2. General Linear Hypothesis

A one-way fixed-effects ANCOVA model with multiple covariates can be expressed as

(1) $\begin{array}{c}{Y}_{\mathrm{ij}}={\mu }_i+\sum _{k=1}^P{X}_{\mathrm{kij}}{\beta }_k+{\epsilon }_{\mathrm{ij}},\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} Y_{ij} =\upmu _{i} +\sum \limits _{k=1}^P {X_{kij} \upbeta _{k} +\upvarepsilon _{ij} ,} \end{aligned}$$\end{document}

where $Y_{\mathrm{ij}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Y_{ij}$$\end{document} is the score of the jth subject in the ith treatment group on the response variable, ${\mu }_i$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu _{i}$$\end{document} is the ith group intercept, $X_{\mathrm{kij}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_{kij}$$\end{document} is the score of the jth subject in the ith treatment group on the kth covariate, ${\beta }_k$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upbeta _{k}$$\end{document} is the slope coefficient of the kth covariate, and ${\epsilon }_{\mathrm{ij}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upvarepsilon _{ij}$$\end{document} is the independent $N(0,{\sigma }^2)$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N(0, \upsigma ^{2})$$\end{document} error with $i=1,\dots ,G(\ge 2),j=1,\dots ,N_i$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i = 1,\ldots , G\, (\ge 2), j = 1, \ldots , N_{i}$$\end{document} , and $k=1,\dots ,P(\ge 1)$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k = 1, \ldots , P\, (\ge 1)$$\end{document} . The least-square estimator for the ith intercept ${\mu }_i$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu _{i}$$\end{document} is given by

(2) $\begin{array}{c}{\hat{\mu }}_i={\overline{Y}}_i-\sum _{k=1}^P{\overline{X}}_{ki·}{\hat{\beta }}_{k,}\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\hat{\upmu }}_{i} ={\bar{Y}}_{i}-\sum \limits _{k=1}^P \bar{X}_{ki\cdot }{\hat{\upbeta }}_{k,} \end{aligned}$$\end{document}

where ${\overline{Y}}_i=\sum _{j=1}^{N_i}{Y}_{\mathrm{ij}}/N_i$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bar{Y}}_{i} =\mathop \sum \limits _{j=1}^{N_i}Y_{ij}/N_{i}$$\end{document} , $\hat{\beta }={({\hat{\beta }}_1,\dots ,{\hat{\beta }}_P)}^T=S_{\mathrm{XX}}^{-1}{S}_{\mathrm{XY}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{{\varvec{\upbeta }}}} = ({\hat{\upbeta }}_{1}, \ldots , {\hat{\upbeta }}_{P})^{\mathrm {T}} = {\mathbf{S}}_{XX}^{-1} {\mathbf{S}}_{XY}$$\end{document} , $S_{\mathrm{XX}}=\sum _{i=1}^G\sum _{j=1}^{\mathrm{Ni}}(X_{\mathrm{ij}}-{\overline{X}}_i){(X_{\mathrm{ij}}-{\overline{X}}_i)}^T$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{S}_{XX} =\mathop \sum \limits _{i=1}^G \mathop \sum \limits _{j=1}^{Ni} (\mathbf{X}_{ij} - \bar{\mathbf{X}}_{i})(\mathbf{X}_{ij} - \bar{\mathbf{X}}_{i})^{\mathrm{T}}$$\end{document} , $S_{\mathrm{XY}}=\sum _{i=1}^G\sum _{j=1}^{\mathrm{Ni}}(X_{\mathrm{ij}}-{\overline{X}}_i)(Y_{\mathrm{ij}}-{\overline{Y}}_i)$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{S}_{XY} =\mathop \sum \limits _{i=1}^G \mathop \sum \limits _{j=1}^{Ni} (\mathbf{X}_{ij} - \bar{\mathbf{X}}_{i})(Y_{ij} - \bar{Y}_{i})$$\end{document} , $X_{\mathrm{ij}}={(X_{1ij},\dots ,X_{\mathrm{Pij}})}^T$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{X}_{ij} = (X_{1ij}, {\ldots }, X_{Pij})^{\mathrm{T}}$$\end{document} , ${\overline{X}}_i=\sum _{j=1}^{\mathrm{Ni}}{X}_{\mathrm{ij}}/N_i={({\overline{X}}_{1i·},\dots ,{\overline{X}}_{Pi·})}^T$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\mathbf{X}}_{i }=\mathop \sum \limits _{j=1}^{Ni} \mathbf{X}_{ij}/N_{i} = (\bar{X}_{1i\cdot }, {\ldots }, \bar{X}_{Pi\cdot })^{\mathrm{T}}$$\end{document} , and ${\overline{X}}_{ki·}=\sum _{j=1}^{\mathrm{Ni}}{X}_{\mathrm{kij}}/N_i$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{X}_{ki\cdot }=\mathop \sum \limits _{j=1}^{Ni} X_{kij}/N_{i}$$\end{document} . Accordingly, the least-squares estimators ${\hat{\mu }}_i$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{\upmu }}_{i}$$\end{document} of ${\mu }_i$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu _{i}$$\end{document} have the following distributions:

(3) $\begin{array}{c}{\hat{\mu }}_i\sim N({\mu }_i,{\sigma }^2\{1/N_i+{\overline{X}}_i^T{S}_{\mathrm{XX}}^{-1}{\overline{X}}_i\})\text{and}\mathrm{Cov}({\hat{\mu }}_i,{\hat{\mu }}_{i^{\prime }})={\sigma }^2{\overline{X}}_i^T{S}_{\mathrm{XX}}^{-1}{\overline{X}}_{i^{\prime }}\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\hat{\upmu }}_{i} \sim N(\upmu _{i}, \upsigma ^{2}\{1/N_{i}+\bar{\mathbf{X}}_i^{\mathrm{T}} \mathbf{S}_{XX}^{-1} \bar{\mathbf{X}}_{i}\})\hbox { and }{ Cov}({\hat{\upmu }}_{i}, {\hat{\upmu }}_{{i}^\prime } )=\upsigma ^{2}\bar{\mathbf{X}}_i^{\mathrm{T}} \mathbf{S}_{XX}^{-1} \bar{\mathbf{X}}_{{i}^\prime } \end{aligned}$$\end{document}

for $i\ne i^{\prime }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i \ne i^\prime $$\end{document} , i and $i^{\prime }=1,\dots ,G$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i^\prime = 1,\ldots , G$$\end{document} . Because the covariances between regression-adjusted estimators $\{{\hat{\mu }}_1,\dots ,{\hat{\mu }}_G\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{{\hat{\upmu }}_{1}, {\ldots }, {\hat{\upmu }}_{G}\}$$\end{document} are generally not zero, they should not be treated as independent variables. For notational simplicity, the prescribed properties are expressed in matrix form:

(4) $\begin{array}{c}\hat{\mu }\sim N_G(\mu ,{\sigma }^{2}V),\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \hat{\varvec{\upmu }} \sim N_{G}({{\varvec{\upmu }}}, \upsigma ^{2}{} \mathbf{V}), \end{aligned}$$\end{document}

where $\hat{\mu }={({\hat{\mu }}_1,\dots ,{\hat{\mu }}_G)}^T$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{\varvec{\upmu }} = ({\hat{\upmu }}_{1}, {\ldots }, {\hat{\upmu }}_{G})^{\mathrm{T}}$$\end{document} , $\mu ={({\mu }_1,\dots ,{\mu }_G)}^T$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\varvec{\upmu }}} = (\upmu _{1}, {\ldots }, \upmu _{G})^{\mathrm{T}}$$\end{document} , $V=D+{\overline{X}}^T{S}_{\mathrm{XX}}^{-1}\overline{X}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{V} = \mathbf{D} + \bar{\mathbf{X}}^{\mathrm{T}}{} \mathbf{S}_{XX}^{-1} \bar{\mathbf{X}}$$\end{document} , $D=\text{Diag}(1/N_1,\dots ,1/N_G)$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{D} = \hbox {Diag}(1/N_{1}, {\ldots }, 1/N_{G})$$\end{document} is the $G\times G$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G \times G$$\end{document} diagonal matrix with diagonal elements $\{1/N_1,\dots ,1/N_G\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{1/N_{1}, {\ldots }, 1/N_{G}\}$$\end{document} , and $\overline{X}=({\overline{X}}_1,\dots ,{\overline{X}}_G)$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\mathbf{X}}= (\bar{\mathbf{X}}_{1}, {\ldots }, \bar{\mathbf{X}}_{G})$$\end{document} .

