=Paper=
{{Paper
|id=Vol-2755/paper3
|storemode=property
|title=To Project or Not to Project: In Search of the Pathway to Object Orientation
|pdfUrl=https://ceur-ws.org/Vol-2755/paper3.pdf
|volume=Vol-2755
|authors=Arno Pasternak,Johannes Fischer
|dblpUrl=https://dblp.org/rec/conf/issep/Pasternak020
}}
==To Project or Not to Project: In Search of the Pathway to Object Orientation==
https://ceur-ws.org/Vol-2755/paper3.pdf
To Project or Not to Project:
In Search of the Pathway to Object Orientation
Arno Pasternak1 and Johannes Fischer2
1
Fritz-Steinhoff-Schule Hagen arno.pasternak@cs.tu-dortmund.de
2
Technische Universität Dortmund johannes.fischer@cs.tu-dortmund.de
Abstract. We study two different ways to learn object-oriented pro-
gramming in higher secondary education:
The first way is to structure the lessons around an overarching project.
First, a smaller object-oriented program is introduced. During the learn-
ing process, this program and the associated tasks become more and
more complex and difficult in structure and size. In the second way, the
individual lessons are structured around various small modules, each in-
troducing new aspects. We call this approach LULUM (Learning Using
Little Useful Modules). We evaluated the approaches with two groups of
students from a comprehensive school. In addition, we compared these
results with those of another group from a grammar school.
Our results are as follows:
When comparing the two approaches, the LULUM approach has shown
to be more successful. The question nevertheless arises as to whether it
makes sense to lay a relatively solid foundation in imperative program-
ming before introducing the fundamentals of object orientation.
Keywords: Higher secondary education, Object Orientation, Project,
LULUM, Modules, CS for all
1 Introduction
Computational Thinking[15] consisting of the key concepts of abstraction and
automation, has been identified as one of the human ways of thinking and pro-
vides essential insights similar to e.g. mathematical, physical and social scientific
thinking.
The great power of CS is to transfer any kind of problem with its possible
solution to the machine level by an abstraction process and thus enables it to be
executed. This is known to be a difficult process, as it has to combine the steps
modeling and programming.
1.1 Short historical Overview of the Development of Programming
languages
Two different paths to programming can be identified in the history of CS:
Copyright © 2020 for this paper by its authors. Use permitted under Creative Commons
License Attribution 4.0 International (CC BY 4.0).
One path has been taken from the machine level to the problem level. Since
programming at the machine level with its very simple memory structures for
numbers and characters was too intransparent and bug-prone, the development
of ever more complex languages moved further away from the machine level. In
the beginning, the focus was primarily on the formulation of processes and less
on data, whose structure was still very close to the machine due to the mostly
mathematical problems. In a second stage of development these data structures
became more and more complex. These languages mostly belong to the group of
imperative languages [14,11,12]. In school, programming as well as modelling is
primarily done according to the imperative paradigm. If at all, model computers
were realized according to this concept.
The second way started from the mathematical description of problems. The
concrete realization on a machine was actually not so important. The result
was declarative programming. In terms of data structure, the list was the only
built-in structure. All other data structures can be represented by this struc-
ture [14,11,10]. CS in school did not follow this path. Lists were introduced
into the school curriculum exclusively as variables with dynamically allocated
memory in imperative programming languages.
The concept of object-oriented programming takes a special role. The expan-
sion and, at the same time, the restriction compared to already existing modular
ideas consists of the reduction of modularity to data capsules, generally called
classes. These were first introduced in the Simula language [4,3] as an extension
of the record types in languages such as Pascal. As a language-defining construct,
this idea has been realized in Smalltalk. This language reduces all programming
to the Object Message Principle [8]. The process control thus becomes part of the
data structuring. In this sense, the programming paradigm can then be called
its own [14,4,3,8,11]. This approach has not been accepted in mainstream lan-
guages, despite assertions to the opposite. In other languages, object orientation
has been added to the data structuring that was common until then and has not
changed the mostly imperative flow control. The gain of such an object-oriented
view is then primarily a profit when designing large systems, and no longer as a
programming paradigm. Most of today’s so-called object-oriented languages are
therefore essentially imperative languages. In school, the contents of imperative
modeling and programming have been transferred to object-oriented description
on an imperative basis.
