Developing students’ transferable knowledge is a primary goal of education. In complex domains such as statistics, however, students often develop brittle and inflexible understanding that does not transfer successfully to real world problems or new contexts. This dissertation explores one approach to facilitating students’ development of transferable knowledge. Guided by the practicing connections framework (Fries et al., 2020), I start with the assumption that for knowledge to transfer it must be coherent. Then, I implement and test a core-concept intervention designed to produce more coherent knowledge by making explicit, for students, the connections between specific content and the coherent domain structure. Students in two ten-week introductory statistics classes (N=526) were randomly assigned, within class, to either receive the intervention, delivered as a series of 24 instructional videos embedded in their textbook, or not. All students used a new research-instrumented, interactive online textbook, Coursekata Statistics and Data Science: A Modeling Approach, capable of interleaving video, R-coding exercises, text, and assessment items. A series of 9 baseline videos were presented to students in both conditions, allowing for comparison of general video watching behavior across conditions. Within class, all instruction and assessment other than the intervention was identical across conditions; across classes, in-class instruction and assessment differed. Results from Study 1 reveal that students did not watch much of the embedded videos: students’ average proportion watched across videos hovered around 0.30, regardless of number of videos available or type of video presented (baseline vs. intervention). No overall effect of condition on transfer or course learning assessments emerged. However, a positive relationship does appear between the minutes of intervention videos watched, controlling for prior knowledge, and students’ performance on the final two sets of textbook chapter review questions that assess cumulative learning in novel contexts. When also controlling for minutes of baseline videos watched, the results are a bit more mixed, but still positive. Study 2 largely replicates the findings of Study 1: students watched overall lower, but similar, amounts of video across conditions, and the positive learning gains associated with intervention video watching persist in this sample. While the effect sizes are small, in the order of 1-3 percentage point score increases for every ten minutes of video watched, these results suggest that they are nevertheless robust across different instructors and samples.

The idea that people learn from sensorimotor experiences, whether through performing actions themselves or observing others, has garnered increasing attention from researchers in psychology, cognitive science, computer science, and education. In teaching and learning research, a key question is whether these sensorimotor experiences can help students acquire abstract concepts in complex domains. Past research has revealed promising evidence in various domains such as mathematics and physics regarding the benefit of incorporating some sort of bodily actions into learning. However, our understanding of how different types of bodily experiences impact learning is still nascent. Questions remain about the effect, mechanism, and practical application of using embodied experiences to help learners learn abstract knowledge in complex domains. These inquiries lead to a series of laboratory experiments and classroom interventions that I will present across three chapters, each written as a discrete empirical article that either has been published or is in preparation for publication. Across three chapters, the work was conducted in the field of statistics and data science education, which was picked because the concepts are intrinsically abstract and difficult, but they simultaneously do not require a sophisticated mathematical background. The first question focuses on the effect of observing bodily actions. Whereas abundant evidence has demonstrated the effect of performing actions, the concept of observing actions is less explored. Would simply observing hands-on representations lead to an increase in learning? This question is answered in Chapter 1, a published work that demonstrates the efficacy of observing hands-on representations in improving students’ understanding of randomness and the shuffle() function in R programming used to simulate randomness. The second question focuses on the mechanism underlying the effect—an embodied representation has more sensorimotor engagement and visuospatial concreteness than an abstract representation, but does sensorimotor engagement offer a unique benefit beyond visuospatial concreteness? Chapter 2 is a manuscript under review that reports on a laboratory experiment designed to isolate the effect of sensorimotor engagement. The findings suggest that sensorimotor engagement offers a unique benefit beyond visuospatial concreteness by helping learners develop more robust visuospatial representations. The last question relates to the practical application of different types of embodied interventions when we have learners with diverse levels of prior knowledge in the classroom. Theories in embodied cognition, along with other empirical evidence in both motor and learning domains, suggest that humans rely on their knowledge of their own bodies to understand other people’s movements. This insight prompted me to ask whether learners’ prior knowledge would moderate the type of embodied intervention (i.e. performing versus observing) on learning. The third Chapter reports on the design of a curriculum-linked embodied intervention to implement embodied activities over the entire school term of a college-level introductory statistics course. Students were randomly assigned the role of a performer or an observer. The findings provided support for the Perform-First hypothesis, showing that compared to observing, performing hands-on activities diminished the correlation between prior knowledge and post-test performance. Overall, this body of work extends the theory of embodied learning and offers practical insights for teachers and curriculum developers about how to implement embodied interventions into their educational materials and instructions.

Learning to program is increasingly important. Yet, it is becoming clear that most students struggle when learning to program (McCracken et al., 2001). This is leading to a divide where some people can program but many others can’t. Prior research has traced poor student outcomes to their early experiences learning programming. Still, little is known about how different programming tasks might impact the processes involved in learning programming. In this dissertation, I extend prior research on students’ experiences when learning computer programming and build on this research by testing the causal influences of different learning tasks on students’ emotions, motivation, and learning. In a randomized experiment, I manipulated the instructional tasks used to introduce programming and investigated students’ emotional trajectories, motivation, and learning outcomes. Participants randomly assigned to predict the outcome of pre-provided code showed more positive emotional trajectories, increased motivation, and greater learning compared to students randomly assigned to modify or produce their own code.

Historically, the study of instruction and the study of cognitive processes involved in learning have proceeded in different fields with little overlap. Issues of random assignment and experimental control have made studies of cognition in educational settings difficult, expensive, and prone to failure, while well-controlled experiments often occur in contrived laboratory situations. More recent efforts have attempted to bridge these issues by using more educationally relevant materials in laboratory studies, and by bringing well-controlled experiments into the classroom. The recent proliferation of instructional video is one tool that provides an inexpensive, easily controllable, and ecologically valid way to study the effects of instruction on cognition and learning.

Much of the existing research on multimedia learning has focused on cognitive processes and instructional principles to improve learning, but there is little attention paid to the semantic understanding that students develop when they read or watch a lesson. The text comprehension literature offers a model for how understanding develops, and may be fruitful for studying video instruction. In three studies, I investigate the usefulness of text comprehension as a model for understanding instruction.

In the first study, I assess the effect of adding information that increases the local coherence of a lesson. These experiments show that additional information improves retention for some question types, but it is not clear whether the effect is due to increased salience of the particular information rather than local coherence per se. In the second study, I investigate the effect of segmenting and labeling portions of a lesson. These 5 experiments fail to show consistent patterns of results, but highlight the importance of participants’ prior knowledge and engagement with the lesson. In the last study, I evaluate the effects of advance organizers on learning. This experiment highlights the importance of the specific materials used and suggests that well-cited findings from the literature may not be easily replicable. Though these studies provide mixed results overall, they contribute to our understanding about how people learn from video by highlighting the importance of the materials we teach, the tests we use, and the prior knowledge of the participants.

Background

Results

Conclusions