The adjusted group means are the expected group responses evaluated at the grand covariate means:

(5) $\begin{array}{c}{\mu }_i^{\ast }={\mu }_i+\sum _{k=1}^P{\overline{X}}_{k··}{\beta }_k\text{for}i=1,\dots ,G,\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \upmu _i^*=\upmu _{i}+\sum _{k=1}^P {\bar{X}}_{k\cdot \cdot }\upbeta _{k}\quad \hbox { for }i = 1,\ldots , G, \end{aligned}$$\end{document}

where ${\overline{X}}_{k··}=\sum _{i=1}^G\sum _{j=1}^{\mathrm{Ni}}{X}_{\mathrm{kij}}/N_T$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{X}_{k\cdot \cdot }=\mathop \sum \limits _{i=1}^{G}\mathop \sum \limits _{j=1}^{Ni} X_{kij}/N_{T}$$\end{document} , $k=1,\dots ,P,$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k = 1,\ldots , P, $$\end{document} and $N_T=\sum _{i=1}^G{N}_i$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{T}=\mathop \sum \limits _{i=1}^G N_{i}$$\end{document} . A natural and unbiased estimator of the adjusted group mean ${\mu }_i^{\ast }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu _i^*$$\end{document} is

(6) $\begin{array}{c}{\hat{\mu }}_i^{\ast }={\hat{\mu }}_i+\sum _{k=1}^P{\overline{X}}_{k··}{\hat{\beta }}_k={\overline{Y}}_i-\sum _{k=1}^P{\hat{\beta }}_k({\overline{X}}_{ki·}-{\overline{X}}_{k··}).\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\hat{\upmu }} _i^*= {\hat{\upmu }}_{i}+\sum _{k=1}^P {\bar{X}}_{k\cdot \cdot } {\hat{\upbeta }}_{k}={\bar{Y}}_{i} - \sum _{k=1}^P {\hat{\upbeta }}_{k}({\bar{X}}_{ki\cdot } - {\bar{X}}_{k\cdot \cdot }). \end{aligned}$$\end{document}

Then, the least-squares estimators ${\hat{\mu }}_i^{\ast }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{\upmu }}_i^*$$\end{document} of the adjusted group means ${\mu }_i^{\ast }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upmu _i^*$$\end{document} have the following distributions:

$\begin{array}{c}{\hat{\mu }}_i^{\ast }\sim N({\mu }_i^{\ast },{\sigma }^2\{1/N_i+{({\overline{X}}_i-M)}^T{S}_{\mathrm{XX}}^{-1}({\overline{X}}_i-M)\})\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\hat{\upmu }} _i^*\sim N(\upmu _i^*, \upsigma ^{2}\{1/N_{i} + ({\bar{\mathbf{X}}}_{i} - \mathbf{M})^{\mathrm{T}}{} \mathbf{S}_{XX}^{-1} ({\bar{\mathbf{X}}}_{i} - \mathbf{M})\}) \end{aligned}$$\end{document}

and

(7) $\begin{array}{c}\mathrm{Cov}({\hat{\mu }}_i,{\hat{\mu }}_{i^{\prime }})={\sigma }^2{({\overline{X}}_i-M)}^T{S}_{\mathrm{XX}}^{-1}({\overline{X}}_{i^{\prime }}-M),\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {Cov}({\hat{\upmu }}_{i}, {\hat{\upmu }}_{{i}^\prime } )=\upsigma ^{2}({\bar{\mathbf{X}}}_{i} - \mathbf{M})^{\mathrm{T}}{} \mathbf{S}_{XX}^{-1} ({\bar{\mathbf{X}}}_{{i}^\prime } - \mathbf{M}), \end{aligned}$$\end{document}

where $M=\sum _{i=1}^G\sum _{j=1}^{\mathrm{Ni}}{X}_{\mathrm{ij}}/N_T={({\overline{X}}_{1··},\dots ,{\overline{X}}_{P··})}^T$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{M}=\mathop \sum \limits _{i=1}^G \mathop \sum \limits _{j=1}^{Ni} \mathbf{X}_{ij}/N_{T} = (\bar{X}_{1\cdot \cdot }, {\ldots }, \bar{X}_{P\cdot \cdot })^{\mathrm{T}}$$\end{document} for $i\ne i^{\prime }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i \ne i^\prime $$\end{document} , i and $i^{\prime }=1,\dots ,G$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i^\prime = 1,\ldots , G$$\end{document} . The vector of adjusted group mean estimators ${\hat{\mu }}^{\ast }={({\hat{\mu }}_1^{\ast },\dots ,{\hat{\mu }}_G^{\ast })}^T$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{\varvec{\upmu }}^* = ({\hat{\upmu }}_1^{*}, {\ldots }, {\hat{\upmu }}_G^*)^{\mathrm{T}}$$\end{document} has the distribution

(8) $\begin{array}{c}{\hat{\mu }}^{\ast }\sim N_G({\mu }^{\ast },{\sigma }^2{V}^{\ast }),\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\hat{\varvec{\upmu }}}^*\sim N_{G}({{\varvec{\upmu }}}^*, \upsigma ^{2}{} \mathbf{V}^*), \end{aligned}$$\end{document}

where ${\mu }^{\ast }={({\mu }_1^{\ast },\dots ,{\mu }_G^{\ast })}^T$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varvec{\upmu }}^* = (\upmu _1^*, {\ldots }, \upmu _G^*)^{\mathrm{T}}$$\end{document} , $V^{\ast }=D+{(\overline{X}-{M1}_G^T)}^T{S}_{\mathrm{XX}}^{-1}(\overline{X}-{M1}_G^T)$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{V}^* = \mathbf{D} + ({\bar{\mathbf{X}}} - \mathbf{M1}_G^{\mathrm{T}})^{\mathrm{T}}{} \mathbf{S}_{XX}^{-1} ({\bar{\mathbf{X}}} - \mathbf{M1}_G^{\mathrm{T}} )$$\end{document} , and $1_G$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{1}_{G}$$\end{document} is a $G\times 1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G \times 1$$\end{document} column vector of all 1’s.

To test the general linear hypothesis about treatment effects or adjusted mean effects in terms of

(9) $\begin{array}{c}{\text{H}}_0:C{\mu }^{\ast }=0_c\text{versus}{\text{H}}_1:C{\mu }^{\ast }\ne 0_c,\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \hbox {H}_{0}{:} \,\mathbf{C}\upmu ^* = \mathbf{0}_{c}\hbox { versus H}_{1}{:}\, \mathbf{C}\upmu ^* \ne \mathbf{0}_{c}, \end{aligned}$$\end{document}

where C is a $c\times G$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$c \times G$$\end{document} contrast matrix of full row rank and $0_c$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{0}_{c}$$\end{document} is a $c\times 1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$c \times 1$$\end{document} null column vector, the Wald test statistic is of the form

