Longitudinal and Panel Sequence Workflows
In gp3sequences: Transparent Analysis of Ordered Categorical Sequences

knitr::opts_chunk$set(collapse = TRUE, comment = "#>")
library(gp3sequences)

Scope

Panel sequences are repeated ordered-state records from the same independent unit. The workflow preserves the panel identifier, occasion, sequence identity, and preprocessing decisions. Distance between occasions is a structural change measure; it is not evidence of learning, adaptation, or causality by itself.

Synthetic data

base <- data.frame(
  participant_id = rep(paste0("p", 1:4), each = 8L),
  occasion = rep(rep(c(1, 2), each = 4L), times = 4L),
  sequence_id = rep(paste0("s", 1:8), each = 4L),
  sequence_order = rep(1:4, times = 8L),
  state = c(
    "A", "B", "C", "D", "A", "B", "D", "D",
    "A", "C", "C", "D", "A", "C", "D", "D",
    "D", "C", "B", "A", "D", "C", "A", "A",
    "D", "B", "B", "A", "D", "B", "A", "A"
  ),
  stringsAsFactors = FALSE
)
head(base)

Prepare and audit the panel

panel <- prepare_sequence_panel(
  base,
  panel_id_col = "participant_id",
  occasion_col = "occasion"
)
panel$index

A unique panel/occasion combination is required by default. This prevents two sequences from being silently treated as the same repeated observation.

Summarise occasions and states

panel_summary <- summarise_sequence_panel(panel)
panel_summary$occasions
head(panel_summary$states)

Quantify within-panel change

changes <- compare_sequence_panel_changes(
  panel,
  method = "levenshtein",
  normalise = "max_length"
)
changes

Alternative distance methods use the same explicit arguments as compute_sequence_distance(). The result compares consecutive occasions within each panel only.

plot_sequence_panel_changes(changes, metric = "distance", type = "individual")
plot_sequence_panel_changes(changes, metric = "distance", type = "summary")

Reporting

Report the panel unit, occasion ordering, distance method, normalisation, sequence counts at each occasion, and any missing occasions. Treat change as a structural description unless a separate design supports stronger inference.