point more systematically in future research with novice and
expert programmers.
We sensed from these programmers that programming did not teach them
to think logically or approach problems in a more systematic manner.
This 's the way they approached problems prior to learning to program,
and thus was one of the reasons why they found programming so
appealing. It remains an open question whether children who do not have
this cognitive style to begin with would develop a more formal and
systematic approach to problem solving as the result of learning to
program. However, it seems that many novice programmers apply their
preferred mode of problem solving to programming and thus do not
necessarily or automatically develop new ways of thinking by virtue of
their programming experiences.
It was clear that for these programmers, knowledge of programming

entailed knowing how a programming language works, not just knowing
the individual commands. Several adult programmers commented that
learning a new language became a trivial undertaking for them since there
are only a few distinct classes of languages.Once they understood how
several worked, it was easy to map this understanding onto new
languages. In contrast to the experts' grasp of the principles underlying
a language, novices frequently know only what individual commands mean,
and thus often cannot tell what a program does (Kurland, Clement,
Mawby & Pea, 1984). The difference between novices and these experts
lies in the expert's understanding of the relationship between commands
and the rules of the language which determine the control structure of
the program. Programming instruction from the very beginning might
benefit from greater stress on flow-of-control and other structural issues,
with less focus on individual commands and generating programs without
any regard for style or elegance.
Finally, thorny policy issues surrounding the place of programming in the

standard precollege curriculum are currently besetting our education
system. While by no means definitive, this study coupled with previous
work with novice programmers in classroom settings suggests that current
orientations towards teaching programming at the precollege level may be
seriously misguided or unrealistic. It is undeniable that there are
fundamental programming concepts which students must understand in
order to participate in the emerging information age (Sheingold, Hawkins
& Kurland, 1984). Students need to grasp these fundamental concepts in
order to effectively interact with computers, whether they choose to
id2
program, do word processing, manipulate a data base, construct elaborate
graphics, create musical compositions, or do any other of the myriad
activities for which computers are appropriate. The question thus comes
down to, what do expert programmers know that novices could know
without having to themselves become expert programmers. The challenge
that faces computer educators today is to identify these concepts and find
ways to teach them, whether it is through programming or other
computer-related activities.
63 - 20
REFERENCES
Becker, H.J. (1982) Microcomputers in the classroom: Dreams and

realities. Report No. 310. Center for Social Organization of Schools,
The John Hopkins university, Baltimore, MD.
Clement, C., Kurland, D.M., Mawby, R. & Pea, R.D. (1984, August).
Analogical reasoning and computer programming. Paper presented at
the Conference on Thinking, Cambridge, MA.
Dalbey, J., & Linn, M.C. (1984, April). Making pre-College instruction
in programming cognitively demanding: issues and interventions.
"Paper presented at the American Educational Research Association
Annual Meeting, New Orleans, LA.
Johns, R.P. (May 28, 1984) Don't judge a programmer by expertise alone.
Computerworld, 18.
Kurland, D.M. & Cahir, N. (1984). The development of computer
programming expertise: An interview study of expert adult
Unpublished manuscript, Bank Street College of
Education,
Education, Center for Children and Technology.
Kurland, D.M., Clement, C., Mawby, R. & Pea, R.D. (August, 1984).
Mapping the cognitive demands of learning to program. Paper
presented at the Conference on Thinking, Cambridge, MA.
Kurland, D.M., Mawby, R., & Cahir, N. (1984, April). The development
of rammin Paper presented at the annual meeting of
Research Association, New Orleans, LA.
Mawby, R. (1984, April). Determininstudent's understanding of
programming concepts. Paper the anntiriIneetink
Molzbeeger, P. (1983)Aesthetics and programming. (in A. Janda (Ed),
Chi'83 Conference Proceedings. Boston, MA.: ACM. (pp. 247 -249).
Pea, R.D. & Kurland D.M. (1983). On the cognitive prerequsites of
learning computer programming. (Tech. Step. No. 18). New York:
Bank Street College of Education, Center for Children and
Technology.
Pea, R.D. & Kurland D.M. (in press) On the cognitive effects of
learning computer programming: A critical look. New Ideas in
Psychology.
Sheingold, K., Hawkins, J. & Kurland, D.M. Classroom software for the
information age. (Tech. No. 23). New York: Bank Street
Rep.
College of Education, Center for Children and Technology.
Soloway, E., Erlich, K., Sonar, ,T., & Greenspan, J. (1982) What do
novices know about programming? In B. Sneiderman & A. Badre
(Eds.), Directions in human-computer interactions. Hillsdale, NJ.:
Aplex.
Watt, D. (1984). Creating Logo cultures. Pre-Procedings of the 1984
National Logo Conference, 25-29.
60
- 22
ISSUES AND PROBLEMS IN STUDYING
TRANSFER EFFECTS OF PROGRAMMING
Kate Ehrlich
Honeywell Information Systems
Waltham, Mass.
Valerie Abbott
Yale University
Dept. of Psychology
New Haven, Ct.
William Salter
Bolt, Beranek, Newman, Inc.
Cambridge, Mass.
Elliot So loway
Yale University
Dept. of Computer.Science
Address correspondence to:
New Haven, Ct.
Kate Ehrlich
Honeywell Information Systems
200 Smith St.
Waltham, Ma. 02154
This work was supported by the National Science Foundation, under NSF Grant IST-81.14840.
66
Ehrlich, Abbott, Salter, Soloway Page 1
ABSTRACT
It is commonly believed that programming helps students develop new thinking skills that they
can use to solve problems in a variety of problem-solving domains. We examined whether
programming skills transfer by testing a group of college students before and after they had
taken their first, introductory programming course. We also tested a group of students who were
not enrolled in any programming course. The study focused on procedural skills. These skills are
linked to understanding the steps that are needed to solve a problem as contrasted with simply
memorizing formulae. Many students fail to adopt a procedural, active style of problem-solving.
The study examined whether students can transfer some of the procedural skills they develop in a
programming course to other non-programming problems. The results of the study offered some
support for the transfer of procedural skills. However, the results were not conclusive due to
unexpected problems with the control group of subjects. These results form the basis of a
discussion of issues and problems related to studying transfer effects from programming.
Ehrlich, Abbott, Salter, So loway Page 2
1. INTRODUCTION
Educators are responding to the growing importance of computers and computer literacy by
stressing the need to integrate computers and, especially programming courses, into the school
curriculum. The motivation for the emphasis on computer education is two-fold. On the one
hand it is firmly believed that students who can add programming skills to their other scholastic
achievements have a greater chance than students without these skills of finding employment
when they leave school. On the other hand, the emphasis on giving students training in
programming reflects an implicit belief that programming indirectly improves students' problem-
solving skills. This paper addresses some of the :slues raised by the second of these goals: the
educational benefits of programming. In particular, we will focus on one of the more farureaching
and contentious claims: that programming teaches students thinking skills which can be
transferred to other problem-solving domains
Some of the belief that programming skills can transfer comes from the idea that programming
emphasizes the "bow" of problem-solving. That is, programming is believed to teach students
general methods for solving problems ( [7]). Although programming is a skill that is not easily
acquired e.g., 11, 10, 9, 31, the difficulty of learning to program seems to reinforce the belief that
those students who have successfully grasped the fundamental concepts of programming have
learned some general problem-solving skills as well. Previous research on transfer (e.g., 17, 81)
offers some interesting conjectures and anecdotes to support the importance of teaching
programming to young children to better position them to master more traditional math and
science subjects. However, the empirical support for transfer is weak.
Programming constructs such as assignment statements provide powerful metaphors for the
active process of transforming input values into output values. Moreover programming teaches
students to solve problems by breaking the problem down into small components and it teaches
students how to represent the steps involved in solving a problem. In these ways, programming,
at least in procedural languages such as Pascal and Basic helps students develop their procedural
skills. Furthermore, in earlier empirical studies we found that programmers were better able to
write algebraic equations in the context of a computer program when compared to writing
algebraic equations in a standard algebra context (18, 41). Given this prima facie evidence for
the benefits of programming, we felt it worthwhile to tackle the transfer issue directly.
The study reported in this paper focused on trying to obtain evidence that procedural skills do
transfer from programming to other domains. We examined the issue by testing students on a
set of algebra word problems. The kind of problems we used were based on previous research,
which will be reviewed in the next section. This research identified a set of algebra word
problems that elicited errors that could be traced to the adoption of a descriptive rather than a
procedural approach to solving the problem. Our intent is to examine whether programming, by
Ehrlich, Abbott, Salter, So loway Page 3
enconraging students to be mare procedural in their problem-solving, can improve performance

on thesulgebra word problem's.
2. PREVIOUS RESEARCH
2.1. Algebra Word Problems

In some recent research, Clement and his colleagues 121found that students seemed to have a lot
of difficulty in translating simple algebra word problems from the description of the problem in
English into an algebraic equation. Two typical problems are shown in Table 1. Among college
freshman engineering students, 37% missed the first problem while 73% missed the second. In
their experiments, Clement and his colleagues were able to eliminate difficulty with algebraic
manipulation and tricky wording as major sources for the errors. The errors made on problems 1
and 2 were largely of one kind; most were revereals: 8SeeP instead of See6P and 4Cee5S instead
or 5C 4S. The consistency of these error patterns across these and other problems, and analysis
4
tacking aloud protocols, argues against the idea that they were caused by carelessness, and
,,,o,;ets that they stem from conceptual bugs.
Clement and his colleagues carried out a number of videotapell interviews in order to understand
the source of these errors 12]. Using this technique they were able to identify a number of
strategies which led to the reversal error. What seemed to underlie the strategies, was that
students seemed to be adopting a descriptive approach to the problem. The incorrect equation
and the descriptive approach contrasts with the correct equation S or, which needs to be
viewed as expressing an active operation being performed on one number (the number of
professors) LI order to obtain another number (the number of students).
The results of these studies lend themselves to the following analysis:

Students are making errors can algebra word problems because they are adopting a
descriptive rather than a procedural approach.
Programming provides an environment in which students are encouraged to develop a
procedural view of problem solving 18i. In particular, we found when programmers
were asked to provide a computer program as a solution to "students & professors"
type problems. they were more, often correct than when asked to simply provide an
algebraic equation. Note carefully that the key line in the computer program is
precisely the same algebraic eqetation that would be required for a correct solution to
the algebra word problem!
Then claim lead to the following prndiet:or I; the kind of skills we have identified as being
associated with progratereing do transfer' to ()thee domains we should find that students do better
J
Ehrlich, Abbott, Salter, Soloway
PROBLEM 1
Given the following statement:
*There ate six times as many students as professors at this University".
Write an equation to represent the above statement. Use S for the number of students and P for
the number of professors.
Result: 03% correct

Typical wrong answer: 6S .. P
PROBLEM 2
Giver the following statement:
"At Mindy's restaurant, for every four people who order cheesecake, there are five people who
order strudel."
Write an equation to represent the above ststemcat. Use C for the number of cheesecakes
ordered and S for the number of strudels ordered.
Result: 27% correct

Typical wrong answer: 4C 0., 5S
Table 1:
EXAMPLES OF ALGEBRA WORD PROBLEMS
on algebra word problems after completing a programming course as compared with their
performance before they have taken the course.
2.2. Pilot Study