1.2 Educational Consequences of OOP
The more extensive and complex the structures of a programming language are,
the more complex its syntax tends to become. For the learnes, this means that
they have to master many ideas at a very early stage when being introduced to
these languages. For example, the use of classes as a special modular concept re-
quires a broad understanding of procedures and functions, including the concept
of return values.
From a didactic point of view, this requires careful reflection on how and
when the concepts of modeling and programming should be taught [6]. In the
past, teaching was mainly limited to the advanced levels of the school education
and therefore the teaching of these structures was limited to a very short period
of time. We know that teaching subjects are more effective if they are spirally
processed several times over a number of years. Regarding the teaching of OOM
and OOP, there is hardly any experience in this area [1].
In the future, teaching CS will be increasingly integrated into the lower grades
of the school and in some circumstances extended to the primary level. Therefore
it makes sense to think more about when which concept should be introduced
in the classroom.
1.3 Object Oriented Modeling and Programming in School
Since most of the lessons are currently taught in the higher secondary levels, we
limit ourselves to these grades. The aim is to introduce the basic ideas of OOM
and OOP. Currently this is mostly done in the programming language Java or
in languages similar to Java.
Of course, formulating the goals and choosing a programming language is not
sufficient. We need both didactically and methodically well justified procedures.
Of course, we do not want to teach a programming language reducing to the
syntax of a programming manual, but prefer to do it with examples, which, if
possible, should be taken from real life contexts, to make the students familiar
with the subject. A central didactical question here is whether it makes sense to
do this in a scenario typical to the work of a computer scientist, i.e. a project,
or whether it is better to choose smaller examples according to the individual
concepts to be taught, so as not to overburden the students.
1.4 Research Questions
1. Are there significant differences in learning outcomes between project-based
learning and learning with more appropriate examples? Therefore we exam-
ined two comparable groups at a comprehensive school. We limit ourselves
to relatively small projects, as they were shown to be more successful in an
earlier study [5].
2. Are there differences in the increase in learning compared with established
subjects? Thus at the end of the survey, we additionally examined our CS
students on their mathematical knowledge on fundamental facts.
3. Are there differences between grammar school and comprehensive school in
respect of learning success? We surveyed students at a grammar school to
find out how strongly the specific conditions at the examined comprehensive
school influence learning success in CS.
2 Teaching Framework
In Germany CS is not a compulsory subject, but it is an elective subject at many
schools in higher secondary education. In many cases, the courses are attended
by students who have a rather strong interest in computer science. However,
our goal is that all students are taught in CS. This will certainly have an im-
pact on the overall pace of teaching and additionally on the curriculum. This
requires the practical implementation of the ’Computer Science Teachers Asso-
ciation’ (CSTA) from the year 2016: ’It is intended to introduce the principles
and methodologies of CS to all students, whether they are college bound or
career bound after high school’ [2, p.7]. Thus the principles of Computational
Thinking [15] are implemented.
The following characteristics apply to the school in which one of the authors
teaches: Almost half of the students choose computer science, of which again
almost half are female. Almost 34 of the students have a migration background.
For most students CS is a ’normal’ subject, which means that they do not spend
more time on CS than on other subjects. Students spend very little time doing
their homework.
We chose two different approaches for teaching object-oriented modeling and
programming (OOM and OOP), as outlined next.
2.1 Approach 1: Small Projects
Sentance and Waite describe an approach called PRIMM [13], which consists
of the steps Predict, Run, Investigate, Modify, Make. Some of their ideas were
borrowed from Kölling and Rosenberg [9], but they work with (relatively) large
projects, whereas we preferred small ones.
This group started with a project with about 65 lines of code, which simulated
a card game. The game initially consisted of a stack class, modeling a stack
of cards. The individual cards had a unique number as value. In the actual
application, two stacks were created, and cards were drawn alternately from the
stacks and checked.
The basic actions are: The program was first read and analyzed. Then small
modifications and later extensions of the class were made. The target situation
of the card game was: Two players of the players class play a simple card
game. Each has a deck of cards of the stack class, which consists of a fixed
number of cards from the playing card class.
2.2 Approach 2: Learning using little useful modules (LULUM)
In this approach the concepts were introduced with small modules in form of
classes. If possible, the modules were taken from the students’ real life. However,
the contexts differed. Otherwise, the methodological procedure did not differ
from that in approach 1.
