Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

Handling Heterogeneity in Programming Courses for

Freshmen

Dino Capovilla, Marc Berges, Andreas Mühling and Peter Hubwieser


TUM School of Education
Technische Universität München
Arcisstr. 21, 80333 München, Germany
Email: {dino.capovilla, berges, andreas.muehling, peter.hubwieser}@tum.de

Abstract—One of the biggest challenges of the computer- with no prior experience are forming the largest part of the
science department at our university is handling the enormous drop-outs in other studies: “Little or no pre-college exposure to
heterogeneity of freshmen concerning both their previous pro- computer science programming” is given as the primary reason
gramming abilities and their learning behaviors due to their for dropping out of CS studies [3]. Besides the differences in
biographical and social background. programming experience, the expectations that students have of
In this paper, we present the design and evaluation of a computer science can vary enormously [4]. These are typically
preliminary programming course based on the teaching method formed by the students’ social background and can also result
of Mastery Learning that is particularly suited for groups of in different ways of learning or approaching CS studies. This,
students characterized by considerable diversity. Trained peer in turn, can lead to great differences in “problem understanding
tutors closely guide the participants through a step-by-step and reflection about the problem and the solution” [5].
programming exercise. We tested the method in several courses
that ran for two and a half days four weeks before the start of To support freshmen and prevent them from dropping out,
lectures. We collected data from two different surveys(N = 200 we offer special courses on a voluntary basis. These courses
and N = 300, respectively). last two and a half days and take place four weeks before the
First, we quantified the considerable differences concerning start of the regular CS1 lectures. Here, we are focusing on
the prior experience in programming of the participants. Second, the students’ self-perception of their programming ability. The
we succeeded to show that the outcome of our method is methodology is based on the idea of Mastery Learning and
independent from different sensory preferences and different a very intensive support from peer tutors. These tutors were
computer-usage behaviors of the students. Third, the results of instructed to guide the participants in small steps through a set
the survey demonstrate that our method is suited to increasing of programming assignments. The workbook mainly contained
the self-perception of programming ability. This helps freshmen reproducible examples and exercises and the project was split
to overcome initial self-doubts when beginning their CS studies. into several small parts. Proceeding to the next assignment was
only allowed after mastering the current one.
I. I NTRODUCTION
In a study presented in this paper we measure the impact
Beginning CS students arrive at university with a diverse
and success of this approach. We investigated how the success
background in programming experiences. Also, their respective
of our course depends on a student’s sensory preference and
social background further increases diversity e.g. concerning
computer-usage behavior and how it influences the (subjec-
their ability to adapt to a new way of learning when coming
tively judged) gain in knowledge and perceived success.
from secondary schools into university. Handling this hetero-
geneity is one of the biggest challenges of the CS-department
II. BACKGROUND AND C OURSE D ESIGN
at our university. This heterogeneity is, for example, due to
students’ specific prior experiences and approaches to learning A. Context and Situation
[1]. These can also be influenced by various disabilities. Due
All students who enrolled for a major in CS were invited
to the implementation of the “UN Convention on the Rights of
by the CS department to take part in our course. Also, the
Persons with Disabilities”, we expect the number of students
websites of the university informed potential freshmen about
with disabilities to increase substantially in our courses, as
the courses. Participants had to register online in order to
the UN-BRK [2] demands that universities must accommodate
take part. During the registration, we asked them about their
people with disabilities as part of their regular study programs.
prior programming experience. This was done by specifically
The research questions we are trying to answer in this asking them how many lines the biggest program has that they
paper are twofold. First, we want to quantify the extent of developed. They could choose between three levels: (L1) “I
heterogeneity concerning several influencing factors. Second, have no experience at all”, (L2) “less than 100 lines of code”,
we want to find out, whether or not the method of Mastery and (L3) “more than 100 lines of code”.
Learning can be successfully applied to alleviate the differ-
We formed groups based on this self-assessment in order to
ences before the start of lectures.
prevent less experienced students from getting frustrated by the
The programming abilities of our freshmen range from speed of experienced participants in their groups - a problem
no experience at all to almost professional expertise. This that has been noted for CS1 courses by e.g. [6]. Conversely,
heterogeneity becomes especially relevant, since the students more experienced students may then progress at a faster pace
IEEE,
c 2015. This is the author’s version of the work. It is posted here by permission of IEEE for your personal use.
