4.3.1 Replication
By replication we mean the application of each treatment of interest to more than one
experimental unit (animal, panellist etc.). Replication has two benefits. Firstly, the
population mean response for a particular treatment is estimated by averaging the
observed responses across all the units receiving that treatment. The greater the number
of individual observations contributing to such a treatment sample mean, the greater the
precision with which the treatment population mean is estimated or, equivalently, the
greater the power of the experiment in detecting differences between treatment
population means.
Secondly, two units that receive the same treatment will not necessarily yield the same
response, as variation in response between identically treated units is due to experimental
error. By examining the variation in response between units within treatments, an
estimate of the experimental error variance can be made. This forms a baseline against
which all other apparent differences can be measured, in particular, differences between
the various treatment sample means.