LICUR and ISOA person-oriented pattern classification

Description

Clusters people's multivariate profiles at successive times and tests which movements between profile types occur more or less often than chance.

The person-oriented approach treats a person's pattern of scores, not single variables, as the unit of analysis. LICUR first removes a small residue of unclassifiable outliers, clusters profiles separately at each time point, and cross-tabulates cluster membership between adjacent times. Configural frequency analysis then identifies transitions that occur more often than chance (types) and less often (antitypes); the same linking can join profiles from different domains. ISOA instead pools every person-at-a-time profile, classifies them all together into one time-invariant set of typical states, checks that the classification holds at every time, and then studies changes in state frequencies and individual movement between states, displayed as stream charts. Both methods show stable and changing individuals and sequences that never occur, which variable-oriented averages hide.

Attributes

Loss type
Unmigrated
Effort
Weeks

Steps

  1. Choose the profile variables (for example usage, satisfaction and confidence scores) and the time points.
  2. LICUR: remove a residue of cases that fit no cluster (typically a few per cent), then cluster profiles separately at each time point.
  3. LICUR: cross-tabulate cluster membership between adjacent times and test each cell against chance with configural frequency analysis to find types and antitypes.
  4. ISOA: pool all person-occasion profiles, classify them into a single set of typical states, and check that the states are equally well represented at each time.
  5. ISOA: track state frequencies over time and individual transitions, and draw stream (Sankey-style) charts.
  6. Interpret stable individuals, common pathways and sequences that never occur, and relate them to background variables.

Sources

  • Handbook of Child Psychology Vol 1 (Lerner et al., 2015)

Also known as

  • LICUR
  • Linking of Clusters after Removal of a Residue
  • ISOA
  • I-States-as-Objects Analysis
  • LICUR and ISOA: person-oriented classification of patterns over time

Categories

  • Analytics & Behaviour
  • Statistics & Modelling