Inferences on location parameters based on independent multivariate skew normal distributions

Ma, Ziwei; Wang, Tonghui; Wei, Zheng; Zhu, Xiaonan

doi:10.1108/AJEB-03-2022-0034

Purpose

The purpose of this study is to extend the classical noncentral F-distribution under normal settings to noncentral closed skew F-distribution for dealing with independent samples from multivariate skew normal (SN) distributions.

Design/methodology/approach

Based on generalized Hotelling's T² statistics, confidence regions are constructed for the difference between location parameters in two independent multivariate SN distributions. Simulation studies show that the confidence regions based on the closed SN model outperform the classical multivariate normal model if the vectors of skewness parameters are not zero. A real data analysis is given for illustrating the effectiveness of our proposed methods.

Findings

This study’s approach is the first one in literature for the inferences in difference of location parameters under multivariate SN settings. Real data analysis shows the preference of this new approach than the classical method.

Research limitations/implications

For the real data applications, the authors need to remove outliers first before applying this approach.

Practical implications

This study’s approach may apply many multivariate skewed data using SN fittings instead of classical normal fittings.

Originality/value

This paper is the research paper and the authors’ new approach has many applications for analyzing the multivariate skewed data.

1. Introduction

Although the normal distribution is a standard assumption for modeling observations in general, practitioners and researchers prefer more flexible models that account for the non-normality when the data collected in finance and econometric fields. The family of skew normal (SN) distributions, introduced by Azzalini (1985) for the univariate case, Azzalini and Valle (1996) for the multivariate case and Chen and Gupta (2005) for the matrix variate case, becomes a popular parametric family in statistical analysis of real data which account for asymmetry. There are several successful applications using SN, like modeling skewness premium of a financial asset by Carmichael and Coën (2013), addressing “wrong skewness” problems in stochastic frontier models by Wei et al. (2021). Here just list a few, an updated review was given by Adcock and Azzalini (2020).

Based on the definition given in Arellano-Valle et al. (2005), a p-dimensional random vector Y is said to be SN distributed with the location parameter vector $μ \in R^{p}$ ⁠, the scale parameter matrix Σ (a p × p positive definite matrix), and the shape parameter vector $λ \in R^{p}$ ⁠, denoted as $Y \sim S N_{p} (μ, Σ, λ)$ ⁠, if its probability density function (pdf) is given by

f_{Y} (y) = 2 ϕ_{p} (y; μ, Σ) Φ (λ^{'} Σ^{- 1 / 2} (y - μ)), y \in R^{p},

(1)

where ϕ_p(∙; μ, Σ) is the pdf of the p-dimensional normal distribution with the mean vector μ and the covariance matrix Σ, and Φ(∙) is the cumulative distribution function (cdf) of the univariate standard normal distribution. Extensions of Equation (1) are investigated by many researchers (see Azzalini and Capitanio, 1999; Wang et al., 2009; Young et al., 2016; Li et al., 2018).

For the univariate SN family, constructing plausibility regions for skewness parameter was discussed by Zhu et al. (2017) using inferential models (IMs). The joint plausibility regions for location parameter and skewness parameter were studied by Ma et al. (2018) when scale parameter is known using IMs, and the joint plausibility regions for location parameter and scale parameter were constructed by Zhu et al. (2018) when skewness parameter is given. For multivariate SN model, the confidence regions for location parameter are obtained by Ma et al. (2019). In this work, we study the difference of location parameters based on independent multivariate SN distributions so that the generalized Hotelling's T², and noncentral closed skew F-distributions are used. Under the assumption of equal but unknown scale parameters, the confidence regions for differences of location parameters of the multivariate SN model are proposed. Simulation studies show that the proposed confidence regions have higher relative coverage frequency rates than those in classical normal model for skewed data.

