PICA: A Data-driven Synthesis of Peer Instruction and Continuous Assessment

TL;DR

PICA presents a data-driven synthesis of Peer Instruction and Continuous Assessment by pairing students for collaborative CA tasks using a five-dimensional score vector derived from the most recent independent CA attempt, forming complementary pairs via Euclidean distance. Ten quiz dyads over a 15-week course reveal that collaborative b-quizzes produce significantly higher learning gains than remote, but do not show significant gains on subsequent individual CA tasks; nevertheless, engagement and peer interactions increase, hinting at the method's potential to foster small learning communities. The study demonstrates the feasibility of data-informed pairing, highlights limitations due to nonrandomized groupings, and outlines concrete avenues for refinement, such as qualitative feedback, richer student profiles, and scalable pairing strategies, to enhance learning equity and collaboration in STEM education.

Abstract

Peer Instruction (PI) and Continuous Assessment(CA) are two distinct educational techniques with extensive research demonstrating their effectiveness. The work herein combines PI and CA in a deliberate and novel manner to pair students together for a PI session in which they collaborate on a CA task. The data used to inform the pairing method is restricted to the most previous CA task students completed independently. The motivation for this data-driven collaborative learning is to improve student learning, communication, and engagement. Quantitative results from an investigation of the method show improved assessment scores on the PI CA tasks, although evidence of a positive effect on subsequent individual CA tasks was not statistically significant as anticipated. However, student perceptions were positive, engagement was high, and students interacted with a broader set of peers than is typical. These qualitative observations, together with extant research on the general benefits of improving student engagement and communication (e.g. improved sense of belonging, increased social capital, etc.), render the method worthy for further research into building and evaluating small student learning communities using student assessment data.

Figures (6)

• Figure 1: Six quiz dyads with the specific concept covered by each quiz question. An 'a' quiz is taken by students independently outside of class time. The subsequent 'b' quiz covers the same concepts but is taken by students collaborating in pairs during class time. Coloring indicates questions deliberately crafted to test students on the same concept or topic in order to later quantify the effects of collaboration.
• Figure 2: A sample distance matrix of the $n(n-1)/2$ pairwise distances between the five-dimensional quiz score vectors for a subset of six students (from $n=34$ students in total). Each row and column corresponds to a student (identifiers have been excluded). The number and color in a cell represent the Euclidean distance between score vectors.
• Figure 3: Weekly schedule of the course in which the PICA method was tested and the first quiz dyad was administered. (A) and (C) represent class meetings while (B) represents the roughly 48-hour period in which Quiz 1a is available to students. (D) is when the instructor executes code to download Quiz 1a results and generate student pairings (for students present in classroom). (E) is the 25-30 minute period in which students in class collaborate on Quiz 1b, and remote students take it independently.
• Figure 4: Normalized gain (left) and modified normalized gain (center), and a boxplot of MNG (right) for each quiz dyad and each student. Each plot represents students taking a b-quiz independently and remotely (orange, $n=68$) and students taking a b-quiz collaboratively in class (blue, $n=232$).
• Figure 5: Modified normalized gain (MNG) split by students whose a-quiz score was lower (yellow)/higher (green) than their partner's (left). These same students' MNG is also plotted against the a-quiz distance to their partner (center, right) with the approximate confidence interval of the relationship shaded in the respective color.