Interaction screening by partial correlation

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

Interaction effects between predictors can play an important role in improving prediction and model interpretation for regression models. However, it is both statistically and computationally challenging to discover informative interactions for high dimensional data. Variable screening based on marginal information is popular for identifying important predictors, but it is mainly used for main-effect-only models. In this paper, we study interaction screening for high dimensional quadratic regression models. First, we show that the direct generalization of main-effect screening to interaction screening can be incorrect or inefficient, as it overlooks the intrinsic relationship between main effects and interactions. Next, we propose a main-effect-adjusted interaction screening procedure to select interactions while taking into account main effects. This new unified framework can be employed with multiple types of correlation measures, such as Pearson correlation coefficients, nonparametric rank-based measures including Spearman's and Kendall's correlation coefficients. Efficient algorithms are developed for each correlation measure to make the screening procedure scalable to high dimensional data. Finally, we illustrate performance of the new screening procedure by simulation studies and an application to a retinopathy study.

Original languageEnglish (US)
Pages (from-to)317-325
Number of pages9
JournalStatistics and its Interface
Volume11
Issue number2
DOIs
StatePublished - Jan 1 2018

Fingerprint

Partial Correlation
Screening
Main Effect
Interaction
High-dimensional Data
Correlation coefficient
Predictors
Regression Model
Pearson Correlation
Interaction Effects
High-dimensional
Efficient Algorithms
Simulation Study
Prediction
Model

Keywords

  • High dimensional data
  • Interaction effects
  • Marginal statistic
  • Quadratic regression
  • Rank correlation
  • Variable screening

ASJC Scopus subject areas

  • Statistics and Probability
  • Applied Mathematics

Cite this

Interaction screening by partial correlation. / Niu, Yue; Hao, Ning -; Zhang, Hao.

In: Statistics and its Interface, Vol. 11, No. 2, 01.01.2018, p. 317-325.

Research output: Contribution to journalArticle

@article{3477219b7be4407e8ecbeaf67c71096a,
title = "Interaction screening by partial correlation",
abstract = "Interaction effects between predictors can play an important role in improving prediction and model interpretation for regression models. However, it is both statistically and computationally challenging to discover informative interactions for high dimensional data. Variable screening based on marginal information is popular for identifying important predictors, but it is mainly used for main-effect-only models. In this paper, we study interaction screening for high dimensional quadratic regression models. First, we show that the direct generalization of main-effect screening to interaction screening can be incorrect or inefficient, as it overlooks the intrinsic relationship between main effects and interactions. Next, we propose a main-effect-adjusted interaction screening procedure to select interactions while taking into account main effects. This new unified framework can be employed with multiple types of correlation measures, such as Pearson correlation coefficients, nonparametric rank-based measures including Spearman's and Kendall's correlation coefficients. Efficient algorithms are developed for each correlation measure to make the screening procedure scalable to high dimensional data. Finally, we illustrate performance of the new screening procedure by simulation studies and an application to a retinopathy study.",
keywords = "High dimensional data, Interaction effects, Marginal statistic, Quadratic regression, Rank correlation, Variable screening",
author = "Yue Niu and Hao, {Ning -} and Hao Zhang",
year = "2018",
month = "1",
day = "1",
doi = "10.4310/SII.2018.v11.n2.a9",
language = "English (US)",
volume = "11",
pages = "317--325",
journal = "Statistics and its Interface",
issn = "1938-7989",
publisher = "International Press of Boston, Inc.",
number = "2",

}

TY - JOUR

T1 - Interaction screening by partial correlation

AU - Niu, Yue

AU - Hao, Ning -

AU - Zhang, Hao

PY - 2018/1/1

Y1 - 2018/1/1

N2 - Interaction effects between predictors can play an important role in improving prediction and model interpretation for regression models. However, it is both statistically and computationally challenging to discover informative interactions for high dimensional data. Variable screening based on marginal information is popular for identifying important predictors, but it is mainly used for main-effect-only models. In this paper, we study interaction screening for high dimensional quadratic regression models. First, we show that the direct generalization of main-effect screening to interaction screening can be incorrect or inefficient, as it overlooks the intrinsic relationship between main effects and interactions. Next, we propose a main-effect-adjusted interaction screening procedure to select interactions while taking into account main effects. This new unified framework can be employed with multiple types of correlation measures, such as Pearson correlation coefficients, nonparametric rank-based measures including Spearman's and Kendall's correlation coefficients. Efficient algorithms are developed for each correlation measure to make the screening procedure scalable to high dimensional data. Finally, we illustrate performance of the new screening procedure by simulation studies and an application to a retinopathy study.

AB - Interaction effects between predictors can play an important role in improving prediction and model interpretation for regression models. However, it is both statistically and computationally challenging to discover informative interactions for high dimensional data. Variable screening based on marginal information is popular for identifying important predictors, but it is mainly used for main-effect-only models. In this paper, we study interaction screening for high dimensional quadratic regression models. First, we show that the direct generalization of main-effect screening to interaction screening can be incorrect or inefficient, as it overlooks the intrinsic relationship between main effects and interactions. Next, we propose a main-effect-adjusted interaction screening procedure to select interactions while taking into account main effects. This new unified framework can be employed with multiple types of correlation measures, such as Pearson correlation coefficients, nonparametric rank-based measures including Spearman's and Kendall's correlation coefficients. Efficient algorithms are developed for each correlation measure to make the screening procedure scalable to high dimensional data. Finally, we illustrate performance of the new screening procedure by simulation studies and an application to a retinopathy study.

KW - High dimensional data

KW - Interaction effects

KW - Marginal statistic

KW - Quadratic regression

KW - Rank correlation

KW - Variable screening

UR - http://www.scopus.com/inward/record.url?scp=85043349763&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85043349763&partnerID=8YFLogxK

U2 - 10.4310/SII.2018.v11.n2.a9

DO - 10.4310/SII.2018.v11.n2.a9

M3 - Article

AN - SCOPUS:85043349763

VL - 11

SP - 317

EP - 325

JO - Statistics and its Interface

JF - Statistics and its Interface

SN - 1938-7989

IS - 2

ER -