The organization of this paper is listed below. In Section 2, the definition of matrix variate SN distribution is introduced and some useful properties of sampling distribution on difference of sample means are derived. In Section 3, the confidence regions on the difference of location parameters by pivotal method are proposed when scale parameters from two populations are assumed to be equal but unknown. A group of simulation studies, which illustrate the effectiveness of our proposed methods, are given in Section 4, followed by a real data example in Section 5. The conclusion is given in Section 6.

2. Matrix variate SN distributions and sampling distributions

Let M_n×k be the set of all n × k matrices over the real field $R$ and $R^{n} = M_{n \times 1}$ ⁠. The transpose of a matrix A is denoted as A′. The n × n identity matrix is denoted as I_n, the constant vector ${(1, \dots, 1)}^{'} \in R^{n}$ is denoted as 1_n, and ${\bar{J}}_{n} = \frac{1}{n} 1_{n} 1_{n}^{'}$ ⁠. For $B = {(b_{1}, b_{2}, \dots, b_{n})}^{'} \in M_{n \times k}$ with $b_{i} \in R^{k}$ ⁠, let $Vec (B) = {(b_{1}^{'}, b_{2}^{'}, \dots, b_{n}^{'})}^{'} \in R^{n k}$ ⁠. For positive definite matrix T ∈ M_n×n, we use T⁻¹, T^1/2 and T^−1/2 to denote, respectively, the inverse, symmetric square root of T, and symmetric square root of T⁻¹. For B ∈ M_m×n, C ∈ M_n×p, we use B ⊗ C to denote the Kronecker product of B and C. Through this paper, N(0, 1) represents the standard normal distribution and bold phase letters represent vectors.

Definition 2.1.

Ye et al. (2014) The n × p random matrix Y is said to have a SN matrix distribution with location matrix M, scale matrix V ⊗Σ and skewness parameter matrix γ ⊗λ′, denoted by $Y \sim S N_{n \times p} (M, V \otimes Σ, γ \otimes λ^{'})$ , if $y \equiv Vec (Y) \sim S N_{n p} (μ, V \otimes Σ, γ \otimes λ)$ , where M ∈ M_n×p, V ∈ M_n×n, $μ = Vec (M)$ , $γ \in R^{n}$ and $λ \in R^{p}$

Suppose that $X_{1} \in M_{n_{1} \times p}$ and $X_{2} \in M_{n_{2} \times p}$ are two independent sample matrices such that

X_{i} \sim S N_{n_{i} \times p} (1_{n_{i}} \otimes μ_{i}^{'}, I_{n_{i}} \otimes Σ_{i}, 1_{n_{i}} \otimes λ_{i}^{'})

(2)

for i = 1, 2. We are interested in analyzing the difference vector μ_d = μ₁ − μ₂. The sampling distributions of the sample mean and sample covariance matrix are given by Ma et al. (2019) in following Lemma.

Lemma 2.1.

Ma et al. (2019) Let $Y \sim S N_{n \times p} (1_{n} \otimes μ^{'}, I_{n} \otimes Σ, 1_{n} \otimes λ^{'})$ . Then,

\bar{Y} = {(\frac{1}{n} 1_{n}^{'} Y)}^{'} \sim S N_{p} (μ, \frac{Σ}{n}, \sqrt{n} λ)

and

(n - 1) S = Y^{'} (I_{n} - {\bar{J}}_{n}) Y \sim W_{p} (n - 1, Σ)

are independently distributed, where

W_{p} (n - 1, Σ)

represents thep-dimensional Wishart distribution withn − 1 degrees of freedom and the mean Σ.

By Lemma 2.1, we have

{\bar{X}}_{i} = {(\frac{1}{n_{i}} 1_{n_{i}}^{'} X_{i})}^{'} \sim S N_{p} (μ_{i}, \frac{Σ_{i}}{n_{i}}, \sqrt{n_{i}} λ_{i})

(3)

and

(n_{i} - 1) S_{i} = X_{i}^{'} (I_{n_{i}} - {\bar{J}}_{n_{i}}) X_{i} \sim W_{p} (n_{i} - 1, Σ_{i})

(4)

for i = 1, 2. It is natural to use the statistic ${\bar{X}}_{d} = {\bar{X}}_{1} - {\bar{X}}_{2}$ to inference on μ_d.

