Premium
INTERPRETATION OF THE RESULTS OF COMMON PRINCIPAL COMPONENTS ANALYSES
Author(s) -
Houle David,
Mezey Jason,
Galpern Paul
Publication year - 2002
Publication title -
evolution
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.84
H-Index - 199
eISSN - 1558-5646
pISSN - 0014-3820
DOI - 10.1111/j.0014-3820.2002.tb01356.x
Subject(s) - principal component analysis , covariance , trait , biology , statistics , automatic summarization , covariance matrix , intuition , variance (accounting) , mathematics , computer science , artificial intelligence , philosophy , accounting , epistemology , business , programming language
Common principal components (CPC) analysis is a new tool for the comparison of phenotypic and genetic variance‐covariance matrices. CPC was developed as a method of data summarization, but frequently biologists would like to use the method to detect analogous patterns of trait correlation in multiple populations or species. To investigate the properties of CPC, we simulated data that reflect a set of causal factors. The CPC method performs as expected from a statistical point of view, but often gives results that are contrary to biological intuition. In general, CPC tends to underestimate the degree of structure that matrices share. Differences of trait variances and covariances due to a difference in a single causal factor in two otherwise identically structured datasets often cause CPC to declare the two datasets unrelated. Conversely, CPC could identify datasets as having the same structure when causal factors are different. Reordering of vectors before analysis can aid in the detection of patterns. We urge caution in the biological interpretation of CPC analysis results.