This project studies how student-generated drawings aid learning of abstract science concepts in a lab setting.
Have you ever wondered why images stick in your mind longer than words? Our brains are hardwired to process visual information with remarkable efficiency. Research shows that visual aids not only make learning easier by breaking down complex ideas, but also help illustrate spatial relationships, pinpoint knowledge gaps, and keep us captivated
Studies reveal that drawing can significantly enhance understanding of how physical systems work. However, its effectiveness hinges on factors like the accuracy of the drawings, the learner’s prior knowledge, and the level of guidance provided while drawing
In two experiments, we randomly assigned college students to different learning tasks. Some studied or copied an instructor’s drawing, others completed a scaffolded drawing worksheet, and some drew their own representations on a blank sheet while reading about abstract concepts or physical systems. We found that, when it came to abstract lessons, unguided drawing significantly boosted retention but not transfer. However, for lessons on physical systems, drawing didn’t improve retention or transfer and students reported higher levels of intrinsic cognitive load.
These insights suggest that while drawing aids learning, its impact depends on the nature of the content and the context in which it’s used. So next time you’re learning a new abstract concept, consider picking up a pen and sketching it out—you might just find it sticks with you longer. However, if you are learning about a physical system studying an instructor provided visual representation is preferred.
Visual representations play a crucial role in learning by summarizing verbal information, illustrating spatial relationships, and enhancing memory retention
According to the Cognitive Theory of Multimedia Learning, effective learning involves selecting, organizing, and integrating new information with prior knowledge
The Dual Channel Processing Theory, a well-supported cognitive theory, explains that we process verbal and non-verbal information through separate channels, forming unique connections between new material and our existing knowledge
Learner-generated drawings are visual representations created by students to achieve educational goals
Research on learner-generated drawings has yielded mixed results, influenced by factors such as prior knowledge and the level of guidance provided during the drawing process. Early studies showed weak effects favoring drawing to learn. For example, researchers found that drawing or paraphrasing while learning about electrochemistry had varied impacts based on the level of detail of students’ drawings
Most studies on learner-generated drawings focus on concrete, observable systems. However, not all scientific concepts have definitive visual representations. Topics like dark matter, black holes, and natural selection are either theoretical, unfold over long periods, or lack a physical presence. Our research explores whether the efficacy of drawing to learn depends on the content of the lesson.
We hypothesize that drawing will enhance learning of abstract concepts more effectively than concrete ones. For abstract lessons, we expect learning to increase as students generate more of their own drawings. For concrete lessons, we predict that guided drawing will most benefit learning by reducing cognitive demands.
A convenience sample of 238 undergraduate students was gathered from the University of California, San Diego Psychology Subject Pool. All participants received partial course credit for their participation. Twenty-five participants were excluded from data analysis, leaving a final sample of 213 participants. Of the 213 participants, 43 identified as male, 165 identified as female, and five identified as non-binary. A majority of participants were in their early twenties with a mean age of 20.27 (SD = 1.93 years). Fifty-seven students participated in the study condition, 54 in the copy condition, 54 in the complete condition, and 48 in the draw condition.
We used a between-subjects design to manipulate participants’ drawing experience and measure their learning from a lesson about black holes. Participants were randomly assigned to one of four drawing conditions, which varied by the degree to which they generated their illustrations: copying a provided illustration (“copy”), completing a partial illustration (“complete”), free drawing their own illustration (“draw”), or a control condition that involved no drawing (“study”). Learning was measured using multiple choice and open response questions designed to assess both the retention and transfer of the lesson content. We also measured participants’ prior knowledge about physics and astronomy, their visual imagery ability, and their cognitive load during the learning activity to use as covariates in our analyses. The survey was designed to prevent participants from accessing the entire study at once or revisiting previous sections. While there was no time limit, participants were expected to complete the experiment within an hour.
The materials for this study included a Qualtrics survey, a passage about black holes, illustrations to support the passage, passage comprehension tests (multiple choice and open response), and measures of individual differences (prior knowledge, visual imagery ability, cognitive load). All materials were presented to participants via a Qualtrics survey, which they accessed online via desktop computers in a lab setting.
Prior Knowledge: Participants self-reported their knowledge and confidence in physics and astronomy using 5-point Likert scales. The knowledge scale ranged from “I know nothing at all” to “I know a great deal,” and the confidence scale ranged from “not confident at all” to “extremely confident.” The prior knowledge score was calculated by multiplying the knowledge and confidence ratings for both subjects and summing the results.
Black holes Lesson: All participants read an educational passage about black holes, adapted from The Cosmic Perspective textbook
Illustrations: Participants in the copy, study, and complete conditions received illustrations produced using Adobe Photoshop 2019 and Pages. Inspired by The Cosmic Perspective textbook and Pearson Mastering Astronomy platform
Passage Comprehension Tests: Three types of tests were developed: two analogous 15-question multiple-choice tests, one 4-question open response retention test, and one 4-question open response transfer test. The multiple-choice tests included 8 perfect analogs and 7 strong pairs of questions. Retention questions asked students to summarize lesson information, while transfer questions required application of knowledge to new scenarios. Answers were scored based on identified idea units, with inter-rater reliability for retention and transfer tests being r(212)=0.995, p <0.0001 and r(212)=0.798, p <0.0001, respectively.