Overall. In a preliminary study [41 we compared the performance of a group of 31 students who
had just completed an introductory programming course with a similar group of 26 students who
had no programming experience. We Asve both groups of students a test which included a
number of algebra word problems of the kind shown in Table 1. The results were consistent with
our predictions; the students who took the programming course performed better than the
students who had no expsure to programming. The programmers averaged 69% correct whereas
the non-programmers tr-lraged only 54% correct (ANOVA: F114 IM 4.40, p < 0.05).
Problem Decoml asition. We also obtained results which pointed to specific effects related to
programming. In the earlier study on algebra word problems, Clement et al. 12j found that
students performed more accurately on problems in which only one of the variables was multipled
by a number (e.g., S SP) as compared with problems in which both variables were multiplied
by a number (e.g., 5C 4S). Both of these problems were included in the preliminary study.
For the sake of brevity we refer to the first, simpler problems as an integral problem, and the
second problem as a non-integral problem. The data from the two problems are shown in Table
2. These data indicate that although the nonprogrammers are still performing worse on the non-
integral than on the integral problems, the programmers are performing equally well on both.
One explanation for the better performance by the programmers is that they developed skill in
problem decomposition and that they were applying that skill to the non-integral problems.
Why should programming help problem decomposition? One answer is that students are taught
to solve problems by breaking them down into small manageable components. For instance, we
have observed students who solved a non-integral problem by writing the equation as
X=C5
S=X/4
instead of writing the equation, S 5/4 C directly. The skill that is exercised here is both
decomposition and composition. The student must be able to see that a single equation can be
broken down into two parts, and that a correct solution can be composed by using an additional
variable to connect the separate parts. The advantage of decomposing a problem in this way,
whether it is done explicitly or not, is that it is often easier to generate two simple equations than
a single, more complex one.
Equations as active operations. The second result which demonstrated influence of the
programming course occurred on a set of problems which were constructed to elicit improvements
71
NON-PROGRAMMERS PROGRAMMERS
Number of subjects (N = 26) (N = 31)
GENERATE
Correct %Correct %Correct
NON - INTEGRAL Answer
At Mindy's restaurant for every 4 people 4S--5C 39% 67%
who ordered cheesecake, there were
people who ordered strudel.
INTEGRAL
At a Yankees game for every 3 hot dog H=3C 62.5% 68%
sellers there is a Coke seller.
TASK: Write an equation to represent the statement
Table 2: PERCENT CORRECT FOR INTEGRAL AND NON-INTEGRAL

VERSIONS OF EQUATION GENERATION PROBLEMS: PILOT STUDY
in performance. Equations for the word problems given in Table 1 can be written equivalently as
a ratio: C/S 4/5; as a multiple expression: SC 4S; or with a single vitriolic on one side of
the equation: C 4/5 S. In an unpublished study, we found that there was a strong correlation
between the form in which students wrote their equation and the accuracy of that equation.
Equations written as multiples were commonly incorrect (i.e., 4C 5S), while equations written
in either of the other two forms were more commonly written correctly. Indeed, the previous
researcg on algebra word problems noted that many of the incorrect equations were associated
with a simple word-order match strategy in which the order of numbers and variables in the
equation matches the order these were mentioned in the problem description. This strategy, of
course, will generate an incorrect multiple equation, 4C low 5S. We reasoned that if students were
making errors because of this strategy, showing them an equation not written as a multiple might
elicit better performance. To test the effect of the form of the equation on performance, we gave
students partial equations to complete; examples of the problems and the results are shown in
Table 3.
Two important results emerged. Firstly, we found that students were more accurately on
equations that were written in the form of a ratio than on equations written as a multiple. This
result is perhaps not surprising given that the problems we were using are properly classified as
ratio problems. The result is more important, however, for its demonstration that the completion
task can elicit more accurate performance, and, by implication, that the task is sensitive to
changes that might result from a programming course. The effect of a programming course on
algebra word problems can be seen in the next result.
Secondly, we found th.it the programmers performed better than the non-programmers on
problems in which the equation was written with a single variable on one side. This is the form
in which equations and assignment statements arc most often written in a program. Moreover,
assignment statements in a program convey the notion of an active operation in which the output
variable (e.g., S in the assignment statement, S : 5/4 ' C) is assigned the result of the
operation on the input variable (e.g.,the result of 5/4 * C). One interpretation of this result is
that programmers are exposed not only to a particular form of equation, but that they are
exposed to the notion of an equation (as an assignment statement) being associated with an
active operation. This interpretation amounts to the claim that programming helps students
overcome their descriptive approach to equations and instead encourages them to take a more
active, procedural approach.
73
Ehrlich, Abbott, Salter, So loway
NON-PROGRAMMERS PROGRAMMERS
Umber of subjects (11 26) (N = 31)
Equation Correct %Correct %Correct

given hewer
COMPLETE THE EQUATION
A: SINGLE VARIABLE FORM OF EQUATION

When the slot ',chine it Rosin's her gives
jackpot. there are 6 nickels for every
5 quarters. - r41 N - Q 35% 61%
5
8: RATIO FORM OF EQUATION
Im Fairmont Hills there are 8 plumbers for P P
every 3 electricises. - - u - 65% 77%
E 3 E
C: MULTIPLE FORM OF EQUATION
When Elizabeth Taylor goes to Tiffemys
she bays 3 rubies for every 2 eserelds. u TE 2R = 3E 35% 48%
TASK: Complete the given 'cooties by replacing the question Aerks
Table 31 PERCENT CORRECT RESPONSES FOR

THE EQUATION COMPLETION PROBLEMS: PILOT STUDY
AV AILABLE
BEST COPY
2.3. Summary of Prem. lona Research

Previous research suggests that there is a class of problems, algebra word problems, in which
students make errors because they adopt a descriptive rather than a procedural approach to the
problem. Programming, as we argued earlier, encourages students to develop a more active,
procedural approach to problem solving. Students who have taken a programming course should
be able to transfer the skills they have learned in that course to other problems, in just those
cases where a procedural-based approach is needed. Initial support for this claim is encouraging.
In the pilot study described above, we reported some evidence to suggest that two aspects of
problem-solving problem complexity and an active approach to equations, did seem to be
improved by programming experience.
3. CURRENT STUDY
The present study was designed to follow up the preliminary results from the pilot study in a
more experimentally controlled setting. The study focused on the following question:
1. Do students improve their performance on algebra word problems as a result of
taking a programming course?
2. Can we attribute any improvement in performance to the development of particular
skills or strategies?
skills associated with problem decomposition
strategies associated with viewing an equation as an active operation
generality of transfer across problem types
3.1. Materials
The first question was addressed by comparing performance for each student over the whole test.
The test consisted of 36 items; these items are described below.
The second question has three parts: EVoblem Decomposition, Problem Strategy,
Problem Type.
Problem Decomposition. We examined skills associated with problem decomposition by
varying the complexity of the problem. In addition to the integral and non-intepal versions
used in the pilot study, we added a third level of complexity, called combination problems. In
these problems, both an integral and a non-integral equation are required to solve the problem.
Examples of the problems are given in Table 4. If programming teaches problem decomposition
and students can transfer that skill, we should find that students from the programming course
show more improvement on the complex problems compared with the simple problems.
Problem Strategy. We examined whether students will view equations in a more active way
after taking a programming course. To examine this question we used the same completion task
Ehrlich, Abbott, Salter, Seloway
PROBLEM 1: INTEGRAL
Write an equation (or equations) to represent the following statement:
At the bookstore, for every 9 copies of the Times there is a copy of

the Gazette?
PROBLEM 2: NON-INTEGRAL
"M Mindy's restaurant for every 4 people who ordered cheesecake there
were 5 people who ordered strudel?
PROBLEM 3: COMBINATION
"The candy store sells 4 bars of chocolate for every 3 ice-creams it sells
and it also sells 5 ties as many bars of chocolate as candies."
Table 41
EXAMPLES OF RATIO PROBLEMS AT THREE LEVELS OF COMPLEXITY
we used in the pilot study. In this task, students are given a partial equation to complete instead
of generating an entire equation. The test included 12 of these completion problems. As in the
previous study, the three alternate forms of writing an equation were used. These forms are:
ratio, single variable and multiple. Each form occurred equally often (i.e., times) acmes the
total of 12 problems. Examples of the probLms are shown in Table 5. If students approach
problems in a more active manner we should find that this change is picked up by their response
to the tingle variable forms of equations because this form of equation is most similar to an
assignment statement in a program. Moreover, if programming encourages a different approach
to problem solving rather than simply exposing students to a particular form, we should find
improvement on both the single variable forms and on the multiple forms.
Problem Type. In the pilot study we used only one kind of algebra word problem; those
involving ratios. In order to broaden the scope of our inquiry into transfer effects, the present
study included a set of algebra word problems that involved percentages. Percentages are
commonly used in algebra word problems, for example, to calculate interest or to assess the
amount of different ingredients in a mixture e.g., 15). Percent problems are of particular interest
here because they can be constructed to have the same surface form as the ratio problems, even
though their underlying structure and method of solution may be different. An example is:
"The number of paper bags on the ground at the park is 45% of the number of people visiting
the park that day."
The correct equation could be written:

PB 45% V
where PB represents the number of paper bags and V represents the number of visitors in the
park. We varied the complexity of the percent problems in the same way as for the ratio
problems. Examples of the problems are shown in Table 6.
There are a number of superficial similarities between the percent and ratio problems. However,
there is one very important difference. In the percent problems, a word order match strategy
(i.e., a non-procedural approach) can be used to generate a correct equation; this strategy will
lead to an incorrect equation for the ratio problems. If programming gives students a general
improvement in performance, we should find that students improve on both problems as a result
of taking a programming course. Alternatively, if programming is associated with teaching
procedural skills then there should be more improvement for the ratio problems since these
problems benefit more then the percent problems from a procedural approach.
77
RATIO FORM
"In Fairmont Hill., there are 5 plumbers for every 5 electricians.'
Let P represent the number of plumbers and let E represent the number of
electricians. Complete the equation given below by replacing the question
marks.
E
Y P
SINGLE LEMER FORM

'At the last company cocktail party there were 3 people who drank
wine for every 7 people who drank beer.'
Let W represent the number of wine drinkers and let B represent the number of
beer drinkers. Complete the equation given below by replacing the question
marks.
I
B -W
I
MULTIPLE FORM
'When Elizabeth Taylor goes to Tiffany. she buys 5 rubies for every
6 emeralds.*
Let B represent the number of rubies and let E represent the number of
emeralds. Complete the equation given below by replacing the question marks.
YE mitt;
Table 5:
EXAMPLES OF PROBLEMS USED IN COMPLETION TASK
BEST COPY AVAILABLE

7
Ehrlich, Abbott, Salter, So 'away
PROBLEM I: INTEGRAL
"The number of paper bags on the ground at the park is 45% of the number
of people visiting the park that day.'
PROBLEM 2: NON-INTEGRAL
"70% of the number of long distance calls that Mary made accounted for
30% of the total number of calls she made."
PROBLEM 3: COMBINATION
"20% of the students at a high school own cars.