(10) $\begin{array}{c}{W}^{\ast }={\left(C{\hat{\mu }}^{\ast }\right)}^T{\left({\mathrm{CV}}^{\ast }{C}^T\right)}^{-1}\left(C{\hat{\mu }}^{\ast }\right)/\left\{(G-1){\hat{\sigma }}^2\right\}\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} W^* = (\mathbf{C}\hat{\varvec{\upmu }}^*)^{\mathrm{T}}(\mathbf{CV}^*\mathbf{C}^{\mathrm{T}})^{-1}(\mathbf{C}\hat{\varvec{\upmu }}^*)/\{(G - 1){\hat{\upsigma }}^{2}\} \end{aligned}$$\end{document}

where ${\hat{\sigma }}^2=\mathrm{SSE}/\nu$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{\upsigma }}^{2} = {SSE}/\upnu $$\end{document} , $\mathrm{SSE}=\sum _{i=1}^G\sum _{j=1}^{\mathrm{Ni}}{(Y_{\mathrm{ij}}-{\overline{Y}}_i)}^2-S_{\mathrm{XY}}^T{S}_{\mathrm{XX}}^{-1}{S}_{\mathrm{XY}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${SSE} = \mathop \sum \limits _{i=1}^G \mathop \sum \limits _{j=1}^{Ni} (Y_{ij} - \bar{Y}_{i})^{2} - \mathbf{S}_{XY}^{\mathrm{T}} \mathbf{S}_{XX}^{-1} \mathbf{S}_{XY}$$\end{document} , and $\nu =N_T-G-P$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upnu =N_{T} - G - P$$\end{document} . Note that the contrast matrix is confined to satisfy ${C1}_G=0_c$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{C1}_{G} = \mathbf{0}_{c}$$\end{document} . Hence, the general linear hypothesis of ${\text{H}}_0:C{\mu }^{\ast }=0_c$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {H}_{0}{:}\, \mathbf{C}\upmu ^* = \mathbf{0}_{c}$$\end{document} versus ${\text{H}}_1:C{\mu }^{\ast }\ne 0_c$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbox {H}_{1}{:} \,\mathbf{C}\upmu ^* \ne \mathbf{0}_{c}$$\end{document} is equivalent to

(11) $\begin{array}{c}{\text{H}}_0:C\mu =0_c\text{versus}{\text{H}}_1:C\mu \ne 0_c.\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \hbox {H}_{0}{:}\, \mathbf{C}{\varvec{\upmu }} = \mathbf{0}_{c}\hbox { versus H}_{1}{:}\, \mathbf{C}{\varvec{\upmu }} \ne \mathbf{0}_{c}. \end{aligned}$$\end{document}

Also, the Wald test statistic can be rewritten as

(12) $\begin{array}{c}{W}^{\ast }={\left(C\hat{\mu }\right)}^T{\left({\mathrm{CVC}}^T\right)}^{-1}\left(C\hat{\mu }\right)/\left\{(G-1){\hat{\sigma }}^2\right\}.\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} W^* = (\mathbf{C}\hat{\varvec{\upmu }})^{\mathrm{T}}(\mathbf{CVC}^{\mathrm{T}})^{-1}(\mathbf{C}\hat{\varvec{\upmu }})/\{(G - 1)\hat{\upsigma }^{2}\}. \end{aligned}$$\end{document}

The Wald-type test has great practical and pedagogical appeal than the test procedure under the full-reduced-model formulation. Because of its simplicity and generality, the associated properties are derived and presented in the subsequent illustration. Under the null hypothesis with $C\mu =0_c$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{C}{\varvec{\upmu }} = \mathbf{0}_{c}$$\end{document} , the test statistic $W^{\ast }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^*$$\end{document} has an F distribution

(13) $\begin{array}{c}{W}^{\ast }\sim F(c,\nu ),\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} W^* \sim F(c, \upnu ), \end{aligned}$$\end{document}

where $F(c,\nu )$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F(c, \upnu )$$\end{document} is an F distribution with c and $\nu$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upnu $$\end{document} degrees of freedom, $\nu =N_T-G-P$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upnu =N_{T} - G - P$$\end{document} , and $N_T=\sum _{i=1}^G{N}_i$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{T}=\mathop \sum \limits _{i=1}^G N_{i}$$\end{document} . Hence, $H_0$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_{0}$$\end{document} is rejected at the significance level $\alpha$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upalpha $$\end{document} if $W^{\ast }>F_{c,\nu ,\alpha }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^* > F_{c,\,\upnu ,\,\upalpha }$$\end{document} , where $F_{c,\nu ,\alpha }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{c,\,\upnu ,\,\upalpha }$$\end{document} is the upper $(100·\alpha )$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(100\cdot \upalpha )$$\end{document} th percentile of the F distribution $F(c,\nu )$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F(c, \upnu )$$\end{document} . For fixed covariate values of $\{X_{\mathrm{ij}},j=1,\dots ,N_i$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\mathbf{X}_{ij}, j = 1, {\ldots }, N_{i}$$\end{document} and $i=1,\dots ,G\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i = 1,\ldots , G\}$$\end{document} , the test statistic $W^{\ast }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^*$$\end{document} has the general distribution

(14) $\begin{array}{c}{W}^{\ast }\sim F(c,\nu ,\Lambda ),\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} W^* \sim F(c, \upnu , \Lambda ), \end{aligned}$$\end{document}

where $F(c,\nu ,\Lambda )$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F(c, \upnu , \Lambda )$$\end{document} is a non-central F distribution with c and $\nu$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upnu $$\end{document} degrees of freedom and non-centrality parameter

(15) $\begin{array}{c}\Lambda ={\left(C\mu \right)}^T{\left({\mathrm{CVC}}^T\right)}^{-1}\left(C\mu \right)/{\sigma }^2.\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Lambda = (\mathbf{C}{\varvec{\upmu }})^{\mathrm{T}}(\mathbf{CVC}^{\mathrm{T}})^{-1}(\mathbf{C}{\varvec{\upmu }})/\upsigma ^{2}. \end{aligned}$$\end{document}

The associated power function of the general linear hypothesis is readily obtained as

(16) $\begin{array}{c}\Psi \left(\Lambda \right)=P\{F(c,\nu ,\Lambda )>F_{c,\nu ,\alpha }\}.\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Psi (\Lambda )=P\{F(c, \upnu , \Lambda ) > F_{c,\,\upnu ,\,\upalpha }\}. \end{aligned}$$\end{document}

3. Random Covariate Models

The prescribed statistical inferences about the general linear hypothesis are based on the conditional distribution of the covariate outcomes. As noted in Gatsonis and Sampson (Reference Gatsonis and Sampson1989), Mendoza and Stafford (Reference Mendoza and Stafford2001), and Sampson (Reference Sampson1974), the actual values of covariates cannot be known in advance just as the primary responses. It is vital to treat the covariates as random variables and to derive the distribution of the test statistic over possible values of the covariate variables. Moreover, Elashoff (Reference Elashoff1969) and Harwell (Reference Harwell2003) emphasized that the statistical assumptions underlying the ANCOVA include the random assignment of subjects to treatments and the covariate variables are independent of the treatment effects. Moreover, the normal covariate setting is commonly employed to provide a fundamental framework for analytical derivation and theoretical discussion in ANCOVA studies as in Elashoff (Reference Elashoff1969) and Harwell (Reference Harwell2003). Thus, it is constructive to assume the covariates have independent and identical normal distribution

(17) $\begin{array}{c}{X}_{\mathrm{ij}}\sim N_P(\theta ,\Sigma ),\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathbf{X}_{ij} \sim N_{P}({{\varvec{\uptheta }}}, {{\varvec{\Sigma }}}), \end{aligned}$$\end{document}

where $\theta$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\varvec{\uptheta }}}$$\end{document} is a $P\times 1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P \times 1$$\end{document} vector and $\Sigma$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\varvec{\Sigma }}}$$\end{document} is a $P\times P$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P \times P$$\end{document} positive-definite variance–covariance matrix for $i=1,\dots ,G$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i = 1,\ldots , G$$\end{document} , and $j=1,\dots ,N_i$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$j = 1, {\ldots }, N_{i}$$\end{document} .