The difference between two independent SN distributed random vectors follows a closed SN distribution, which is reviewed below.

Definition 2.2.

(Gonz $\overset{́}{a}$ lez-Far $\overset{́}{i}$ as et al. (Gonzalez-Farias et al., 2004) A random vector $Y \in R^{p}$ is said to have closed SNdistribution (CSN), denoted as $C S N_{p, q} (μ, Σ, D, v, Δ)$ , if its pdf is

f_{p, q} (y; μ, Σ, D, v, Δ) = C ϕ_{p} (y; μ, Σ) Φ_{q} (D (y - μ); v, Δ), y \in R^{p},

where

C^{- 1} = Φ_{q} (0; v, Δ + D Σ D^{'})

,p ≥ 1,q ≥ 1,

μ \in R^{p}

,

Σ \in M_{p \times p}^{+}

,D ∈ M_q×p,

v \in R^{q}

,

Δ \in M_{q \times q}^{+}

and

ϕ_{k} (\cdot; η, Ω)

,

Φ_{k} (\cdot; η, Ω)

are the pdf and cdf of ak-dimensional normal distribution.

For simplicity, we assume that ν = 0 so that Y ∼ CSN_p,q(μ, Σ, D, Δ). The following two properties of CSN can lead to the distribution of ${\bar{X}}_{d}$ ⁠.

Lemma 2.2.

(Gonz $\overset{́}{a}$ lez-Far $\overset{́}{i}$ as et al. (Gonzalez-Farias et al., 2004) Let Y ∼ CSN_p,q(μ, Σ, D, Δ)

For an arbitrary constant $b \in R^{p}$ ,

Y + b \sim C S N_{p, q} (μ + b, Σ, D, Δ)

(5)

For nonzero real number $c \in R$ ,

c Y \sim C S N_{p, q} (c μ, c^{2} Σ, c^{- 1} D, Δ);

(6)

Let $Y_{i} \sim C S N_{p, q_{i}} (μ_{i}, Σ_{i}, D_{i}, Δ_{i})$ , for i = 1, 2, be independently distributed. Then,

Y_{1} + Y_{2} \sim C S N_{p, q_{1} + q_{2}} (μ_{1} + μ_{2}, Σ_{1} + Σ_{2}, D^{*}, Δ^{*})

(7)

where

D^{*} = (\begin{matrix} D_{1} Σ_{1} {(Σ_{1} + Σ_{2})}^{- 1} \\ D_{2} Σ_{2} {(Σ_{1} + Σ_{2})}^{- 1} \end{matrix}), Δ^{*} = (\begin{matrix} A_{11} & A_{12} \\ A_{21} & A_{22} \end{matrix})

and

\begin{matrix} A_{11} & = & Δ_{1} + D_{1} Σ_{1} D_{1}^{'} - D_{1} Σ_{1} {(Σ_{1} + Σ_{2})}^{- 1} Σ_{1} D_{1}^{'}, \\ A_{22} & = & Δ_{2} + D_{2} Σ_{2} D_{2}^{'} - D_{2} Σ_{2} {(Σ_{1} + Σ_{2})}^{- 1} Σ_{2} D_{2}^{'}, \\ A_{12} & = & - D_{1} Σ_{1} {(Σ_{1} + Σ_{2})}^{- 1} Σ_{2} D_{2}^{'} . \end{matrix}

In term of CSN, ${\bar{X}}_{i} \sim C S N_{p, 1} (μ_{i}, \frac{Σ_{i}}{n_{i}}, n_{i} λ_{i}^{'} Σ_{i}^{- 1 / 2}, 1)$ for i = 1, 2. Thus, we obtain the distribution of ${\bar{X}}_{d}$ ⁠.