40% of the students with cars make up 80% of the students with jobs."
Table as
EXAMPLES OF PERCENT PROBLEMS AT THREE LEVELS OF COMPLEXITY
Ehrlich, Abbott, Salter, Subway Page 8
3.2. Design
The test consisted of 12 completion problems and 24 problems which required students to either
generate an equation or a numerical solution when supplied with the value of one of the
variables. For the 24 'generate' problems 2 factors were independently varied: problem
complexity (integral, non-integral, combination) and problem type (ratio, percent). Crossing
these 2 factors yielded 3 x 2, i.e., 6 experimental conditions. There were 4 problems/condition,
giving s total of 24 test items. The main measure of performance was accuracy.
Two tests were constructed each with 36 items; one called Test A and the other called Test
B. The two tests differed only in the particular lexical items that were used in the problems; in
all other respects the tests were identical. Subjects were given one test at the beginning of the
semester (e.g.,Test A) and the other test at the end of the semester (e.g.,Test B). Subjects were
randomly assigned either Test A or Test B for the first testing session.
3.3. Procedure
The study was run using the following procedure. Subjects were college students who were
enrolled in an introductory programming course. The language they were taught was Pascal. A
complete test of the 38 algebra word problems was administered to each student at the beginning
of the semester, before they had taken the programming course, and again at the end of the
semester after they have finished the course. In order to be confident that the results we obtain
are due to the programming course rather than to other factors involved in a test-retest situation,
we also gave the same test to a group of students who were not enrolled in a programming
course. This second 'control' group were carefully selected so that their background matched
that of the experimental group as closely as possible. In particular we asked all students to allow
the university to release to us their SAT scores so that we had some independent assessment of
their general ability level.
3.4. Subjects
Experimental Subjects. These subjects were recruited from an introductory programming
course in the computer science department at a large state university. The course we recruited in
was open to students throughout the university and was taught at a level appropriate for
students who were not majoring in computer science. The department did offer a parallel course
that semester for computer science majors. Subjects were paid $5 for participating in the first
pre-test and an additional $15 if they returned to take part in the second post-test.
A total of 132 subjects participated in the initial pre-test. 92 of these subjects, i.e., 70%,
returned for the post-test session at the end of the semester. Because the study was designed to
test the effect of programming on performance on algebra word problems, we eliminated those
subjects who had taken previous programming courses at college level from further consideration.
By this criterion we rejected 10 out of the 92 subjects who had completed both the pre-test and
the post-test. We were thus left with a awe group of 82 subjects who had completed both the
pie -test and the post-test and who had no previous significant programming experience.1
Control Subjects. This group of subjects were recruited from upper level psychology courses,
specificIlly a course on motivation and a cognitive psychology course. Subjects were given
experimental credit for participating in the first, pre-test and experimental credit + 110 for
returning to take the second, post-test.
A total of 4f subjects participated in the initial pre-test. 2 of these subjects turned out to be
enrolled in the programming course and have been included in the data for the experimental
subjects. Thus, there were 44 control subjects who took the initial pre-test. 32 of these subjects,
i.e., 739 , returned for the post-test session at the end of the semester. From this initial pool of
32, we eliminated those subjects who had significant prior programming experience, using the
same criterion as we used for the experimental subjects. By these criteria, we eliminated 4
subjects. We were thus left with a core group of 28 subjects who had completed both the pre-
test and the post-test and who had no previous significant programming experience and who were
not currently enrolled in a programming course.
In terms of background, 79% of the programmers had arts majors with the remaining 21%
having majors in science or engineering. All the non-programmers had non science majors. Thus
most of our subjects, both experimental and control, were not science majors.
4. RESULTS
Before going through the results we need to point out that after carrying out the study, we found
that we had an uneven distribution of men and women in our two groups of subjects. The
genderes were evenly distributed for the programmers: there were 42 women and 40 men.
However, of our core group of 28 control subjects, 24 of them were women and 4 of them were
men. If it were the case that men and women responded the same way across all conditions, we
would be justified in ignoring gender and simply evaluating performance on the basis of the pre-
test and the post-test for the two groups of subjects. However, as will become apparent, we
found differences in performance between men and women throughout all the data analyses.
Thus, it must be stressed, that although we will present the data for all groups of subjects, the
data from the male control group are not reliable because the sample site is too small, and the
!We did not exclude subjects who said they had taken a programming course such as BASIC in high school, since
our previous research has led us to believe that these courses do not exert sufficient influence on a student's problem
solving to warrant the exclusion of that student.
Si
variances in performance too large to allow us to perform meaningful statistical comparisons. We

have, however, included the data in the tables to give some !lenge of how these 4 people
performed.
4.1. Overall
The first set of results we will present relate to our first question:
Do students improve their performance on algebra word problem, as a result of taking
a programming course?
The data which are presented in Table 7 indicate that any support for that question has to be
qualified. In particular, the data indicate that the men did improve as a result of taking the
programming course but the women did not. An analysis of the data confirm that observation;
when level of improvement was compared, the men programmers showed more improvement than
the women (ANOVA: F1, 80 mg 7.4, p < 0.01). The men showed a highly significant
improvement in performance when tested separately from the women, (t test: t35 4.96, p <
0.0005). It. must be stressed, however, that the data we are discussing is improvement in
performance. The women are performing as well, in fact better, than the men on the initial test.
That is, the data imply that men and women are equally capable of solving algebra word
problems, hut, for some reason, men are more likely to improve their performance.
We need to add a further qualification to our result. In the absence of an adequate control group
of men aon-programmers, we cannot draw a definitive conclusion about the contribl:tion of the
programming course to the improvement in performance. Indeed, the data in Table 7 show that
there was some improvement for the men non-programmers, although that improvement was not
reliable (t test: t3 1.88, n.s.) because of the small sample and the extremely high variance in
their data.
The data show that the women programmers did not improve their performance on the test.
One might want to conclude from those data that the women programmers, did not transfer their
skills and learning from programming to the algebra word problems. An alternative explanation
is that the women have higher SAT scores on average than the men (see Table 7). Thus there
may be more women than men who are already performing near their optimum level on the pre-
test and hence who are not able to improve their performance. This alternative explanation was
not borne out by our analysis. We found that SAT scores accounted for only .1% of the variance
in the improvement scores, while gender accounted for 9%. That is, a subject's SAT score was
not a very good predictor of improvement, but the gender of the subject did predict
improvement.2 Thus, their higher SAT scores do not account for the failure of the women to
7We also looked at the data for the individual subjects and found that 17 of the 42 women programmers actually
did worse on the post-test than on the pre-test. Only 7 of the 40 men showed a decrement in performance.
ALL PROBLEMS
PROGRAMMERS NON-PROGRAMMERS
MALE FEMALE MALE FEMALE
(N = 40) (N = 42) (N = 4) (N = 24)
SAT 620 560 540 500
PRE-TEST 46% (4%) 55% (4%) 54% (12%) 44% (6%)
POST-TEST 61% (4%) 57% (4%) 66% (16%) 45% (5%)
DIFF 13% 2% 12% 1%
Table 7: Percent correct for problems overall.

The total number of problems was 38.
The standard errors are given in parentheses.
improve.
We also found that, in general, the men were more consistent than the women, particularly with
respect to the relation between performance in the programming course and amount of
improvement. It. might be expected that students with higher grades in the programming course
should show more improvement b..tame they have a better understanding of programming to
transfe4 (assuming that the grades are measuring programming understanding). This expectation
was borne out for the men; those with higher grades in the course showed more improvement
than those with lower grades (correlation an 0.38, p < 0.05). There was, however, no correlation
between course grade and improvement, for the women (correlation 3. 0.01). It should be noted
that the women received the same grades in the course, on average, as the men.
In Table 8, we show the pre-test and post-test scores for the men and women as a function of
grade. The me:Y3 SAT score for the subjects at each grade level is also listed. The data for the
men show a highly consistent and regular pattern of performance across grade, SAT, pre-test
score, post -test score and difference (i.e., improvement) score. The pattern for the women is
much less consistent.
What can we conclude from these data? One conclusion certainly seems to be that our test is not
eliciting the same evidence of transfer from the women programmers as from the men. There are
a number of requirements for programming skills to transfer. One is that the student acquire a
good enough understanding of programming to be able o change previous problem-solving
behavior. A second is that the student perceive, either consciously or unconsciously, that there is
some commonality between the algebra word problem and programming. If a student keeps
knowledge of each subject area locked in separate mental compartments, there is no
communication for transfer to take place. A third requirement is that sufficient time has to
elapse for well-rehearsed behaviors such as those used in the algebra word problems, to be
replaced by the newer behaviors learned in the programming course. The women programmers
may have a different agenda than the men witL respect to any of these requirements.
The differences between the men and women could of course, reflect some idiosyncracies of the
particular students we tested, and certainly further studies need to be conducted to verify our
findings. However, to the extent that the present results are valid they offer some intriguing
differences between men and women that only show up on certain types of problems and when
students are tested for changes in performance.
MALES
N= PRE-TEST POST-TEST DIFF

Mel n
GRADE SAT
TOP 545 12 21.8 27.6 5.8
HIGH 525 9 17.7 24.8 7.1
MEDIUM 501 12 15.3 18.5 3.2
LOW 507 6 11.8 11.2 -0.6
unknown 1
FEMALES
fil= PRE-TEST POST-TEST DIFF

Mean
GRADE SAT
TOP 587 16 25.4 26.2 0.8
NIGH 539 8 15.5 14.4 -1.1
MEDIUM 565 13 17.8 19.6 1.8
LOW 457 4 13.7 9.4 -4.2
unknown 1
Table 8:
The mean pre-test and post-test scores for male and female programmers
as a function of grade in programming course
Ehrlich, Abbott, Ss her, So 'away Page 12
4.2. Particular skills and strategies

The second question we posed was:
Can we attribute any improvement in performance to the development of particular
skills or strategies?
We proposed to examine this question in the context of problem decomposition, problem strategy
and problem type.
4.2.1. Problem Decomposition

The ratio and percent problems were presented at three levels of complexity - integral, non-
integral and combination kinds of equations. The data for these prphiemcireAhown o__,
The men programmers show the predicted pattern of performance; the more complex the problem
the more they improve. However, we also found some improvement for the men non-
program mem.. Spc.cifir,a4, tliia group s'uowoci a similar pattern of improvement to the men
programmers. Moreover, for the non-integral problems the improvement WM significant (t test:
t3 4.9, p < 0.02). Thus, we cannot conclude that it is just programming that helps people to
cope with complexity; the men non-programmers had no programming experience and were still
able to improve. However, as we have pointed out before, the data from this group does not
provide a reliable indication of the performance of men who have no programming experience.
We thus must conclude by saying that programming may have some impact on how well people
cope with complexity, but further studies need to be done to tease out just how much
programming helps.
4.2.2. Problem Type

The test included 12 problems involving ratios and a matching set of 12 problems that involved
percentages. The ratio and percent problems were constructed to match as closely as possible in
all ways except for whether the problem was one that involved percentages or one that involved
ratios. However, one of the important differences between these two types of problem is that a
procedural approach benefits the ratio problems more that, the percent problems.
When we looked at the data for these two problem types we first found a dramatic difference in
performance between men and women; the data are shown in Table 10 and Table 11. On the
ratio problems the men programmers improved (by 15%) while the women programmers showed
no improvement. However, on the percent problems the women improved (by 9%) but the men
did not. An ANOVA confirmed the three way interaction between gender (male/female) problem
type (ratio/percent) and test time (pre/post) (ANOVA: F1, 80 mu 57.11, P < 0.002). Note that
the women programmers showed more improvement than the men on the percent problems even
though their initial performance was higher than the men. These data further highlight that men
and women differ in their style, rather than their accuracy of solving 1..oblems.
Ehrlich, Abbott, Salter, Solowsy
COMPLEXITY
PROGRAMM1,18 8 - M
M.
MALE (N = 40) FEMALE (N = 42)

NON- NON-
INTEGRAL INTEGRAL COMBINATION INTEGRAL INTEGRAL COMBINATION
PRE 69% (3%) 50% (4%) 84% (5%) 69% (4%) 63% (4%) 40% (4%)
POST 69% (5%) 61% (4%) 47% (5%) 69% (4%) 61% (5%) 47% (5%)
DIFF 0% 11% I3% 0% -2% 7%
NON-PROGRAMMERS

NON- NON-
INTEGRAL INTEGRAL COMBINATION INTEGRAL INTEGRAL COMBINATION
PRE 69% (6%) 53% (9%) 38% (20%) 54% (6%) 47% (7%) 30% (6%)
POST 72% (16%) 78% (11%) 63% (14%) 58% (5%) 50% (6%) 31% (7%)
DIFF 3% 25% 25% 4% 3% 1%
Table fils Percent correct as a function of the complexity of the problem.

