The Fertilizer Experiment That Measured the Closet
Mr. Okafor's class tests whether a new fertilizer increases bean plant height. Ten seeds go into fertilized potting mix on the classroom windowsill; ten go into unfertilized mix on a shelf inside the supply closet. All twenty are watered on the same schedule with the same volume, planted in identical pots from the same seed packet. After four weeks, the fertilized plants average 18.4 cm and the unfertilized plants average 11.2 cm. The class writes that the fertilizer added 7.2 cm of growth. A student points out that the closet shelf is about 4 °C cooler and gets no direct light. The class has time to repeat the experiment exactly once before the science fair.
1. Identify the design flaw and explain precisely what it does to the 7.2 cm conclusion.
Read the worked answer
The flaw is confounding: fertilizer is not the only thing that differs between groups. The windowsill group also got direct light and roughly 4 °C more warmth.
What that does to the conclusion: three variables (fertilizer, light, temperature) all changed together. The 7.2 cm difference is real as a measurement, but it cannot be assigned to any one of them - light alone could plausibly produce a gap that size in four-week bean seedlings, and temperature could contribute independently or interact with light.
Also note what was controlled: seed source, pot type, water volume, and watering schedule - those are done correctly and should be kept. The class's error is not sloppiness everywhere; it is one uncontrolled axis (the growing location) that happens to be the most powerful one for plants.
Bottom line: the data support "plants on the windowsill with fertilizer grew taller than plants in a dark, cool closet," not "fertilizer added 7.2 cm."
2. Design the single repeat the class can afford. State the variables and the physical setup.
Read the worked answer
Fix the location; move the treatment. Put all twenty pots in the same place - one windowsill run, or better, one bench under identical light - and vary only the fertilizer.
Setup:
- 20 identical pots, same seed packet, same soil base, same water volume and schedule (all already in place).
- Randomly assign 10 pots to fertilized mix and 10 to unfertilized mix rather than choosing which pots go where by convenience - this protects against a subtle bench-position effect.
- Keep the closet out of the design entirely; if the class wants to study light, that becomes a separate experiment.
- Measure all plants on the same day with the same ruler, from soil line to tallest leaf, ideally with the measurer unaware of which pots are which.
Independent variable: fertilizer. Dependent variable: height in cm at 4 weeks. Controlled: light, temperature, water, pot, seed lot, measurement method.
Cost: free - the repeat requires no new materials, only a location change, so the one available repetition is enough to remove the confound.
3. A classmate argues they should just keep the closet group and "add fertilizer to the closet plants too." What would that design let the class conclude, and what would it not?
Read the worked answer
Interpret the proposal carefully: fertilized closet plants plus unfertilized closet plants gives a second, independent 2-group comparison - fertilizer tested at low light and low temperature.
What it would let them conclude: whether the fertilizer effect holds in cool, dim conditions. If fertilized plants win in the closet too, the effect is not dependent on bright windowsill light, which is a genuinely stronger claim than the original flawed design.
What it would not do:
- It would not rescue the original windowsill-vs-closet pair. That comparison stays confounded and should be discarded, not pooled with the new data.
- With n = 10 per group and one repetition, it cannot support claims about how much growth the fertilizer adds, only whether a difference appears.
Judgment call: with only one repetition available, spend it removing the confound (same location) rather than expanding to a second location. A clean answer to "does fertilizer matter here?" beats a messy answer to "does fertilizer interact with light?" - and the second question can wait for a design with more replication.