clustassoc_like_typology_validation()

clustassoc_like_typology_validation() checks whether sequence cluster solutions at different k values account for the association between a covariate and sequence dissimilarities. It is analogous to WeightedCluster clustassoc().

Intuitively: if adding cluster labels greatly reduces the remaining pseudo-R² of the covariate, the typology has captured much of the sequence variation associated with that covariate. If little changes, the clustering has not absorbed that association.

Function Usage

clustassoc_like_typology_validation(
    diss,
    covariate,
    clustering_labels_by_k,
    *,
    sample_weights=None,
    covariate_is_categorical=False,
    verbose=False,
)

R / Literature Parameter Mapping

Sequenzo	R	Notes
diss	Distance matrix input to clustassoc	Square n × n
covariate	Covariate tested for association	Numeric or categorical
clustering_labels_by_k	Cluster labels at each k	Dict keyed by number of clusters
Pseudo-R² computation	TraMineR::dissmfacw() via distance_multifactor_anova	Multifactor distance ANOVA

Entry Parameters

Parameter	Required	Type	Description
diss	✓	ndarray	Square symmetric distance matrix (n × n).
covariate	✓	array-like	Length-n covariate aligned with diss rows/columns.
clustering_labels_by_k	✓	dict	Map k → length-n cluster label vector.
sample_weights	✗	array-like / None	Optional 1D weights of length n.
covariate_is_categorical	✗	bool	If True, one-hot encode the covariate.
verbose	✗	bool	Print progress per k.

What It Returns

A DataFrame with one row per k in clustering_labels_by_k (sorted by k):

Column	Description
k	Number of clusters.
pseudoR2_original	Pseudo-R² of covariate on distances (no clustering in model).
pseudoR2_remaining_after_clustering	Pseudo-R² of covariate after adding cluster dummies.
association_unaccounted_share	pseudoR2_remaining / pseudoR2_original.
association_accounted_share	1 - association_unaccounted_share. Higher values suggest the typology explains more of the covariate–sequence association.

If pseudoR2_original is 0, unaccounted/accounted shares are NaN.

association_accounted_share can be slightly negative if the covariate pseudo-R² increases after adding clustering—usually from numerical error, model order, weights, or a poor cluster solution. Treat such cases as no evidence that the typology accounts for the association.

Example

from sequenzo import clustassoc_like_typology_validation

validation = clustassoc_like_typology_validation(
    diss=distance_matrix,
    covariate=education,
    clustering_labels_by_k={
        3: labels_k3,
        5: labels_k5,
        8: labels_k8,
    },
    covariate_is_categorical=True,
)

print(validation)

R Counterpart

• Closest R function: WeightedCluster::clustassoc()
• Mapping note: Uses Sequenzo distance_multifactor_anova (dissmfacw-style) for pseudo-R² terms.

Notes

• diss must be square; covariate and each label vector must have length n = diss.shape[0].
• Cluster labels are one-hot encoded with drop_first=True.
• Useful after sequence clustering when you need to know how much of a covariate–sequence association is explained by the typology.

Authors

Code: Yuqi Liang

Documentation: Yuqi Liang

References

Bolano, D., & Studer, M. (2020). The link between previous life trajectories and a later life outcome: A feature selection approach.

Unterlerchner, L., Studer, M., & Gomensoro, A. (2023). Back to the features. Investigating the relationship between educational pathways and income using sequence analysis and feature extraction and selection approach. Swiss journal of sociology, 49(2), 417-446.