There were a total of 8 problems for each level of complexity.
.,,....10*,
Ehrlich, Abbott, Salter, So lowsy Page 13
The improvement in performance by the women on the percent problems cannot be attributed to
their programming experience, however. A comparison of the level of improvement between the
women programmers and the women nonprogrammers revealed that there was nu difference
between them (ANOVA: F1, =6 0.133); the non-programmers improved as much as the
programmers. The men programmers showed no improvement in performance. These data
suggest that programming makes little or no contribution to performance on percent problems.
Both the men programmers and non-programmers showed some improvement on the ratio
problems. The improvement for the men non-programmers was not statistically reliable (t test:
t3 1.36, n.s.) due to the high variance in the data.
The comparison of ratio and percent problems is consistent with the claim that programming
helps students develop a procedural approach to problem-solving. This conclusion follows from
the argument that a procedural approach should improve performance on the ratio problems
only, which is what we found, albeit for the men programmers only, and with some question
about the contribution of programming rather than gender.
4.2.3. Problem Strategy

Both the ratio and percent problems discussed above were presented in the context of a generate
task; students were given a statement and asked to generate an equation to represent that
statement. In another set of ratio problems we prompted students for the correct equation by
giving them a partial equation and asking them to complete it. The equation was variously
written in the form of a ratio, with a single variable on one side of the equation, or in a multiple
form where each variable is multiplied by a single number. The data for these 12 completion
problems are shown in Table 12. These data show dramatic improvement in performance for the
men programmers (50%) (t test: t30 1. 4.95, p < 0.0005). There was no improvement in
performance at all for any of the other groups of subjects. In Table 13 we also show the data
broken down by the form of the solution. Here it can be seen that the men programmers
improved most on the multiple and single variable forms of the equations. A.64in, none of the
other groups showed any substantial improvement in performance.
These data provide the strongest evidence so far for transfer effects from programming. Even
though we cannot rule out other explanations, at least some of the improvement on these
problems represents transfer from programming. It is also interesting to note that the men
improved more on this completion task than on the set of ratio problems presented earlier which
used a generate task. The difference between these two results was not reliable (t test: t39
1.62).
The completion problems, particularly those problems in which equations are presented in the
form of a single variable c one side of the equation, represent the situation of nearest transfer in
PERCENT PROBLEMS
MALE (N=40) FEMALE (N=42; MALE (N=4) FEMALE (N = 24)
Correct Correct Correct Correct

PRE 60% (5%) 61.5% (5%) 77% (12%) 46% (7%)
POST 61% (5%) 70% (4%) 94% (2%) 57% (7%)
RIFF 1% 8.6% 17% 11%
Table 11: Percent correct performance for PERCENT problems.

There were a total of 12 problems in the set.
RATIO PROBLEMS
MALE (N=40) FEMALE (N=42) MALE (N=4) FEMALE (N = 24)
Correct Correct Correct Correct

PRE 43% (4%) 63% (5%) 31% (16%) 42% (6%)
POST 58% (6%) 48% (6%) 48% (24%) 36% (7%)
RIFF 15% -5% 17% -6%
Table 10: Percent correct performance for RATIO problems.

COMPLETION TASK - FORM OF EQUATION
PROGRAMMERS
RATIO SL MULTIPLE RATIO SL MULTIPLE
PRE 63% (6%) 40% (5%) 26% (6%) 71% (6%) 46% (6%) 38% (6%)
POST 77% (5%) 61% (6%) 54% (6%) 66% (6%) 49% (6%) 42% (6%)
DIFF 14% 21% 28% -5% 3% 4%
NON-PROGRAMMERS
RATIO SL MULTIPLE RATIO SL MULTIPLE
PRE 75% (18%) 63% (16%) 25% (14%) 65% (8%) 34% (7%) 33% (8%)
POST 81% (19%) 38% (24%) 50% (29%) 65% f8%) 34% (7%) 27% (8%)
DIFF 6% -25% 25% OS 0% -6%
Table 13: Percent correct performance for the COMPLETION task.

TLere were a total of 4 problems for each form.
Time standard errors are given in parentheses.
COMPLETION TASK
MALE (N=40) FEMALE (N=42) MALE (N=4) FEMALE (N = 24)
Correct, Correct Correct Correct

PRE 43% (45) 52% (5%) 54% (14%) 44% (6%)
POST 64% (5%) 52% (5%) 56% (21%) 42% (6%)
DIFF 21% 0% 2% -2%
Table 12: Percent correct performance for the COMPLETION task.

this study. It is thus not surprising that we should find the greatest transfer where there is the
greatest similarity between the items in the test and the kind of problems the student gets in the
programming course. However, more subtle skills must also be transferred from programming to
algebra word problems to account for the exceptional improvement in performance on the
problems which were presented in the multiple form. Students are not likely to see equations
written in this form in computer programs. However, this form of writing an equation is the form
that is "most strongly associated with making errors. The students who improved their
performance when the equation was presented in this form may very well have realized how to
write correct equations in a single variable form. They may have then completed the multiple
form equations by first changing it into the single variable form, solved it that way, and then
changed it back into a multiple.
Our conclusion from these data is that they provide quite strong evidence not only that a
programming course can improve performance, but that programming encourages students to
develop skills associated with viewing equations as active procedures rather than as descriptions.
Both the data from the comparison of the problem types and these data from the completion
problems are consistent with that conclusion.
5. DISCUSSION
There is a strong belief that students who learn programming for the first time should be able to
transfer some of their new programming skills to other disciplines. However, it has proven
extremely difficult to provide empirical evidence for this belief in part because there is a lot of
uncertainty over where to look for transfer effects.
In the present study we took a very focused approach to the transfer issue by examining whether
a programming course improved procedural skills. In particular we examined whether
programming helps students decompose problems, whether it helps them develop a more active
approach to problem-solving and whether the kind of transfer we were examining was specific to
particular non-procedural strategies. To further demonstrate the influence of programming on
individual students, we tested the same students before and after they had taken their first,
introductory programming course, on a set of algebra word problems. The performance of these
programmers was compared with a similar group of students who had not taken a programming
course.
Our data were inconclusive with respect to the transfer issue. Unexpectedly, we found a large
gender difference which translated into the suggestion that men were far more likely than women
to show transfer effects. That is, more of the men improved their performance on the set of
algebra problems than did the women. Moreover, clue to an inadequate control group for the
men programmers, it was not possible to demonstrate whether this improvement wz1 due to
93
r
gender differences or to programming experience. However, even if we take a pessimistic view of
the data and treat the means from the control group of men as being representative of men non-
programmers in general, there is some evidence for one of the procedural skills; the men
programmers did adopt a more active approach to problem-solving after taking the programming
course as compared with their previous performance. The improvement was most clearly
demonstrated when the problem prompted the students to take a more active approach rather
than when this approach was expected to emerge spo-' neously.
It seems unlikely that students who take a new course in any subject do not learn something new
that can be applied to other old, more familiar subjects. Despite the apparent strong face
validity of transfer effects from programming, it seems very difficult to provide strong empirical
evidence in its favor. Based on what we found in this study, future studies of transfer effects will
need to pay more attention to individual differences, in addition to focusing on the specific types
of skills that are learned in programming that might be transferred.
References
11) Bonar, J., Ehrlich, K., Soloway, E.
Collecting and Analyzing On-Line Protocols from Novice Programmers
Behavioral Research Methods and Instrumentation 14:203-209, 1982.
121 Clement, J., Lochhead, J., and Monk, G.
Translation difficulties in learning mathematics.
American Mathematical Monthly 88:215-40, 1981.
131 Ehrlich, K., Soloway, E.
An Empirical Investigation of the Tacit Plan Knowledge in Programming.
Technical Report 82-30, Dept. of Computer Science, T ale University, 1982.
141 Ehrlich, K., Soloway, E., Abbott, V.
Transfer Effects From Programming To Algebra Word Problems: A Preliminary Study.
Technical Report 257, Dept. of Computer Science, Yale University, 1983.
Hinsley, D. A.,Hayes, J. R., and Simon, H. A.

From words to equations: Meaning and representation in algebra word problems.
Erlbaum, Hillsdale, NJ, 1977,
Howe, J.A,*1., O'Shea, T. and Plane, J.

Teaching Mathematics Through Loki.^ Programming.
Technical Report 115, University of Edinburgh, Artificial Intelligence, 1979.
Papert, S.
Mindatorms, Children, Computers and Powerful Ideas.
Basic Books, 1980.
Soloway, E., Lochhead, J., Clement, J.

Does Computer Programming Enhance Problem Solving Ability? Some Positive
Evidence on Algebra Word Problems.
R. Seidel, R. Anderson, B. Hunter (Eds.), Academic Press, New York, NY, 1982b, pages
171-215.
191 Soloway, E., Sonar, .1., Woolf, B., Barth, P., Rubin, E., and Ehrlich, K.
Cognition and programming: Why Your Students Write Those Crazy Programs.
In Proceedings of the National Educational Computing Conference. NECC, No. Denton,
Tx., 1981.
1101 Soloway, E., Ehrlich, K., Bonar, J., Greenspan, J.
What Do Novices Know About Programming?
Ablex, Inc., 1082, .
95
WHAT WILL IT TAKE TO LEARN THINKING SKILLS
THROUGH COMPUTER PROGRAMMING?
Roy D. Pea
Imagine yourself as a visitor to a traditional farming society in West

Africa. You have arrived as a cross-cultural psychologist to study
whether and how literacy affects the way people think. Let us begin by
asking why you are here.
The acquisition of literacy had long been claimed to promote the

development of intellectual skills. Prowinent historians and psychologists
had long argued that written language has many important properties that
distinguish it from oral language, and that the use of written language
leads to the development of highly general thinking abilities, such as
logical reasoning and abstract thinking. Piagetian studies in other
cultures had made clear that the kind of abs.`:ract thinking associated with
formal operations did not develop in or1 cultures. By contrast, when
one looked at cultures that used written language, various cognitive tasks
revealed high logical competencies.
But you had observed that studies bearing on this claim had always been
done in societies such as Senegal or Mexico, where literacy and schooling
were confounded. Perhaps schooling was responsible for these changes in
thinking, rather than the use of written language per se.
This essay also appeared in the Preproceedings of the National Lcgo

Conference, MIT, Cambridge, MA, Julie 1984, under the title of "Symbol
systems and thinking skills: Logo in context." I would like to take this
opportunity to thank the Spencer Foundation and the National Institute of
Education (Contract #400-830016) for supporting our research program on
the development of Logo programming and its relation to other cognitive
skills. My colleagues at the Center for Children and Technology have
been a continuing source of encouragement and stimulation.
The reason you have travelled to Africa is that you plan to test, for the
first time, the cognitive effects of literacy independently of schooling.
The society you are studying--the Vai--does not transmit literacy in the
Vai written language through formal schooling. Their reading and writing
are practiced and learned only through the activities of daily life.
The Vai invented their written language a mere 150 years ago, and have
continued to pass literacy on to their children without schooling.
Like all the psychologists before you, you have brought along suitcases
filled with psychological tests and materials for experiments on concept
formation and verbal reasoning. Results from performances by the Vai
with and without written language experience will tell you whether
possessing literacy affects the way they think. You then carry out your
research.
As you look over your results from several years of work, you find no
general cognitive effects of being literate in the Val script. For example,
the literate Vai were no better than the nonliterate Vai in categorization
skills or in syllogistic reasoning. Literacy or se did not appear to
produce the general cogil:tive effects on higher thinking skills you
expected.
So you mull over this fact for some time. How could this be? The
arguments were so plausible for why written language would affect the
way people think. You wonder -- could the studies be done more
carefully?
Before continuing this research strategy, you realize that there is a