Theorem 2.1.

Let ${\bar{X}}_{d} = {\bar{X}}_{1} - {\bar{X}}_{2}$ with ${\bar{X}}_{i} \sim C S N_{p, 1} (μ_{i}, \frac{Σ_{i}}{n_{i}}, n_{i} λ_{i}^{'} Σ_{i}^{- 1 / 2}, 1)$ for i = 1, 2. Then,

{\bar{X}}_{d} \sim C S N_{p, 2} (μ_{d}, Σ_{d}, D_{d}, Δ_{d})

(8)

where

\begin{matrix} μ_{d} = μ_{1} - μ_{2}, & Σ_{d} = \frac{Σ_{1}}{n_{1}} + \frac{Σ_{2}}{n_{2}} \\ D_{d} = (\begin{matrix} λ_{1}^{'} Σ_{1}^{1 / 2} Σ_{d}^{- 1} \\ - λ_{2}^{'} Σ_{2}^{1 / 2} Σ_{d}^{- 1} \end{matrix}), & Δ_{d} = (\begin{matrix} A_{11} & A_{12} \\ A_{21} & A_{22} \end{matrix}) \end{matrix}

(9)

with

\begin{matrix} A_{11} & = & 1 + n_{1} λ_{1}^{'} λ_{1} - λ_{1}^{'} Σ_{1}^{1 / 2} Σ_{d}^{- 1} Σ_{1}^{1 / 2} λ_{1}, \\ A_{22} & = & 1 + n_{2} λ_{2}^{'} λ_{2} - λ_{2}^{'} Σ_{2}^{1 / 2} Σ_{d}^{- 1} Σ_{2}^{1 / 2} λ_{2}, \\ A_{12} & = & A_{21} = λ_{1}^{'} Σ_{1}^{1 / 2} Σ_{d}^{- 1} Σ_{2}^{1 / 2} λ_{2} . \end{matrix}

Proof.

By part (2) and (3) of Lemma 2.2, the desired result follows immediately. □

Remark.

If λ₂ = 0, i.e. X₂ following multivariate normal distribution with mean μ₂ and covariance $\frac{Σ_{2}}{n_{2}}$ ⁠, the distribution of difference ${\bar{X}}_{d}$ has the form $X_{d} \sim C S N_{p, 1} (μ_{d}, Σ_{d}, λ_{1}^{'} Σ_{1}^{1 / 2} Σ_{d}^{1 / 2}, A_{11})$ which can be further expressed as $X_{d} \sim S N_{p} (μ_{d}, Σ_{d}, Σ_{1}^{1 / 2} λ_{1} / A_{11}^{1 / 2})$ ⁠.

Figure 1 presents the contour of bivariate closed SN for various combinations of shape parameter parameters D with different scale parameter $Σ = (\begin{matrix} 1 & ρ \\ ρ & 1 \end{matrix})$ ⁠. Specifically, the gray contours show D = 0 with ρ = 0, 0.5 and −0.5; the green contours show $D = (\begin{matrix} 0 & 0 \\ 3 & 0 \end{matrix})$ with ρ = 0, 0.5 and −0.5; the blue contours show $D = (\begin{matrix} 1 & - 5 \\ 0 & 0 \end{matrix})$ with ρ = 0, 0.5 and −0.5 the red contours show $D = (\begin{matrix} 1 & - 5 \\ 3 & 0 \end{matrix})$ with ρ = 0, 0.5 and −0.5. From another point of view, these contour plots present bivariate normal distribution (gray), SN distribution (green and blue) and closed SN distribution (red).

Figure 1

View large Download slide

Contour plot of bivariate normal, SN and close SN distributions with various values of parameters

3. Inference on difference of location parameters

In this section, the inference on the difference of location parameter is proposed when the scale parameter Σ₁ and Σ₂ are unknown but assumed to be equal, say Σ₁ = Σ₂ = Σ. The main result is based on the generalized Hotelling's T² under multivariate SN setting.