Under the multinormal distribution of $\{X_{\mathrm{ij}}\sim N_P(\theta ,\Sigma ),j=1,\dots ,N_i$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\mathbf{X}_{ij} \sim N_{P}({{\varvec{\uptheta }}}, {{\varvec{\Sigma }}}), j = 1, {\ldots }, N_{i}$$\end{document} and $i=1,\dots ,G\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i = 1,\ldots , G\}$$\end{document} , it is straightforward to show (Gupta and Nagar Reference Gupta and Nagar1999, Theorem 2.3.10 and Theorem 3.3.6) that $Z=\overline{X}{C}^T{\left({\mathrm{CDC}}^T\right)}^{-1/2}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{Z} = \bar{\mathbf{X}}{} \mathbf{C}^{\mathrm{T}}(\mathbf{CDC}^{\mathrm{T}})^{-1/2}$$\end{document} has a matrix normal distribution and $S_{\mathrm{XX}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{S}_{XX}$$\end{document} has a Wishart distribution

(18) $\begin{array}{c}Z\sim N_{P,c}(0,\Sigma \otimes I_c)\text{and}{S}_{\mathrm{XX}}\sim W_P(N_T-G,\Sigma ),\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathbf{Z} \sim N_{P, c}(\mathbf{0}, {\varvec{\Sigma }}\otimes \mathbf{I}_{c})\hbox { and }{} \mathbf{S}_{XX} \sim W_{P}(N_{T} - G, {\varvec{\Sigma }}), \end{aligned}$$\end{document}

where $I_c$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{I}_{c}$$\end{document} is an identity matrix of dimension c. Accordingly, both $T={\{S_{\mathrm{XX}}+{\mathrm{ZZ}}^T\}}^{-1/2}Z$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{T} = \{\mathbf{S}_{XX} + \mathbf{ZZ}^{\mathrm{T}}\}^{-1/2}{} \mathbf{Z}$$\end{document} and $T^{\ast }=T\xi$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{T}^* = \mathbf{T}{\varvec{\upxi }}$$\end{document} have an inverted matrix variate t-distribution (Gupta and Nagar Reference Gupta and Nagar1999, Section 4.4):

(19) $\begin{array}{c}T\sim {\mathrm{IT}}_{P,C}(\nu +1,0,I_P,I_c)\text{and}{T}^{\ast }\sim {\mathrm{IT}}_{P,1}(\nu +1,0_P,I_P,\Gamma ),\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathbf{T} \sim {IT}_{P,C}(\upnu + 1, \mathbf{0}, \mathbf{I}_{P}, \mathbf{I}_{c})\hbox { and }{} \mathbf{T}^* \sim {IT}_{P,1}(\upnu + 1, \mathbf{0}_{P}, \mathbf{I}_{P}, \Gamma ), \end{aligned}$$\end{document}

where $\xi ={\left({\mathrm{CDC}}^T\right)}^{-1/2}\left(C\mu \right)/\sigma$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varvec{\upxi }} = (\mathbf{CDC}^{\mathrm{T}})^{-1/2}(\mathbf{C}{\varvec{\upmu }})/\upsigma $$\end{document} and $\Gamma ={\xi }^T\xi ={\left(C\mu \right)}^T{\left({\mathrm{CDC}}^T\right)}^{-1}\left(C\mu \right)/{\sigma }^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Gamma = {\varvec{\upxi }}^{\mathrm{T}}{\varvec{\upxi }} = (\mathbf{C}{\varvec{\upmu }})^{\mathrm{T}}(\mathbf{CDC}^{\mathrm{T}})^{-1}(\mathbf{C}{\varvec{\upmu }})/\upsigma ^{2}$$\end{document} . Moreover, $A^{\ast }=T^{{\ast }^T}{T}^{\ast }/\Gamma$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A^* = \mathbf{T}^{*^{\mathrm{T}}}{} \mathbf{T}^*/\Gamma $$\end{document} has a matrix variate beta type I distribution (Gupta and Nagar Reference Gupta and Nagar1999, Theorem 5.2.4) or a Beta distribution

(20) $\begin{array}{c}{A}^{\ast }\sim B_1^I(P/2,(\nu +1)/2)\equiv \text{Beta}\{P/2,(\nu +1)/2\}.\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} A^* \sim B_1^I (P/2, (\upnu + 1)/2) \equiv \hbox {Beta}\{P/2, (\upnu + 1)/2\}. \end{aligned}$$\end{document}

Following these results, standard matrix algebra shows that non-centrality parameter $\Lambda$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda $$\end{document} defined in Equation 15 has the alternative form

(21) $\begin{array}{c}\Lambda ={\xi }^T(I_c-T^{T}T)\xi =\Gamma B^{\ast },\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Lambda = {\varvec{\upxi }}^{\mathrm{T}}(\mathbf{I}_{c} - \mathbf{T}^{\mathrm{T}}{} \mathbf{T}){\varvec{\upxi }} = \Gamma B^*, \end{aligned}$$\end{document}

where $B^{\ast }=(1-A^{\ast })\sim \text{Beta}\{(\nu +1)/2,P/2\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B^* = (1 - A^*) \sim \hbox {Beta}\{(\upnu + 1)/2, P/2\}$$\end{document} . In connection with the effect size measures in ANOVA, the first component $\Gamma$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varvec{\Gamma }}$$\end{document} in $\Lambda$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda $$\end{document} is rewritten as

(22) $\begin{array}{c}\Gamma =N_T{\gamma }^2\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Gamma = N_{T}\upgamma ^{2} \end{aligned}$$\end{document}

where ${\gamma }^2={\sigma }_{\gamma }^2/{\sigma }^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upgamma ^{2}=\upsigma _{\upgamma }^{2} /\upsigma ^{2}$$\end{document} , ${\sigma }_{\gamma }^2={\left(C\mu \right)}^T{\left({\mathrm{CQC}}^T\right)}^{-1}\left(C\mu \right)$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma _{\upgamma }^{2} = (\mathbf{C}{\varvec{\upmu }})^{\mathrm{T}}(\mathbf{CQC}^{\mathrm{T}})^{-1}(\mathbf{C}{\varvec{\upmu }})$$\end{document} , $Q=\text{Diag}(1/q_1,\dots ,1/q_G)$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{Q} = \hbox {Diag}(1/q_{1}, {\ldots }, 1/q_{G})$$\end{document} , $q_i=N_i/N_T$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q_{i}=N_{i}/N_{T}$$\end{document} for $i=1,\dots ,G$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i = 1,\ldots , G$$\end{document} . Consequently, the non-centrality term $\Lambda$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda $$\end{document} has a useful formulation

(23) $\begin{array}{c}\Lambda =N_T{\gamma }^2{B}^{\ast }.\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Lambda = N_{T}\upgamma ^{2}B^*. \end{aligned}$$\end{document}

It should be pointed out that Gupta and Nagar (Reference Gupta and Nagar1999) only provides the generic definition and analytic properties of an inverted matrix variate t-distribution. Their results are applied and extended here to the context of ANCOVA. Accordingly, under the random covariate modeling framework, the $W^{\ast }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^*$$\end{document} statistic has the two-stage distribution

(24) $\begin{array}{c}{W}^{\ast }|B^{\ast }\sim F(c,\nu ,\Lambda )\text{and}{B}^{\ast }\sim \mathrm{Beta}\{(\nu +1)/2,P/2\}.\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} W^*{\vert }B^* \sim F(c, \upnu , \Lambda )\hbox { and }B^* \sim {Beta}\{(\upnu + 1)/2, P/2\}. \end{aligned}$$\end{document}

The exact power function can be formulated as

(25) $\begin{array}{c}{\Psi }_E\left(\Lambda \right)=E_B\left[P\{F(c,\nu ,\Lambda )>F_{c,\nu ,\Lambda }\}\right],\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Psi _{E}(\Lambda ) = E_{B}[P\{F(c, \upnu , \Lambda ) > F_{c,\,\upnu ,\, \Lambda } \}], \end{aligned}$$\end{document}

where the expectation $E_B$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_{B}$$\end{document} is taken with respect to the distribution of $B^{\ast }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B^*$$\end{document} .