In this second group, a class for the representation of fractions was intro-
duced as a first example. Another smaller example realized an exchange office
for different currencies. Afterwards a class for the processing of accounts with
passwords was edited. Since the context was changed in each case, it was always
possible to work with shorter program texts.
2.3 Group Structure
In the 2018/2019 school year two parallel groups were set up. Both groups were
almost the same number of students, 20 and 19 respectively. The split of the
students into the two courses was made for administrative reasons and therefore
practically randomly with concern to the knowledge in computer science. How-
ever, the proportion of female students was, at one third, smaller than in recent
years, but still significantly higher than at other equivalent schools.
All students had lessons with another teacher in the previous school year.
They had no CS teaching before in the lower secondary education. Therefore,
to get to known each other at the beginning of the school year, a teaching unit
with an introduction to LaTeX was held. Subsequently, the (imperative) pro-
gramming with the data structure array introduced in the previous school year
was repeated before the introduction to object-oriented programming according
to one of the approaches described above was started. This introductory unit
had a length of 10 teaching weeks in both courses with three hours per week
each.
3 Description of Empirical Research
3.1 Description of the Questionary
We asked at several points for a description of the terms field, class, attribute,
method, object, abstract class, abstract data type, OO modeling and programming:
before the series of lessons (pre), at the end of the 10-weeks unit (post), and after
another 6 weeks (long-term) . In addition, everytime we asked for the modeling
of a car class for a car shop. So, in essence, only reproductive knowledge on
the lower levels of learning taxonomies were necessary.
The students’ statements were translated into a numerical value of a Lik-
ert scale from 1 to 6 for the assessment No answer, wrong, mostly wrong or
incomplete, partly wrong / partly right, mostly right, right.
For a reasonable evaluation of the results we evaluated the effect size d ac-
cording to Hattie [7] for the individual subquestions. In this way we cannot only
measure the results of the two groups, but also whether a successful learning
effect can be observed at all. According to Hattie, the goal is to achieve an effect
size of d ≥ 0.4 in a one-year course of about four hours per week. Since our
unit took only a few weeks, even correspondingly smaller values are of interest.
3.2 Differences between the two Teaching Approaches
We investigated the two groups Project and LULUM. Let us first look at the
learning outcomes of the students.
We can recognize in Fig. 1 that almost all students are not familiar with the
terms used in object-oriented programming at the beginning of the lesson. Fig. 2
shows that in addition to the students who have made only little progress, some
students did indeed progress. It can also be seen that the learning increase of
CS−Survey 2018/2019 Pre * Project/LULUM CS−Survey 2018/2019 July * Project/LULUM
60
60
Project Project
LULUM LULUM
50
50
40
40
Students' Percentage
Students' Percentage
30
30
20
20
10
10
0
0
no answer wrong mostly wrong part/part mostly correct correct no answer wrong mostly wrong part/part mostly correct correct
Students' Answers Students' Answers
Fig. 1. Pre Answers (before teaching unit): Fig. 2. Post Answers (after 10 weeks):
Project/LULUM Project/LULUM
the students in the LULUM group is slightly above that of the project group.
The question arises whether these results are statistically significant.
3.3 Evaluation of Selected Statistical Results
An evaluation of the effectiveness of an educationally relevant measure can be
made with the effect size according to Hattie [7, p.8]. With this metric, different
groups can be compared on a linear scale with regard to a measure.
Effect size Total class operation object car class
Project 0.48 1.51 0.28 0.33 0.74
LULUM 0.80 2.13 1.24 0.78 0.54
Table 1. Effect size d for the different groups.
d describes the increase in performance in each of the
different groups from the beginning (pre) to the end
(post) of the intervention.
Table 1 shows the learning effects (individually for the two groups) for the
four test items between the beginning (pre test) and the end of this unit (post
test) introducing OOM and OOP, and also a summary over the four items (’To-
tal’). It shows a significantly better learning effect (ddif f = 0.8 − 0.48 = 0.32) in
favour of the LULUM group compared to the project group.
A second variant for measuring the effect size is to perform a measurement
between two groups. Table 2 shows the differences between the two groups at
the beginning and end of the lecture.
