Why the project is built this way

The closest published, assessed analogue to this project is Project EDDIE: short data modules built on an A-B-C fading scaffold, where part A is heavily guided and part C is close to open-ended, wrapped in the 5E learning cycle of engagement, exploration, explanation, expansion and evaluation. The four stages here do the same thing over twelve weeks instead of one lab.

The support in the Data Library fades on purpose. Stages 1 and 2 have a first-plot resource for most rows, Stage 3 has the thresholds table, Stage 4 has the Lab 8 script and nothing else. By Stage 4 you should not need to be told which plot to make.

• Carey, Farrell, Hounshell & O’Reilly (2020), Ecology and Evolution 10(22), 12515–12527

  21 classrooms at 17 colleges and universities, 277 undergraduate and graduate students, paired before-and-after analysis on about 172.

  Simulation-modelling proficiency rose from 1.59 to 2.27 on a 1 to 5 scale (p < .001, effect size 0.65) and confidence from 1.60 to 2.21 (0.61). General Lake Model proficiency rose 1.29 to 2.21 (0.73), confidence 1.39 to 2.16 (0.67). Knowledge of ecosystem models rose 2.17 to 3.11 (0.72). Evidence of systems thinking in free-response answers rose from about 3 per cent to 9 per cent (p = .02) for model benefits, but not significantly for challenges (p = .484).

  Caveat. These are self-reported measures, and the largest gains came from the students who started lowest. They are evidence that a scaffolded module moves novices, not that it moves everyone.

  doi.org/10.1002/ece3.6757

• O’Reilly et al. (2017), BioScience 67(12), 1052–1061

  Six modules across eight courses, 1,380 students.

  Significant improvement in competence with spreadsheet software and in conceptual understanding of how to use large, complex datasets to address scientific problems.

  Caveat. Two upper-level courses whose students already scored high showed no significant change. That is a ceiling effect, and it is the reason the scaffolding here is pitched at genuine novices rather than at the strongest people in the room.

  doi.org/10.1093/biosci/bix118

• Soule, Darner, O’Reilly et al. (2018), Journal of Geoscience Education 66(2), 97–108

  The same modules across two instructional formats, in-person and larger-enrolment.

  Effective for developing quantitative literacy, statistical reasoning and concept understanding in both settings; the authors conclude the modules are sufficiently flexible to work in either.

  Caveat. The exact before-and-after numbers for this study sit behind a paywall and have not been checked here. The conclusion is quoted; the numbers are not.

  doi.org/10.1080/10899995.2018.1411708

The worked-example effect is one of the most replicated findings in instructional psychology. For a novice, studying a worked example produces better retention and transfer, at lower cognitive load, than solving an equivalent problem unaided (Sweller & Cooper, 1985, and many replications since). Pairing an example with a problem works as well as examples alone, and both beat problem-solving alone.

The effect reverses as you gain expertise: once you know what you are doing, being walked through it slows you down. That is the expertise-reversal effect, and it is why the support fades rather than continuing at full strength.

The symposium panel are people who work on this system. They are there to respond to your reasoning, and they do not grade you.

That split is deliberate. Ecological service-learning improves classroom climate and students’ sense that they can act on environmental problems (Ecosphere, 2022), and community and citizen-science work improves understanding of how science actually proceeds. But the same literature is clear that real client work has no single right answer and that partner judgements vary a great deal. Using practitioners for feedback and not for marks is the defensible way to get the benefit without making your grade depend on whose day it is.

You arrive in this course from earth science, chemistry and biology, with very uneven exposure to R. Groups are balanced on that prior background rather than left to form themselves, so that no group is carrying all the coding and no group is carrying none of it.

Tools like CATME Team-Maker and the open-source gruepr do this from a short survey of skills, schedule and background; gruepr runs locally, balances or diversifies across roughly fifteen attributes, and can avoid isolating a single person by gender or background in a group. Self-selected teams tend to reproduce existing friendships and, with them, existing skill gaps.

Three honest limits, because the same standard applies to the course as to your assumptions register.

• The EDDIE evidence is mostly self-reported proficiency and confidence, not independently measured skill.
• EDDIE modules are short exercises for individuals or small groups. This is a twelve-week gated team project. The analytical moves are the same; the scale, the negotiation and the handover between stages are not, and no study here tested those.
• The strongest gains in that literature came from the students who started with the least. If you arrive already fluent in R, the scaffolding is not aimed at you, and you should skip ahead rather than sit through it.

Provided-data projects in limnology are not unusual, but most comparable courses have students collect their own field data over a term and write it up. What is less common, across the programs we looked at, is the combination of four gated stages with a written handover between them, a published data library with checked links, a practitioner panel that gives feedback without grading, and peer assessment used at full weight.

The closest analogues are problem-based engineering programs, where a semester-long group project with supervisor gates and an external examiner is the norm, and graduate water-resources practicums that deliver a report to a real agency client. Both of those run at higher weight and over longer periods than the 15 per cent this project carries.