Tracing Students' Experimentation Strategies Through Simulation‐Based Scientific Investigation

Designing controlled experiments is central to scientific investigation, yet little is known about how students change their experimentation strategies as they work through a simulation‐based task and how different patterns of strategy use relate to investigation‐relevant reasoning. To address these gaps, we examined the strategies used by N = 681 middle and high school students in the United States as they used the Wildfire Explorer simulation to investigate how vegetation type affects wildfire spread. Students' time‐stamped interaction logs captured how they configured and conducted simulation trials. We used k‐means clustering to identify strategy profiles, full‐path sequence mining and heuristic process mining to characterize how strategies were organized over time, and multilevel models to examine associations with observation accuracy, evidence quality, and explanation level. We identified four strategy profiles characterized by low‐frequency use of coded strategies, one‐spark experimentation, mixed use dominated by three‐spark noncontrastive configurations, and near‐exclusive use of standard vary‐one‐thing‐at‐a‐time configurations. Adding profile membership improved model fit for all three reasoning outcomes after accounting for students' background characteristics and teacher‐level clustering. Relative to the low‐frequency profile, the standard vary‐one‐thing‐at‐a‐time profile was positively associated with all three outcomes; the two intermediate profiles showed more selective associations. Together, these findings show that students' experimentation differed in both variable control and the generation of observable comparative evidence, variation that would be obscured by a simple optimal‐versus‐suboptimal classification. These profiles provide a differentiated account of students' investigation practices and identify patterns that could be examined in future studies of instructional support.