Repeated Measures Workflow

How to analyze repeated measurements from the same subjects in Licklider, from identifying the data structure to choosing between paired tests, Friedman, and mixed models.

Repeated measures data occurs when the same subjects are measured more than once — across time points, treatment conditions, or both. Because observations from the same subject are correlated, they cannot be treated as independent, and standard group comparison tests do not apply directly.

This page describes how to handle repeated measures data in Licklider.

This workflow produces concrete outputs rather than a method name alone: depending on the design, expect a paired repeated-measures result, a Friedman omnibus test result, or a mixed-model coefficient table, plus the linked design checks and a longitudinal figure such as a spaghetti plot when requested.

Identifying repeated measures data

Before running the analysis, Licklider needs to know that the data has a repeated structure. The key indicator is a subject ID column — a column that identifies which subject each row belongs to — where the same ID appears in more than one row.

If your dataset has a subject ID column, make sure it is identified in the Data Contract → see Variable and ID Mapping.

When Licklider detects that the same ID appears across multiple conditions or time points, it will ask you to confirm whether the design is repeated measures or independent samples → see Pseudoreplication Detection.

That confirmation step lowers the risk of treating the same subject as if it had contributed independent rows. It does not remove all risk: if the subject ID is missing, declared at the wrong level, reused inconsistently, or collapsed before upload, Licklider may not be able to recognize the repeated structure from the table alone.

Data format

Long format (recommended)

One row per observation. Columns should include:

  • Subject ID
  • Time point or condition
  • Outcome value

Wide format

One row per subject, with one column per time point. Wide format can be reshaped to long format in the Prep panel before analysis → see Table Shape: Wide vs Long.

Choosing the right analysis

Two time points or two conditions

A paired comparison is appropriate. Licklider selects a paired t-test (for normal differences) or Wilcoxon signed-rank test (for non-normal) automatically when subject IDs are consistent across two conditions.

For this path, expect a paired repeated-measures result centered on the test statistic and p-value, with reporting expectations carried by the paired test path rather than by a mixed-model table.

Request it with: "Compare before and after" or "Paired comparison between conditions A and B."

Three or more time points — non-parametric

The Friedman test evaluates whether the outcome differs across time points without assuming normality. Use this when normality cannot be assumed and covariates are not needed.

For this path, expect an omnibus repeated-measures result centered on the Friedman statistic and p-value. It is a good fit when the main question is whether anything changes across repeated conditions, not which subject-specific trajectory best explains the data [1].

Request it with: "Run a Friedman test across the three time points."

Three or more time points — with covariates or complex design

A generalized linear mixed model (GLMM) with a random intercept accounts for the correlation between repeated observations from the same subject. Use this when:

  • Covariates need to be included
  • The design is unbalanced (subjects have different numbers of observations)
  • The outcome is binary (use Binomial family)

For this path, the returned output is more model-like: expect fixed-effect coefficients, standard errors, test statistics, p-values, 95% confidence intervals, convergence status, and numerical warnings as documented on the GLMM: Gaussian and Binomial page.

Request it with: "Fit a mixed model with subject as a random effect" or "Use a GLMM to account for repeated measurements."

For more detail → see Repeated Measures and Mixed Models and GLMM: Gaussian and Binomial.

The reason Licklider separates these paths is statistical as well as practical: paired tests are enough when exactly two matched observations exist, Friedman is useful when a rank-based omnibus test is preferred without covariates, and GLMM is the more appropriate path when repeated observations are incomplete, unbalanced, or need covariate adjustment [1, 2, 3].

For currently documented method availability, see Supported Methods Matrix.

Line plots for longitudinal data

For data with multiple time points, a line plot with individual trajectories (spaghetti plot) shows within-subject change more clearly than group means.

Request it with: "Show a spaghetti plot" or "Plot individual trajectories over time."

The goal is not decoration. Individual trajectories make within-subject change visible and can reveal patterns that disappear when only group summaries are shown.

What Licklider automates and what it cannot infer

Licklider automates several parts of this workflow:

  • It can detect repeated subject IDs across conditions or time points
  • It can ask you to confirm whether the design is repeated or independent
  • It can route two-condition repeated designs toward a paired analysis automatically
  • It can guide wide tables toward long-format preparation before analysis

Those protections help, but they are not a complete substitute for study-design judgment. Licklider cannot determine automatically whether the subject ID column was declared at the correct biological level, whether visit order or time spacing changes the scientific interpretation, whether missing visits are ignorable, or whether a random-intercept model is sufficient when subject-specific slopes matter.

Those limits matter because the result can still look technically plausible while answering the wrong question. A repeated design can be misread as independent if subject structure is hidden, and a simple random-intercept model can be too coarse when the scientific question depends on subject- specific trajectories.

Common problems

Same ID appears in multiple groups unexpectedly

If subject IDs appear across groups but the design is intended to be independent, Licklider will flag this as a potential pseudoreplication concern. Confirm whether the design is paired or independent → see Paired vs Unpaired Guard.

Wide format data

If each time point is a separate column, reshape to long format before requesting the analysis → see Table Shape: Wide vs Long.

Repeated structure is only partly visible in the table

If uploads contain summary rows instead of one row per subject visit, or if the subject ID is missing or misdeclared, Licklider may not be able to detect that the observations are correlated. In that situation, p-values can look more convincing than the design really supports.

Design rationale and references

Licklider keeps repeated-measures guidance simple at the workflow level because most readers first need to answer a design question: "Is this paired, rank-based repeated, or model-based repeated data?" The workflow therefore starts with structure recognition, then routes toward the smallest method that matches the question instead of forcing every reader into a mixed-model specification up front.

This is also why the page does not present a full repeated-measures ANOVA workflow as the default. Friedman remains useful as a distribution-light omnibus repeated-measures test [1], while mixed models are preferred when follow-up is irregular, covariates matter, or outcomes are not all Gaussian [2, 3].

References

  1. Friedman, M. (1937). The use of ranks to avoid the assumption of normality implicit in the analysis of variance. Journal of the American Statistical Association, 32(200), 675-701.
  2. Laird, N. M., & Ware, J. H. (1982). Random-effects models for longitudinal data. Biometrics, 38(4), 963-974.
  3. 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.

What this page does not cover

  • Group comparison without repeated measures → see Group Comparison Workflow
  • GLMM implementation details → see GLMM: Gaussian and Binomial
  • Pseudoreplication detection → see Pseudoreplication Detection
  • Classical repeated-measures ANOVA with sphericity testing and Greenhouse-Geisser or Huynh-Feldt correction
  • Random-slope models and richer covariance structures as documented outputs on this page