Pseudoreplication Detection

Pseudoreplication treats non-independent measurements as if they were independent. It is not a subtle statistical error —it is a structural flaw that inflates n, deflates SE, and makes effects appear more reliable than they are. It is also extremely common in cell biology, animal experiments, and multi-site clinical research.

STEP 1 —The Pitfall

Classic pseudoreplication patterns: analyzing cells from the same culture dish as independent observations; treating technical replicates (multiple wells from the same animal) as biological replicates; treating sister cohort animals (from the same litter, cage, or breeding pair) as fully independent when their shared environment creates non-independence.

The inflation factor for n is equal to the number of technical replicates per biological unit. With 3 animals and 10 cells per animal, pseudoreplication claims n=30 when the biological n is 3. The SE is inflated by ~0 ~3.2, and p-values can be artificially reduced by an order of magnitude.

[Image placeholder: Hierarchical data structure diagram showing 3 animals x4 slices each x5 cells per slice = 60 total measurements. Correct analysis tree: n_biological=3. Pseudoreplication analysis: n_inflated=60. Arrow showing the 20-fold inflation in n and the ~0 ~4.5xdeflation in SE.]

Hierarchical nesting: 60 total observations from 3 biological units. Pseudoreplication claims n=60; correct n=3.

STEP 2 —Journal Requirement

ARRIVE 2.0 requires explicit statement of the unit of analysis and the unit of randomization. Nature Methods: "n refers to biologically independent samples." eLife editorial policy now rejects papers where pseudoreplication is detected in review. Most neuroscience and cell biology journals have strengthened their requirements following the pseudoreplication awareness movement (Lazic et al., 2010; 2018).

STEP 3 — What Licklider Currently Provides

What the current product does

Card 17 flags possible pseudoreplication when observation and data-contract context suggest that samples may not be independent.
The user is asked to confirm whether the data should be treated as repeated/non-independent, as independent samples, or left unresolved for descriptive-only handling.
The sampling_independence_status records the user's resolution, and confidence levels are tracked.
Export-facing behavior and disclosure text can change depending on how the risk is resolved. Unresolved cases can affect admissibility for claim-bearing output.

What the product does not do

The product does not automatically detect hierarchical nesting trees or calculate biological n vs. measurement n.
The product does not compute inflation factors.
The product does not offer automated aggregation to biological unit level or mixed-model specification as resolution buttons.
Pseudoreplication risk is surfaced through heuristic signals and user confirmation, not through statistical proof.

Known limitations

Confidence depends on the available observation structure and data-contract signals, so weaker cases should be read as prompts for review rather than automatic verdicts. The product can help surface and gate the issue, but the experimental design still needs human interpretation.

For the detailed quality-checks documentation, see Pseudoreplication Detection in the quality-checks docs.

[Image placeholder — not current product UI. Illustrative concept of a pseudoreplication detection panel.]

Illustrative concept only. The current product surfaces pseudoreplication risk via Card 17 and confirmation flow, not via this specific panel layout.

STEP 4 — Draft Output (Draft / Needs review)

The following is an example of how a researcher might write a pseudoreplication disclosure, not output that Licklider generates automatically.

Each animal contributed multiple technical replicates (n = 5 cells per
animal; 3 animals per group). To avoid pseudoreplication, data were
aggregated by averaging technical replicates within each animal prior
to statistical analysis. Statistical comparisons were performed on
animal-level means (n = 3 biological replicates per group).
Individual cell-level data are shown in the figure; statistics are
based on biological n.