Cognitive Load: Cognitive load was measured using a 10-item instrument assessing intrinsic, extraneous, and germane load
Visual Imagery: Visual imagery ability was assessed using the Vividness of Visual Imagery Questionnaire (VVIQ), which asks participants to rate the vividness of imagined scenes on a 5-point Likert scale
The purpose of our experiment was to determine if more generative illustration activities would improve retention and transfer of knowledge about black holes. We hypothesized that participants who created their own illustrations would perform better on comprehension measures compared to those who studied provided illustrations. From the initial 238 participants, 213 were included in the final analysis after excluding those who did not engage in the illustration activity, failed attention checks, or did not complete the study.
Participants demonstrated good visual imagery ability (M = 3.761, SD = 0.574) on a scale from 1 to 5, but reported low prior knowledge in physics and astronomy (M = 9.310, SD = 6.813) on a scale from 0 to 50. Their pre-test scores on a multiple-choice test specific to black holes were also low (M = 4.188 out of 12, SD = 1.963). There were no significant differences across conditions for visual imagery ability, self-reported prior knowledge, or pre-test scores, indicating a baseline similarity among participants.
We analyzed the effects of illustration generativity on multiple-choice gains, open-response retention scores, and open-response transfer scores. No significant differences were found in multiple-choice gains or transfer scores across conditions. However, we did observe significant differences in open-response retention scores (F(3,209) = 7.41, p < 0.001). Participants who drew their own illustrations (M = 6.688, SD = 2.93) scored significantly higher on retention compared to those who copied illustrations (M = 4.204, SD = 2.750, p < 0.0001) and those who completed partial illustrations (M = 4.796, SD = 2.595, p = 0.0037). The difference between drawing and studying was near significance (p = 0.052).
Additional analyses were conducted to explore mechanisms by which drawing might enhance learning. We examined reading time, cognitive load, condition enjoyment, and word count:
Reading Time: Participants in the drawing condition spent more time engaging with the lesson (M = 16.994 min, SD = 5.809) compared to other conditions (study, copy, complete), which could contribute to their better performance on the retention test. This was confirmed by a significant linear regression (F(1,211)=6.97, p = 0.008), with reading time predicting retention test scores.
Cognitive Load: We found no overall differences in cognitive load across conditions. However, significant differences were observed in extraneous load (F(3,209) = 15.82, p < 0.0001) and germane load (F(3,209) = 6.20, p = 0.0005). Participants in the complete condition reported higher extraneous load, while those in the study condition reported higher germane load compared to other groups.
Condition Enjoyment: Participants in the study condition enjoyed the activity more (M = 3.965/5 Likert scale, SD = 0.778) than those in other conditions, with significant differences observed between study and complete conditions, and between copy and complete conditions.
Word Count: No significant differences were found in total word count or unique word count used to answer open response questions across conditions, suggesting that the amount of writing did not account for differences in retention scores.
The findings of our study suggest that drawing to learn theoretical science can enhance the recall of key points addressed in a passage. Participants in the drawing condition significantly outperformed all other groups on the open response retention test. However, they did not show superior performance on the multiple-choice test or the transfer test compared to other visualization conditions. The most notable difference in performance on the open response retention task was observed between the draw and copy conditions.
These patterns of effects contradict previous findings that suggested students perform better on retention tests when their visualization is guided or inspired by an instructor. While our data indicate that the visualization strategy does not impact students’ transfer ability, it is possible that providing participants in future studies with more time could reveal an effect of visualization strategy. Additionally, exploratory analyses reveal that time spent on the passage may be a crucial factor in supporting retention performance and highlight that self-reported germane load is significantly predicted by the visualization condition.
There are limitations to our study that future research should address. Importantly, a lab setting does not represent true learning environments where students are genuinely motivated to learn. Our participants were volunteers who participated to receive partial course credit, regardless of their performance in the experiment, which means they may not have been genuinely invested in learning about the formation and properties of black holes.
Furthermore, certain visualization tasks may have disproportionately discouraged participants from completing the task. For example, because the study illustration condition requires no active engagement, participants assigned to this condition may have passively scanned the illustrations rather than studying them carefully. Similarly, students assigned to complete an illustration may have found this task odd or elementary, discouraging them from actively completing the worksheet. Feedback from our pilot study suggests that participants did not enjoy completing the illustration and were generally confused by it.
Future studies should test these visualization manipulations in true classroom environments where theoretical science concepts are taught (e.g., advanced physics, astronomy, and evolutionary biology courses). This approach will help determine the effectiveness of drawing as a learning strategy in more realistic and motivating educational settings.
This research was conducted at the University of California, San Diego in the LIME Lab between 2018 and 2020 and served as my honors thesis project. I would like to extend my heartfelt gratitude to Dr. Geller for providing invaluable mentorship and support throughout my time at UCSD. Your guidance was instrumental in the success of this project and my professional development.