Notably, the omnibus test of the equality of treatment effects is a special case of the general linear hypothesis by specifying the contrast matrix as $C_D\mu =0_{(G-1)}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{C}_{D}{\varvec{\upmu }} = \mathbf{0}_{(G - 1)}$$\end{document} where

(26) $\begin{array}{c}{C}_D=(1_{(G-1)},-I_{(G-1)})\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathbf{C}_{D} = (\mathbf{1}_{(G - 1)}, -\mathbf{I}_{(G - 1)}) \end{aligned}$$\end{document}

is a $(G-1)\times G$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(G - 1) \times G$$\end{document} contrast matrix of full row rank. The component ${\gamma }^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upgamma ^{2}$$\end{document} in the non-centrality term $\Lambda$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda $$\end{document} is simplified as

(27) $\begin{array}{c}{\delta }^2={\sigma }_{\delta }^2/{\sigma }^2,\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \updelta ^{2}=\upsigma _{\updelta }^{2} /\upsigma ^{2}, \end{aligned}$$\end{document}

where ${\sigma }_{\delta }^2={\sum }_{i=1}^G{q}_i{({\mu }_i-\tilde{\mu })}^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma _{\updelta }^{2} =\sum _{i=1}^G q_{i}(\upmu _{i} - \tilde{\upmu })^{2}$$\end{document} and $\tilde{\mu }={\sum }_{i=1}^G{q}_i{\mu }_i$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{\upmu }=\sum _{i=1}^G q_{i}\upmu _{i}$$\end{document} . The corresponding non-central component $\Lambda$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda $$\end{document} is expressed as

(28) $\begin{array}{c}{\Lambda }_D=N_T{\delta }^2{B}^{\ast }.\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Lambda _{D}=N_{T}\updelta ^{2}B^*. \end{aligned}$$\end{document}

The power function of the omnibus F test of treatment differences is simplified as

(29) $\begin{array}{c}{\Psi }_E\left({\Lambda }_D\right)=E_B\left[P\{F(G-1,\nu ,{\Lambda }_D)>F_{(G-1),\nu ,\alpha }\}\right].\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Psi _{E}(\Lambda _{D})=E_{B}[P\{F(G - 1, \upnu , \Lambda _{D}) > F_{(G-1),\,\upnu ,\,\upalpha }\}]. \end{aligned}$$\end{document}

Note that ${\sigma }_{\delta }^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma _{\updelta }^{2} $$\end{document} reduces to the form ${\sigma }_{\delta }^2=\sum _{i=1}^G{({\mu }_i-\overline{\mu })}^2/G$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma _{\updelta }^{2} =\mathop \sum \limits _{i=1}^G (\upmu _{i} - \bar{\upmu })^{2}/G$$\end{document} with $\overline{\mu }=\sum _{i=1}^G{\mu }_i/G$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{\upmu }=\mathop \sum \limits _{i=1}^G \upmu _{i}/G$$\end{document} when $q_i=1/G$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q_{i} = 1/G$$\end{document} for all $i=1,\dots ,G$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i = 1, {\ldots }, G$$\end{document} . Hence, ${\delta }^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\updelta ^{2}$$\end{document} has the same form as the signal to noise ratio $f^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f^{2}$$\end{document} in ANOVA (Fleishman Reference Fleishman1980) for balanced designs. Although the prescribed application of general linear hypothesis is discussed only from the perspective of a one-way ANCOVA design, the number of groups G may also represent the total number of combined factor levels of a multi-factor ANCOVA design. Hence, using a contrast matrix associated with a specific designated hypothesis, the same concept and process of assessing treatment effects can be readily extended to two-way and higher-order ANCOVA designs.

4. Sample Size Determination

It is essential to note that the power function ${\Psi }_E$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{E}$$\end{document} depends on the group intercepts $\{{\mu }_1,\dots ,{\mu }_G\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{ \upmu _{1},\ldots , \upmu _{G}\}$$\end{document} and variance component ${\sigma }^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^{2}$$\end{document} through the non-centrality $\Lambda$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda $$\end{document} or the effect size ${\gamma }^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upgamma ^{2}$$\end{document} , but not the covariate coefficients $\{{\beta }_1,\dots ,{\beta }_P\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{ \upbeta _{1}, \ldots , \upbeta _{P}\}$$\end{document} . Also, under the prescribed stochastic assumptions for the covariate variables, the multivariate normal distribution leads to the unique conditional property on a beta distribution in the general distribution of the test statistic $W^{\ast }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^*$$\end{document} . Due to the fundamental property of the contrast matrix, the resulting distribution and power function do not depend on the mean vector $\theta$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varvec{\uptheta }} $$\end{document} and variance–covariance matrix $\Sigma$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varvec{\Sigma }} $$\end{document} of the multinormal covariate distribution. To determine sample sizes in planning research designs, the power functions ${\Psi }_E$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{E}$$\end{document} can be applied to calculate the sample sizes $\{N_{E1},\dots ,N_{\mathrm{EG}}\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{ N_{{E1}},\ldots , N_{EG}\}$$\end{document} needed to attain the specified power $1-\beta$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1 - \upbeta $$\end{document} for the chosen significance level $\alpha$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upalpha $$\end{document} , contrast matrix C, intercept parameters $\{{\mu }_1,\dots ,{\mu }_G\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{ \upmu _{1},\ldots , \upmu _{G}\}$$\end{document} , variance component ${\sigma }^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^{2}$$\end{document} , and the number of covariates P.

For an ANCOVA design with a priori designated sample size ratios $\{r_1,\dots ,r_G\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{r_{1},\ldots , r_{G}\}$$\end{document} with $r_i=N_i/N_1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r_{i} = N_{i}/N_{1}$$\end{document} for $i=1,\dots ,G$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i = 1, \ldots , G$$\end{document} . The required computation is simplified to deciding the minimum sample sizes $N_{E1}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{{E}{1}}$$\end{document} (with $N_{\mathrm{Ei}}=N_{E1}·r_i,i=2,\dots ,G)$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{Ei} = N_{{E}{1}}\cdot r_{i}, i = 2,\ldots , G)$$\end{document} required to achieve the selected power level with the power functions ${\Psi }_E$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{E}$$\end{document} . Using the embedded functions in popular software systems, optimal sample sizes can be readily computed through an iterative process. The SAS/IML (SAS Institute 2017) and R (R Development Core Team 2017) programs employed to perform the suggested power and sample size calculations are available as supplementary material. The proposed power and sample size procedures for the general linear hypothesis tests of ANCOVA subsume the results in Shieh (Reference Shieh2017) for a single contrast test as a special case. Notably, the derivations and manipulations of an inverted matrix variate t are more involved than that of a Hotelling’s $T^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T^{2}$$\end{document} distribution as demonstrated in Shieh (Reference Shieh2017).

Alternatively, a simple procedure for the comparison of treatment effects has been described in Cohen (Reference Cohen1988, pp. 379–380). Unlike the proposed two-stage distribution, it is suggested that $W^{\ast }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^*$$\end{document} has a simplified F distribution

(30) $\begin{array}{c}{W}^{\ast }\sim F(G-1,\nu ,{\Lambda }_A),\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} W^*\sim F(G - 1, \upnu , \Lambda _{A}), \end{aligned}$$\end{document}

where ${\Lambda }_A=N_T{\delta }^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda _{A} = N_{T}\updelta ^{2}$$\end{document} . The corresponding power function is of the form

(31) $\begin{array}{c}{\Psi }_A\left({\Lambda }_A\right)=P\{F(G-1,\nu ,{\Lambda }_A)>F_{(G-1),\nu ,\alpha }\}.\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Psi _{A}(\Lambda _{A}) = P\{ F(G - 1, \upnu , \Lambda _{A}) > F_{(G-1),\,\upnu ,\,\upalpha }\} . \end{aligned}$$\end{document}

