Diane Jass Ketelhut
8 min
This exploratory study investigates whether seventh-grade students' initial self-efficacy—their belief in their ability to succeed in scientific inquiry—predicts their data-gathering behaviors within a virtual environment. The researcher analyzed data from 96 students navigating 'River City,' a multi-user virtual environment (MUVE) where students act as scientists to investigate an epidemic. Using individual growth modeling, the study tracked students across three visits to assess their total scientific inquiry behaviors and the diversity of data sources they accessed.
The study found a clear initial correlation between self-efficacy and scientific inquiry behaviors: students entering with higher self-efficacy engaged in more data-gathering activities than those with lower self-efficacy. However, over the course of three sessions, this gap disappeared. By the end of the intervention, initial self-efficacy no longer predicted the frequency of data-gathering. Furthermore, the diversity of sources students chose to explore was not influenced by their self-efficacy levels. Interestingly, the data suggested that gender may influence the trajectory of engagement, with girls showing higher growth rates in data-gathering behaviors over time compared to boys.
These findings suggest that immersive virtual environments might serve as a powerful 'equalizer' in science education. By providing an authentic, inquiry-based context, MUVEs may help bridge the engagement gap between students who arrive with high confidence in their scientific abilities and those who do not. If sustained, this suggests that novel technological platforms could help counteract the traditional erosion of confidence students often face in science classrooms, potentially fostering broader participation in scientific inquiry regardless of a student's prior self-perception.
Sam: Students in open worlds wander a lot. How do you turn wandering into a measure of inquiry?
Alex: That's the one idea that makes the study work. A server logs every move and click, with a timestamp. She mapped each loggable action onto the national science standards' definition of inquiry. Visiting new places counts as making observations. Asking a resident "what's new" counts as posing questions. Taking a water sample counts as using tools. Opening a library book counts as consulting sources.
Sam: So the standards' definition becomes a checklist, and the server fills it in.
Alex: Yes, with eight indicators. She built two outcomes from them. One is the total number of data-gathering behaviors per visit. The other is diversity: how many of the eight categories a student drew on at all. Repeat clicks on the same source mostly counted once, since the point is new information.
Sam: Which visits did she look at? A first visit is mostly figuring out the controls.
Alex: Right, and she drops that one. Students visited six times over about three weeks. The first visit was orientation. The last two were for testing hypotheses in experimental and control versions of the town. She analyzed visits two through four, the open-ended investigation in winter, spring, and summer. Then she used individual growth modeling, which fits each student's own trajectory across those visits. Teacher, gender, and prior knowledge were controls.
Sam: So what happened overall, before we bring confidence in?
Alex: On the first analyzed visit, students averaged about thirteen data-gathering behaviors. They added nearly two more on each visit after that. She reads this as an argument against pure novelty. If shiny technology were the only draw, you'd expect engagement to fade, not climb.
Sam: Three visits isn't long for novelty to wear off, though.
Alex: She makes that point herself. She calls for repeating it with a longer version of River City.
Sam: Then the main question. Did confidence matter?
Alex: At the start, yes, though the first estimate is weak. In a model without gender, each point of self-efficacy meant almost two more behaviors initially. But that sits right at p equals point one, which she flags as suggestive. Once gender enters the model, the starting gap looks clearer. Boys with high self-efficacy gathered about sixteen pieces of evidence on that first visit, boys with low self-efficacy about eleven. Girls showed a similar five-point gap, a bit lower overall.
Sam: And then the trajectories diverge or converge?
Alex: They converge, unevenly. High-confidence boys barely changed across visits. Low-confidence boys grew. High-confidence girls grew slightly. Low-confidence girls grew fastest of the four groups. By the third investigative visit, entry self-efficacy no longer predicted how much data anyone gathered.
Sam: So the gap that was there at the start simply washed out.
Alex: In this sample, yes. And something else appeared. On that last visit, girls averaged about eighteen pieces of evidence and boys about fifteen. That gender difference is borderline, at p of about point oh nine.
Sam: What about the second outcome, the variety of sources?
Alex: Students started out using about two and a half of the eight categories, and that number grew over time. Self-efficacy never predicted it, in any model she fit. That cuts directly against the expectation that low-confidence students avoid exploring.
Sam: Here's my obvious objection. Prior knowledge and confidence usually travel together. Couldn't this just be students who know more doing more?
Alex: She hit exactly that. The content pretest and self-efficacy correlated at about point five. With both in the model, neither was significant. So she dropped prior knowledge from the fitted models. That means the self-efficacy effect at entry can't be cleanly separated from knowing more science.
Sam: Does she explain why the gap closes?
Alex: She offers possibilities and labels them as questions. Students may have rated their confidence based on classroom science, then found a game-like world felt different. They might have higher confidence with game-like technology. The motivation of the environment might override the usual effect. Or immersion as a scientist might actually shift self-efficacy. But she measured self-efficacy only once, before entry, so none of these is tested.
Sam: And the girls result? That's the one people will quote.
Alex: She treats it as intriguing, not settled. The town was deliberately designed with features meant to appeal to girls. There are more female residents than male, and the mayor and university president are women. So she wonders whether the pattern is specific to River City.
Sam: Let's be blunt about trust, then. What's the limitation that matters most?
Alex: It's exploratory and correlational, and several findings rest on p-values around point oh nine or point one. There are only three waves of data. And a quarter of students had just two, partly because one teacher skipped the second world entirely. It's also one fairly homogeneous school, and one treatment group. Prior gaming experience wasn't controlled. Most importantly for practice, she didn't link data gathering to learning outcomes.
Sam: So who should sit down with the full paper, and where should they start?
Alex: Anyone studying self-efficacy, game-based or virtual learning, or growth modeling of log data. Start with the results on how self-efficacy and gender shape the trajectories together. Then read the limitations, which are frank about how fragile those estimates are. The measures section is worth a look for how clicks were mapped onto inquiry.
Sam: And for everyone else, the one line to keep?
Alex: Low confidence shaped where these students started, not, in this sample, where they ended up.
Sam: A small study with a question big enough to chase.