The R Book

Replication and Pseudoreplication

To qualify as replicates, measurements must have the following properties:

• They must be independent.
• They must not form part of a time series (data collected from the same place on successive occasions are not independent).
• They must not be grouped together in one place (aggregating the replicates means that they are not spatially independent).
• They must be of an appropriate spatial scale;
• Ideally, one replicate from each treatment ought to be grouped together into a block, and each treatment repeated in many different blocks.
• Repeated measures (e.g. from the same individual or the same spatial location) are not replicates (this is probably the commonest cause of pseudoreplication in statistical work).

Pseudoreplication occurs when you analyse the data as if you had more degrees of freedom than you really have. There are two kinds of pseudoreplication:

• temporal pseudoreplication, involving repeated measurements from the same individual;
• spatial pseudoreplication, involving several measurements taken from the same vicinity.

Pseudoreplication is a problem because one of the most important assumptions of standard statistical analysis is independence of errors. Repeated measures through time on the same individual will have non-independent errors because peculiarities of the individual will be reflected in all of the measurements made on it (the repeated measures will be temporally correlated with one another). Samples taken from the same vicinity ...