mescla.emma.rank

How many components to retain – and therefore how many end-members.

The number of end-members is k + 1, where k is the retained rank: a k-dimensional mixing subspace is bounded by a simplex with k+1 vertices (Christophersen & Hooper, 1992). Choosing k is therefore the single most consequential decision in an EMMA, and the “rule of one” alone is a weak basis for it. Every criterion here is offered side by side, and rank_summary() prints them together so the choice is made with eyes open. The decisive evidence is usually the residual structure (see mescla.emma.diagnostics), not any of these rules.

Functions

broken_stick	Retain components exceeding the broken-stick expectation.
cumulative_variance_rule	Smallest k whose components explain at least threshold of the variance.
parallel_analysis	Horn's parallel analysis: retain components beating uncorrelated noise.
rank_summary	Scree table with every retention criterion side by side.
rule_of_one	Number of eigenvalues >= 1 (Kaiser's criterion; Joreskog et al., 1976).