Distribution

When specifying a mixture model (which is done when using Create > Segments) the following options are available for specifying the shape of the mixing distribution.

Finite Assumes that there is no variation within each class. This is the default.

Multivariate Normal – Spherical Forces the classes to be “spherical” (i.e., with equal variances for every variable).

Multivariate Normal – Diagonal Estimates separate variances for each variable (i.e., in the case of Number and Number - Multi questions, this is equivalent to transforming the data to z-scores).

Multivariate Normal – Block Diagonal Permits correlation between levels of the same attribute, but not between attributes.

Multivariate Normal – Full Covariance Permits correlations between all attributes and attribute levels.

These assumptions are ordered from most restrict to least restrictive.

Pool variance Constrains the variance and covariance assumptions to be common across classes. For example, if Multivariate Normal – Spherical and Pool variance are both selected, a single variance parameter is employed for the entire model.

See also

• Statistical Model for Latent Class Analysis
• Latent Class Analysis and Mixture Models

Further reading: Market Segmentation Software