A dataset may contain hundreds of measurements and still have only a handful of independent experimental units.
The distinction matters because statistical inference depends on biological independence, not simply on the number of observations recorded by an instrument or generated by repeated measurements.
Cells from the same culture dish, technical replicates from the same biological sample, multiple fields from the same tissue section, repeated measurements from the same animal, or several organoids derived from one donor may provide valuable information about variability. But they do not automatically represent independent biological replicates.
When dependent observations are treated as independent samples, the apparent sample size can become artificially inflated. Standard errors may shrink, confidence in the estimated effect may appear stronger than the design supports, and statistical significance may emerge from replication structure rather than from genuine independent evidence.
The correct experimental unit is therefore determined by the level at which the experimental intervention or biologically meaningful sampling occurs.
This does not mean that technical or nested measurements should be discarded. They can often be incorporated through averaging, hierarchical models, mixed-effects approaches, or other statistical strategies that preserve their information without pretending they are independent.
The practical question is simple but powerful: if one biological source were removed, how many genuinely independent sources of evidence would remain?
Clarifying the experimental unit is not a reporting technicality. It can determine whether an effect is statistically interpretable and, in some cases, whether the central scientific conclusion is supportable at all.