3.1 Some related distributions

At first, we consider the distribution of $S = {({\bar{X}}_{d} - μ_{d})}^{'} Σ_{d}^{- 1} ({\bar{X}}_{d} - μ_{d})$ ⁠. The following definition and lemma by Zhu et al. (2019) are useful to derive the distribution of S.

Definition 3.1.

Zhu et al. (2019) Let X ∼ CSN_p,q(μ, I_p, D, Δ_q). The distribution of X′X, denoted by $X^{'} X \sim C S χ_{p}^{2} (λ, δ_{1}, δ_{2}, Δ_{q})$ , is called a noncentral closed skew chi-square distribution with degrees of freedom p, noncentrality parameter λ = μ′μ, skewness parameters δ₁ = Dμ, δ₂ = DD′ and parameter Δ_q

Lemma 3.1.

Zhu et al. (2019) Let X ∼ CSN_p,q(μ, Σ, D, Δ) and Q = X′WX with a nonnegative definite W ∈ M_p×p. If Σ^1/2WΣ^1/2 is idempotent of rank k, then $Q \sim C S χ_{k}^{2} (λ, δ_{1}, δ_{2}, Ω_{q})$ , where λ = μ′Wμ, δ₁ = DΣWμ, δ₂ = DΣWΣD′ and Ω_q = Δ + D(Σ −ΣWΣ)⁻¹D′

Based on Theorem 2.1 and Lemma 3.1, we obtain the following result.

Proposition 3.1.

Let $S = {({\bar{X}}_{d} - μ_{d})}^{'} Σ_{d}^{- 1} ({\bar{X}}_{d} - μ_{d})$ . Then, $S \sim C S χ_{p}^{2} (0, 0, δ_{2}, Ω)$ with

$δ_{2} = \frac{n_{1} n_{2}}{n_{1} + n_{2}} (\begin{matrix} λ_{1}^{'} λ_{1} & λ_{2}^{'} λ_{1} \\ λ_{2}^{'} λ_{1} & λ_{2}^{'} λ_{2} \end{matrix})$ and $Ω = (\begin{matrix} 1 + \frac{n_{1}^{2} λ_{1}^{'} λ_{1}}{n_{1} + n_{2}} & \frac{n_{1} n_{2} λ_{2}^{'} λ_{1}}{n_{1} + n_{2}} \\ \frac{n_{1} n_{2} λ_{2}^{'} λ_{1}}{n_{1} + n_{2}} & 1 + \frac{n_{2}^{2} λ_{2}^{'} λ_{2}}{n_{1} + n_{2}} \end{matrix})$ ⁠.

Proof.

From part (i) of Lemma 2.2, we have ${\bar{X}}_{d} - μ_{d} \sim C S N_{p, 2} (0, Σ_{d}, D_{d}, Δ_{d})$ ⁠. Since $Σ_{d}^{1 / 2} Σ_{d}^{- 1} Σ_{d}^{1 / 2} = I_{p}$ is an idempotent of rank p, the desired result follows immediately. □

Remark.

Comparing with one sample case, the distribution of quantity $n {(\bar{X} - μ)}^{'} Σ^{- 1} (\bar{X} - μ) \sim χ_{p}^{2}$ for $\bar{X} \sim S N_{p} (μ, \frac{Σ}{n}, \sqrt{n} λ)$ is free of the skewness parameter λ. Here, the distribution of S follows noncentral closed SN distribution given above which depends on the parameters δ₂ and Ω. Readers are referred to check out Figures 5 and 6 in Zhu et al. (2019) for the density curves of CSχ²(0, 0, δ₂, Ω).