It is easily seen from the model assumption given in Equation 1 that ${\sigma }_Y^2=\text{Var}\left(Y_{\mathrm{ij}}\right)={\beta }^T\Sigma \beta +{\sigma }^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma _Y^2 = \hbox {Var}(Y_{ij}) ={\varvec{\upbeta }}^{\mathrm {T}} {\varvec{\Sigma }} {\varvec{\upbeta }} + \upsigma ^{2}$$\end{document} and $\rho =\mathrm{Corr}(Y_{\mathrm{ij}},\sum _{k=1}^P{X}_{\mathrm{kij}}{\beta }_k)={\beta }^T\Sigma \beta /{\{{\sigma }_Y^2·{\beta }^T\Sigma \beta \}}^{1/2}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = {Corr}(Y_{ij}, \mathop \sum \limits _{k=1}^{P}X_{kij}\upbeta _{k}) ={\varvec{\upbeta }} ^{\mathrm {T}} {\varvec{\Sigma }} {\varvec{\upbeta }} /\{ \upsigma _Y^2 \cdot {\varvec{\upbeta }}^{\mathrm {T}} {\varvec{\Sigma }} {\varvec{\upbeta }} \}^{{1/2}}$$\end{document} where $\beta ={({\beta }_1,\dots ,{\beta }_P)}^T$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varvec{\upbeta }} = (\upbeta _{1}, \ldots , \upbeta _{P})^{\mathrm {T}}$$\end{document} . Hence, the advantage of ANCOVA over ANOVA in the reduction of error variance from ${\sigma }_Y^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma _Y^2 $$\end{document} to ${\sigma }^2=(1-{\rho }^2){\sigma }_Y^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^{2} = (1 - \uprho ^{2})\upsigma _Y^2$$\end{document} by a factor $(1-{\rho }^2)$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(1 - \uprho ^{2})$$\end{document} . For ease of illustration, the power function of the omnibus F test of treatment differences in ANOVA is also presented here:

(32) $\begin{array}{c}{\Psi }_O\left({\Lambda }_O\right)=P\{F(G-1,N_T-G,{\Lambda }_O)>F_{(G-1),(N_T-G),\alpha }\},\end{array}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Psi _{O}(\Lambda _{O}) = P\{ F(G - 1, N_{T} - G, \Lambda _{O}) > F_{(G-1),\,(N_T-G),\,\upalpha }\} , \end{aligned}$$\end{document}

where ${\Lambda }_O=(1-{\rho }^2){\Lambda }_A$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda _{O} = (1 - \uprho ^{2})\Lambda _{A}$$\end{document} . With the reduction of error variance from ${\sigma }_Y^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma _Y^2 $$\end{document} to ${\sigma }^2=(1-{\rho }^2){\sigma }_Y^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^{2} = (1 - \uprho ^{2})\upsigma _Y^2$$\end{document} , it is evident that ${\Lambda }_O\le {\Lambda }_A$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda _{O} \le \Lambda _{A}$$\end{document} . Hence, the computed power ${\Psi }_O$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{O}$$\end{document} is generally less than ${\Psi }_A$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{A}$$\end{document} when all other factors are fixed despite the marginal difference between the two error degrees of freedom $N_T-G$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{T} - G$$\end{document} and $\nu =N_T-G-P$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upnu = N_{T} - G - P$$\end{document} .

The prevailing procedure of Cohen (Reference Cohen1988) provides a direct application of the ANOVA formula in combination with a reduced degrees of freedom and a correlation-adjusted variance. It is computationally simple because the simple formulation ${\Psi }_A$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{A}$$\end{document} depends only on a non-central F distribution. On the other hand, one critical disadvantage of this method is that the F distribution and the associated sample size formula do not fully take into account the distributional features of covariates. A direct comparison of the two non-centrality components in Equations 28 and 30 reveals that ${\Lambda }_D<{\Lambda }_A$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Lambda _{D} <\Lambda _{A}$$\end{document} because $0<B^{\ast }<1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0< B^* < 1$$\end{document} . This indicates that the power function ${\Psi }_A$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{A}$$\end{document} tends to over-estimate the true power ${\Psi }_E$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{E}$$\end{document} and it also leads to an under-estimated sample size in attaining a desired power level. Notably, the suggested exact procedure is of pedagogical importance and involves a beta mixture of non-central F distributions. These theoretical examinations assure that the proposed technique has analytical superiority over the current method of Cohen (Reference Cohen1988). Their practical accuracy will be demonstrated in the succeeding empirical assessments.

5. Numerical Assessments

To further demonstrate the contrasting features and practical consequences of the proposed approach and existing methods, detailed empirical appraisals are conducted to examine their performance in power and sample size calculations. For ease of comparison, the numerical illustration considered in Maxwell and Delaney (Reference Maxwell and Delaney2004, pp. 441-443) for sample size planning and power analysis is utilized as the fundamental framework.

In particular, Maxwell and Delaney (Reference Maxwell and Delaney2004) described an ANOVA design with $G=3$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G = 3$$\end{document} , group intercepts $\{{\mu }_1,{\mu }_2,{\mu }_3\}=\{400,450,500\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{ \upmu _{1}, \upmu _{2}, \upmu _{3}\}=\{ 400, 450, 500\}$$\end{document} , and error variance ${\sigma }_Y^2=10,000$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma _Y^2 = 10,000$$\end{document} . Then, an ANCOVA model is introduced with the inclusion of an influential covariate variable X with $\rho =\mathrm{Corr}(X,Y)=0.5$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = {Corr}(X, Y) = 0.5$$\end{document} to partially account for the variance in the response variable Y. The corresponding unexplained error variance ${\sigma }^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^{2}$$\end{document} in ANCOVA is reduced as ${\sigma }^2=(1-{\rho }^2){\sigma }_Y^2=7,500$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^{2}= (1 - \uprho ^{2})\upsigma _Y^2 = 7,500$$\end{document} . To detect the treatment differences, they showed that the total sample sizes required to have a nominal power of 0.80 are 63 and 48 for the balanced ANOVA and ANCOVA designs, respectively. Thus, the ANCOVA design has the potential benefits to attain the same power with nearly 25% fewer subjects than an ANOVA. It should be noted that the power formulas ${\Psi }_A$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{A}$$\end{document} and ${\Psi }_O$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{O}$$\end{document} given in Equations 31 and 32, respectively, were applied for sample size calculations in Maxwell and Delaney (Reference Maxwell and Delaney2004). To show a profound implication of the sample size procedures, extensive simulation study was performed under a wide range of model configurations.

First, the number of covariates and the population correlation between the response and covariate variables are extended to $P=1,\dots ,10$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P = 1, \ldots , 10$$\end{document} and $\rho =0.1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.1$$\end{document} , 0.5, and 0.9. In each combined case of P and $\rho$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho $$\end{document} , the required total sample sizes $N_{\mathrm{TO}},N_{\mathrm{TA}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{TO}, N_{TA}$$\end{document} , and $N_{\mathrm{TE}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{TE}$$\end{document} are computed with the power functions ${\Psi }_O,{\Psi }_A$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{O}, \Psi _{A}$$\end{document} and ${\Psi }_E$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{E}$$\end{document} for the ANOVA, approximate ANCOVA, and exact ANCOVA methods, respectively. Throughout this numerical investigation, the significance level and nominal power are chosen as $\alpha =0.05$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upalpha = 0.05$$\end{document} , and $1-\beta =0.80$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1 - \upbeta = 0.80$$\end{document} , respectively. Note that the effect sizes associated with $\rho =0.1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.1$$\end{document} , 0.5, and 0.9 are ${\delta }^2=0.1684$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\updelta ^{2}= 0.1684$$\end{document} , 0.2222, and 0.8772, respectively. Second, to assess the potential impact of different and smaller effect sizes, the intercept parameters are modified as $\{{\mu }_1,{\mu }_2,{\mu }_3\}=\{410,450,490\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{ \upmu _{1}, \upmu _{2}, \upmu _{3}\}=\{ 410, 450, 490\}$$\end{document} in the second set of numerical investigations. The resulting effect sizes are ${\delta }^2=0.1077$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\updelta ^{2} = 0.1077$$\end{document} , 0.1422, and 0.5614 for $\rho =0.1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.1$$\end{document} , 0.5, and 0.9, respectively. Overall, these considerations result in a total of 60 different combined configurations. For $\{{\mu }_1,{\mu }_2,{\mu }_3\}=\{400,450,500\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{ \upmu _{1}, \upmu _{2}, \upmu _{3}\}=\{ 400, 450, 500\}$$\end{document} , the computed total sample sizes $N_T$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{T}$$\end{document} are summarized in Tables 1, 2 and 3 for $\rho =0.1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.1$$\end{document} , 0.5, and 0.9, respectively. On the other hand, the corresponding results of $\{{\mu }_1,{\mu }_2,{\mu }_3\}=\{410,450,490\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{ \upmu _{1}, \upmu _{2}, \upmu _{3}\}=\{ 410, 450, 490\}$$\end{document} are presented in Tables 4, 5 and 6.