Not for redistribution. The definitive version was published in 3rd International Conference on Learning and Teaching in
Computing and Engineering (LaTiCE), 2015.
http://dx.doi.org/10.1109/LaTiCE.2015.18
in order to prevent them from getting bored. Each group had content into manageable units. The students process these units
about 11 participants. Table I shows the number of participants autonomously in a self-allocated time period. The learning
and the number of groups for each level in the years of 2012 success is examined by an instructor after each unit. If the
and 2013. student masters the unit, she proceeds with the following
TABLE I. N UMBER OF PARTICIPANTS FOR EACH LEVEL OF PRIOR
one. Otherwise, the instructor helps her by giving further
PROGRAMMING EXPERIENCE . T HE NUMBER OF GROUPS IS GIVEN IN explanations or suggesting additional exercises. In our case,
PARENTHESES . the units are presented in form of a workbook composed
Level 2012 2013
of theoretical parts, examples, exercises, and corresponding
L1 97 (10) 161 (15) solutions. We favored this approach over a “‘minimal input”
L2 46 (5) 80 (5) design, as it as been observed to be problematic - especially
L3 55 (6) 76 (6)
Σ 198 (21) 317 (26) for science education [14].

The heterogeneity regarding the prior programming expe- To successfully create a sense of achievement for the stu-
rience is rather large, as our results show. The results of Table dents, the “achievement-related self-concept”, which is influ-
I taken together with the results of a questionnaire (see next enced by the “affective entry characteristics” [15] are of major
section) are shown in Table II, both in absolute and in relative importance. Bloom assumes that the learning outcome depends
terms. up to one quarter on these affective entry characteristics.
Such characteristics are formed essentially by the judgments
TABLE II. S ELF - ASSESSMENT OF PRIOR PROGRAMMING EXPERIENCE : of parents, teachers and peers [15]. It can be assumed that
R EG . = N UMBER OF REGISTRATIONS ; Q UEST. = Q UESTIONNAIRE programming is seen by many as some form of “talent” - or
Level Reg. 2012 Reg. 2013 Quest. 2013 may at least often be associated with intelligence. This would
L1 97 (49%) 161 (51%) 93 (44%) explain that “not being able to program” is a typical reason
L2 46 (23%) 80 (25%) 51 (24%)
L3 55 (28%) 76 (24%) 68 (32%) given by drop-outs.
Σ 198 317 212
As Dweck and Leggett show [16], learning outcomes are
heavily influenced by the subjectively perceived possibility
B. Mastery Learning of changing one’s talents. They show that if learners believe
Several studies have shown that freshmen consider it very that intelligence (aptitude) is not malleable but fixed, their
difficult to learn how to program [7], [8], [9]. Since program- perceived personal ability to solve an assignment will be
ming requires practical training and application of theoretically much lower than if they believe the opposite. This has also
learned concepts (investigated more closely in [10]), it is to be been shown to influence beginning students in their choice in
expected that inexperienced programmers are facing a greatly favor or against a CS-major [6]. The self-perceived ability is
increased workload at the beginning of their studies. Also, influenced essentially by the judgments of parents, teachers
the beginning of studies at university marks a breakpoint for and peers, too. This underlines the crucial role of teachers
students as they’re coming from secondary schools that work because their comments and behavior will foster or obstruct
rather differently. the students’ confidence in their aptitudes.
The primary purpose of our courses is not to increase the Our method of creating worksheets to present the necessary
actual programming abilities in order to make the group of background information to the students allows the aforemen-
CS1 students more homogeneous. This is hardly possible in tioned prerequisite of an individualized presentation that is
two and a half days and we also don’t want to get ahead of the suited to each student’s needs - as long as they’re able to read
CS1 course. Instead, the aim of the course is to increase the and write. Graphical representations can be supplemented by
self-perceived programming ability of all students independent textual descriptions for visually impaired persons, or - as in
of their prior knowledge and their biographical and social our case - left out completely. By supplying the worksheets
backgrounds. Particularly, inexperienced freshmen should be in a digital form, all sorts of special input and output devices
motivated to learn programming hands-on right from the can be used with them, accommodating the special needs of
beginning and (even more important) associate programming students with (visual) impairments. By using simple speech
with a positive experience. (or translating the worksheets into English) students from
To implement such a course, two prerequisites must be foreign countries or deaf or hard of hearing people can be
met: First, the topics must be presentable in an individualized accommodated.
