When a Student Investigation Is Confounded in Grade 7 Science
During a grade 7 investigation, groups design their own fair test of the question 'Does the amount of fertilizer affect plant height?' One group of four sets up three plants: 0 mL, 5 mL, and 10 mL of fertilizer solution. But the 10 mL plant sits on the windowsill while the other two sit on a shelf three meters away, and the 10 mL plant is in a larger pot. After three weeks the 10 mL plant is 4 cm taller, and the group writes, 'More fertilizer makes plants grow taller. Our data prove it.' Two other groups used similar but unconfounded designs and found no difference between treatments. The teacher has one class period before the unit's argument-writing assessment.
1. What do you say in the moment? How do you respond to the flawed design without simply telling them the answer?
Read the worked answer
I do not say 'your design is confounded.' I ask questions that make the confound visible to them, and I keep the design theirs:
- 'Walk me through everything that is different between your three plants besides the fertilizer.' Let them list. They will produce light and pot size.
- 'Which of those differences could also change how tall a plant gets?'
- 'Given that, what would your data have to look like before you could say the fertilizer caused the 4 cm?'
- 'You cannot rerun three weeks. What could you do with the time you have?' - they can compare their setup with the two unconfounded groups and revise the claim, not the experiment.
I also flag the word 'prove' once, without a lecture: 'What word would a scientist use instead of prove, and why?'
The goal is that the group, not the teacher, names the alternative explanation. If a group cannot get there, I supply the question ('could the windowsill alone explain 4 cm?'), never the conclusion.
2. How would you use the disagreement between this group and the two unconfounded groups to teach evidence and error?
Read the worked answer
The disagreement is better material than a clean result. Put all three data sets on the board and pose the problem: 'Group A's plants differ by 4 cm. Groups B and C's do not. What is the most likely explanation?'
Then have students generate competing explanations - fertilizer, light, pot size, seed variation - and ask which single explanation accounts for all three data sets at once. Fertilizer does not: if fertilizer mattered, B and C should show an effect too. Light and pot size do explain A's result and only A's result. That is the reasoning I want, not a definition.
Only after students have done that do I attach the vocabulary - confounding variable, control, replication - to the thing they just discovered. Vocabulary should label the reasoning, not substitute for it.
A second move: ask why three groups' data are more informative than one group's, which builds the idea of replication without a lecture on the scientific method.
3. What assessment items would show understanding of fair tests rather than recall of definitions?
Read the worked answer
Items should require transfer to a scenario they have not seen and cannot answer with a memorized sentence:
- Novel confound: 'A student tests whether music affects plant growth. The rock-music plant is in the closet; the classical plant is by the window. Name one alternative explanation for any difference and describe the smallest change that would fix it.'
- Design revision: give a flawed procedure and ask students to rewrite it - scored on whether they identify the variable held constant.
- Claim–evidence fit: give two data sets and ask which claim each supports and what neither can support.
- Language of certainty: ask students to rewrite 'our data prove' as a claim a scientist would make, and to say what further evidence would strengthen it.
What I would not count as evidence: defining 'controlled variable' correctly in a sentence. A student can recite the definition and still design the windowsill experiment, so the assessment must require the design decision itself.