Description
Search coded event streams for recurring sequences that occur in the same order with consistent time intervals, building hierarchically into longer hidden patterns.
T-pattern detection searches a time-stamped or ordered event record for pairs of events A and B where B follows A within a critical interval more often than chance would predict. Detected pairs become new units, and the search repeats so that patterns nest hierarchically into longer sequences that may be invisible to observers because other events intervene. Unlike lag sequential analysis it does not require events to be adjacent, and unlike sequential pattern mining it tests interval consistency against a null model. The THEME software runs the detection, with randomisation or shuffling checks to estimate how many patterns would arise by chance. Comparing pattern sets across sessions shows how the organisation of behaviour changes over time. Settings such as minimum occurrences and significance level strongly affect results, so they must be reported, and patterns still need interpretation.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Unmigrated
- Stage
- Discovery, Evaluation
- Data
- Quantitative
- Effort
- Weeks
Steps
- Code sessions with timestamps or at least event order preserved, using a tool such as LINCE.
- Check agreement between coders before analysis.
- Export to THEME and set parameters (minimum occurrences, significance level, interval type); report them.
- Run detection and compare against randomised or shuffled data to estimate chance patterns.
- Compare detected patterns across sessions, groups or time points.
- Interpret how pattern structure and complexity change, checking examples in the raw recordings.
Sources
- The Sage Handbook of Mixed Methods Research Design (Poth, 2023)
Also known as
- T-pattern analysis
- TPA
- Hidden temporal pattern detection
Categories
- Analytics & Behaviour
- Statistics & Modelling