16. Pseudoreplication

When Sample Size Is Illusory

1 Introduction

Pseudoreplication is one of the most persistent causes of invalid inference in ecology and biology. It occurs when observations that are not truly independent are analysed as if they were independent replicates.

The result is usually an underestimate of uncertainty and an inflated chance of declaring an effect that is not actually supported by the design.

2 Key Concepts

The chapter is built around the following design principles.

• Pseudoreplication occurs when non-independent observations are treated as independent replicates.
• Experimental units define the true level of replication.
• Subsamples are not replicates of a treatment effect.
• Dependence can be spatial, temporal, or hierarchical.
• Good analysis cannot rescue a fundamentally pseudoreplicated design.

3 Main Idea

The crucial question is always:

What is the experimental unit?

If treatment is applied at the level of ponds, plots, sites, or animals, then those are the units of replication. Multiple subsamples taken within a unit do not create additional independent replicates of the treatment effect.

4 Common Forms

• Simple pseudoreplication: many subsamples from one treated unit are analysed as replicates.
• Temporal pseudoreplication: repeated observations through time on one unit are treated as independent replicates.
• Spatial pseudoreplication: nearby observations inherit shared structure but are analysed as if fully independent.

5 Why It Matters

Pseudoreplication is not a minor technicality. It is a design problem that can invalidate the inferential logic of a test.

What often looks like a large sample size may in fact be only:

• one pond,
• one plot,
• one bay,
• one lake,
• one experimental arena.

6 Practical Rule

Count replicates at the level where the treatment or condition varies independently.

7 Source

This chapter replaces the earlier file:

• 15-pseudoreplication.qmd