radically different way to think about your project. When you arrived
you took for granted that literacy would have its general effects, and
then looked to see if it did so by testing for general intellectual benefits.
But with several years of survey and ethnographic observations under
your belt, you have come to better understand the tasks that Vai
literates encounter in their eve.~; day practices of literacy. How does this
relate to your experiments?
What you decide you could do instead is to actually look to see how
literacy is practiced in the Vai culture. What is done with the written
language? And then you ask a very different type of research question:
How could what the Vai people do specifically with written language affect
their thought processes? You decide, in other words, to let your
fieldwork on literacy practices dictate the design of your "outcome" tasks.
You thereby gain a great deal of precision in your hypotheses for the
cognitive effects of literacy.
This reorientation literally turns on its head your paradigm of looking for
general cognitive effects of literacy. You have abandoned the approach
of making general predictions from developmental theory to effects on
general intelligence. You now start with concrete observations of literacy
behavior and build up to a general functional theory of the cognitive
effects of specific literacy practices.
With this new approach you find that the Vai use their written language
primarily for letter-writing, and for recording lists and making technical
'farming plans.
You then begin a new phase of research, to tease out cognitive effects of
specific literacy practices rather than literacy per se. You design new
tasks for assessing literacy effects that draw on related skills to those
required by the practices you observe, but which involve different
materials.
What you find when guided by this new functional perspective are
dramatic cognitive effects of literacy. But they are more restricted in
nature. For example, letter writing, a common Vai literacy practice,
requires more explicit rendering of meaning than that called for in face to
face talk. So you refine a communication task where the rules of a novel
board game must be explained to someone unfamiliar with it, either face to
face or by dictating a letter for an absent person. You find, lo and
behold, that performances of Vai literates are vastly superior on either
version of this task to those of nonliterates.
This is no mere parable. It is an account of an extensive five-year
research project carriee out by Professors Sylvia Scribner and Michael
Cole (1981). It is the account of an intellectual voyage not so far
removed from what children are learning with Logo programming. We can
fruitfully apply the schema of this Vai story to questions about the
cognitive effects of programming. And here I believe, is where one will
find what will be needed to learn thinking skills through programming.
Here, too, there are persuasive and intuitively appealing arguments for
why people should become better thinkers by virtue of the use of a
powerful symbol system such as the Logo programming language. It is
alleged that children will acquire general cognitive skills such as planning
abilities, problem solving heuristics, and reflectiveness on the revisionary
character of the problem solving process itself. The features of
programming literacy assumed here include the necessarily explicit nature
of writing program instructions, the strategic and planful approaches
ingredient to modular program design, and experience with the logic of
conditionals, flow of control, and with program debugging.
But for programming languages, unlike written natural languages, we do

not have the benefit of known historical and cultural changes that appear
to result hi part from centuries of use of the written language. The
symbol systems provided by programming languages are relatively new.
They have certainly changed the world; we now live in an information age
because of achievements made possible by these languages. But what
does it mean for how individuals think and learn?
Let us move our West African story to the context of the American
Classroom. Here again we enter as psychologists, looking for general
cognitive effects, much like the first literacy questions of the African
enterprise.
Of course we assume that we know what kind of a mind-altering substance

programming is (having been so affected ourselves), and we assume that
"programming intelligence" and the kinds of programming activities carried
out by adults will affect children too.
- 4 -9
But we should give pause--for we have entered another culture. What
will children do with a programming language in a discovery-learning
situation, in Logo's "learning without curriculum" pedagogy (Papert,
1980), without benefit of being shown what kinds of things can be done,
or being taught about the powers of the system or of thinking skills?
Nonetheless, without benefit of such hindsight, what do our psychologists

in the Logo classroom do? Here we refer to our own work. They, too,
look for programming's "effects," guided by somewhat the same kind of
thinking that possessed the first phase of the Vai studies. The primary
difference was that instead of testing for increments in general
intelligence, or concept formation, they thought they were looking at more
specific effects, quite plausibly linked to programming activities.
Planning skills were the central focus, not abstract reasoning, which is
only indirectly related to programming.
The psychologists' reasoning went something like this: Both rational

analyses of programming and observations of adult programmers show that
planning is manifested in programming in important ways. Once a
programming problem is formulated, the programmer often maps out a
program plan or design that will then be written in programming code.
Expert programmers spend a good deal of their time in planning program
design, and have many planning strategies available, such as problem
decomposition, modular documentation, subgoal generation, retrieval of
known solutions, and evaluative analysis and debugging of program
components (e.g. Pea & 1983).
Our psychologists studying the cognitive effects of Logo created planning

tasks to reveal the development of different planning strategies, and of
skills at plan revisions analogous to program revisions. But in two
different studies, after a year of Logo programming, no effects of
programming on performances in these planning tasks were found (Pea &
Kurland, 1984). Children improved with age and practice on the planning
tasks, but non-programmers did just as well after a year's time as did
Logo programmers. Once again, like the researchers in West Africa, we
must reflect on our first set of assumptions for framing the research
questions, and reconsider the meaning of our research findings.
Let us take a different, functional or activity-based approach to

programming. Consider "programming" not as a unitary 'even, whose
features we know by virtue of how adults do it at its best, nor as what
it looks like in its ideal text-book forms. Let us look at programming as
a set of practices that emerge in a complex goal-directed cultural
framework of thought, emotion, and action.
Viewed in this way, by analogy to the Vai studies on literacy practices,

we see that programming is as various and complex an activity matrix as
literacy. Just as one may use one's literacy in Vai society to make
laundry lists rather than analyze and reflect on the logical structures of
written arguments, so one may achieve much more modest activities in
programming than dialectics concerning the processes of general problem
solving, planning, precise thinking, debugging, and the discovery of
powerful ideas. One may, in particular, write linear brute-force code for
drawing in turtle graphics.
Stated baldly, from a functional perspective we may see that powerful

ideas are no more attributes I -herent "...a" Logo than powerful ideas are
inherent "in" written languag.:. Each may be put to a broad range of
purposes. What one does with Logo--or written language--or any symbol
system, for that matter--is an open matter. One must come to these
powerful ideas and potentially fertile grounds for developing general
thinking skills through discovery, or through learning with the guidance
of others. Independent discovery and practice of Logo recursion, for
example, may be a very rare spontaneous occurrence. The Vai in Africa
have not spontaneously got onto the logical features of written language,
philosophy, and textual analysis that written language allows. Likewise,
most of our students--from grade school up through high school--have
not spontaneously got onto the programming practices, such as planning
to reuse procedures as building blocks in other programs, use of
conditional or recursive structures, or careful documentation and
debugging, that Logo allows.
- 6 - 101
For the Vai society, one could imagine introducing new logical and
analytic uses of their written language. Similarly, one could imagine
introducing children the Logo programming practices many educators
*a..
have heretofore taken for granted will emerge. In either case, we would
argue that without some functional significance to the activities for those
who are learning the new practices, there is unlikely to be successful,
transferrable learning. Serving some purpose--whether being able to
solve problems one could not otherwise, satisfying an intrinsic interest in
complex problem solving, or achieving solidarity with a peer group who
define their identity in part by "doing" Logo or written language--is a
necessary condition for the symbolic activities we are interested in
promoting to be ones our learners find a commitment to.
It is my hunch that wherever we see children using Logo in the ways its
designers hoped, and learning new thinking and problem solving skills, it
is because someone has provided guidance, support, and ideas for how
the language could be used. They will have pointed the way through
examples, rules, and help in writing programs and discussing the
powerful ideas. To call these rich activities "learning without curriculum"
is misleading, and an overly narrow view of what constitutes curriculum,
for any projected path toward greater competency that another person
helps arrange can be thought of as a curriculum.
There are many profound consequences of this more general account of

what is involved in thinking about Logo as potential vehicle for promoting
thinking and problem solving skills. A functional approach to
programming recognizes that we need to create a culture for Logo, in
which students, peers and teachers talk about thinking skills, display
them aloud for others to share and learn from, a culture that continually
reveals how programming is a vehicle for learning general thinking skills,
and that builds bridges to thinking about other domains of school and
life. Such thinking skills, as played out in programming projects, would
come to play functional roles in the lives of those in this culture. Dialog
and inquiry about thinking and learning processes would become second
nature, and the development of general problem solving skills so
important in an information age would be a common achievement of
102
students. This vision could, in principle, be realized. I imagine that
important cognitive effects of programming, or of literacy are possible,
but only when certain uses of these symbol systems are practiced, not
the ones most engaged in today. There is far too much faith today that
Logo carries with it guarantees of cognitive outcomes, and there is
already evidence that when these changes are not found, educators will
be prematurely discouraged.
Where are we left after these two continents of travel? With reason for
optimism. There are many streams of Logo activities and research that
should go on, for plurality and diversity provide exciting grounds for
emergent ideas. Communication among groups of students, teachers,
psychologists, and computer scientists will help in the formation of a
broad community exploring these issues. These streams will no doubt
embody a diversity of assumptions about what will best help create the
culture of Logo I have referred to, in which one will be more likely to
find the cognitive effects on thinking skills so many take for granted.
Similar Logo cultures may arise that center on math learning, or
programming. Each is likely to require the construction of extensive
microworlds for learning more specific than the Logo language itself.
It is uplifting that there are so many positive energies in education

today. The enthusiasm for Logo as a vehicle of cognitive change is an
exhilarating part of the new processes of education one can see emerging.
Cultures with thinking tools like Logo can be created. But we must first
recognize that we are visitors in a strange world--at the fringe of
creating a culture of education that takes for granted the usefulness of
the problem solving tools provided by computers, and the kind of
thinking and learning skills that the domain of programming makes so
amenable to using, refining, and talking about together.
References
Papert, S. (1980). Mindstorms. New York: Basic Books,
Pea, R. D., & Kurland, D. M. (1983). On the cognitive prerequisites of

learning computer programming. Interim Report to the National
Institute of Education (Contract #400-83-0016).
Pea, R. D., & Kurland, D. M. (1984). Logo programming and the

development of planning_ skills. Technical Report No. 16. New
York, NY: Center for Children and Technology, Bank Street College
of Education.
Scribner, S. , & Cole, M. (1981). The psychology of literacy.

Cambridge, MA: Harvard University Press.
-9-104
Making Programming Instruction
Cognitively Demanding: An Intervention Study
John Dalbey
Francoise Tourniaire
Marcia C. Linn
Assessing the Cognitive Consequences of
Computer Environments for Learning tACCCEL)
Lawrence Hall of Science
University of California
Berkeley, California 94720
This document has been produced under Contract No, 3E

0400830017 from the National Institute of Education, U.S.
Department of Education to the University of California ano
Far west Laboratory for Educational Research and Divelop-
ment However, it does not necessarily reflect the views f
the Institute, the Department, the University or the Labora-
tart'.
BEST COPY AVAILABLE

Introduction
A plethora of recent reports on the status of education

in the United States today call attention to the lack :t
higher cognitive outcomes from classroom Instruction. Tne
report of the National Commission on Excellence in Educa-

tion, °A Nation at Risk," describes a pressing need for edu-
cational reform to create a "learning society." Sivilarly,
the National Science Board, in a report entitled °Educating
Americans for the 21st Century,' has called for the "new
basics," or the thinking skills required to cope with rapid
technological and scientific advances. These and other
reports emphasize tne need for instruction which fosters
problem solving and prepares students to deal with new tecm-

nolcgical tools as they become available. Computer learning
environments have the potential for imparting some of these
important higher cognitive skills.
This paper reports the evaluation of instructional pro-

visions designed to foster h;gher cognitive skill in a cou-
puter programming course. This intervention explicitly
encourages novice programmers to engage in the problen-
solving skill of planning.
LualltilssA of tt: Cansatc guIcgauu

Several features of the computer learning environment
can, if capitalized upon, increase the quantity and quality
of cognitively demanding activities offered in schools. Tne
BEST COW AVAJIAIILE

106
3
ACCCEL project has identified six features of this environ-

ment which mak* it capable of providing cognitively demand-
ing activity. Three are characteristic of many school
environments, and three are somewhat unique to computer
learning.
The first feature common to both some school and sane

computer learning environments Is camalsilge Students can
solve complex problems when using computers. For examples
students can solve problems which require the manageeent of
large amounts of information, such as plotting graphs or
computing compound Interest. The second feature is 1n117.