manner in order to reach all students irrespective of their prior
knowledge and social background. At least within the bounds In contrast to typical Mastery Learning approaches, we
of “beginning CS students”, so we can assume, e.g. that they greatly emphasized the role of the tutor and also reduced
are able to read. Second, the course must be designed in a way the group size considerably. The tutor is needed to help the
that gives every student the possibility of experiencing a sense students gain a sense of achievement and also to support those
of achievement. learners, who are more oriented towards verbal communication
in their sensory preferences. As shown by Cottam et al.
Our method is based on the theory of Bloom’s Mastery [17], peer tutors are well suited for this task. They found in
Learning1 [13]. Mastery Learning works by dividing learning their study of the effect of peer tutoring in CS1/2 significant
1 There are different progressive educational approaches similar to Bloom’s. differences in the exam grades and consider peer tutoring
The basic idea can be attributed to Washbourne [11]. Keller developed the as an intervention to improve performance and retention in
PSI Personal System of Instruction [12] at almost the same time as Bloom. computer-related courses.
C. Course Description having produced a working program alleviates the (sometimes
overwhelming) complexity of programming that beginners
The courses that have been investigated in the study pre-
often feel. Instead we want to achieve an increase in their per-
sented here have taken place between four and two week before
ceived programming abilities. If this is achieved, we hope that
the start of regular lectures. There were four identical runs of
“programming” is associated with a positive social experience
the course taking place in direct succession. Each course lasts
and the sense of an achievement.
15 hours, split up into 5 sections of 3 hours each. The courses
took place in a computer lab at our faculty. The students The students were free to work individually or to collabo-
worked with Java and the NetBeans IDE and could also bring rate with other participants. In order to avoid that participants
their own laptop. Java and NetBeans was chosen because the may not be able to accomplish individually what their group
CS1 course that the students are attending afterwards are using can [20], everyone had to develop an independent program.
the same technology. They had to upload the source code on our server at lunch
Based on several studies that describe the concept of break and at the end of every day. Finally, each participant
“arrays” to be very hard for novice programmers (e.g. [18], created up to five consecutive versions of his program. The
[19]), we were looking for an assignment simple enough to uploaded code was used to calculate the time needed to
be understood within the short amount of time and involving complete the assignment. This was possible by subtracting the
basic concepts of programming with particular attention placed start time from the time stamp of the last code package. We
on arrays. We chose to have the students implement the board only used an aggregated form of this data in the form of 0 to
game “Mastermind” as it fulfills these requirements. 6 hours, 6 to 9 hours, 9 to 12 hours and 12 hours and more.

In order to make this assignment suitable for Mastery The workbook is based on the textbook Java: How to pro-
Learning, we split it into a basic version and two extensions gram [21]. It is composed of 13 pages and divided in 7 units.
divided in several steps. This is the summary of the assignment The first four units contain a description of the assignment
given: to solve and the basic rules and concepts of programming in
Java. These four units are presented with many examples and
Basic version: small exercises with solutions. Additionally, the students had
access to the internet. Unit 5 consists of writing an executable
1) Set a four digit number like 1453 or 9467. This program with some basic functionality. We considered the
number is hard coded in the source code. The player achievement of this level as fulfilling the expectations. Unit
has to find out this number. 6 and unit 7 were considered as exceeding the expectations
2) The player enters a four-digit number as a guess. and exceeding the expectations by far, respectively.
After each guess the computer responds with how
We offered the workbook in two languages (German and
many of the guessed digits are contained in the
English) and also offered access to a digital version besides the
number and how many digits are already at the correct
printed one. As pointed out above, we explicitly left out graph-
position.
ical representations and visualizations due to the problems
3) Step 2 will be repeated until the player guessed the
associated with visually impaired students. As visualizations
right number or a maximum of 12 repetitions has
are a common teaching aid in computer science, it is not
been reached.
hard to find it elsewhere, if desired. The peer tutors helped
Extension level I: the students to find such material online and in books.

