A 10 year retrospective on research on Media Computation: ICER 2013 preview

I get to teach our Media Computation in Python course, on Georgia Tech’s campus, in Spring 2014.  I’ve had the opportunity to teach it on study abroad, and that was wonderful. I have not had the opportunity to teach it on-campus since 2007.  Being gone from a course for seven years, especially a big one with an army of undergraduate TA’s behind it, is a long time. The undergraduate TA’s create all the assignments and the exams, in all of the introductory courses in the College of Computing.  Bill Leahy, who is teaching it this summer semester, kindly invited me to meet with the TA’s in order to give me a sense for how the course works now.

It’s a very different course than the one that I used to teach.

• I mentioned the collage assignment, which was one of the most successful assignments in MediaComp (and shows up even today in AP CS implementations and MATLAB implementations).  Not a single TA knew what I was talking about.
• The TA’s complained to me about Piazza.  “Nobody posts” and “I always forget that it’s there” and “It seems to work in CS classes, but not for the  other majors.”  I told them about work that Jennifer Turns and I did in 1999 that showed why Piazza and newsgroups don’t work as well as integrated computer-supported collaborative learning, and how that work led to our development of Swikis.  Swikis were abandoned many years ago in MediaComp, even before the FERPA concerns.
• Sound is mostly gone.  Students have to play a sound in one assignment based on turtle graphics.  Students never manipulate samples in a sound anymore.
• I started to explain why we do what we do in MediaComp: Introducing iteration as set operations, favoring replicated code over abstraction in the first half of the semester, avoiding else.  They thought that those were interesting ideas to consider adding to the course.  I borrowed a copy of the textbook from one of them, and read them part of the preface about Ann Fleury’s work.  Lesson: Just because you put it in the book and provide the citation, doesn’t mean that anybody actually reads it, even the TA’s.

It’s a relevant story because I’m presenting a paper at ICER 2013 on Monday 12 August that is a 10 year retrospective on the research on Media Computation.  (I’m making a preview version of the paper available here, which I’ll take down when the ACM DL opens up the ICER 2013 papers.) It was 10 years ago that we posted our working document on creating MediaComp and our 2002 and 2003 published design papers, all of which are still available. We made explicit hypotheses about what we thought Media Computation would do.  The ICER 2013 paper is a progress report.  How’d we do?  What don’t we know?  In hindsight, some seem foolish.

• The Plagiarism Hypothesis:  We thought that the creative focus of MediaComp would reduce plagiarism.  We haven’t done an explicit study, but if we found a difference with statistical significance, it would be meaningless.  Ten years later, still lots of academic misconduct.
• The Retention Hypothesis: Perhaps our biggest win — students are retained better in MediaComp than traditional classes, across multiple institutions.  The big follow-up question: Why?  Exploring that question has involved the work of multiple PhD students over the last decade, helping us understand contextualized-computing education.
• The Gender Hypothesis: We designed MediaComp based on recommendations from people like Jane Margolis and Joanne Cohoon on how to make an introductory CS course that would be successful with women.  Our evidence suggests that it worked, but we don’t actually know much about men in the class.
• The Learning Hypothesis:  We hoped that students would learn as much in MediaComp as in our traditional CS1 class.  Answering that question led to Allison Elliott Tew’s excellent work on FCS1.  The bottom line, though, is that we still don’t know.
• The More-Computing Hypothesis: We thought that non-CS majors taking MediaComp would become enlightened and take more CS classes.  No, that didn’t really happen, and Mike Hewner’s work helped us understand why not.

There are two meta-level points that I try to make in this paper.

• The first is: Why did we think that curriculum could do all of this, anyway?  Curriculum can only have so much effect.  There are lots of other variables in student learning, and curriculum only touches some of those.
• The second is: How did we move from Marco Polo to theory-building?  Most papers at SIGCSE have been classified as Marco Polo (“We went here, and we saw that.”)  MediaComp’s early papers were pretty much that — with the addition of explicit hypotheses about where we thought we’d go.  It’s been those explicit hypotheses that have driven much of the last 10 years of work.  Understanding those hypotheses, and the results that we found in pursuit of those hypotheses, have led us to develop theory and to support a broader understanding of how students learn computing.

Lots of things change over 10 years, and not always in positive directions. Good lessons and practices of the past get forgotten.  Sometimes change is good and comes from lessons learned that are well worth articulating and making explicit.  And sometimes, we got it plain wrong in the past — there are ideas that are worth discarding.  It’s worth reflecting back occasionally and figuring out how we got to where we are.