langs. The computer can challenge the student to solve
problems such as figuring out the best move in a game or
determining the most efficient path through a maze. The
third feature is the provision for 214,1tipl: Aphitions to
problems. Students can write and compare several prograas
which do the same thing. These features are characteristic
of some other school activities, such ss writing reports or
proving geometry theorems.
Three additicnal features of the computer learning

environment are less characteristic of other classroom
learning. First, the computer environment is lailLagllyie
The computer can respond quickly and inforzatively to the
learner. Thus, students can try several approaches for
reformulating their computer program and determine whether

each of tnose approaches is successful. in contrasts it
BEST COPY AVAILABLE

often taxes days or weeks for students to get responses to
their homework assignments in other subjects. Second. tne
computer can provide uncials feedback. Thus, the computer

can tell exactly what happens when data is entered into a
program. In contrast, when students write essays, they fre-

quently get relatively imprecise feedback. For example,
they may learn that their work merits a *e". Third, cox-
puter learning environments are consistent. Tney give tne
same response when the same information is presented. More-

over, they provide the same response for all learners. In
contrast: teachers do not necessarily respond identically to

the same information, either because they are rushed :r dis-
tracted, or because they are tailoring their response to the
perceived needs of the student. When teachers behave as
good tutors, their tailored responses to the students pro-

vide advantages not available in most computer learning
environments. On the other hand, when teachers are dis-

tracted. their inconsistent responses can be less desirable
than those characteristic of computer environments.
eCilgLAa.aail
Currently. the most cognitively demanding activity
readily available on the computer is programming. This
situation is changing as new software l-,, comes available.
Eventually, software which demands higher cognitive skills.
but is free from some of the drawbacks of programming may
become preferable to programming for fostering higher cogni-
BEST COPY AVAILABLE

tive skill (Linn, 1984). Currently, however: pre - college
students who have cognitively demanding interactions with

computers are usually engaged in programming.
In spite of this situation, much programming instruc-

tion lacks a conceptual framework, and is not necessarily
geared to fostering higher cognitive skills (Linn A Fishery
1983) . Until Tecently, most teachers did not have teKtboocs
for programming instruction, but instead amassed materials
somewhat haphazardly (e.g. Becker, 1984). Limited funds
have been available for teacher professional development in
computer education; many districts have followed a "buy

harcware now. plan for its use later" approach. Before pr3-
sramming instruction can achieve reasonable goals, this
situation must be rectified. The current study offers one
approach.
CtACLCIlaltiLl a Iltal3Q122111 EgILLIS

The contrast between the behavior of technological
experts and the activities of students participating in
pre-college programming classes motivated us to conduct this
study. This discrepancy strongly suggests a need for
materials which form a chain of cognitive accomplishments

culminating in technological expertise. As matters scow
stand, some college teachers complain that pre - college
instruction actually interferes with ability to profit from

college computer education programs. Analysis of the nature
of expertise. the characteristics of current pre-college
109
instruction and the potential of the computer learning
environment, provides the primary rationale for this study.
The behavior of experts in computer learning environ-

tents contrasts sharply with the characteristics of current
instruction as revealed by several studies of experts {Kur-
land L Pea, 1984; Linn, 1984; Jeffries, Turner, Poison,

Atwood, 1981). To create the chain of cognitive accomplish-
ments which culminates in technological expertise students
must learn to use the skills experts use everyday. Current
instruction may not provide this opportunity.
One component of expertise is an extensive repertoire

of wpr%raming templates. Templates are stereotypic
prescriptions for a particular aspect of a program. similar
to schemata as described by Norman, Gentner, & Stevens

(1976). Templates can apply to a whole program as exempli-
fled in an 'input-process-output template. Templates can
also apply to a specific function of a progran such as a
loop function as illustrated in Table 1. Research by Kur-
land L Pea 11984] reveals that both expert adult and expert
student programmers can articulate their templates: recog-

nize the relationships between their templates and new ten-
plates. and actively seek new templates.
Technological experts also use a variety of procedural
skills. These we among the skills referred to as the new

basics by recent commission reports. They are a part of tne
set of thinxing or problem solving skills which individuals
UST COPY AVAILASUE

1-10
4.7
need to survive in our society. An important component bf
this skill is planning or the ability to determine an
appropriate sequence of available templates.
In the past, Investigations of expert performance in
formal systems such as solving mechanics problems in physics

has proved informative for educators designing programs to
foster these skills in novices (Larkin, McDermott, Simon I

Simon, 19b0; Keller i Reif. 1980). The current study was
oesigned to examine the effects of coaxing a novice program-
ming course more cognitively demanding.
Literature cn planning solutions to programming prob-
lems by experts indicates that experts engage in two comple-
mentary techniques: top-down design and stepwise refinement
(Brooks, 1980; Atwood & Ramsey. 1978; Jeffries, Turner, Poi-
son, L Atwood, 1981). Top-down design is an approacn which
decomposes a complex problem into subproblems. Experts can
co this effectively, we surmise, because they have a large
repertoire of program templates. Experts use their
knowledge of templates to guide the decomposition process.
Tcp-down design is somewhat iterative in nature. After tne
initial decomposition, each resulting subproblem may require

further decomposition until the problem reaches a manageable
degree of complexity, Experts proceed with top-down design
by selecting appropriate templates for each problem.
In another design technique called stepwise refinement.
experts engage in successive restatements of the problem

8
specification with each step closer to machine level nota-

tions The original problem specification describes in
natural language a process the computer is to perforv.

Stepwise refinement means to translate the process descrlo-
tion into language the machine understands through incremen-
tal stages. Experts can do this well because they are very
familiar with the language tile machine uses. Experts know
the degree of pvecision and the degree of clarity needed to
cescribe the process for a machine solution. Ultimately,
they generate unambiguous statements of their program
cesign.
tAcsicisitrassisl, Qf ilugmall WILAMIDS
Students, those who are just beginning to learn a pro-

f:ramming language, usually differ dratatically depending on
the sort of instruction they have received (Soloway,
Ehrlich, sonar, i Greenspan, 1982i. Members of the ACCGiL

project have observed over 25 junior high BASIC programming
classes. Most classes offer Instruction which emphasizes
the language features, and which often fails to provide
instruction in how to combine the language features into

larger algorithms,
Much of the programming instruction which we observed

could be described as drill and practice in learning pro-
gramming language features. Students are introduced to a
language feature such as the PRINT statement. They write
programs using that statement. Their understanding of the
BUT COPY AVAILABLE 112

program is basically at the level of a single line. They
type in a line and get feedback about their use of the PRI4T
statement. Students respond by typing in a different line
which hopefully corrects the mistake they have made ini-
tially. These students are engaged in drill and practice on
a language feature.
Instruction rarely emphasizes the templates which

experts use for solving programming problems. Students,
therefore: fail to acquire templates to help them decompose
problems and plan problem solutions.
Nov ice programmers are characterized by a 'rush to the
computer." They frequently attempt to go from a statement of

the proolem directly to the program code without any con
sideration of how to design the code. Novices appear to lack
the tools necessary for constructing intermediate states
between the problem specification and the problem program
code. They rarely receive an opportunity to observe their
teachers or expert programmers model the use of planning.
The expert process of stepwise refinement is also nei

ther regLired nor really necessary for most assignments that
novices receive in programming courses. It appears tnat many
novices fail to grasp the notion that programs are detailed
process descriptions which can be refined out of a natural

language description. Instead, novices usually have a
tinker toy model of program construction. They presume that

programs are assembled by piecing the language features
:4'
together. They fall to understand that the natural language
problem description is less precise and more ambiguous than

a problem description in machiine terminology. Therefore,
they do not engage in the activities required for refining
the natural language statement of the problem into a state-
ment which can be decomposed and coded into a problem so1J-

Lion. As a result, when they are asked to solve proble*s
which are more complex than simple translations of known
language features, their solutions are often poorly organ-
ized, incorrect, or inefficient. Thus, the top down design

and stepwise refinemenk which experts use to write prograns
is not taught nor required by most introductory programming

courses at the junior high level.
ACCCEL Explicitness
The ACCCLI. explicitness intervention fosters higher
cognitive skill by providing students with some abstract
templates which they can use for stepwise refinement of the
problem specifications. We set out to provide students with
a mechanism for constructing a problem solution that Was
more detailed than the available problem specification but

less detailed than the actual language statements. Thus, we
wished to encourage students to consider an intermediate
state between the problem specifications and the program
code.
As noted above, most students go directly from tne
SOT COPY AVAILAILE 114

problem specification to the terminal, This frequently
leads to frustration and inefficient trial and error so1J-
tions. however, it should be noted that students are basi-
cally very happy %hen working at the terminal and fail to
associate tneir difficulties in achieving a solution with

their lack of planning. Students believe they need more
terminal time, not that they need to plan, to solve, problens

effectively.
Because of the nature of novice instruction in junior
high, it was impossible to emphasize planning until students

had gained at least a reasonable subset of language
features. Thus, we selected an instructional setting where
students had already been programming for one semester and
had gained enough familiarity with the language that they
could be given problems which required planning. Since
instruction up to this point had not emphasized planning,

cur intervention required a considerable departure from pre-
vious activities. Our intervention interfered with sts-
dents/ access to the computer terminals.
Thus, our approach was to provide abstract templates
which students could use to refine the problem specifica-
tions. We also required students to perform numerous exer-
cises where they took problem specifications and translated
them into the templates that .ire provioed. We did not
require students to code their solutions from the templates.
but rather, provided extensive e.perience in 'gapping problem

so 12 -
Specifications onto the templates.
lattuelltion Euscdurt
The particular design technique we selected to teach

students, structure diagraming, is one of several graphical
methods. It is a method for pictorially representing the
logical organization of the solution to a problem. The suo-
jects were eighth grade students (age 12 and 13) in an urban
school. The students had just begun the second semester of
the first course in programming in BASIC. Most of them were
familiar with elementary statements in BASIC, but were not
yet adept at writing programs.
The instruction consisted of five lessons, each a class

period long. The classroom Instruction was carried cut by
two of the authors. Each lesson was comprised of a lecture
and student exercises. The lecture was a brief (10-20 min.)
explanation azout the design technique, focusing on one par-
ticular aspect. The exercises were prepared by the project
and students worked on them at their desks (not at the con-
puter). The exercises provided examples and problens to oe

designed.
The first lesson provided a motivating example using a
dart game, explained the purpose of the technique, and

introduced three basic program structures or blocks: action,
loop, and decision (see Figure 1). The assignment involved
classifying simple English problem statements according to
UST COPY AVAILABLE 116

the three types of blocks.
The second lesson presented a way to draw a diagram
which corresponded to each block. The assignment required
students to draw diagrams for problems they had classified

in the previous exercises.
In the 'third lesson the diagraming technique was
expandeo to problems that combined blocks. The assignment

involves e,awing diagrams with combined block structures
from 0 problem statement.
The fourth lesson proceeded to advanced problems where
blocks were nested within one another. Students worked
exercises drawing structure diagrams for these complex prop-

lem types.
The fifth lesson explained the conventions for
translating the structure diagrams into BASIC code.
Specific language features were described for coding each
block.
SulleicrIs
Participants in this program were 30 eighth graders in
a racially and socio-economically mixed classroom. They had
just begun the second semester of their first programming
course in BASIC. When we instituted tne intervention, these
students has been introduced to most BASIC language
features, including input, output, loops, and decisions.
117
t
' 1
dm
14-
Most were familiar with elementary statements in BASIC b4t
were not yet adept at writing programs. The class was
self -paces in the sense that students proceeded through a
series or problems at their own speed. The most advanced
students were solving the 85th problem which had been
offered that year, whereas the least advanced students were
still on the 20th problem. These problems were, in general,
less complex than those used in our exercises.
As a result of the intervention, all the students in
the class were able to use structure diagrams for simple
problem specifications. Problems which all students could
represent with structured diagrams are given In Table I.
Students were considerably less successful when they
were asked to produce structure diagrams for broblems
specifications where the superficial grammatical features bf

the text did not match the block types they had been taught.
we spent considerable effort preparing problems as
standardized as possible. They were much less ambiguous
than problem specifications in most textbooks. Even with
these carefully prepared problems, however, students often
failed to identify the salient features that were cues to
the selection of the proper block type. As long as the
superficial form exactly matched the block types, students
were able to proceed. However, if the text of the problem

15-
statement varied from the prototypical form, or required
interpretation, mast students were at a loss as to how to
begin. The problem statements in Table 2 presented diffi-
culties for students. For example, most students did not
realize that problem 1 in Table 2 was a looping problem.
Students also had difficulty in determining how much to

refine their specifications. Thus, students frequently did
not refine tne problem specifications to the level of detail
f the available block types. For example, problem 3 in
Table 2 requires students to write a structure diegram for
computing an average. Many students treated the computation
of the average as an action. They did not realize that con-
puting an average required a looping process. There was an
example proolem showing how averages were computed, but
apparently the students did not benefit from it. The result
as that their solutions were not refined to an appropriate
level cf detail.
A thiro Kind of student difficulty was a failure to
distinguish process from content in the problem statement.

