Results of a course combining pair programming, peer instruction, and media computation over four years. (a) one-year retention for majors
who pass introductory computing. (b) Passing rates for initially enrolled students. (c) one-year retention of initially enrolled students.
2001–2008
2008–2012
100%
71% 73% 71%
90% 87%
89%
of Students Who Pass
20%
40%
60%
80%
100%
64%
100%
82%
79%
81%
82%
82%
82%
Retention (Passing)
40%
60%
80%
60%
63%
Retention (Enrolled)
40%
60%
80%
51%
50%
51%
20%
20%
0%
0%
0%
Males Females Overall
Males Females Overall
Males Females Overall
(a) More passing
students are retained.
(b) More enrolled students
pass the course.
(c) Combined—more enrolled students
are ultimately retained.
those that worked alone, there was no
significant difference. It is important
to note that significantly more students
in the pairing section persisted to the
end and took the final. This resulted in
a higher percentage of students passing the course in the pairing sections.
It also refutes the claim that weak students fail to learn the material because
their partner does all of the work.
media computation
Georgia Tech requires all students to
take a course in computer science, including students in Liberal Arts, Architecture, and Business majors. During
the first four years of this requirement,
the overall pass rate was 78%, which is
quite reasonable. The pass rate for students in Liberal Arts, Architecture, and
Business, however, was less than 50%
on average.
Guzdial and his colleagues created
a new course just for students in Liberal Arts, Architecture, and Business programs. For these students, computing
is more about communication than
calculation. Students in these programs
most often use the computer in order
to communicate with digital media.
Media Computation was an approach to CS1 that explained how digital media are manipulated. Students
learned about loops by changing all
the pixels in a picture to compute a
negative image, or all the samples in a
sound in order to decrease the volume.
Students learned about conditionals by
removing red eye in the image without
changing any other colors, or changing
only part of a sound.
What was most exciting about
Media Computation was that our assignments were defined in terms of
computation, but the choice of what
media to use in the assignments was
up to the students. Students produced
beautiful and creative works of art—in
their CS1 class.
The result on retention was pretty
dramatic. 2 The pass rate for students
in those majors went from below 50%
in the former class to 85% in the Media Computation class. The research
evidence in the computing education
community suggests it is not just media. Giving students a context in which
to apply and understand computing
We believe each
of these approaches
addresses a failing
of traditional
introductory
computing courses.
makes it more relevant, and makes the
students more successful.
Peer instruction
Many of us have had the experience
lecturing to a class where we have explained a key concept with brilliant
clarity. We turn to the class, ask if
there are any questions, and we hear…
crickets. One-half of the class is looking at phones or laptops, one-fourth
looks utterly confused and scared, and
another one-fourth looks bored. No
one asks anything, you think “they’ve
got it” and move on. After the exam
you discover: they didn’t get it.
Peer Instruction, originally developed by Eric Mazur for teaching physics, seeks to remedy this problem by
engaging students in the learning
process. Peer Instruction modifies the
standard “lecture” to revolve around
3–5 questions per lecture. For each of
these questions, students follow the PI
process: individually think about the
problem and answer (often using clickers); discuss the question in groups;
then answer again (using a clicker).
Lastly, the instructor leads a class-wide discussion on the question and
dynamically adjusts their explanation
based on student performance.
Peer Instruction in physics has consistently shown twofold improvements
in student performance on concept
inventory exams versus standard lecture in large multi-institutional stud-