Observation Unit Declaration

One of the most common structural errors in biomedical statistics is treating technical replicates as independent biological observations. Before Licklider runs any analysis, it asks a single foundational question: what does one row in this dataset represent? The answer locks the unit of observation for the entire session.

STEP 1 —The Pitfall: The Unit of Analysis Problem

Pseudoreplication —using multiple measurements from the same biological unit as independent observations —is one of the most cited statistical errors in life sciences research. A 2018 survey of 100 papers in top neuroscience journals found that approximately 55% contained at least one potential pseudoreplication error.

The error typically arises because the distinction between the biological unit and the measurement unit is never explicitly declared. Common scenarios:

• Tissue slices treated as independent animals. Six brain slices from the same mouse are entered as six rows. An unpaired t-test is run with n = 6 per group, when the true biological n is 1 per group.
• Technical replicates treated as biological replicates. Three PCR runs from the same RNA sample appear as three rows. The analysis reports n = 3, implying three independent samples.
• Cells treated as independent animals. Flow cytometry from 10,000 cells per mouse are entered as individual rows. The t-test runs with n = 10,000 and produces p < 0.0001 —for a comparison with 3 mice per group.
• Measurements treated as patients. Multiple imaging timepoints per patient are entered as independent rows, inflating the effective sample size.

STEP 2 —Journal Requirement

NC3Rs ARRIVE 2.0 guidelines (adopted by over 1,000 journals including Nature Methods, PLOS Biology, and Journal of Neuroscience) require explicit reporting of "the number of animals used in each experiment and the number analyzed," with a clear distinction between biological and technical replicates.

Nature Cell Biology's reporting guidelines state: "When reporting data from experiments using cell lines or primary cultures, the number of independent biological experiments (N) should be distinguished from the number of technical replicates within each experiment (n)." eLife's statistics checklist similarly requires authors to "clarify whether the reported n refers to the number of biological replicates or technical replicates."

Without a declared observation unit, there is no structural basis for answering these questions. Licklider's Observation Unit Declaration ensures the answer is captured at the start of every analysis, not reconstructed from memory at submission.

STEP 3 —Licklider's Solution

Input

• An interactive declaration dialog, presented at dataset upload, asking the researcher to specify the observation unit from a curated list: Animal / Donor / Patient / Cell / Tissue Slice / Well / Sample / Measurement / Other (custom label).
• For each non-leaf unit (e.g., "Tissue Slice"), Licklider prompts for the parent unit (e.g., "Animal") and the subject ID column that links measurements to their biological source. This enables automatic detection of nested structure.
• When repeated measurements or nested structure are detected in the uploaded dataset, the declaration is required before any inferential figure can be generated.

Output

• Locked observation unit label. The declared unit is displayed throughout the Licklider interface —in figure panels, guard messages, and export previews —so the researcher is always reminded of the analytical foundation in use.
• Nested structure summary. When a parent unit is declared (e.g., slices nested within animals), Licklider displays the nesting relationship and the recommended aggregation strategy in text form (e.g., "mean per animal → test between animals").
• N disclosure integration. The declared observation unit determines how N is counted in the N Disclosure panel (see 1.2.1). Biological N and technical N are tracked separately.
• Methods text contribution. The declared unit flows directly into the auto-generated Methods text (see STEP 4).

Guard

STEP 4 —Draft Output (Draft / Needs review)

The declared observation unit is automatically reflected in the auto-generated Methods text. Stable declaration IDs for audit trail purposes are planned for a future release.

Example: Tissue slice data, animal as observation unit

The unit of observation was the individual mouse (n = 6 per group;
n = 4 tissue slices per mouse). To prevent pseudoreplication, measurements
from the four slices of each animal were averaged prior to statistical
analysis (slice-level n = 24 per group; animal-level n = 6 per group used
for all inferential tests). All analyses were conducted at the biological
replicate level (animal) as declared in the Licklider Data Contract.

Example: Cell line, biological replicate as observation unit

The unit of observation was the independent biological replicate (n = 4
per condition), defined as a distinct passage of cells treated and
measured on separate occasions. Within each replicate, triplicate
technical measurements were averaged prior to group-level analysis.
Reported n values refer to biological replicates throughout, as declared in the Licklider Data Contract.

Related Features