For example, in problem 4 in Table 2, students were unable
to differentiate between the operations to be performed and
the data tnat the operations were to be perforned on. Thus,
they cid not differentiate between the counting operation
required in problem 9 and the data which was esse-ntially the
report cards.
Finally, a major difficulty was that students failed to
119
- lb -
see the benefits of the structure diagram process for helb-
ing them to solve programming problems. Students preferred
to be on-line with the computers. A few students employed

structure diagrams as an intermediate step between the prob-
lem specification and coding of a problem solution. 4owever,
these stuoents mere in the minority. It seemed that stu-
dents turned to structure diagrams as a last resort when all
other approaches for programming failed. Perhaps+ if these
students were given challenging assignments for Which they
could fino no other solution, they might attempt to use
structure diagrams. This difficulty, as we anticipated, was
cue to the lack of previous emphasis on planning problem
solutions. Our intervention represented a departure from
previous practice in this classroom.
After we completed the intervention, but before we
administered the postest, serious discipline problems arose
in this class. A more stringent class management scheme was
adopted after the intervention. The students rebelled
against the new state of affairs. Therefore, it was felt
that this class was no longer comparable With the other
BASIC classes. Further evaluation of the intervention was
not pursued.
This investigation revealed that students can use
structure diagrams for refining problem specifications+
although they have difficulty as soon as the features of the
120
problem specification deviate from those they have already
encountered. It seems clear that more varied experien:e
mapping statements onto the structure diagrams is required.
Although students can use structure diagrams.

tney have
difficulty using them when the problem
specifications becone
complex. Similarly, they have difficulty
coding solutions
for complex problem specifications. Our experience suggests
that the difficulties in coding may reflect
difficulties in
understanding the problem specification. Thus, students
need additional instruction both to interpret the
problem
specifications and to translate these
interpretations into
program code.
Current modes of instruction fail to communicate tne

value of planning in programming. This lack of appreciation
cf planning steus in part from certain characteristics of
the instruction. First. students initial programming
experiences do not require planning, and tnerefore, the
aovantaies of planning are not apparent to them. Second,
students rind the on-line experience very notivaring and
they prefer to be on-line interacting with the cumputer,
even if they are not making progress in solving problems.
It appears that the interactive nature of the computer
learning environment has not been filen channeled to VIe
higher cognitive skill of planning.
One important reason why students fail to appreciate

planning is that many students can solve even the most
18
difficult problems assigned without planning. The most dif-
ficult problem which students were currently attempting wss
to write a number guesser program. Many students could
envision how to solve this problems mithout spending any
time planning.
Our experience in imOementing the ACCCEL explicitness
treatment suggests some directions that might be considered.
First, rather than beginning programming instruction with
crill and practice on the language features, it mould seem
quite appropriate to begin instruction with comprehension of
program code. Students could be given reasonable sized pro-
grams (10-50 lines of code) and could be encouraged to cone
to understand those programs. Those programs would demon-
strate how planning is used in programming. Students could
see how experts use planning to write a big program. Thus,
students mould have a better understanding of the role of
planning in programming. Second, structure diagrams could
be used to help students comprehend a large program. A
large program could be represented using structure diagrams.

Comprehension of the program could be encouraged by using
structure diagrams to illustrate the templates used by tree
programmer to construct the program. Instruction could then
proceed by demonstrating the top-down design and the succes-

sive refinements used by the expert programmer to construct
the program.
Programming instruction has the potential for fostering
122
- 19 -
the higher cognitive Urns called for by the many recent

reports on tne state of educational practice out, so far,
the potential is not being achieved. Instruction which
builds a chain of cognitive consequences culminating in tne
planning skills uses by expert programmers requires early
and consistent emphasis on these stills. Teachers are
nee ced who can demonstrate planning. Texts are needed which
oelineate the steps between problem specification and pro-
gram code. Research is needed to more clearly understand
the chain of activities which will facilitate the desired
con sequences.
123
TABLE 1
PPOaLEMS WHICH NAPPED

ON TO THE STRUCTUK4L DIAGFAMS
L. Print "RAIDERS ARE CHAMPS" 79 times.

Compute 7 times 17.
7. If Your name is the same as the teacner' print
other;4isa print "DIFFERENT."
4. For each number from 1 to 50, print the number And its
so,Jare root.
Sri an and Jennv ne.....?d

to score a total of 1Z000
El..CTAM in the team competition.
to Input their individLil
scores, add them together, and print Nh.=ther they Non or
hat-. S. Ask the user to enter his or her name. Fcr eacn
letter in the ame, print a star.
TAF:iLE
FROI=BLEM L HiCH WEF,E DIFFICULT,

TO MAP ON TO THE STRUCTURAL DIAGRAMS
Sally has 50 dollars in the bank.

1. She olans to (-!o,..2it
5 dollars each month. For each month o* the 'fear, calcu:ate
her new account balance.
2. Joe can read 65 paces a day. Compute hok.) long it

take him to finish a 75Q page book.
Find the average height of students in your class.

7.7..
the average is grater than 60 inches print "G1ants", nrhr

' print "Shrimos."
4. Given a list of classes and grades -From your report
reach each grada and i t the grads is An A, add wir:4 to
t7.7.;rd:
a total. iCount th,e number o+ A's on your report card.
BEST COPY AVAILABLE
124
Figure'l
PROBLEM ANALYSIS
ACTION LOOP DECISION
DEFINITION: An Action is one or more A Loop is one or more A Decision is making a

instructions that the instructions that the choice among several
computer performs In computer performs re- actions.
sequential order. peatedly.
1AGRAM;
F (Ffalse)
T (T-itrue)
EXAMPLE:
Print "Jack" For each Row on the screen If temperature 75
or
Print your name Print "warm"
Print "hot"
[ Add 31 plus 17
An Action block is a A Loop block shows the loop A Decision block shows the con-
rectangle drawn around "control" in the top part, dition being tested in the top
the action to be per- and the repeated part or part (above the dotted line).
formed. "body" In the enclosed box. The false (F) alternative is
shown next, and the true (T)
alternative is shown at the
bottom.
BIBLIOGRAPHY
Atwood, M. E. and Ramsey, H. R. QmlnitIXe IICUCttall III Mt

aallarsbanilan mad alt22EY Of senuttr ILUSEAMI: An !melt!:
gall= 21 smut= debugging. Tech Report TR-78-A210, U. S.
As-ny Research Institute for the Behavioral and Social Sci-
ences, Alexandra, VA, 1978
Becker, H. J. 502221 Ulel Q. CiLL=PLMQUIICI. 1-5, Baltimore,
MD.: Center for Social Organization of Schools, The Johns
Hopkins University, 1984.
Brooks, R. Studying programmer behavior experimentally: The
problems cf proper methodology. CommumIcatigal of, the Agri,
1990. Za(/), 207-213.
Heiler, J. I. and Re if, F. Etelsriblna affmatims Liu= autism-
10.A..09 _cablem
P dessrlaXian 2hysics. Berke-
ley, CA: Lawrence Hall of Science, 1980. Paper presented at
the annual meeting of the American Educational Research As-
sociation in New York, March 1982
Jeffries, R., Turner, A. T., Poison, P. G., and Atwood, M. E.

Processes involved in designing software. In J. R. Anderson
(Ed.), CQUIlithe skills EMd noir icatalitiaa Hillsdale,
NJ: Lawrence Eribaum Associates, 1981.
Kurland, O. M. and Pea. R. O. Childrtel madcis of =ME=
Liu lags =ma= (Tech. Rep. No. 10), New York: Bank
.
Street College of Education, Center for Children and Tech-
nology, 1983.
Larkin, 4., McDermott, J., Simon, D. P., and Simon, H. A. Expert

and novice performance in solving physics problems. Szi:
crisa, June, 1980, Z211, 1335-1342.
Lim. M. C. Eastcrina muItabls saaltaucacts tram QQTRAIIC

lamming ca. ummenli. Lawrence Hall of Science, University
of California, Berkeley, CA: Project ACCCEL, 1984.
Linn, M. C. and Fisher, C. W. The gap between promise and reali-
ty in computer education: planning a response. Eracstdinas
at daking Cur Schmalz Ea= Ettectlxes A Caatscugz far Emu=
=MS, 1983, San Francisco, CA: Far West Laboratory.
Norman, C., Gentner, D., and Stevens, A. Comments on learning
schemata and memory representation. csmoitian amg instElic=
ti2u, 1976, Hillsdale, NJ: Eribaum.
Soloway, E., Ehrlich, K., Sonar, J., and Greenspan, J. What do
novices know about programming?. latectlans in tumair-
cimuur IncElaigias, 1982, Norwood, NJ: Ablex.
127
BEST COPY AVAILABLE
DISCUSSION
Jan Hawkins
The teaching and learning of programming has quite rapidly become