1) The number should be chosen randomly by the com- Each group was supervised and guided by a peer tutor. As
puter. tutors we employed experienced students that convinced us of
2) The length of the number should be set by the player their CS-abilities as well as their soft skills. All these students
before the start of the game. had already attended a regular 3-day preparation course for
student tutors in which they learned basics of education and
Extension level II: teaching methods. The tutors were given a written description
of our teaching method and their tasks together with a written
1) The program should be amended such that the player “code of conduct”. Additionally, they were briefed by us right
sets the number and the computer tries to guess it. before the start of their courses. We stressed the important
Repetitions should be displayed on the screen. aspects of their job again and answered any questions that
We used a special Java class, that simplifies the program- they had about their tasks. We advised the tutors to give
ming by providing a simple (procedural) API that can be a brief introduction of the assignment, of the use of the
used without going into the details of object-orientation2 . For workbooks, and of the concept of Mastery Learning. Also,
example, the class offers (static) methods like readInt(), they should stress right from the beginning the possibilities
readString(), or write(). of using the internet and additional literature. Above all, the
tutors should keep a close eye on the students with little prior
Of central importance is that the participants should all knowledge and on those with major difficulties. To locate these
directly experience a feeling of self-efficacy when taking their students they used a report form of the individual progress
first steps into programming. Depending on a student’s prior (only for themselves) and were asked to pay attention to their
experience, the peer tutors decide how much (and what kind interactions and behavior.
of) support they are offering. We hope that the feeling of
The forms simply collected, for each student, the units
2 URL will only be included in final paper due to anonymity of study that this person has successfully completed already.
The students had to report to their tutor whenever they had and are typically learning best by concrete experiences and
finished a unit. Therefore, a lack of progress is directly active experimentation. The learning type according to the
observable by the tutor in this form. In these cases the tutors VARK model is tested by a series of questions that are
should specifically offer additional help to the students and presenting real-world, everyday situations (e.g. “someone is
try to identify what problems they are currently facing. We asking for directions, what would you prefer to do?”) and
had a response rate of approximately 70% concerning theses providing four courses of action (e.g. “showing the way on
progress forms. We used them to evaluate the success of the a map”) which are considered to correspond to one of the four
participants. We verified the validity of the data of the progress types. The responses to theses questions are then summed and
forms by examining the code of 10 participants chosen at the maximal value among each of the four types is the type
random that has been uploaded. of the person. If there is no clear maximum, i.e. if there are
several types with the same value, the person is considered to
III. DATA C OLLECTION be of a multi-modal type.
A. Questionnaire B. Exam results
In 2013, we collected personal data of the participants We used the final grades that the students received after
by an online survey (using LimeSurvey3 ). Participation was their CS1 course for our survey as well. We grouped the exam
voluntary and anonymous. 212 of 250 participants left a participants into four groups, three corresponding to the levels
complete response that was also deemed “valid” by us, in the of prior programming experience (L1 - L3) as assessed by
sense that the responses were plausible. our survey and a fourth group for those students who did not
The participants were asked for their gender, course of participate (NP) in our courses. The exam was split into two
studies, the self-assessment of their programming skills (L1- parts - one theoretical and one practical. We evaluated these
L3, as described above), and the grade of their university two parts differently. In order to compare the results of the
entrance qualification. Courses of study with less than 20 single groups applying the Two-Sample Welch t-test [23] we
participants were aggregated as “miscellaneous”. Also, we ag- used the scores achieved in the exam instead of the actual
gregated the grades of their entrance qualification into “high” grades.
and “low” performer - excluding the middle ground in order
to get a clearer picture of the two extremes. IV. R ESULTS
A. Course Performance
We also asked them about their computer-usage behavior.
