Taxometric analysis

Description

Tests whether individual differences form natural categories (taxa) or a continuum before people are analysed or described as types.

Cluster analysis and latent class analysis will produce groups whether or not natural groups exist, and imposing categories on dimensional data loses power and predictive validity. Meehl's taxometric procedures test the question empirically. MAXCOV examines how the covariance of two indicators changes across levels of a third; MAMBAC examines mean differences above and below sliding cut points. Taxonic structure produces peaked or distinctive curves, dimensional structure flat or concave ones. Modern practice compares observed curves with curves from simulated categorical and dimensional data matched to the sample, summarised in a comparison curve fit index. Fraley and Spieker applied the procedures to infant attachment, long described in types, and found mostly continuous variation. The procedures need adequate samples, often several hundred cases, and valid indicators that separate the putative groups.

Attributes

Loss type
Unmigrated
Effort
Days

Steps

  1. Select three or more valid indicators of the construct (for example attitude or behaviour scores thought to define a user type).
  2. Run taxometric procedures such as MAXCOV and MAMBAC on the indicators.
  3. Simulate matched categorical and dimensional comparison data sets and compare observed curves with each, using a comparison curve fit index.
  4. Decide whether the construct is better treated as categories or dimensions, using several procedures for consistency.
  5. Only if categories are supported, apply clustering or latent class models; otherwise describe users on continuous dimensions.

Sources

  • Handbook of Child Psychology Vol 3 (Lerner & Lamb, 2015)

Also known as

  • Taxometric analysis (categories versus continua)
  • Taxometrics
  • MAXCOV
  • MAMBAC

Categories

  • Analytics & Behaviour
  • Statistics & Modelling