an important topic among educators. For years, programming was taught
in a small number of schools as a specialized skill for vocation. It has
recently assumed importance as a central curriculum topic for all students
in many schools, beginning at the elementary level. This is chl'e in part
to the flood of microcomputer technology into schools, but also to the
notion that programming is a learning environment where students will
acquire general problem solving skills. This is a powerful reason for the
rapid assimilation of programming into the curriculum. Without full
commitment to the importance of programming as a skill for all students,
even the possibility that its practice allows students to acquire broadly
applicable skills for solving problems is a reasonable rationale for its
adoption. This rationale is particularly timely today since schools are
widely concerned with evidence that many students fail to learn
problem-solving skills.
Two major questions are raised in this symposium, and interleaved
throughout the papers. First, how can programming be most effectively
taught? Second, what is learned when students engage in programming
practice- -both with respect to programming commands and concepts, and
to broader problem-solving skills? Not surprisingly, as these authors
look more closely at programming practices and possible circumstances of
transfer, the situation quickly becomes more complex than at initial
probing.
Specifically, various sources of complexity are identified in the
papers. As the first two papers point out (Kurland, Clement, Mawby &
Pea; Kurland, Mawby & Cahir) in order to determine the effects of
programming and decide about the best conditions for learning, a better
understanding of just what activities programming entails is needed.
Kurland et al note that programming is a multifaceted activity, not a
unitary topic. The task varies according to language, programming
environment, and social surround. Under the description "programming"
two students may be doing radically different tasks, and learning quite
different skills. In addition, as programming arrives in the classroom,
the goals of the activity are not clear. What are the conditions of
expertise? What should we expect proficient students to be able to do or
know? Neither of these issues has an easy solution. Kurland et al
emphasize the importance of planning a program, of understanding the
structural aspects of the activity, as opposed to the mere production of
code. This preparatory work and understanding of the broad scheme
seems to characterize the work of experts, and contrasts sharply with the
work of students who participated in the research.
Dalbey, Tourniaire, and Linn also identify some of the
characteristics and techniques of expert programmers (e.g. programming
templates, procedural skills), and attempt to embody one of these
planning techniques in the programming curriculum of eighth grade novice
programmers. In contrast to an instructional approach which emphasizes
language features, they provide instruction in structured diagramming -a
representational format for planning a program prior to beginning coding.
While all students were able to learn this technique, many had difficulty
transferring the technique to novel problem types, and many failed to see
-2
-129
the value of this activity. The issue of transfer of skill, then, is
complex even within the domain of programming problems. In addition,
the techniques valued by experts may have a complex relationship to the
types of programming supports that are effective with novices. The
identification and adoption of expert techniques may not be the critical
factor for novices. Understanding the conditions in which students
perceive a programming tool as useful for problem solution would seem
essential in the building of an effective instructional sequence.
While Erlich, Abbott, Salter and Soloway find some evidence of
transfer of procedural programming skills to non-programming algebra
problems, they also note the intricacies of data interpretation. The
influence of gender relative to programming experience they document is
not readily understandable in the improvements noted among male
programming students. These sorts of gender differences have been
widely noted in mathematics achievements (cf. Brush, 1980), and with
computers (cf. Hawkins, 1984). The evidence of gender differences is
not surprising, as performances in these domains seems complexly related
to differential expectations and performance conditions. These resul+s
add an iteresting note of complexity to the understanding of conditions
of transfer.
Pea suggests that research concerning the transfer of skills must be
based on a thorough understanding of the functional activities of a
"culture". How does the group enlist particular skills in the carrying out
Noting the findings of a research
of internally meaningfully activities?
program that failed to find evidence of improvement in planning skills
among elementary students after a year's programming experience, Pea
recommends both changes in the instructional environment and in the
130
places one looks to find generalization of ability. Programming per se is
not a privileged environment in which general skills are "naturally"
acquired. Rather, bridges must be built between the programming work
and other problems in a learning setting mutually constructed by teachers
and students.
These studies take on the difficult task of examining the conceptual and policy
questions raised by the enthusiasm in the educational community for
programming. In addition to the implications of their findings, the
papers articulate paths for yet deeper probing. Among the most salient:
What are the goals of programming work for students at different
levels? One angle of approach was adopted by Kurland et al: what are
experts able to do? This information about end-state, or exp :rt practice,
should help to shape the programming environment that students
encounter. Additional analyses of the characteristics of expertise are
essential. Complementary approaches are also necessary. The
development of goals is an issue of values and priorities in a larger

cultural surround. As teachers rightly note, when programming comes
into the curriculum for all students, something else goes out. In light of
the time commitment and evidence of possible prerequisite abilities, is the
pursuit of programming expertise a reasonable goal appropriate for all
students? If it is a privileged environment for learning problem-solving

skills, its adoption can be well justified. The emerging evidence,
however, is equivocal and the conditions of transfer are certainly more
complex than initial claims imply.
An analogy to the historical conditions of written literacy is useful in
thinking about the goals for programming. Resnick and Resnick (1977)
trace the historical development of definitions of literacy in terms of
reading comprehension. An examination of changing criteria for literacy
in Europe and American reveals that sharp shifts occurred, relative to
social conditions and valued skills. Our current definition of literacy is a
recent one (at most three generations old)--from memorization and reading
in order to decipher, to reading in order to comprehend and develop
problem-solving capacities; from education of an elite, to mass literacy.
The current criterion defines reading for comprehension as a standard for
instruction, and adopts notions of functional literacy for participation in
today's society as goals (e.g. ability to read a newspaper, fill out a
form, read an instructional manual).
Although programming is a much more recent cognitive technology,
underlying some of the curriculum approaches is a notion of general, mass
"literacy" in these skills. The development of goals for this area, I
believe, rests not simply on an analysis of the abilities of experts, but

also on a analysis of necessity and meaning of functional programming
literacy in the broader society. In the rush to develop a programming
curriculum, this analysis for education has yet to be done.
Where does one look for effective instructional variables? There is
a nice correspondence in these papers between promising conditions of
instruction for programming, and possible conditions of generalization.
There is a focus on the need to explore not simply the means for
instructing ;ti. 'nds of a programming language and the
practice of coding, but on Lilt: eeri to teach planning and structure in
programming concepts. This then points to a need to explore
programming environments (e.g., editors, planning aids, pseudo code,
debugging assistants, trace facilities), and means for encouraging
students to think carefully about the problem and the overall structure of
the program prior to entering code. Kurland et al, and Dalbey et al
have already begun work in this direction. Their results indicate the
need for more. Rather than simply comparing the relative advantages of
languages, additional research concerning the effects of various
representations and programming supports for novices is required.
Perhaps it is through structured supports in programming problems that
generalization can be seen by students, and therefore by researchers.
How else can transfer be examined? Thus far, examinations of
transfer of programming skill have largely been based on an abstract
analysis of the skills entailed in accomplishing a working program
(cf. Pea 11 Kurland, 1984). Thus, one looks for transfer of procedural or
planning abilities that are assumed to be evoked in programming practice.
One develops tasks that are intended to evoke these skills--the major
requirement of such tasks is that component skills are invited in
performance. The quality of such skills in students' performances is then
assessed. However, as Pea points out, programming is a functional
activity like other skilled performances of a culture. It takes place in
particular circumstances for particular reasons. Relatively little attention
has been paid thus far to the types of problems or programming goals
that students are asked to engage in. Programming practices were
developed in order to do particular classes of tasks better or differently.
Programming is a topic, but it is also a particular medium or means to
solve problem: that have been identified as approp a.
It might be useful, therefore, to pay careful attention to the types
of problems students are asked to find solutions to through the medium of
programming. Many of the these tasks would have previously been done
through different representations and procedures with different media.
_6
133
(e.g. math problems done with paper, pencil and a calculator). Rather
than designing tasks with the focus on structural similarities to
programming, tasks that are functionally similar to students' programming
work might evoke a different approach that has been learned through
programming. The representations and skills practiced through use of
the technology to solve particular problems might result in new ways of
seeing those classes of problems, independent of the technology.
What are effective social conditions for learning to program? There
is evidence in these studies that social circumstances play an important
role in the programming performances we analyze. This is noted at two
levels in these papers. In terms of the larger culture, functional
activities are important in defining how skills are acquired and applied.
In addition, gender--the expectations and orientation; to performance in
particular circumstances--may be an important variable in learning and fo::
assessing skill. Cognitive questions about development, and educational
questions about implementation must take account of these larger social
circumstances. More needs to be known.
At a more immediate level, the locus of programming instruction is
the classroom. Classrooms are complex social settings in all grades. A
programming curriculum exists within a well-established set of
expectationsinstruction and interaction among classroom members. In
establishing an effective programming agenda for schools, more needs to

be known about the social structure and resources that embody the ideas.
For example, Kurland mentions the example of "code stealing", where one
student borrows sections of code to construct a programcode that he
does not hilly understand. If this reliance on others' work is used
appropriately, it is possible to incorporate the exchange of work as an
_ 134
effective instructional practice. The student, for example, could be
encouraged to construct variations on the program which would require
changes in the code. Whether he knew how to vary it himself, or knew
where to get help, providing situatio.As which focus on variations in
construction and use of a particular section of code could lead to
improved understanding.
These papers are important in further defining the nature of
problems in developing effnctive instructional programs for programming
skills, with an emphasis on using this environment to teach problem
solving. They thus point to important paths for future research.
REFERENCES
Brush, L. E. (1980). Encouraging girls in mathematics. Cambridge, MA:
Abt Books.
Hawkins, J. (1984). Commuters and girls: Rethinking the issues. New
York: Bank Street college of Education, Center for Children and
Technology Technical Report No. 24.
Pea, R. D. & Kurland, D. M. (1984). Towards co hive technolo ies for
writing. Unpublished manuscript, RevrTW)a--
, Center for Children and Technology.
Resnick, D. P. & Resnick, L. B. (1977). The nature of literacy: An
historical exploration. Harvard Educational Review, 47(3), 370-385.

SD 257 441 IR 011 673 Kurland, D. Midian, Ed. Title: Docuxidit Resume

Uploaded by

SD 257 441 IR 011 673 Kurland, D. Midian, Ed. Title: Docuxidit Resume

Uploaded by

DOCUXIDIT RESUME

SD 257 441 IR 011 673

AUTHOR Kurland, D. Midian, Ed.

EDRS PRICE MF01 Plus Postage. PC Not Available from EDRS.

"PERMISSION TO REPRODUCE THIS

TO THE EDUCATIONAL RESOURCES

1984 AERA Annual Meeting

1984 AERA Annual Meeting

D. Midian Kurland 1 INTRODUCTION

This collection of papers was originally presented at a symposium entitled

At the center of the current uneasiness over the role of programming in

The five papers in this symposium examine computer programming from a

The third paper by Ehrlich, Abbott, Salter and Soloway directly

Finally, Jan Hawkins provides a commentary in which she discusses the

As the teaching of programming is initiated at increasingly early grade

Defining programming and assessing its cognitive demands is problematic

One consequence of the fact that programming refers to a configuration of

The beginner& linear style of constructing programs, whether in Logo or

Furthermore, in school classrooms we have often seen students borrow

In learning to program, the instructional environment can reduce certain

Instruction is important in other ways as well. For example, students in

Finally, the social context can mediate the demands placed on an

Analysis of the Cognitive Demands of Modular Programming

We expect that procedural reasoning ability is an important skill

Decentration also may be important skill in programming since

On the basis of this rational analysis we designed a study to investigate

In the second part of the task, we designed two problems, based on a

The procedural flow of control task, therefore, involved a system of

The second demands task was designed to assess decentering as well as

(a) Inaccurate information bug: instructions given were simply

The task sheet included an area labeled "workspace", analogous to the

The task, then, could be solved by planful use of flexible modules of

In the first comprehension tasks the superprocedure was named

Both programming comprehension tasks were designed as paper and pencil

LOGO PROCEDURES: DRAWING OF SCREEN EFFECTS OF

LOGO PROCEDURES: DRAWING OF SCREEN EFFECTS OF

Again analysis of the errors of students who were partially correct

The students' written descriptions of the procedures in both the

Students were much more likely to use "analytic" than "synthetic"

Figure 5 provides more detail on the features of Logo students used to

figze A figure D fig.rce E figare F figure C figure 8 figure G

% bhp did it: 73 .51 .91 .91 .96 .60 .31

The group of figures C, E, and F was chosen by over 90% of the

Examination of correlations with subscores on the programming production

Other subcomponents of the production task that we assumed would

NO CHANGE CATCH NO FIX CATCH SOME FIX. CATCH. & FIX

WRONG .03 .06 ma" .91

AMBIGUOUS .11 .41 .nal .48

TEMPORAL .16 .11 ma* .73

EMBEDDED .29 .10 .40 .21

* This category is not applicable

Correlations of 'Demands' Measures with

In general the correlations of the demands measure were higher with

We examined the relation between math achievement level (assigned on the

Nonetheless, we did find moderate relationships between the ability to

We can best assess cognitive demands of programming when we are clear

In conclusion, we have argued that asking what are the cognitive

Programming has the potential to serve as a fertile domain in which to

New instructional approaches may dramatically facilitate the learning of

The teaching of high level, general purpose languages such as BASIC,

The situation is complicated by the misconception being perpetrated

The work reported here was supported by the National Institute of

What does it take for students to become reasonably competent

Our decision to study expert programmers was prompted by the need to

The study was conducted in two parts. In the first we conducted a

Expert Adult Programmers

All the participants began programming in their late teens or early

Upon receipt of the completed questionnaires, follow-up interviews were

A related finding concerned the relationship of prior technical ability in

A second common theme across the interviews was ro.spondents

"Actually Iwork an incredible number of hours, as most

The amount of time expert computer programmers dedicated to learning

In a related study of the work habits of "super" programmers, Molzberger

The third striking characteristic of these expert programmers was the

While planning and debugging skills were emphasized by the programmers

A clear implication that can be drawn from these interviews is that to