Table 1. Computed sample size, estimated power, and simulated power for the ANOVA, approximate ANCOVA, and exact ANCOVA methods when $G=3$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G = 3$$\end{document} , ${\sigma }_Y^2=10000$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^2_Y = 10000$$\end{document} , $\rho =0.1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.1$$\end{document} , ${\sigma }^2=9900$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^2 = 9900$$\end{document} , $\{{\mu }_1,{\mu }_2,{\mu }_3\}=\{400,450,500\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\upmu _1, \upmu _2, \upmu _3\} = \{400, 450, 500\}$$\end{document} , ${\delta }^2=0.1684$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\updelta ^2 = 0.1684$$\end{document} , Type I error $\alpha =0.05$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upalpha = 0.05$$\end{document} , and nominal power $1-\beta =0.80$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1 - \upbeta = 0.80$$\end{document}

Table 2. Computed sample size, estimated power, and simulated power for the ANOVA, approximate ANCOVA, and exact ANCOVA methods when $G=3$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G = 3$$\end{document} , ${\sigma }_Y^2=10000$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^2_Y = 10000$$\end{document} , $\rho =0.5$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.5$$\end{document} , ${\sigma }^2=7500$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^2 = 7500$$\end{document} , $\{{\mu }_1,{\mu }_2,{\mu }_3\}=\{400,450,500\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\upmu _1, \upmu _2, \upmu _3\} = \{400, 450, 500\}$$\end{document} , ${\delta }^2=0.2222$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\updelta ^2 = 0.2222$$\end{document} , Type I error $\alpha =0.05$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upalpha = 0.05$$\end{document} , and nominal power $1-\beta =0.80$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1 - \upbeta = 0.80$$\end{document}

Table 3. Computed sample size, estimated power, and simulated power for the ANOVA, approximate ANCOVA, and exact ANCOVA methods when $G=3$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G = 3$$\end{document} , ${\sigma }_Y^2=10000$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^2_Y = 10000$$\end{document} , $\rho =0.9$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.9$$\end{document} , ${\sigma }^2=1900$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^2 = 1900$$\end{document} , $\{{\mu }_1,{\mu }_2,{\mu }_3\}=\{400,450,500\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\upmu _1, \upmu _2, \upmu _3\} = \{400, 450, 500\}$$\end{document} , ${\delta }^2=0.8772$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\updelta ^2 = 0.8772$$\end{document} , Type I error $\alpha =0.05$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upalpha = 0.05$$\end{document} , and nominal power $1-\beta =0.80$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1 - \upbeta = 0.80$$\end{document}

Table 4. Computed sample size, estimated power, and simulated power for the ANOVA, approximate ANCOVA, and exact ANCOVA methods when $G=3$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G = 3$$\end{document} , ${\sigma }_Y^2=10000$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^2_Y = 10000$$\end{document} , $\rho =0.1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.1$$\end{document} , ${\sigma }^2=9900$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^2 = 9900$$\end{document} , $\{{\mu }_1,{\mu }_2,{\mu }_3\}=\{410,450,490\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\upmu _1, \upmu _2, \upmu _3\} = \{410, 450, 490\}$$\end{document} , ${\delta }^2=0.1077$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\updelta ^2 = 0.1077$$\end{document} , Type I error $\alpha =0.05$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upalpha = 0.05$$\end{document} , and nominal power $1-\beta =0.80$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1 - \upbeta = 0.80$$\end{document}

Table 5. Computed sample size, estimated power, and simulated power for the ANOVA, approximate ANCOVA, and exact ANCOVA methods when $G=3$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G = 3$$\end{document} , ${\sigma }_Y^2=10000$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^2_Y = 10000$$\end{document} , $\rho =0.5$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.5$$\end{document} , ${\sigma }^2=7500$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^2 = 7500$$\end{document} , $\{{\mu }_1,{\mu }_2,{\mu }_3\}=\{410,450,490\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\upmu _1, \upmu _2, \upmu _3\} = \{410, 450, 490\}$$\end{document} , ${\delta }^2=0.1422$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\updelta ^2 = 0.1422$$\end{document} , Type I error $\alpha =0.05$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upalpha = 0.05$$\end{document} , and nominal power $1-\beta =0.80$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1 - \upbeta = 0.80$$\end{document}

Table 6. Computed sample size, estimated power, and simulated power for the ANOVA, approximate ANCOVA, and exact ANCOVA methods when $G=3$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G = 3$$\end{document} , ${\sigma }_Y^2=10000$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^2_Y = 10000$$\end{document} , $\rho =0.9$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.9$$\end{document} , ${\sigma }^2=1900$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upsigma ^2 = 1900$$\end{document} , $\{{\mu }_1,{\mu }_2,{\mu }_3\}=\{410,450,490\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\upmu _1, \upmu _2, \upmu _3\} = \{410, 450, 490\}$$\end{document} , ${\delta }^2=0.5614$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\updelta ^2 = 0.5614$$\end{document} , Type I error $\alpha =0.05$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upalpha = 0.05$$\end{document} , and nominal power $1-\beta =0.80$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1 - \upbeta = 0.80$$\end{document}

The sample size calculations presented in Tables 1, 2, 3, 4, 5 and 6 reveal that, as expected, the computed sample sizes of the ANOVA procedure remain identical for different number of covariates P when all other factors are fixed. In contrast, the sample size of the exact approach increases with increase in the number of covariates P and with decrease in the effect size ${\delta }^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\updelta ^{2}$$\end{document} when all other configurations are held constant. Likewise, the total sample size produced by the approximate procedure also increases with decrease in the effect size ${\delta }^2$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\updelta ^{2}$$\end{document} . However, the reported sample sizes in Tables 1 and 2 do not vary with P, and the computed sample sizes marginally increase with larger P for the other cases in Tables 3, 4, 5 and 6. More importantly, the total sample sizes $N_{\mathrm{TO}},N_{\mathrm{TA}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{TO}, N_{TA}$$\end{document} , and $N_{\mathrm{TE}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{TE}$$\end{document} associated with the ANOVA, approximate ANCOVA, and exact ANCOVA procedures have a consistent order of $N_{\mathrm{TA}}\le N_{\mathrm{TO}}\le N_{\mathrm{TE}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{TA} \le N_{TO} \le N_{TE}$$\end{document} for all the cases in Tables 1 and 4 with $\rho =0.1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.1$$\end{document} . The order between the two sample sizes $N_{\mathrm{TO}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{TO}$$\end{document} and $N_{\mathrm{TE}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{TE}$$\end{document} is reversed for large magnitudes of $\rho =0.5$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.5$$\end{document} and 0.9 with $N_{\mathrm{TA}}\le N_{\mathrm{TE}}\le N_{\mathrm{TO}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{TA} \le N_{TE} \le N_{TO}$$\end{document} for the situations in Tables 2, 3, 5 and 6. For ease of explication, the estimated powers for the three different sample size procedures are also listed in Tables 1, 2, 3, 4, 5 and 6.

