Paired vs Unpaired Guard
How Licklider determines whether a comparison is paired or independent, what it asks when the design is ambiguous, and how the answer affects test selection.
Whether observations are paired or independent is one of the most consequential choices in a group comparison. Paired designs — where the same subject appears in more than one condition — have lower variance and require different tests than independent-group designs. Using the wrong test for the design leads to either inflated or deflated false positive rates.
Licklider checks the pairing structure of every comparison and asks you to confirm when the design is ambiguous.
How Licklider assesses pairing
When a group comparison is requested, Licklider looks for a subject ID column in the dataset. If the same ID appears in more than one group, the design is treated as potentially paired.
Licklider then checks whether the pairing is consistent: whether each subject appears the same number of times across conditions and whether the structure is compatible with a paired test.
If pairing is clearly present and consistent, the paired interpretation is used directly. If the structure is ambiguous — for example, if some IDs appear in more than one group but the pattern is irregular — Licklider asks you to confirm.
What you are asked to confirm
When the pairing structure is ambiguous, Licklider presents a confirmation step with two options:
Paired observations (recommended) The observations are matched — the same subject, animal, or sample contributes measurements across conditions. Select this when the repeated structure is intentional and the pairing is valid.
Independent groups The observations are independent — each row comes from a distinct biological unit with no intended matching. Select this when subject IDs appear across groups for reasons other than experimental pairing, for example if IDs were reused across unrelated experiments.
How the answer affects test selection
The pairing confirmation directly determines which statistical test runs:
| Design | Normality | Test |
|---|---|---|
| Paired | Normal | Paired t-test |
| Paired | Non-normal | Wilcoxon signed-rank |
| Paired, 3+ groups | Normal | Repeated-measures ANOVA |
| Paired, 3+ groups | Non-normal | Friedman |
| Independent | Normal | Welch's t-test |
| Independent | Non-normal | Mann-Whitney U |
| Independent, 3+ groups | Normal | One-way ANOVA |
| Independent, 3+ groups | Non-normal | Kruskal-Wallis |
If you override the pairing assessment in the Chat, the override is validated against the data structure. If the selected design is not compatible with the data — for example, if you select paired but the IDs are not consistent across groups — Licklider will flag the mismatch.
Where the result appears
The pairing assessment and your confirmation are visible in the Inspector. If the pairing structure is unresolved, the Inspector will indicate that confirmation is required before claim-bearing export is allowed.
What this page does not cover
- How subject ID columns are identified → see ID, Batch, and Timepoint Columns
- How pseudoreplication relates to pairing → see Pseudoreplication Detection
- How repeated measures analyses are structured → see Repeated Measures Model Suggestion