Method map

Choose a sequence method from the structural question you are asking, not from the visual result you hope to obtain.

Decision table

Research question	Primary function(s)	Main object / output	Key decision to report
Are the input sequences valid and explicitly ordered?	audit_sequence_data(), validate_sequence_data()	issue / validation tables	ordering, missing-state, duplicate-position policy
Which states dominate overall or by position?	summarise_sequence_states()	state shares / counts	weighting, duration use, grouping
Which direct transitions dominate?	summarise_sequence_transitions()	transition table	self-transitions, denominator
Which exact adjacent patterns recur?	extract_sequence_ngrams()	motif occurrences	motif length, overlap rule
Which non-adjacent ordered patterns recur?	extract_sequence_subsequences()	subsequences	span / gap constraints
What is a representative aligned path?	create_consensus_sequence()	consensus sequence	alignment, tie policy, missing positions
How different are complete sequences?	compute_sequence_distance()	pairwise distance matrix	distance family, normalisation, costs
Do sequences form useful descriptive clusters?	cluster_sequences(), validate_sequence_clusters()	assignments / validation	k, linkage/method, distance choice
Are clusters stable under resampling?	bootstrap_sequence_clusters()	stability results	bootstrap design, seed, repetitions
Which trajectories represent a cluster?	extract_representative_sequences()	medoids / representatives	representation criterion
What does transition structure look like as a graph?	create_transition_network()	weighted directed network	weight normalisation, self-loops
Do recent states improve next-state prediction?	fit_higher_order_transition_model()	higher-order transition model	order, smoothing, backoff
Is a compact latent-state model useful?	fit_sequence_hmm()	categorical HMM	number of states, seed, convergence
Are there multiple latent sequence components?	fit_sequence_hmm_mixture()	mixture model	components, states, seeded fits
Do multiple channels need joint latent modelling?	fit_multichannel_sequence_hmm()	multichannel HMM	channel representation, state count
Do covariates predict transition probabilities?	fit_covariate_sequence_hmm()	covariate HMM	covariates, reference levels
Is a declared group difference compatible with random assignment?	declare_sequence_comparison_design(), test_sequence_group_difference()	randomization inference	assignment mechanism, statistic
How does structure change across panels / waves?	prepare_sequence_panel(), compare_sequence_panel_changes()	panel summaries	panel identity, wave ordering
How do transition probabilities evolve over time?	fit_time_varying_sequence_model()	time-varying model	smoothness, prediction target

Workflow families

Data

Validation & preparation

Begin here whenever sequence order, repeated states, durations, or metadata may be ambiguous.

Read workflow →

Patterns

Motifs & subsequences

Use exact contiguous motifs for adjacent patterns and bounded subsequences for ordered non-adjacent structure.

Read motif workflow →

Geometry

Distance & clustering

Use whole-sequence dissimilarity when order and path shape matter beyond marginal state frequencies.

Read clustering workflow →

Graphs

Transition networks

Use directed networks when states are nodes and observed first-order movement is the core object.

Read network workflow →

Latent

HMM families

Use latent models only when the statistical abstraction is justified and convergence / label exchangeability are reported.

Read HMM workflow →

Design

Group inference

Use declared comparison designs and randomization logic when moving beyond descriptive group differences.

Read inference workflow →

Questions to answer before analysis

1. What is the sequence unit? Participant, trial, session, scanpath, episode, or another unit?
2. What does order mean? Event order, time bins, aligned positions, or panel wave?
3. Are positions comparable across sequences? If not, aligned consensus and position-wise summaries may be misleading.
4. Is duration meaningful? Decide whether counts, event durations, or both enter the estimand.
5. What structural feature is primary? States, motifs, complete path geometry, transitions, latent states, or time-varying probabilities?
6. What is inferential versus descriptive? A group contrast does not become causal merely because it is statistically tested.
7. Which tuning choices matter? Distance costs, k, motif thresholds, model state count, smoothing, seeds, and bootstrap repetitions should be declared.
8. What would falsify the interpretation? Plan sensitivity checks before inspecting the most attractive plot.

Do not reverse-engineer a method from a visually appealing result

Clusters, networks, HMM states, and motifs are representations of the declared sequence structure. They are not self-validating substantive categories.

Suggested analysis path

raw events
   │
   ├── audit / validate
   ▼
prepared ordered sequences
   │
   ├── descriptive summaries ──► states / transitions / paths
   ├── pattern mining ─────────► motifs / subsequences
   ├── whole-sequence geometry ► distance / clustering / stability
   ├── graph structure ────────► transition networks / higher-order models
   ├── latent structure ───────► HMM / mixture / multichannel / covariate HMM
   └── design-aware inference ─► declared comparison / randomization test

Next: copy complete examples or browse the full methodology articles.