Replication and Randomisation
Martin Bader, Sebastian Leuzinger · 2024
Two fundamental tools that help us with the causal assignment of variance are replication and randomisation. Both are intricately related, and both also relate to the concept of statistical independence. Considerations of statistical independence are important but rarely absolute. Both experimental and observational studies are necessarily confined spatially and temporally and often need to be confirmed via additional, independent data. Higher replication will always result in a higher signal-to-noise ratio, and therefore higher chances to find treatment effects, should they be present. Replication and statistical independence are pivotal concepts in statistical hypothesis testing, they are tightly coupled — replication ensures statistical independence, and pseudo-replication ultimately is caused by a lack thereof. Randomisation is a tool to ensure statistical independence. Replication as the tool to separate signal from noise is simply indispensable, but without properly characterising at which level the experimental units are independent.