As enrollments in computing courses have surged, the ratio of students to faculty has risen at many institutions. Along with many other large undergraduate programs, our institution has adapted to this challenge by hiring increasing numbers of undergraduate tutors to help students. In early computing courses, their role at our institution is primarily to help students with their programming assignments. Despite our institution offering a training course for tutors, we are concerned about the quality and nature of these student-tutor interactions. As instruction moved online due to COVID-19, this provided the unique opportunity to record all student-tutor interactions (among consenting participants) for research. In order to gain an understanding of the behaviors common in these interactions, we conducted an initial qualitative analysis using open coding followed by a quantitative analysis on those codes. Overall, we found that students are not generally receiving the instruction we might hope or expect from these sessions. Notably, tutors often simply give students the solution to the problem in their code without teaching them about the process of finding and correcting their own errors. These findings highlight the importance of tutoring sessions for learning in introductory courses and motivate remediation to make these sessions more productive.

As computer science enrollments continue to surge, exam grading requires significant instructional resources. Online grading platforms have been developed in recent years and have been adopted at a number of institutions; however, their effectiveness compared with traditional grading on paper is not fully known. This study is the first in CS to compare online and paper grading. Comparing overall time to grade, including all factors, online grading doesn't show a consistent advantage compared with paper grading. We observed that online grading is much faster during the actual grading phase, but some of this benefit is offset by the additional overhead prior to grading (e.g., scanning) for online grading. Examining student and grader preferences based on feedback, both groups show a strong preference for the online format, predominately due to the convenience of the online platform. Graders report being able to grade more accurately online with the ability to modify rubrics, but that grading on paper tends to result in more social interactions among graders.

Pair programming has been widely adopted as an effective pedagogical tool in CS1/2. The successful execution of pair programming depends on each partner contributing similar amounts of work. However, anecdotes from past research observed that some student pairs can contribute a drastically unequal amount towards their assignments. We address this through a pilot of a hybrid pair programming model. Here, students complete two halves of the assignment as independent programmers and then in partnerships. We conducted a randomized experiment on this model in a CS1 class and found that while pairs generally performed better than solo students in the checkpoint, pairs in the second stage who had worked independently during the checkpoint portion tend to be more resistant to sudden increases in difficulty. We found no significant difference on selected exam problems between the control and experiment groups. We identified lower contributors within pairs and these students showed improved scores on most exam questions directly related to their independent checkpoint work. Survey feedback on hybrid pair programming is positive. The hybrid model is easily adopted in CS classes and positions students to better face the obstacles presented by the assignment while allowing them to benefit from pair programming.

Programming projects are widely used in CS1 classes to develop students' coding skills. To improve the learning impact of these projects, we propose and study a special project format named two-stage project in an introductory computer science course. For the first stage, students submit their programming projects individually followed by a second stage where they are paired to work on the same project in order to create an improved solution. Through peer collaboration, students review each other's work from the first stage, and write correctly-styled, well-documented, and more thoroughly tested code during the second stage. We used isomorphic assessments before and after the second stage of a project to measure students' understanding of the course material. Results indicate that two-stage projects tend to improve student understanding of course learning objectives. We also studied students' perceptions and experiences with two-stage projects, and their confidence toward computing. Students liked working on two-stage projects because they saw new ways to approach the same problem, and they liked discussions with their peers.

Replication research is rare in CS education. For this reason, it is often unclear to what extent our findings generalize beyond the context of their generation. The present paper is a replication and extension of Achievement Goal Theory research on CS1 students. Achievement goals are cognitive representations of desired competence (e.g., topic mastery, outperforming peers) in achievement settings, and can predict outcomes such as grades and interest. We study achievement goals and their effects on CS1 students at six institutions in four countries. Broad patterns are maintained --- mastery goals are beneficial while appearance goals are not --- but our data additionally admits fine-grained analyses that nuance these findings. In particular, students' motivations for goal pursuit can clarify relationships between performance goals and outcomes.

Collaborative exams have shown promise for improving student learning in computing. Prior studies have focused on benefits for all students, whereas this study seeks to refine our understanding of which students benefit and how group composition impacts that benefit. Using a crossover experimental design, the study first investigates whether students from differing performance levels (low, medium, or high) benefit from the collaborative exam. We find that students in the middle of the class (neither high nor low performers) tend to benefit strongly from the collaborative exam. Second, we explore whether group composition based on performance levels impacts the performance of members of the group. The results suggest more homogeneous groups (i.e., students in the group are at similar performance levels) are beneficial whereas students in groups with high heterogeneity do not experience significant performance differences between the pre-test and post-test.

