Abstract:Physiologists often wish to compare the effects of several different treatments on a continuous variable of interest, which requires an analysis of variance. Analysis of variance, as presented in most statistics texts, generally requires that there be no missing data and often that each sample group be the same size. Unfortunately, this requirement is rarely satisfied, and investigators are confronted with the problem of how to analyze data that do not strictly fit the traditional analysis of variance paradigm. One can avoid these pitfalls by recasting the analysis of variance as a multiple linear regression problem. When there are no missing data, the results of a traditional analysis of variance and the corresponding multiple regression problem are identical; when the sample sizes are unequal or there are missing data, one can use a regression formulation to analyze data that cannot be easily handled in a traditional analysis of variance paradigm and thus overcome a practical computational limitation of traditional analysis of variance. In addition to overcoming practical limitations of traditional analysis of variance, the multiple linear regression approach is more efficient because in one run of a statistics routine, not only is the analysis of variance done but also one obtains estimates of the size of the treatment effects (as opposed to just an indication of whether such effects are present or not), and many of the pairwise multiple comparisons are done (they are equivalent to t tests for significance of the regression parameter estimates). Finally, interaction between the different treatment factors is easier to interpret than it is in traditional analysis of variance.

Contrastive linear regression

Comparison of dimension reduction based logistic regression models for case control genome wide association study： principal components analysis vs. partial least squares

Longitudinal regression of covariance matrix outcomes

Unsupervised learning with contrastive latent variable models

Data-Driven Subgroup Identification for Linear Regression

Identifying Effect Modification of Latent Population Characteristics on Risk Factors with a Sparse Varying Coefficient Regression

Separating common from salient patterns with Contrastive Representation Learning

Testing High Dimensional Covariance Matrices, with Application to Detecting Schizophrenia Risk Genes

Dimensionality Reduction of Genetic Data using Contrastive Learning

Group Sparse Reduced Rank Regression for Neuroimaging Genetic Study.

A Comparison of Approaches to Control for Confounding Factors by Regression Models

Contrastive representations of high-dimensional, structured treatments

Principal regression for high dimensional covariance matrices

Linear Contour Learning: A Method for Supervised Dimension Reduction

Disjoint Contrastive Regression Learning for Multi-Sourced Annotations

Adaptive Contrast for Image Regression in Computer-Aided Disease Assessment

High-dimensional regression in practice: an empirical study of finite-sample prediction, variable selection and ranking

Robust Regression with Compositional Covariates

Rank-N-Contrast: Learning Continuous Representations for Regression

Large-scale electronic-structure theory and nanoscale defects formed in cleavage process of silicon

Multiple linear regression is a useful alternative to traditional analyses of variance