Repeated Measures Model Suggestion

When repeated observations from the same subject are present across time points or conditions, a standard group comparison may not be the most appropriate analysis. Licklider detects this structure and suggests an analysis approach that accounts for it.

This is a suggestion, not an automatic switch. Licklider does not change the analysis model on its own — it identifies the structure, describes the appropriate approach, and records the guidance as part of the figure's disclosure.

When a suggestion appears

Licklider surfaces a model suggestion when the confirmed analysis unit is a biological unit and time points are present:

This suggestion depends on the repeated structure being visible in the uploaded data and confirmed analysis-unit metadata. If subject IDs, timepoint labels, or replicate structure are missing, inconsistent, or collapsed before upload, Licklider may not be able to recognize that the rows are repeated observations from the same unit. In that situation, a repeated design can still be analyzed as if it were independent, which can make the resulting inference look more certain than it should.

Two time points When the same subjects are measured at two time points, Licklider suggests:

A paired analysis (paired t-test or Wilcoxon signed-rank), which accounts for the matched structure
A change score analysis, where the within-subject change is computed and analyzed directly
A paired line plot, which makes the within-subject trajectory visible

Three or more time points When the same subjects appear at three or more time points, Licklider suggests:

A repeated-measures model or mixed model, which accounts for the correlation between observations from the same subject over time
A spaghetti plot, which shows individual trajectories rather than group means

For line charts specifically, a suggestion also appears when the subject-to-timepoint mapping and repeated-measure patterns are consistent with a longitudinal design.

What appears here is guidance output rather than a new test result: a model suggestion, a figure suggestion, and a disclosure path that records whether you acted on the suggestion.

What the suggestion means

The suggestion is guidance, not a requirement. It tells you that the data structure you have is better served by a model that accounts for within-subject correlation than by a model that treats all observations as independent.

Licklider keeps this as a suggestion rather than switching models automatically because repeated-measures structure still requires study-design judgment. The software can detect patterns that look longitudinal, but it should not silently choose between a paired test, a change-score analysis, a repeated-measures model, or a mixed model without you confirming that those rows really represent the same biological unit and that the suggested model matches the scientific question.

Ignoring the suggestion is possible, but doing so means that the correlation structure of the data is not accounted for in the analysis. The figure will include a disclosure noting that the structure was identified and what was suggested.

That disclosure is part of the safety design: it keeps the repeated structure visible even when you choose not to adopt the suggested model immediately.

Acting on the suggestion

To follow the suggestion, describe the intended approach in the Chat:

"Use a paired t-test for this comparison"
"Analyze the change score from baseline"
"Fit a repeated-measures model"

Licklider will update the analysis accordingly.

For complex repeated-measures designs with many time points, covariates, or nested grouping structures, a mixed model or GLMM may be appropriate. These can be requested in the Chat. For more detail → see Repeated Measures and Mixed Models.

Effect on export

The repeated-measures guidance does not by itself block export. The disclosure that a repeated-measures structure was identified — and whether it was acted on — is included in the figure's export output.

This page does not block export automatically because a repeated-measures suggestion is advisory rather than a fully adjudicated design error. Some users may still be in exploratory mode, and some datasets need additional context before the repeated structure can be confirmed. Export blocking is reserved for the cases where the analysis-unit status itself remains unresolved.

If the analysis unit status is unresolved alongside the repeated-measures structure, that unresolved status may block claim-bearing export. Resolving the analysis unit confirmation (see Pseudoreplication Detection) addresses this.

What this page does not cover

How the repeated-measures structure is detected → see Pseudoreplication Detection
How pairing affects two-group test selection → see Paired vs Unpaired Guard
Mixed model and GLMM analyses → see Repeated Measures and Mixed Models

Design Rationale & References

Licklider's design choices

Licklider surfaces repeated-measures model guidance because the main statistical risk is treating correlated within-subject observations as if they were independent. The suggestion therefore acts as a guardrail: it points the reader toward paired, change-score, repeated-measures, or mixed-model analyses before an independence-based comparison is over-interpreted [1, 2].

The product keeps this as a suggestion instead of an automatic switch because repeated structure is not only a pattern-recognition problem. It also depends on study intent, analysis-unit definition, and whether the same identifier truly represents the same biological source. That is why Licklider records the guidance in disclosure output and lets the user choose the final modeling path rather than silently rewriting the analysis.

Methodological foundations

Laird, N. M., & Ware, J. H. (1982). Random-effects models for longitudinal data. Biometrics, 38(4), 963-974.
→ Foundational reference for using models that account for within-subject correlation in longitudinal and repeated-measures data.
Gueorguieva, R., & Krystal, J. H. (2004). Move over ANOVA: progress in analyzing repeated-measures data and its reflection in papers published in the Archives of General Psychiatry. Archives of General Psychiatry, 61(3), 310-317.
→ Explains why repeated-measures and mixed-model approaches are often preferable to treating correlated observations with independence-based methods.

Implementation boundaries

Licklider can suggest a repeated-measures path only when the repeated structure is visible in subject IDs, timepoints, and related analysis-unit information.
If those fields are missing, inconsistent, or encoded outside the uploaded table, Licklider may not detect the repeated structure automatically.
This page documents a suggestion and disclosure workflow, not an automatic model switch.
A suggestion does not guarantee that the exact repeated-measures model you need is already fit; it signals that an independence-based analysis is likely too simplistic for the observed structure.