Current CS1 learning outcomes are relatively general, specifying tasks such as designing, implementing, testing and debugging programs that use some fundamental programming constructs. These outcomes impact what we teach, our expectations, and our assessments. Although prior work has demonstrated the utility of single concept assessments, most assessments used in formal examinations combine numerous heterogeneous concepts, resulting in complex and difficult tasks.

This paper presents results of a large-scale survey of students' experiences in CS classes at two institutions: a small liberal arts college and a large research-focused university. Our work provides a fine-grained view of students' feelings and behaviors in CS classes, from introductory through to upper division courses. We find significant differences between the reported behaviors and feelings of female students compared to male students: female students are less comfortable asking questions in class and interacting with their instructor, and come out of a class with lower confidence in their ability to tutor for the class, despite the fact that they perform just as well as male students. Furthermore, we find some of these differences are consistent or increase across course levels, and could potentially affect students' post-college trajectories. Focusing attention on the student experience in more advanced classes may impact gender differences seen in the transition to the CS workforce.

To convey abstract concepts to students with little to no computing background, faculty members often adopt real world analogies to make CS concepts more relatable. Existing work on the power of analogies in CS education [1-2] has looked into the effectiveness and limitations of metaphors. However, there is no existing forum or repository that educators can share successful and interesting analogies. For this Bird of Feathers session, we invite all interested computer science instructors to share their successful (or unsuccessful) analogies with details on where and how the analogies were used, the related CS concepts, and how a particular analogy might fail to convey the instructor's ideas. We propose to focus on analogies for CS1/CS2 concepts because students in those classes may benefit the most from this teaching technique, and appropriate use of analogies for non-majors might help attract students to the field of computing. This BOF's topic relates to the general trend of teaching diverse groups of students in introductory programming courses and provides an excellent opportunity for CS educators to discuss successes and failures in using metaphors in our classes. The second purpose of this BOF session is to propose a suitable online medium where existing and new analogies can be shared with the CS education community. This BOF can be the beginning of a long-term online collaborative project where faculty members continue to share their metaphors with the community.

Collaborative group exams, including two-stage exams, have received increased attention in other disciplines after studies have shown their value for student learning. In computer science, prior work has shown students may value two-stage exams, but their impact on student learning in computing is unknown. In this randomized, crossover study, student learning on four topics during two-stage midterm exams is examined. Specifically, students are randomly selected to answer questions on a topic either working in groups or individually during the exam. On a quiz two weeks later, those who worked in groups statistically significantly outperform those who worked individually. The metric of performance comparison is normalized learning gain. These are short term gains, however, as the benefit is no longer apparent by the final exam. This is the first study to use controlled experiments to examine the value of the group stage of two-stage exams in computing.

Although computing students may enjoy when their instructors teach using analogies, it is unknown to what extent these analogies are useful for their learning. This study examines the value of analogies when used to introduce three introductory computing topics. The value of these analogies may be evident during the teaching process itself (short term), in subsequent exams (long term), or in students' ability to apply their understanding to related non-technical areas (transfer). Comparing results between an experimental group (analogy) and control group (no analogy), we find potential value for analogies in short term learning. However, no solid evidence was found to support analogies as valuable for students in the long term or for knowledge transfer. Specific demographic groups were examined and promising preliminary findings are presented.

Learning a classifier from private data distributed across multiple parties is an important problem that has many potential applications. How can we build an accurate and differentially private global classifier by combining locally-trained classifiers from different parties, without access to any party’s private data? We propose to transfer the “knowledge” of the local classifier ensemble by first creating labeled data from auxiliary unlabeled data, and then train a global differentially private classifier. We show that majority voting is too sensitive and therefore propose a new risk weighted by class probabilities estimated from the ensemble. Relative to a non-private solution, our private solution has a generalization error bounded by O(ε^-2 M^-2). This allows strong privacy without performance loss when the number of participating parties M is large, such as in crowdsensing applications. We demonstrate the performance of our framework with realistic tasks of activity recognition, network intrusion detection, and malicious URL detection.