Question 4 of 9 Intermediate Scenario
Why should the experimental unit determine your statistical test?
Short answer
The experimental unit is the smallest independently assigned entity. It determines the true number of independent replicates and therefore the statistical uncertainty. Treating multiple cells from one treated animal as independent animals is pseudoreplication and can make a weak result look convincing.
Why
Begin with the treatment assignment, not with the spreadsheet row. The experimental unit is the smallest independent entity that could receive a treatment separately, such as an animal, patient, culture vessel or randomized clinic. The observational unit can be smaller.
For example, measuring 100 cells within each of five independently treated mice produces five experimental units, not 500. Cells within one mouse share treatment, environment and biological history, so their outcomes are correlated. Pretending all 500 measurements are independent underestimates uncertainty.
A defensible analysis might summarize per mouse, use a hierarchical model with mouse-level effects or apply a design-aware alternative. If entire plates were randomized, the plate may be the experimental unit even when thousands of wells were measured. Define the unit, randomization, pairing, nesting and batch structure before choosing a test.
More technical measurements may improve measurement precision but cannot replace independent biological replication.
Common mistakes
- Using cell count, image count or sequencing reads as biological sample size when treatment was assigned at animal or culture level
The interviewer might ask next
How would the analysis change if each animal contributed measurements at three time points?