The LULUM group is better in almost all values. The better value in the
pre-post scoring for developing a car-class is due to the fact that except for
Effect size Total class operation object car class
pre 0.21 0.35 -0.25 0.48 0.73
post 0.40 0.39 -0.29 0.66 0.18
Table 2. Effect size d comparing Project and LULUM.
d describes the difference in performance
between the two groups at the beginning (pre)
and at the end (post) of the intervention.
two students in the project group, who wrote nothing at all in the initial test,
all the students in the LULUM group tried a little to answer, which was then
mostly completely wrong. Otherwise, the LULUM group signs a more significant
learning increase than the project group.
We can observe that the effect size for the pre- and the post-examination
between the two groups increases from d = 0.21 up to d = 0.40. This shows also
here a clearly better learning effect (ddif f = 0.40 − 0.21 = 0.19) in favour of
the LULUM group compared to the project group at the end of the teaching
unit. This value is strongly influenced by the many wrong, mostly wrong and
missing answers. If we ignore the learning progress within the lower levels from
’no answer’ to ’mostly wrong’ and limit it to those students whose answers are
at least partially correct, we still get an effect size of d = 0.27 in comparison to
d = 0.40 by all students in favor of the LULUM group.
Unfortunately, there is a non-negligible number of students who have experi-
enced practically no learning increase. We want to know how the learning effect
of students who really show an increase in learning has developed. If we take the
liberty to exclude five of these students from each group, the learning gain of the
project group is d = 0.52 in direct comparison to d = 1.00 for the LULUM group.
The effect size of the LULUM group compared to the project group increases
from d = 0.27 at the beginning to d = 0.50 at the end of the lesson.
Overall we can answer research question 1 to the extent that there is a sig-
nificant difference in learning outcomes in favour of the LULUM group.
3.4 Other Interesting Descriptive Statistical Results
Results at the end of the activity To evaluate long term learning effects,
Fig. 3 and Fig. 4 show the results of the final test conducted 6 weeks after the
teaching unit. The results are not satisfying at all: overall, more than 50% of the
students’ comments on all questions were wrong, or the students did not answer
at all. Another approximately 15% students answered overwhelmingly wrong. It
is noteworthy that this contradicted the teacher’s feeling during the lesson and
when observing students’ activities on the computer.
It is also surprising that the most basic concept (class) shows even worse
results than concepts that build on it (operation and object). Obviously, the
fundamental idea of describing similar objects in a logical structure has not
really been understood.
CS Survey 2018/2019 ** May CS Survey 2018/2019 ** May
70
40
class
object
60
operation
car class
30
50
Students' Percentage
Students' Percentage
40
20
30
20
10
10
0
0
no answer wrong mostly wrong part/part mostly correct correct no answer wrong mostly wrong part/part mostly correct correct
Students' Answers Students' Answers
Fig. 3. Post Answers Fig. 4. Post CS Answers Summary
about the Items
class, operation, object, example carclass
A problematic situation arose from the fact that students in the lower sec-
ondary education have no teaching experience in CS. Is the rate and quantity of
learning material generally too high, so that no permanent effects are achieved
for half of the students? Is the teaching in CS generally ineffective? Or does this
result have nothing to do with the subject CS?
Comparison CS with Math We verified this by asking the students again
about 2 months after this post-survey, and additionally asked them for compara-
ble mathematical terms – function, indefinite integral, definite integral, example
in/outflow of a sea – to examine their mathematical understanding and know-
ledge at a similar level.
Math Survey 2018/2019 Math Survey 2018/2019
60
40
Function
Indefinite Integral
50
Definite Integral
30
In−/Outflow
40
Students' Percentage
Students' Percentage
30
20
20
10
10
0
0
no answer wrong mostly wrong part/part mostly right right no answer wrong mostly wrong part/part mostly correct correct
Students' Answers Students' Answers
Fig. 5. Math Answers about the Items Fig. 6. Math Answers Summary
function, indefinite integral,
definite integral, example in/outflow
We did not expect any mathematical excellence in this supplementary survey,
but the results according to Fig. 5 and Fig. 6 are also disillusioning in the end.