The question was: “How much time of your total time spent at Regarding the time needed to complete the whole assign-
a computer is allocated to the following: Playing games, Pro- ment we formed four intervals, as described above. We had six
gramming, Using social media, and Other.” For each category, persons leaving the course prematurely. These were assigned
the participants could chose on a 5 point scale ranging from 1 to the last interval. TableIII shows a cross-tabulation of the
= (almost) no time to 5 = (almost) all the time. Each participant prior experience and the amount of time needed to complete
that has given a value of 3 or above to exclusively one activity the task.
was taken by us to “belong” to one of three categories formed TABLE III. T IME NEEDED TO COMPLETE THE TASK DEPENDENT ON
by these three options. The category “other” was only used to PRIOR PROGRAMMING EXPERIENCE . S HOWN ARE THE NUMBER OF
prevent a bias by the participants towards our three activities, PARTICIPANTS IN EACH CATEGORY.
but we disregarded all students belonging to this category. Level [0h..6h] ]6h..9h] ]9h..12h] ]12h..∞]
L1 0 0 6 54
Additionally, the students should assess their subjectively L2 1 6 5 5
perceived learning effect and sense of achievement. We asked L3 5 6 14 16
them to rate their general increase in knowledge, their increase Σ 6 12 25 75

in programming abilities and their subjectively felt sense of


achievement again using a 5-point scale ranging from 1 = The mean value of the time needed to complete the
nothing to 5 = very large. assignment was more than 10 hours (about 700 minutes, with a
standard deviation of 171 minutes). All in all, the values ranged
Finally, the last section contained the 48 items of the from 336 to 911 minutes. Those students who completed
VARK learning style test. The VARK model is based upon every unit (64 persons) have a mean value of 667 minutes
several sensory preferences regarding the reception of infor- (sd = 166) work time. About 5% of the participants completed
mation. Flemming and Mills[22] are distinguishing between the assignment within the first 6 hours (i.e. after the first day).
four different types: visual (V), aural (A), read/write (R), and An additional 10% completed the assignment within the first
kinesthetic (K). A person can also have several preferences, 9 hours and another additional 21% finished after 12 hours
which are called multi-modal combinations between these (i.e. after two days). The remaining 64% either made it within
types. A person that is of learning type A prefers an acoustic the allocated 15 hours, did not complete the assignment or
stimulation and is mostly learning from listening. Learning dropped out.
types V in turn prefer a visual stimulation and are typically
very prone to visualizations. A special type of visually oriented Concerning the individual progress of the participants, we
learners are those of type R - they need a visual stimulation but examined the results of the progress forms filled out by the
preferably in the form of written text. Finally, kinesthetic or K tutors. This allowed us - in particular - to find out how many
types are combining several channels for information reception of the participants reached the basic level as well as the two
additional levels of the course. Table IV shows the data both
3 www.limesurvey.org in absolute and relative quantities.
TABLE IV. E XAMINATION OF THE PROGRESS FORMS . P = NUMBER OF TABLE V. R ESULTS OF THE GROUP COMPARISONS : KG = GAIN IN
PARTICIPANTS ; OK = EXPECTATIONS FULFILLED ; OK + = EXPECTATIONS KNOWLEDGE , AG = GAIN IN PROGRAMMING ABILITIES , SA = SENSE OF
EXCEEDED ; OK ++ = EXPECTATIONS FAR EXCEEDED . ACHIEVEMENT. F OR COURSES OF STUDY, BUSINESS ADMINISTRATION
(BA) WAS COMPARED TO COMPUTER SCIENCE (CS) AND GAMES
Level P ok ok+ ok++ ENGINEERING (GE). T HE COMPUTER USAGE BEHAVIORS ARE
2012 L1 76 72 (95%) 58 (76%) 23 (30%) SOCIAL - MEDIA (SM), PROGRAMMING (PR) AND PLAYING GAMES (PG). *
L2 25 25 (100%) 25 (100%) 19 (76%)
L3 52 49 (94%) 47 (90%) 37 (71%)
= p < 0.05; ** = p < 0.01; *** = p < 0.001
Σ 153 146 (95%) 130 (85%) 79 (52%)
Groups KG AG SA
2013 L1 102 91 (89%) 76 (74%) 36 (35%)
L2 - L3 3.6 < 2.8*** 3.6 > 2.4*** 4.0 > 3.7
L2 14 14 (100%) 13 (93%) 13 (93%)
L3 - L1 2.8 < 3.9*** 2.4 < 3.8*** 3.7 < 3.8
L3 21 21 (100%) 18 (86%) 15 (71%)
female - male 3.8 > 3.4* 3.7 > 3.2** 3.8 > 3.7
Σ 137 126 (92%) 117 (85%) 64 (47%)
BA - CS 3.9 > 3.2* 4.0 > 3.0*** 3.5 < 3.9
BA - GE 3.9 > 3.4 4.0 > 3.2* 3.5 < 4.0
SM - PR 3.6 > 3.0 3.5 > 2.6* 3.6 < 4.0
PG - SM 3.6 > 3.0 3.3 < 3.5 4.0 > 3.6*
B. Exam Results
Interestingly, our data shows that the participants of the
course show a significantly better result in their CS1 exam For the computer-usage, 69 of the 213 students were
at the end of the first semester when compared to the others. assigned to the pattern of playing games. 18 were assigned
However, this may be due many reasons beyond the effect of to programming and 67 to using social media. All the others
the course per se, so the results have to be treated with some were not assignable to any of the three types unambiguously.
