Partial least squares prediction in high-dimensional regression

R. D Cook, Liliana Forzani

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

We study the asymptotic behavior of predictions from partial least squares (PLS) regression as the sample size and number of predictors diverge in various alignments. We show that there is a range of regression scenarios where PLS predictions have the usual root-n convergence rate, even when the sample size is substantially smaller than the number of predictors, and an even wider range where the rate is slower but may still produce practically useful results. We show also that PLS predictions achieve their best asymptotic behavior in abundant regressions where many predictors contribute information about the response. Their asymptotic behavior tends to be undesirable in sparse regressions where few predictors contribute information about the response.

Original languageEnglish (US)
Pages (from-to)884-908
Number of pages25
JournalAnnals of Statistics
Volume47
Issue number2
DOIs
StatePublished - Apr 1 2019

Fingerprint

Partial Least Squares
Predictors
High-dimensional
Regression
Prediction
Asymptotic Behavior
Sample Size
Partial Least Squares Regression
Diverge
Range of data
Convergence Rate
Alignment
Roots
Tend
Scenarios
Partial least squares
Asymptotic behavior
Sample size

Keywords

  • Abundant regressions
  • Dimension reduction
  • Sparse regressions

Cite this

Partial least squares prediction in high-dimensional regression. / Cook, R. D; Forzani, Liliana.

In: Annals of Statistics, Vol. 47, No. 2, 01.04.2019, p. 884-908.

Research output: Contribution to journalArticle

Cook, R. D ; Forzani, Liliana. / Partial least squares prediction in high-dimensional regression. In: Annals of Statistics. 2019 ; Vol. 47, No. 2. pp. 884-908.
@article{62c19a83ae7e4932b96b243ce53e314d,
title = "Partial least squares prediction in high-dimensional regression",
abstract = "We study the asymptotic behavior of predictions from partial least squares (PLS) regression as the sample size and number of predictors diverge in various alignments. We show that there is a range of regression scenarios where PLS predictions have the usual root-n convergence rate, even when the sample size is substantially smaller than the number of predictors, and an even wider range where the rate is slower but may still produce practically useful results. We show also that PLS predictions achieve their best asymptotic behavior in abundant regressions where many predictors contribute information about the response. Their asymptotic behavior tends to be undesirable in sparse regressions where few predictors contribute information about the response.",
keywords = "Abundant regressions, Dimension reduction, Sparse regressions",
author = "Cook, {R. D} and Liliana Forzani",
year = "2019",
month = "4",
day = "1",
doi = "10.1214/18-AOS1681",
language = "English (US)",
volume = "47",
pages = "884--908",
journal = "Annals of Statistics",
issn = "0090-5364",
publisher = "Institute of Mathematical Statistics",
number = "2",

}

TY - JOUR

T1 - Partial least squares prediction in high-dimensional regression

AU - Cook, R. D

AU - Forzani, Liliana

PY - 2019/4/1

Y1 - 2019/4/1

N2 - We study the asymptotic behavior of predictions from partial least squares (PLS) regression as the sample size and number of predictors diverge in various alignments. We show that there is a range of regression scenarios where PLS predictions have the usual root-n convergence rate, even when the sample size is substantially smaller than the number of predictors, and an even wider range where the rate is slower but may still produce practically useful results. We show also that PLS predictions achieve their best asymptotic behavior in abundant regressions where many predictors contribute information about the response. Their asymptotic behavior tends to be undesirable in sparse regressions where few predictors contribute information about the response.

AB - We study the asymptotic behavior of predictions from partial least squares (PLS) regression as the sample size and number of predictors diverge in various alignments. We show that there is a range of regression scenarios where PLS predictions have the usual root-n convergence rate, even when the sample size is substantially smaller than the number of predictors, and an even wider range where the rate is slower but may still produce practically useful results. We show also that PLS predictions achieve their best asymptotic behavior in abundant regressions where many predictors contribute information about the response. Their asymptotic behavior tends to be undesirable in sparse regressions where few predictors contribute information about the response.

KW - Abundant regressions

KW - Dimension reduction

KW - Sparse regressions

UR - http://www.scopus.com/inward/record.url?scp=85060156842&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85060156842&partnerID=8YFLogxK

U2 - 10.1214/18-AOS1681

DO - 10.1214/18-AOS1681

M3 - Article

AN - SCOPUS:85060156842

VL - 47

SP - 884

EP - 908

JO - Annals of Statistics

JF - Annals of Statistics

SN - 0090-5364

IS - 2

ER -