Is optimal matching the same as sequence analysis?

No. Optimal matching is one method for comparing sequences within a broader family of descriptive and analytical approaches.

Must every sequence have the same length?

Not always, but unequal observation windows and censoring affect interpretation and may require suitable comparison methods.

Can two groups have the same state proportions at each time but different individual histories?

Yes. Aggregate snapshots do not show exactly who moved between states, so different transitions can produce the same time-specific proportions.

AI Glossary / Data science and analytics / Intermediate

Sequence analysis in social sciences

Definition

Sequence analysis in the social sciences studies ordered histories of states or events to understand how social processes unfold. It examines features such as order, timing, duration and transitions, rather than reducing a person's or organization's history to a single snapshot.

What it means

A career can be represented as monthly states of education, employment and unemployment. A service journey can be represented as events such as application, review and completion. These representations preserve aspects of the process that totals alone can hide.

The question determines the representation. A state sequence records the condition at successive positions, often at regular time intervals. An event sequence records occurrences and their order, potentially with timestamps. Duration information can also be represented as spells: a state together with how long it lasts.

Sequence analysis is a broad approach, not one algorithm. It can include visualization, descriptive summaries, comparison of histories and grouping of similar trajectories. Its conclusions depend on how states, time and similarity are defined.

How it works

Define the population, observation window, time origin and state or event categories. Decide whether histories align by calendar date, age, time since enrollment or another meaningful reference. These choices can change what counts as early or late.

Prepare longitudinal records while distinguishing missing observations from substantive states. Summarize time spent in states, transitions or other features and inspect sequence plots. An aggregate state-distribution plot shows the composition at each time but may hide individual transitions.

To compare whole histories, choose a dissimilarity measure suited to the question. Optimal matching is one option; other measures emphasize different aspects of timing, order or duration. Clustering can then identify candidate groups, but the clusters require interpretation and sensitivity checks.

Return to the cases and context. A common trajectory is not necessarily desirable, and a rare one is not automatically an error. Describing sequence patterns does not by itself establish the causes of transitions.

Three practical examples

A labor researcher compares education-to-employment histories across cohorts while accounting for the observation window.
A public-service evaluator studies the order and duration of stages in an application process.
An organisational researcher examines how teams move through training, pilot use and routine adoption of an AI tool, retaining interruptions rather than recording only the final status.

Think of it this way

Think of comparing complete stories rather than counting how often each character appears. Two stories can contain the same characters but differ in order, timing and outcome.

The analogy has limits because sequence analysis uses explicitly coded observations and formal comparisons. The analyst's categories are a simplified representation of social experience, not a complete account of each person's motives or circumstances.

Worked example

Consider two invented six-month histories using T for training, E for employment and U for unemployment. Person A has T, T, E, E, U, U. Person B has U, U, T, T, E, E.

Both spend two months in each state, so a table of total time by state would describe them identically. Their sequences differ: A moves from training into employment and later unemployment, while B moves from unemployment through training into employment. Their final observed states also differ.

Now suppose these six positions mean months since program entry for A but calendar months for B. Comparing them as if their timing origins matched would introduce another problem. The analyst must align the histories according to the research question.

This example shows why state totals, ordering and time alignment provide different information. It does not establish that training caused either person's employment outcome.

Illustrative case study

Illustrative fictional case: Meadow Skills studies participant journeys through a training service. Its initial grouping treats a missing monthly record as unemployment, creating apparent interruptions in several histories.

The team reviews the source records, distinguishes unavailable information from observed labor-market status and documents the treatment of gaps. It compares alternative similarity measures because its substantive question concerns the timing of transitions after enrollment, not merely whether the same states ever occurred.

It uses representative histories to explain the resulting groups and avoids labeling them as fixed personality types. The fictional lesson is that sequence coding and comparison choices shape the result as much as the final grouping algorithm.

Common misconceptions and mistakes

Do not confuse sequence analysis with simply counting events. Do not assume that every sequence position represents the same amount of time unless the representation establishes that.

Avoid treating missingness as an ordinary state without justification. Do not assume there is one universally best distance measure or clustering solution. A state-distribution chart can look stable even when many individuals change states, and a sequence pattern alone does not establish causality or explain why people followed it.

Business use

Sequence analysis can help organizations understand journeys, transitions and repeated process stages. Begin with a question that genuinely depends on history, such as whether delays accumulate before or after a particular stage.

Document categories, time origin, observation windows and missing-data rules. Examine sensitivity to similarity measures and grouping choices, then connect summaries to representative cases. If AI narrates the patterns, require it to distinguish observed sequences from inferred motivations. The analysis should describe pathways without turning a statistical grouping into an unsupported judgment about an individual.

Check your understanding

Two people each spend three months in training and three in employment. Does this establish that their six-month trajectories are the same?

Answer: No. The order, timing and number of transitions can differ. The sequence is needed to distinguish those histories.

Related terms

Longitudinal data
Optimal matching
Clustering
Censoring
state sequence
event sequence
spell duration
transition
life-course analysis

Learn next

Read longitudinal data and optimal matching. Represent a short journey both as successive states and as state-duration spells. Compare two histories with identical state totals, then explain which aspects of timing and order matter to the question you want to answer.

Sources

Gabadinho and colleagues, Analyzing and Visualizing State Sequences in R with TraMineR: https://www.jstatsoft.org/article/view/v040i04

TraMineR, sequence-analysis scope and descriptive tools: https://traminer.unige.ch/index.shtml

TraMineR, research on sequence dissimilarity, timing and duration: https://traminer.unige.ch/documentation.shtml

TraMineR, state and spell representations: https://traminer.unige.ch/doc/seqformat.html

The two histories and training-service case are fictional teaching examples.

Previous: Sequence Next: Sequence effect