These papers were all presented at the 12th Annual International Computing Education Research conference in Melbourne earlier this month, and give a good sense of what CS education researchers are looking at and what they're finding.
Elizabeth Patitsas, Jesse Berlin, Michelle Craig, and Steve Easterbrook: "Evidence That Computer Science Grades Are Not Bimodal", 10.1145/2960310.2960312.
Although it has never been rigourously demonstrated, there is a common belief that CS grades are bimodal. We statistically analyzed 778 distributions of final course grades from a large research university, and found only 5.8% of the distributions passed tests of multimodality. We then devised a psychology experiment to understand why CS educators believe their grades to be bimodal. We showed 53 CS professors a series of histograms displaying ambiguous distributions and asked them to categorize the distributions. A random half of participants were primed to think about the fact that CS grades are commonly thought to be bimodal; these participants were more likely to label ambiguous distributions as "bimodal". Participants were also more likely to label distributions as bimodal if they believed that some students are innately predisposed to do better at CS. These results suggest that bimodal grades are instructional folklore in CS, caused by confirmation bias and instructor beliefs about their students.
This was the most thought-provoking paper of the conference, as it shows pretty conclusively that people are seeing evidence for a "geek gene" where none exists. Mark Guzdial has called this belief the biggest myth about teaching computer science; I hope this paper will finally put some nails in its coffin.
Kelly Rivers, Erik Harpstead, and Ken Koedinger: "Learning Curve Analysis for Programming: Which Concepts do Students Struggle With?", 10.1145/2960310.2960333.
The recent surge in interest in using educational data mining on student written programs has led to discoveries about which compiler errors students encounter while they are learning how to program. However, less attention has been paid to the actual code that students produce. In this paper, we investigate programming data by using learning curve analysis to determine which programming elements students struggle with the most when learning in Python. Our analysis extends the traditional use of learning curve analysis to include less structured data, and also reveals new possibilities for when to teach students new programming concepts. One particular discovery is that while we find evidence of student learning in some cases (for example, in function definitions and comparisons), there are other programming elements which do not demonstrate typical learning. In those cases, we discuss how further changes to the model could affect both demonstrated learning and our understanding of the different concepts that students learn.
The authors present "…a preliminary exploration of the application of knowledge-based learning curve analysis on programming data with the goal of extending the promise of educational data mining to programming." Doing this requires them to identify knowledge components, figure out how to measure progress toward correctness, and a host of other details. Their discussion of how they operationalized these issues was the most valuable part of the paper for me, since it allowed me to see very clearly what their assumptions and definitions are, and where my own thinking is vague, fuzzy, or contradictory. It's intriguing work, and I look forward to their next paper.
Soohyun Nam Liao, Daniel Zingaro, Michael A. Laurenzano, William G. Griswold, and Leo Porter: "Lightweight, Early Identification of At-Risk CS1 Students", 10.1145/2960310.2960315.
Being able to identify low-performing students early in the term may help instructors intervene or differently allocate course resources. Prior work in CS1 has demonstrated that clicker correctness in Peer Instruction courses correlates with exam outcomes and, separately, that machine learning models can be built based on early-term programming assessments. This work aims to combine the best elements of each of these approaches. We offer a methodology for creating models, based on in-class clicker questions, to predict cross-term student performance. In as early as week 3 in a 12-week CS1 course, this model is capable of correctly predicting students as being in danger of failing, or not, for 70% of the students, with only 17% of students misclassified as not at-risk when at-risk. Additional measures to ensure more broad applicability of the methodology, along with possible limitations, are explored.
There's no point getting people into computer science classes if they then drop out. Here, a group from several different institutions show how to build a statistical model that uses clicker data collected during the course of peer instruction to identify students who are likely to drop out only three weeks into an introductory course (i.e., early enough to intervene). Crucially, they conclude that, "Should an instructor not use Peer Instruction…simply asking the same multiple-choice questions before or after class could produce similarly accurate results." While I remain nervous about widespread monitoring students in classrooms, it's clear in this case that doing so could help a lot of people we might otherwise lose.