caution. The data shows that the “‘non-programmers” - on average -
judge their gain in knowledge and their sense of achievement
More specifically, the CS1 exam consists of two separate as higher than the programmers.
parts, namely a practical and a theoretical one. We were able For the VARK learning types, we were able to assess a
to match the data of 160 participants of the practical part to single type of 110 of the 213 course participants. The others
their respective self-assessments at the beginning of our course. were multi-modal types and disregarded for analysis. Overall,
For the theoretical part we were able to match 156 students. we had 20% of the kinesthetic type, 10% of the read/write
When comparing the results to the group of students that did type, 9% of the aural type, and 4% of the visual type. However,
not take part in our courses (487 and 429, respectively) the there were no significant differences to be found between any
following shows: For the practical part, the course participants of these groups.
reached (on average) 84% of the available points, whereas the
non-participants reached 78%. The difference between means V. D ISCUSSION
is highly significant with p < 0.003. Interestingly, for the
theoretical part, there is no significant difference between the As expected, the heterogeneity concerning the prior pro-
groups - participants reached an average of 63% and non- gramming experience are considerable, as Table II shows.
participants reached 62% in the mean. Roughly half of the freshmen have no experience at all,
whereas about a fourth have already considerable prior pro-
gramming experience. The amount of time that the students
C. Comparison of Different Groups of Students took to complete the course assignment (Table III) as well as
their individual paths of progress (Table IV) both reflect this
In order to compare how well our course design works heterogeneity.
in the light of heterogeneity, we compared different groups
dependent on their prior programming experience and on their Some students were able to complete the assignment in less
computer-usage behaviors. Specifically, we compared the three than 6 hours while others were not finished after 15 hours.
levels of experience with each other, the gender, their course Also, there are relatively more students with prior program-
of study and the three categories of computer-usage. We took ming experience that reached the final level of exceeding our
the results of our questionnaire, i.e. the self-assessed gain expectations by far. Clearly, we cannot fully account for the
in programming knowledge and abilities and their sense of amount of help that the peer tutor has given each student
achievement, as described above. The results are shown in along the way to the final solution. An “over-eagerness” of
Table V. For better readability, only those comparisons that peer tutors in their helping younger students has been reported
show a significant difference in one of our tests are shown. before [20]. Nevertheless, it is reasonable to expect the prior
In other words, since the comparison between the groups L1 programming experience to reflect back onto the amount of
and L2 are not shown, there were no significant differences time needed and on the level that is reached by a student. Our
regarding any of the three items we investigated. data shows exactly this.
There were no differences found between the low- and It is interesting to note that the course participants are
high-performers (using the grade of their university entrance actually performing better in the practical part of their CS1
exam). We selected 41 as high- and 49 as low-performers. exam. Clearly, we cannot deduce anything from this fact per
se, but it may clearly show an artifact of having attended our
Concerning the course of study, we had 213 students of course. It might also be that our course (as it is voluntarily)
computer science, 61 students of games engineering, and 30 only attracts the eager and active students. However, in this
students of business administration. According to the results, case, we would also expect to see a difference in the theoretical
the students of business administration felt a significantly part of the exam. One possible explanation is that our course
higher gain in knowledge and abilities when compared to the attracts students that have a practical approach to programming
CS students. only.