3.2 Confidence region of μ_d

In this subsection, we will extend the Hotelling's T² statistic from multivariate normal setting to the multivariate SN setting, called the generalized Hotelling's $T^{2} = \frac{n_{1} n_{2}}{n_{1} + n_{2}} {({\bar{X}}_{d} - μ_{d})}^{'} S_{p}^{- 1} ({\bar{X}}_{d} - μ_{d})$ ⁠, to construct the confidence regions for the difference of location vector where $(n_{1} + n_{2} - 2) S_{p} = (n_{1} - 1) S_{1} + (n_{2} - 1) S_{2}$ ⁠. First we need to derive the distribution of S_p, then extend the F-distribution to closed skew F-distribution which can describe the distribution of T² under the multivariate SN setting.

Proposition 3.2.

Let $(n_{1} + n_{2} - 2) S_{p} = (n_{1} - 1) S_{1} + (n_{2} - 1) S_{2}$ with S₁ and S₂ defined by equation (4) Then, (n₁ + n₂ − 2)S_p ∼ W_p(n₁ + n₂ − 2, Σ)

Proof.

By Lemma 2.1, (n_i − 1)S_i ∼ W_p(n_i − 1, Σ) for i = 1, 2 are independently distributed. Thus, the well-known properties of Wishart distribution for sums and scale transformation lead to the desired result. □

To obtain the distribution of T², we need the following well-known result (Lemma 3.2, Mardia et al. (1980), Theorem 3.4.7) and extended version of the F-distribution, called closed skew F-distribution, Definition 3.2, which was introduced by Zhu et al. (2019).

Lemma 3.2.

If $H \sim W_{p} (m, Σ)$ , m > p, then the ratio $a^{'} Σ^{- 1} a / a^{'} H^{- 1} a \sim χ_{m - p + 1}^{2}$ for any fixed p-vector a

Definition 3.2.

Zhu et al. (2019) Let $U_{1} \sim C S χ_{n_{1}}^{2} (λ, δ_{1}, δ_{2}, Δ_{m})$ , $U_{2} \sim χ_{n_{2}}^{2}$ , and U₁ and U₂ are independent. The distribution of $F = \frac{U_{1} / n_{1}}{U_{2} / n_{2}}$ is called the noncentral closed skew F-distribution with degrees of freedom n₁ and n₂, and parameters λ, δ₁, δ₂ and Δ_m, denoted by $F \sim C S F_{n_{1}, n_{2}} (λ, δ_{1}, δ_{2}, Δ_{m})$

Based on above definition, the pdf of noncentral closed skew F-distribution can be obtained below.

Proposition 3.3.

Let $F \sim C S F_{n_{1}, n_{2}} (λ, δ_{1}, δ_{2}, Δ_{m})$ . The pdf of F is given by

f_{F} (x; λ, δ_{1}, δ_{2}, Δ_{m}) = \int_{0}^{\infty} \frac{n_{1} v}{n_{2}} f_{1} (\frac{n_{1} x v}{n_{2}}; λ, δ_{1}, δ_{2}, Δ_{m}) f_{2} (v) d v

(10)

wheref₁(∙; λ, δ₁, δ₂, Δ_m) andf₂(∙) are pdf of

C S χ_{n_{1}}^{2} (λ, δ_{1}, δ_{2}, Δ_{m})

and

χ_{n_{2}}^{2}

, respectively.

Proof.

Let $F = \frac{U_{1} / n_{1}}{U_{2} / n_{2}}$ with $U_{1} \sim C S χ_{n_{1}}^{2} (λ, δ_{1}, δ_{2}, Δ_{m})$ ⁠, $U_{2} \sim χ_{n_{2}}^{2}$ independently distributed. The joint density of (U₁, U₂) is $f_{(U_{1}, U_{2})} (u, v) = f_{1} (u) f_{2} (v)$ where f₁(∙) and f₂(∙) are pdf of U₁ and U₂, respectively. Then change of variables $x = \frac{n_{2} u}{n_{1} v}$ ⁠, h = v, for x > 0, h > 0. The Jacobian of this transformation is n₁v/n₂. So integrating with respect to v over 0 < v < ∞ leads to desired results.