Also in Mathematics taught by various teachers, almost half of the students show
that they did not achieve the mathematical goals. This is surprising, given that
students have been taking math as a subject since they started school more than
10 years ago. The concept of function has been spiralled through more than five
years and still more than 50% of the students cannot sufficiently explain this
concept.
Fig. 7 compares the results and we get the answer to research question 2:
Hardly any difference can be observed between CS and Mathematics: about half
of the students have learned nearly nothing in both subjects. These results in
both Mathematics and CS cannot satisfy students, teachers and above all society.
It is to be feared that this is also true in a similar way for at least some of the
other subjects.
If these results are transferable to the majority of students and subjects, it
is necessary to identify which topics and problems should be offered in many
subjects so that most students show a greater learning effect than at present.
Comparison of Students from Different School Systems The question
arises whether this basic behaviour applies generally to all schools. Many parents
try to enroll their children directly at the grammar school, so that the compre-
hensive school often lacks the intellectually more capable students. This could
explain why the standards that apply to all schools are only fulfilled by a part
of the students at our school.
In order to examine this, parallel to the survey at the examined comprehen-
sive school we also interviewed students at a Gymnasium who had taken CS as a
school subject there. These results are not fully comparable, as we do not have an
overview of which subjects were taught at this school and how intensively. How-
ever, the same curriculum applies to all students and teachers. The composition
of the courses at the grammar School compared to the comprehensive school is
not entirely comparable. Both the percentage of girls and the percentage of stu-
dents with foreign background is more than twice as high at the comprehensive
school.
We can see in Fig. 8 that in both Mathematics and CS the students’ perfor-
mance is significantly better than in the comprehensive school. However, even
here more than 40% of the students remain in the unsatisfactory range. Since
the surveyed grammar school also has an advanced placement course (AP) in
CS, the figures in CS are not quite adequate.
If one looks only at the base course students in CS in Fig. 9, it becomes clear
that the performance in CS is also worse: more than 50% of the students do not
achieve sufficient performance, either.
In summary, research question 3 can be answered that students tend to do
better at grammar school. However, if only the students in the basic courses are
taken into study, the results come closer.
Math/CS Survey 2018/2019
40
Math
CS
30
Students' Percentage
20
10
0
no answer wrong mostly wrong part/part mostly correct correct
Students' Answers
Fig. 7. Comparing Mathematics with CS (comprehensive school)
Math/CS Survey Gym 2018/2019 Math/CS Gk Survey Gym 2018/2019
40
40
Math Math
CS CS
30
30
Students' Percentage
Students' Percentage
20
20
10
10
0
0
no answer wrong mostly wrong part/part mostly correct correct no answer wrong mostly wrong part/part mostly correct correct
Students' Answers
Fig. 8. Math/CS answers (gram. school) Fig. 9. Math/CS (gram. school w/o AP)
3.5 Interpretation of the Statistical Results
At the beginning of the investigation, about 90% of the students were not able
to answer the questions about object-oriented programming correctly. This per-
centage drops to about 60% in the project group and just over 40% in the
LULUM group. There is no doubt that the students have learned quite a bit
overall. However, about half of the students have learned practically nothing.
The questions were largely reproductive in nature. The only non-reproductive
question was answered even worse.
The approach for both groups was object-oriented programming. The goals
were the same, only the didactic focus was different. The claim of a general
education school for all students is to teach essential ideas in such a way that
the majority is able to understand the ideas to be taught, at least in principle.
This has obviously not been fulfilled.
A possible explanation is that the idea behind these lessons is too difficult for
a large part of the students at this time with their personal learning background.
It cannot be denied that the relationship between the modeling level and the
implementation level in object-oriented modeling and programming is only com-
prehensible to those who already have considerable experience in programming.
If this is correct, this learning process must be preceded by learning exactly
these structures in whatever form in the context of a spiral curriculum in the
sense of the ideas of Bruner [1].
Comparing the results of the survey at the comprehensive school with those
at the grammar school, this interpretation is substantiated. There, too, it can be
recognized that for a large proportion of the students, especially those who do
not attend a more in-depth education in CS, the object-oriented approach does
not produce the desired outcomes.
At the comprehensive school it is also shown that students who have got
to know different aspects of object-oriented modelling and programming with
different smaller examples (LULUM) have a significantly higher learning effect.
This is true for the overall group as well as the limitation to those students who
have at least partially understood the ideas correctly.