When considering the subjectively felt gain in knowledge design is reflecting the very specific constraints of offering a
and abilities, we can see that particularly the students with no time limited short course right between the students secondary
or little prior programming experience see the most increase school life and their new life at university. Nevertheless, our
(Table V). This is in accordance with literature. For example, results show that the design works well over range of factors
in [24] it is reported that Mastery Learning is showing the of diversity, like the sensory preference of the students.
highest impact for less apt students. In our case the prior
programming experience reflects this. As there has been no
significant difference between the two lower levels of prior R EFERENCES
experience, it seems reasonable to merge these two groups in [1] D. Capovilla, J. Krugel, and P. Hubwieser, “Teaching algorithmic think-
subsequent runs of this course. ing using haptic models for visually impaired students,” in Learning and
Teaching in Computing and Engineering (LaTiCE), 2013. IEEE, 2013,
The significant difference that was identified between fe- pp. 167–171.
males and males is explained by their respective prior pro- [2] UN General Assembly, “Convention on the rights of persons
gramming experience: Only 6% of all female participants have with disabilities: resolution / adopted by the general assem-
chosen the highest level of prior experience, wheres 45% of bly,” 2007, a/RES/61/106, www.unhcr.org/refworld/docid/45f973632.
the male students have chosen this level. This may also explain html (03.10.14).
the difference between the students of business administration [3] M. Biggers, A. Brauer, and T. Yilmaz, “Student perceptions of computer
and computer science, since it can be expected that the non- science: a retention study comparing graduating seniors with cs leavers,”
in SIGCSE Bull., vol. 40(1). New York, USA: ACM, 2008, pp. 402–
CS students will, on average, have less prior programming 406.
experience than the CS students. The same holds for the
[4] M. Knobelsdorf and C. Schulte, “Computer science in context: path-
computer-usage behaviors. Those students who fall into the ways to computer science,” in Proceedings of the Seventh Baltic Sea
programming category will most probably also have more prior Conference on Computing Education Research, ser. Koli Calling ’07,
programming experience, on average. vol. 88. Darlinghurst, Australia: Australian Computer Society, Inc.,
2007, pp. 65–76.
As there are no significant differences for the several [5] A. J. Gomes, A. N. Santos, and A. J. Mendes, “A study on students’
types of sensory preference, we can conclude that the single behaviours and attitudes towards learning to program,” in Proc. 17th
most influencing factor on the perceived gain in knowledge ACM Annu. Conf. Innovation and Technology in Computer Science
and ability is the prior programming experience. Whenever a Education, ser. ITiCSE ’12. New York: ACM, 2012, pp. 132–137.
significant difference has been found, the groups that have been [6] C. M. Lewis, K. Yasuhara, and R. E. Anderson, “Deciding to major
compared will most probably reflect this difference in prior in computer science: A grounded theory of students’ self-assessment
of ability,” in Proceedings of the Seventh International Workshop on
experience (e.g. gender). All groups that will most probably not Computing Education Research, Rhode Island, USA, 8-9 August 2011,
reflect this difference (e.g. visually oriented learning type) are K. Sanders, M. E. Caspersen, and A. Clear, Eds. New York: ACM,
not showing a significant difference concerning the perceived 2011, pp. 3–10.
knowledge gain. [7] J. Bennedsen and M. E. Caspersen, “Failure rates in introductory
programming,” SIGCSE Bull, vol. 39, no. 2, pp. 32–36, 2007.
So, future versions of this course should best focus even [8] S. Bergin and R. Reilly, “The influence of motivation and comfort-
more on the prior programming experience, for example in level on learning to program,” in Proceedings of the 17th Workshop of
the form of different assignments. It is positive to note that the Psychology of Programming Interest Group - PPIG’05, 2005, pp.
the perceived sense of achievement has been rated consistently 293–304.
high over all participants. As this has been the main goal of [9] A. Robins, J. Rountree, and N. Rountree, “Learning and teaching
the course, we see our design as successful. programming: A review and discussion,” Journal of Computer Science
Education, vol. 13 no. 2, pp. 137–172, 2003.