Kyle J. Harms, Jason Chen, and Caitlin Kelleher: "Distractors in Parsons Problems Decrease Learning Efficiency for Young Novice Programmers", 10.1145/2960310.2960314.
Parsons problems are an increasingly popular method for helping inexperienced programmers improve their programming skills. In Parsons problems, learners are given a set of programming statements that they must assemble into the correct order. Parsons problems commonly use distractors, extra statements that are not part of the solution. Yet, little is known about the effect distractors have on a learner's ability to acquire new programming skills. We present a study comparing the effectiveness of learning programming from Parsons problems with and without distractors. The results suggest that distractors decrease learning efficiency. We found that distractor participants showed no difference in transfer task performance compared to those without distractors. However, the distractors increased learners cognitive load, decreased their success at completing Parsons problems by 26%, and increased learners' time on task by 14%.
A Parsons Problem is a programming challenge in which the learner is given the statements they need to solve a problem, but in a jumbled order. They must then put those statements together to get the right answer. I've been using them for about a year in the classes I teach, and finding them really useful.
We know from prior work that high school students who completed partially written programs later constructed better quality programs than students who authored programs from scratch Here, the authors explored what happens to learning when Parsons problems include distractors of various kinds. Crucially, they found early that there was little point including irrelevant and tangential distractors, so they focused on distractors that encouraged learners to follow familiar but suboptimal solution paths.
Their conclusion is that distractors of this kind in code puzzles provide no clear benefit while reducing learning efficiency. As a practicing teacher, this is exactly the kind of result that I look for from researchers: a plausible idea has been shown empirically to have no educational benefit. While this single result is not revolutionary, the accumulation of such results is what we need to build the body of pedagogical content knowledge (PCK) that computing education needs.
Craig S. Miller and Amber Settle: "Some Trouble with Transparency: An Analysis of Student Errors with Object-oriented Python", 10.1145/2960310.2960327.
We investigated implications of transparent mechanisms in the context of an introductory object-oriented programming course using Python. Here transparent mechanisms are those that reveal how the instance object in Python relates to its instance data. We asked students to write a new method for a provided Python class in an attempt to answer two research questions: 1) to what extent do Python’s transparent OO mechanisms lead to student difficulties? and 2) what are common pitfalls in OO programming using Python that instructors should address? Our methodology also presented the correct answer to the students and solicited their comments on their submission. We conducted a content analysis to classify errors in the student submissions. We find that most students had difficulty with the instance (self) object, either by omitting the parameter in the method definition, by failing to use the instance object when referencing attributes of the object, or both. Reference errors in general were more common than other errors, including misplaced returns and indentation errors. These issues may be connected to problems with parameter passing and using dot-notation, which we argue are prerequisites for OO development in Python.
Python is my favorite programming language, but it does have its
warts. One that irritated me when I first switched from C++ and
Java to Python is the need to include
in method definitions. I therefore wasn't surprised to see that 53%
of novices forget to include
self in method headers,
along with other problems associated with the explicit use
self to refer to "this" object. As the authors say:
Even a minor oversight could have learning implications given the feedback that the Python interpreter provides. Below is the error message students see when they omit the self parameter:
TypeError: lookup() takes 1 positional argument but 2 were given
Because the error message does not mention 'self' or refer to the instance object placed before the calling method, students who simply forgot the instance object among the parameters may have difficulty interpreting this message to correct their omission.
Like the previous paper, this is exactly the kind of practical, evidence-based advice I look for as a teacher. The authors of this paper go on to say in a section title "Guidance for Instructors":
Given these difficulties, instructors are advised to have students practice the fundamentals of parameter passing and dot-notation as a prerequisite to class definitions. For students to see how Python transforms the method call by adding the instance object as one of the parameters, student understanding of parameter passing needs to be effortless.
I'll be adjusting my lesson plans accordingly.
Originally posted at Never Work in Theory.