A penalized likelihood approach to rotation of principal components

Research output: Contribution to journalArticle

Abstract

A new paradigm for enhancing the interpretability of principal components through rotation is presented within the framework of penalized likelihood. The rotated components are computed as the maximizers of a Gaussian-based profile log-likelihood function plus a penalty term defined by a standard rotation criterion. This method enjoys a number of advantages over other methods for principal component rotation, notably (1) the rotation specifically targets ill-defined principal components, which may benefit the most from rotation, and (2) the connection with likelihood allows assessment of the fidelity of the rotated components to the data, thereby guiding the choice of penalty parameter. The method is illustrated with an application to a small functional dataset. Efficient computation of the penalized likelihood solution is possible using recently developed algorithms for optimization under orthogonality constraints.

Original languageEnglish (US)
Pages (from-to)867-888
Number of pages22
JournalJournal of Computational and Graphical Statistics
Volume14
Issue number4
DOIs
StatePublished - Dec 1 2005
Externally publishedYes

Fingerprint

Penalized Likelihood
Principal Components
Penalty
Interpretability
Likelihood Function
Orthogonality
Fidelity
Likelihood
Paradigm
Principal components
Target
Optimization
Term

Keywords

  • Functional data
  • Ill-defined components
  • Orthogonal rotation
  • Profile likelihood
  • Simplified component technique
  • Varimax

ASJC Scopus subject areas

  • Statistics and Probability
  • Discrete Mathematics and Combinatorics
  • Statistics, Probability and Uncertainty

Cite this

A penalized likelihood approach to rotation of principal components. / Park, Trevor H.

In: Journal of Computational and Graphical Statistics, Vol. 14, No. 4, 01.12.2005, p. 867-888.

Research output: Contribution to journalArticle

@article{251d475cc42e40f1be3266f08e698150,
title = "A penalized likelihood approach to rotation of principal components",
abstract = "A new paradigm for enhancing the interpretability of principal components through rotation is presented within the framework of penalized likelihood. The rotated components are computed as the maximizers of a Gaussian-based profile log-likelihood function plus a penalty term defined by a standard rotation criterion. This method enjoys a number of advantages over other methods for principal component rotation, notably (1) the rotation specifically targets ill-defined principal components, which may benefit the most from rotation, and (2) the connection with likelihood allows assessment of the fidelity of the rotated components to the data, thereby guiding the choice of penalty parameter. The method is illustrated with an application to a small functional dataset. Efficient computation of the penalized likelihood solution is possible using recently developed algorithms for optimization under orthogonality constraints.",
keywords = "Functional data, Ill-defined components, Orthogonal rotation, Profile likelihood, Simplified component technique, Varimax",
author = "Park, {Trevor H}",
year = "2005",
month = "12",
day = "1",
doi = "10.1198/106186005X78134",
language = "English (US)",
volume = "14",
pages = "867--888",
journal = "Journal of Computational and Graphical Statistics",
issn = "1061-8600",
publisher = "American Statistical Association",
number = "4",

}

TY - JOUR

T1 - A penalized likelihood approach to rotation of principal components

AU - Park, Trevor H

PY - 2005/12/1

Y1 - 2005/12/1

N2 - A new paradigm for enhancing the interpretability of principal components through rotation is presented within the framework of penalized likelihood. The rotated components are computed as the maximizers of a Gaussian-based profile log-likelihood function plus a penalty term defined by a standard rotation criterion. This method enjoys a number of advantages over other methods for principal component rotation, notably (1) the rotation specifically targets ill-defined principal components, which may benefit the most from rotation, and (2) the connection with likelihood allows assessment of the fidelity of the rotated components to the data, thereby guiding the choice of penalty parameter. The method is illustrated with an application to a small functional dataset. Efficient computation of the penalized likelihood solution is possible using recently developed algorithms for optimization under orthogonality constraints.

AB - A new paradigm for enhancing the interpretability of principal components through rotation is presented within the framework of penalized likelihood. The rotated components are computed as the maximizers of a Gaussian-based profile log-likelihood function plus a penalty term defined by a standard rotation criterion. This method enjoys a number of advantages over other methods for principal component rotation, notably (1) the rotation specifically targets ill-defined principal components, which may benefit the most from rotation, and (2) the connection with likelihood allows assessment of the fidelity of the rotated components to the data, thereby guiding the choice of penalty parameter. The method is illustrated with an application to a small functional dataset. Efficient computation of the penalized likelihood solution is possible using recently developed algorithms for optimization under orthogonality constraints.

KW - Functional data

KW - Ill-defined components

KW - Orthogonal rotation

KW - Profile likelihood

KW - Simplified component technique

KW - Varimax

UR - http://www.scopus.com/inward/record.url?scp=29544438456&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=29544438456&partnerID=8YFLogxK

U2 - 10.1198/106186005X78134

DO - 10.1198/106186005X78134

M3 - Article

AN - SCOPUS:29544438456

VL - 14

SP - 867

EP - 888

JO - Journal of Computational and Graphical Statistics

JF - Journal of Computational and Graphical Statistics

SN - 1061-8600

IS - 4

ER -