4 Summary
School has the task of conveying the essential ways of thinking. Apart from, for
example, mathematical, physical and social-scientific thinking, this also includes
Computational Thinking. According to Jeannette Wing this consists of abstrac-
tion and automation [15]. Abstraction is implemented by modeling, automation
by programming. At the modeling level, the world is represented by abstract data
types, which are modified by recursive and/or iterative flow structures. Object-
oriented modeling and programming OOM/P has become widely accepted as a
concrete technology in industrial practice.
A dominating position in didactics is to practice this approach in schools as
well. Due to the complexity of the programming language, it is therefore im-
portant to determine how this is practised in school. Therefore, we examined
whether working with small projects or with several independent examples (LU-
LUM) is more suitable.
It can be seen across all students that the assumption that the view of object
orientation, which is a very appropriate approach from the perspective of pro-
fessional computer science, causes more difficulties for students the less they are
involved in the subject CS. It is impossible to determine whether this is caused
by the inherent difficulties of object orientation, or by the high demands of the
syntax of object-oriented languages.
However, the proportion of students who did experience learning growth
is significantly larger in the LULUM group. We only examined two groups of
about 20 students each. If these results are also valid for relatively large groups
of students, a rethinking of CS didactics is necessary. Maybe in a CS for All
curriculum programming should not start with the OOM and OOP, but at much
easier requirements? Are CSUnplugged teaching units helpful in this respect?
And what is the influence of the chosen programming language.
References
1. Bruner, J.S.: The process of education. Harvard University Press (1960)
2. CSTA STANDARDS TASK FORCE: [Interim] CSTA K–12 Computer Science
Standards: Revised 2016. Tech. rep., New York, NY, USA (2016)
3. Dahl, O.J., Myhrhaug, B., Nygaard, K.: Some features of the simula 67 language.
In: Proceedings of the Second Conference on Applications of Simulations. pp. 29–
31. Winter Simulation Conference (1968)
4. Dahl, O.J., Nygaard, K.: Simula: An algol-based simulation language. Commun.
ACM 9, 67–78 (1966)
5. Fischer, J., Pasternak, A.: Comparing approaches for learning abstraction and au-
tomation by object orientation. In: Jasutė, E., Pozdniakov, S. (eds.) 12th Interna-
tional conference on informatics in schools * Situation, evaluation and perspectives.
pp. 39–47. University of Cyprus, Larnaca (2019)
6. Flores, P., Torres, J., Fonseca-Delgado, R.: Design decisions under object-oriented
approach: A thematic analysis from the abstraction point of view. pp. 89–97 (11
2019)
7. Hattie, J.: Visible Learning: A Synthesis of Over 800 Meta-Analyses Relating to
Achievement. Taylor & Francis (2008)
8. Kay, A.C.: The early history of smalltalk. SIGPLAN Not. 28(3), 69–95 (1993)
9. Kölling, M., Rosenberg, J.: Guidelines for teaching object orientation with java.
In: Proceedings of the 6th Annual Conference on Innovation and Technology in
Computer Science Education. pp. 33–36. ITiCSE ’01, ACM, New York, NY, USA
(2001)
10. Queinnec, C.: Lisp in Small Pieces. Cambridge University Press (1996).
https://doi.org/10.1017/CBO9781139172974
11. Sammet, J.E. (ed.): HOPL-II: The Second ACM SIGPLAN Conference on History
of Programming Languages. Association for Computing Machinery, New York, NY,
USA (1993)
12. Sammet, J.E. (ed.): HOPL III: Proceedings of the Third ACM SIGPLAN Confer-
ence on History of Programming Languages. Association for Computing Machinery,
New York, NY, USA (2007)
13. Sentance, S., Waite, J.: Primm: Exploring pedagogical approaches for teach-
ing text-based programming in school. In: Proceedings of the 12th Workshop
on Primary and Secondary Computing Education. pp. 113–114. WiPSCE ’17,
ACM, New York, NY, USA (2017). https://doi.org/10.1145/3137065.3137084,
http://doi.acm.org/10.1145/3137065.3137084
14. Wexelblat, R.: History of Programming Languages. Elsevier Science (2014)
15. Wing, J.M.: Computational thinking. Communications of the ACM 49(3), 33–35
(2006)