Influence of context on the architecture of student-generated models in an intro-bio course
Scientific models are specialised representations that explain or predict a concept, process, or phenomenon. Educators and researchers agree that models are of great importance in the generation, evaluation, and communication of scientific knowledge. As a consequence, models have been included in the standards and required curricula for science at K-12 and university levels in multiple countries.
As a tool, models lend themselves to both authentic instruction and assessment. However, student-constructed models can give us insights into student thinking and reasoning that are not captured in multiple choice or even narrative responses. Features of model architecture have been used in previous studies to gain insights into aspects of students’ cognitive structures (CS), such as the robustness or connectedness of their understanding. Previous studies have reported influences of contextual features on students’ reasoning about evolution, but only in their narrative responses. In this study, we ask whether item feature context (i.e., variables in a question prompt) impacts the architecture of students’ constructed models of evolution by natural selection.
As a tool, models lend themselves to both authentic instruction and assessment. However, student-constructed models can give us insights into student thinking and reasoning that are not captured in multiple choice or even narrative responses. Features of model architecture have been used in previous studies to gain insights into aspects of students’ cognitive structures (CS), such as the robustness or connectedness of their understanding. Previous studies have reported influences of contextual features on students’ reasoning about evolution, but only in their narrative responses. In this study, we ask whether item feature context (i.e., variables in a question prompt) impacts the architecture of students’ constructed models of evolution by natural selection.
Questions:
- How does context affect the architecture of the student generated models?
- How much of this variation can be explained by demographics, such as prior achievement (GPA) and class level?
Methods:
In a large intro-biology course, we administered four isomorphic prompts of the form:
“(Taxon) has (trait). How would biologists explain how a (taxon) with (trait) evolved from an ancestral (taxon) without (trait)?” (Table 1)
Table 1: Contextual variation in the isomorphic prompts
In a large intro-biology course, we administered four isomorphic prompts of the form:
“(Taxon) has (trait). How would biologists explain how a (taxon) with (trait) evolved from an ancestral (taxon) without (trait)?” (Table 1)
Table 1: Contextual variation in the isomorphic prompts
Each student provided model-based responses to two prompts (same type of trait, different taxa; (Figs 1 a & b). We analysed models for aspects of model architecture like size (number of structures, arrows, and propositions) and complexity (Web-Causality Index, WCI). We then analysed the data to explore the effects of demographic variables (performance and class level) on variation in model architecture.
Analysis and Results:
Of 194 students enrolled in the class, 66 constructed models in response to both Human (H) and Cheetah (C) prompts and were therefore included in the analysis. ). Students’ models varied in terms of size (number of structures, arrows and propositions) and complexity (WCI). (Figs 1 a & b)
Of 194 students enrolled in the class, 66 constructed models in response to both Human (H) and Cheetah (C) prompts and were therefore included in the analysis. ). Students’ models varied in terms of size (number of structures, arrows and propositions) and complexity (WCI). (Figs 1 a & b)
Figure 1 a
|
Figure 1 b
|
Context did not significantly influence model size (Fig 2). Both Human and Cheetah models had similar number of structures, arrows, and propositions.
Figure 2: Student generated models did not vary significantly in terms of size
|
Legend:
C - Cheetah H - Human |
However, models about cheetahs were more complex than models about humans and models about humans were more likely to be linear (zero complexity) (Figs 3 a & b).
Figure 3 (a): Student-generated models did showed some variation with respect to complexity
|
Figure 3 (b): The mean WCI was higher for cheetah models
|
The size and complexity of student-generated models varied with context and tritile.
Figure 4 : Variation in size with tritile
|
Figure 5: Variation in complexity with tritile
|
Conclusions:
Our results indicate that contextual features (here, taxon) are eliciting differences in model architecture. While the context of the prompt did not significantly impact model size, complexity did vary with context. This could indicate that while students are using the same number of concepts to explain natural selection in both humans and cheetahs, their cognitive structures are more connected when reasoning about non-human animals.
Middle-achieving students constructed models that were unaffected by context, both in terms of size and of complexity. Students with lower GPAs had the highest variation in complexity based on context, and they had the highest mean complexity when responding to cheetah prompts. Additionally, they had low sized models for both the contexts. High-achieving students had both low size and complexity.
Class levels did not contribute to the variation seen.
Our results indicate that contextual features (here, taxon) are eliciting differences in model architecture. While the context of the prompt did not significantly impact model size, complexity did vary with context. This could indicate that while students are using the same number of concepts to explain natural selection in both humans and cheetahs, their cognitive structures are more connected when reasoning about non-human animals.
Middle-achieving students constructed models that were unaffected by context, both in terms of size and of complexity. Students with lower GPAs had the highest variation in complexity based on context, and they had the highest mean complexity when responding to cheetah prompts. Additionally, they had low sized models for both the contexts. High-achieving students had both low size and complexity.
Class levels did not contribute to the variation seen.