To justify the accuracy of sample size determination, Monte Carlo simulation studies were performed for the prescribed 60 design settings. With the computed sample sizes, parameter configurations, and nominal power, estimates of the true power were computed via Monte Carlo simulation of 10,000 independent data sets. For each replicate, $N_{\mathrm{TO}},N_{\mathrm{TA}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{TO}, N_{TA}$$\end{document} , and $N_{\mathrm{TE}}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{TE}$$\end{document} normal outcomes are generated with the ANCOVA models. Because the power function ${\Psi }_E$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{E}$$\end{document} is irrelevant to the mean vector $\theta$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varvec{\uptheta }$$\end{document} and variance–covariance matrix $\Sigma$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varvec{\Sigma }$$\end{document} of the designated covariate distribution, the covariates are assumed to have independent and identical multinormal distribution $N_P(0_P,I_P)$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{P}(\mathbf{0 }_{P}, \mathbf{I }_{P})$$\end{document} where $0_P$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{0 }_{P}$$\end{document} is a $P\times 1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P \times 1$$\end{document} null column vector and $I_P$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf{I }_{P}$$\end{document} is an identity matrix of dimension P. The regression coefficients are chosen as ${\beta }_1=\dots ={\beta }_P={\beta }^{\ast }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upbeta _{1} = \ldots = \upbeta _{P} = \upbeta ^*$$\end{document} and ${\beta }^{\ast }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upbeta ^*$$\end{document} is a designated value so that the resulting correlation $\rho =0.1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.1$$\end{document} , 0.5, and 0.9. Next, the Wald test statistic $W^{\ast }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^*$$\end{document} was computed and the simulated power was the proportion of the 10,000 replicates whose test statistics $W^{\ast }$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^*$$\end{document} exceed the corresponding critical value $F_{2,\nu ,0.05}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{2,\,\upnu ,\,0.05}$$\end{document} . The simulated power and error are also summarized in Tables 1, 2, 3, 4, 5 and 6 for all the ANCOVA designs. To illustrate the contrasting behavior of the three contending techniques, the induced errors for $\rho =0.1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.1$$\end{document} , 0.5, 0.9 in Tables 1, 2 and 3 are also plotted in Figs. 1, 2, and 3, respectively.

Figure 1. Errors of power estimation for $G=3$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G = 3$$\end{document} and $\rho =0.1$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.1$$\end{document}

Figure 2. Errors of power estimation for $G=3$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G = 3$$\end{document} and $\rho =0.5$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.5$$\end{document}

Figure 3. Errors of power estimation for $G=3$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G = 3$$\end{document} and $\rho =0.9$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uprho = 0.9$$\end{document}

According to the power comparisons, the ANOVA method generally does not provide accurate sample size calculations for an ANCOVA design. Unsurprisingly, the only exceptions occurred when the number of covariates is small and the correlation between the covariates and the response variable is close to zero as in Tables 1 and 4. The approximate ANCOVA method consistently gives larger power estimate than the simulated power for all cases considered here. The discrepancy noticeably increases with the number of covariates and the magnitude of effect size. The resulting errors can be as large as 0.0845, 0.1105, and 0.3049 associated with the scenarios of $P=10$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P = 10$$\end{document} in Tables 1, 2 and 3, respectively. For the relative smaller effect sizes in Tables 4, 5 and 6, the performance of the approximate ANCOVA formula has improved with the errors of 0.0506, 0.0628, and 0.2710 for the cases of $P=10$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P= 10$$\end{document} . Consequently, the overestimation problem of the power function ${\Psi }_A$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{A}$$\end{document} suggests that the computed sample sizes are generally inadequate to achieve the designated power level.

Regarding the accuracy of the proposed exact ANCOVA approach, the corresponding results in Tables 1, 2, 3, 4, 5 and 6 show that the differences between the estimated and simulated powers are fairly small. The largest absolute error is 0.0106 for the two cases of $P=8$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P = 8$$\end{document} and 5 in Tables 2 and 5, respectively. All the other 58 cases in Tables 1, 2, 3, 4, 5 and 6 have an absolute error less than 0.01. These numerical results imply that the proposed exact approach outperforms the ANOVA method and the approximate ANCOVA procedure for all design configurations considered here. Therefore, the suggested power and sample size calculations can be recommended for general use.

6. An Example

A documented example of Maxwell and Delaney (Reference Maxwell and Delaney2004) is presented and extended next to demonstrate the usefulness of the suggested power and sample size procedures and accompanying software programs for the omnibus test of treatment effects in ANCOVA designs.

Specifically, Maxwell and Delaney (Reference Maxwell and Delaney2004, Table 9.7, p. 429) provided the data for assessing the effectiveness of different interventions for depression. There are 10 participants with random assignment in each of the three intervention groups of (1) selective serotonin reuptake inhibitor (SSRI) antidepressant medication, (2) placebo, or (3) wait list control. The measurements are the pretest and posttest Beck Depression Inventory (BDI) scores of depressive individuals. The primary interest of the ANCOVA study is on the group differences of posttest BDI measurements using the pretest BDI scores as covariates. The results show that the estimates of adjusted group means and error variance are $\{{\hat{\mu }}_1^{\ast },{\hat{\mu }}_2^{\ast },{\hat{\mu }}_3^{\ast }\}=\{7.5366,11.9849,13.9785\}$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{{\hat{\upmu }}_1^*,{\hat{\upmu }}_2^*,{\hat{\upmu }}_3^*\}=\{ 7.5366, 11.9849, 13.9785\}$$\end{document} and ${\hat{\sigma }}^2=29.0898$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{\upsigma }}^2 = 29.0898$$\end{document} , respectively. The omnibus F test statistic of treatment differences is $W^{\ast }=3.73$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^* = 3.73$$\end{document} , which yields a p-value of 0.0376. Therefore, the test result suggests that the intervention effects are significantly different at $\alpha =0.05$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upalpha = 0.05$$\end{document} . Although this is not the focus in the illustration of Maxwell and Delaney (Reference Maxwell and Delaney2004), it can be computed from an ANOVA of posttest scores that the variance estimate is ${\hat{\sigma }}_Y^2=39.6185$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{\upsigma }}_Y^2= 39.6185$$\end{document} . Hence, the sample squared correlation between the posttest and pretest BDI scores is ${\hat{\rho }}^2=1-{\hat{\sigma }}^2{\hat{\sigma }}_Y^2=1-29.0898/39.6185=0.2658$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{\uprho }}^2 = 1 -{\hat{\upsigma }}^2{\hat{\upsigma }}_Y^2 = 1 - 29.0898/39.6185 = 0.2658$$\end{document} . The observed value of the ANOVA F test of group differences is $F^{\ast }=3.03$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F^* = 3.03$$\end{document} with a p-value of 0.0647. At the significance level 0.05, the omnibus test of no intervention group difference on the posttest BDI scores cannot be rejected. Although null hypothesis significance testing is useful in various applications, it is important to consult the recent articles of Wasserstein and Lazar (Reference Wasserstein and Lazar2016) and Wasserstein et al. (Reference Wasserstein, Schirm and Lazar2019) for the recommended principles underlying proper use and interpretation of statistical significance and p-values.

In view of the prospective nature of advance research planning, the general guidelines suggest published findings or expert opinions can offer reliable information for the vital characteristics of future study. Accordingly, it is prudent to adopt a minimal meaningful effect size in order to enhance the generalizability of the result and the accumulation of scientific knowledge. For illustration, the prescribed summary statistics of the three-group depression intervention study are employed as population adjusted mean effects and variance component. The suggested power procedure shows that the resulting power for the omnibus test of group differences is ${\Psi }_E=0.6145$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Psi _{E} = 0.6145$$\end{document} when the significance level $\alpha$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upalpha $$\end{document} equals to 0.05. Because the computed power is substantially smaller than the common levels of 0.80 or 0.90, this implies that the group sample size $N=10$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N = 10$$\end{document} does not provide a decent chance of detecting the potential differences between treatment groups. To determine the proper sample size, the proposed sample size computations showed that the balanced group sample sizes of 15 and 19 are required to attain the nominal power of 0.8 and 0.9, respectively. The total sample sizes $N_T=45$ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{T} = 45$$\end{document} and 57 are substantially larger than 30 of the exemplifying design. Essentially, it requires 50% and 90% increases of the sample size to meet the common power levels of 0.80 and 0.90, respectively. These design configurations are presented in the user specifications of the SAS/IML and R programs presented in the supplemental programs. Researchers can easily identify these statements and then modify the input values in the computer code to incorporate their own model characteristics.