[10] M. Berges, A. Mühling, and P. Hubwieser, “The gap between knowledge
VI. C ONCLUSION and ability,” in Proceedings of the 12th Koli Calling International
Conference on Computing Education Research - Koli Calling ’12.
Coming back to the research questions in the introduction, ACM Press, 2012, pp. 126–134.
we can state that a considerable amount of heterogeneity arises [11] C. Washbourne, “Educational measurement as a key to individualizing
from prior programming experience, differing computer usage instruction and promotion.” Journal of Educational Research, vol. 5,
behaviors, and learning styles. Clearly, this list is not exhaus- pp. 195–206, 1922.
tive. However, as our data show, the single most influencing [12] F. S. Keller, “Good-bye, teacher,” Journal of Applied Behavior Analysis,
factor seems to be their prior programming experience, which vol. 1, no. 1, pp. 79–89, 1968.
is valuable information for subsequent changes of the course [13] B. S. Bloom, “Reasoning about naming systems,” Evaluation Comment,
design. vol. 1, no. 2, pp. 1–5, 1968.
[14] P. A. Kirschner, J. Sweller, and R. E. Clark, “Why minimal guidance
Concerning the second research question, we have pre- during instruction does not work: An analysis of the failure of con-
sented the design of a course based on Mastery Learning that structivist, discovery, problem-based, experiential, and inquiry-based
was used in order to alleviate the difficult start of CS studies teaching,” Educational Psychologist, vol. 41, no. 2, pp. 75–86, 2006.
at our faculty. The difficulty arises from several facts, among [15] B. S. Bloom, Human characteristics and school learning. New York:
them, the heterogeneity concerning their prior knowledge and McGraw-Hill, 1976.
abilities in programming. Instead of focusing on particular top- [16] C. S. Dweck and E. L. Leggett, “A social-cognitive approach to
ics that the students should learn, our course placed emphasis motivation and personality,” Psychological Records, vol. 95, pp. 256–
273, 1988.
on conveying a sense of self-efficacy and achievement. We
[17] J. A. Cottam, S. Menzel, and J. Greenblatt, “Tutoring for retention,”
hoped that this serves as a motivation in the first weeks of in Proceedings of the 42nd ACM technical symposium on Computer
their regular CS1 course and therefore decreases the effect of science education, ser. SIGCSE ’11. New York and NY and USA:
a lack of prior programming experience. Clearly, the course ACM, 2011, pp. 213–218.
[18] A. Ehlert and C. Schulte, “Empirical comparison of objects-first [20] J. Carter, D. Bouvier, R. Cardell-Oliver, M. Hamilton, S. Kurkovsky,
and objects-later,” in Proceedings of the fifth international workshop S. Markham, O. W. McClung, R. McDermott, C. Riedesel, J. Shi, and
on Computing education research workshop, ser. ICER ’09. New S. White, “Motivating all our students?” in Proceedings of the 16th
York, NY, USA: ACM, 2009, pp. 15–26. [Online]. Available: annual conference reports on Innovation and technology in computer
http://doi.acm.org/10.1145/1584322.1584326 science education - working group reports, ser. ITiCSE-WGR ’11. New
[19] S. A. Hansen, “Analyzing programming projects,” SIGCSE Bull., York and NY and USA: ACM, 2011, pp. 1–18.
vol. 41, no. 1, pp. 377–381, Mar. 2009. [Online]. Available: [21] P. J. Deitel and H. M. Deitel, Java: How to program, 9th ed. Upper
http://doi.acm.org/10.1145/1539024.1508999 Saddle River and N.J: Prentice Hall, 2012.
[22] N. D. Fleming and C. Mills, “Not another inventory, rather a catalyst
for reflection,” To Improve the Academy, vol. 11, 1992.
[23] Student, “The probable error of a mean,” Biometrika, vol. 6(1), pp.
1–25, 1908.
[24] C.-L. C. Kulik, J. A. Kulik, and R. L. Bangert-Drowns, “Effectiveness
of mastery learning programs: A meta-analysis,” Review of Educational
Research, vol. 60 no. 2, pp. 265–299, 1990.

You might also like