Understanding When Error Obscures Biological Effects
Precision — how consistent are repeated measurements? Darts landing close together = high precision = small error bars.
Accuracy — how close are measurements to the true value? Darts centred on the bullseye = high accuracy.
These are independent. A mis-calibrated pH probe can give precise but inaccurate results. Error bars on a graph measure precision, not accuracy.
Tight cluster on the bullseye. Small error bars. Result is reliable and correct.
Consistent but off-target — e.g. a sensor with a fixed calibration offset. Small error bars but a wrong conclusion.
Wide scatter centred on target. Large error bars. More replicates improve confidence.
Wide scatter AND off-target. Large error bars and a wrong conclusion. A method problem, not just a replicate problem.
With a signal-to-noise ratio of 7.3:1, this experiment can reliably detect the biological effect.
Why this works:
Click to load these scenarios into the visualisation above
Effect: 0.6 pH units (strong photosynthesis)
Method Error: ±0.026 pH (spectrophotometry)
Execution: ±0.03 pH (careful technique)
Result: Effect is 15× larger than error → Highly reliable conclusion
Effect: 0.3 pH units (weak response)
Method Error: ±0.1 pH (pH meter drift)
Execution: ±0.15 pH (variable technique)
Result: Effect only 1.7× larger than error → Questionable conclusion
Effect: 0.15 pH units (minimal response)
Method Error: ±0.1 pH (visual comparison)
Execution: ±0.2 pH (poor standardisation)
Result: Error exceeds effect → Cannot draw conclusion
Effect: 0.4 pH units (moderate)
Method Error: ±0.05 pH (good method)
Execution: ±0.12 pH (student variation)
Replicates: Only 2
Result: Need more data → Statistical power too low
Effect: 0.25 pH units (small but real)
Method Error: ±0.026 pH (spectrophotometry)
Execution: ±0.05 pH (some variation)
Replicates: 6 (good design)
Result: Averaging improves confidence → Reliable detection
Effect: 0.5 pH units (strong)
Method Error: ±0.1 pH (moderate precision)
Execution: ±0.05 pH (excellent technique)
Result: Good technique can't fix poor method, but effect still detectable → Conclusion possible
A large biological effect (e.g., blue light producing ΔpH = 0.6) can be detected even with moderate error. A tiny effect (ΔpH = 0.1) may be undetectable even with excellent precision. Choose experimental conditions that produce large effects when possible.
Errors combine via root-sum-of-squares, not simple addition. This means the largest error source dominates. If method error is ±0.1 and execution is ±0.05, total error is ±0.11 (not ±0.15). Improving execution can't fix a fundamentally imprecise method.
When effect size is at least twice the error magnitude, you can usually draw conclusions. Between 1:1 and 2:1 is questionable. Below 1:1, the effect is "lost in the noise" and no reliable conclusion is possible.
With n replicates, effective error reduces by √n. Three replicates reduce error by 1.73×, six replicates by 2.45×. This is why the 96-well plate format (enabling duplicates/triplicates) is so powerful - it directly improves signal-to-noise ratio.
If students are using Method 2 (pH meters, ±0.1 pH error) and observe a small effect (ΔpH = 0.15), they may blame "student error" when they can't detect it reliably. The truth: the method itself cannot resolve such small effects. Even perfect execution won't help.
If your hypothesis predicts large effects (ΔpH > 0.5), Method 2 (pH meters) may suffice. If predicting subtle effects (ΔpH < 0.2), you must use Method 3 (spectrophotometry). The method must match the expected effect size.
With enough replicates, even tiny effects become "statistically significant." But ask: is ΔpH = 0.05 biologically meaningful for algal photosynthesis? The converse is also true: a large, important effect might not be statistically significant with too few replicates or too much error.
Better methods reduce error but may increase complexity/cost. More replicates improve confidence but require more time/resources. Good experimental design balances precision needed (based on expected effect size) with practical constraints.