By Lemma 3.2 and Definition 3.2, we obtain the distribution of T² as follows.

Theorem 3.1.

For two independently distributed random matrices

X_{i} \sim S N_{n_{i} \times p} (1_{n_{i}} \otimes μ_{i}^{'}, I_{n_{i}} \otimes Σ, 1_{n_{i}} \otimes λ_{i}^{'}) for i = 1,2,

whereλ_i's are known. Let

{\bar{X}}_{i}

,S_ibe given byand,

{\bar{X}}_{d} = {\bar{X}}_{1} - {\bar{X}}_{2}

, and

S_{p} = (n_{1} - 1) S_{1} + (n_{2} - 1) S_{2}

. Then, the distribution of

T^{2} = \frac{n_{1} n_{2}}{n_{1} + n_{2}} {({\bar{X}}_{d} - μ_{d})}^{'} S_{p}^{- 1} ({\bar{X}}_{d} - μ_{d})

is given by

\frac{n_{1} + n_{2} - p - 1}{p (n_{1} + n_{2} - 2)} T^{2} \sim C S F_{p, n_{1} + n_{2} - p - 1} (0,0, δ_{2}, Ω),

(11)

where

δ_{2} = \frac{n_{1} n_{2}}{n_{1} + n_{2}} (\begin{matrix} λ_{1}^{'} λ_{1} & λ_{2}^{'} λ_{1} \\ λ_{2}^{'} λ_{1} & λ_{2}^{'} λ_{2} \end{matrix})

and

Ω = (\begin{matrix} 1 + \frac{n_{1}^{2} λ_{1}^{'} λ_{1}}{n_{1} + n_{2}} & \frac{n_{1} n_{2} λ_{2}^{'} λ_{1}}{n_{1} + n_{2}} \\ \frac{n_{1} n_{2} λ_{2}^{'} λ_{1}}{n_{1} + n_{2}} & 1 + \frac{n_{2}^{2} λ_{2}^{'} λ_{2}}{n_{1} + n_{2}} \end{matrix})

⁠.

Proof.

Rewrite T² as

T^{2} = \frac{n_{1} n_{2}}{n_{1} + n_{2}} \frac{{({\bar{X}}_{d} - μ_{d})}^{'} Σ^{- 1} ({\bar{X}}_{d} - μ_{d})}{{({\bar{X}}_{d} - μ_{d})}^{'} Σ^{- 1} ({\bar{X}}_{d} - μ_{d}) / {({\bar{X}}_{d} - μ_{d})}^{'} S_{p}^{- 1} ({\bar{X}}_{d} - μ_{d})} .

Since S_p and ${\bar{X}}_{d} - μ_{d}$ are independent by Lemma 2.1, the conditional distribution of

w = (n_{1} + n_{2} - 2) {({\bar{X}}_{d} - μ_{d})}^{'} Σ^{- 1} ({\bar{X}}_{d} - μ_{d}) / {({\bar{X}}_{d} - μ_{d})}^{'} S_{p}^{- 1} ({\bar{X}}_{d} - μ_{d}) \sim χ_{n_{1} + n_{2} - p - 1}^{2}

given ${\bar{X}}_{d} - μ_{d}$ by Proposition 3.2 and Lemma 3.2. It is clear that $w \sim χ_{n_{1} + n_{2} - p - 1}^{2}$ since the conditional distribution does not depend on ${\bar{X}}_{d} - μ_{d}$ ⁠. On the other hand, since $Σ_{d}^{- 1} = \frac{n_{1} n_{2}}{n_{1} + n_{2}} Σ^{- 1}$ ⁠, then $\frac{n_{1} n_{2}}{n_{1} + n_{2}} {({\bar{X}}_{d} - μ_{d})}^{'} Σ^{- 1} ({\bar{X}}_{d} - μ_{d}) = {({\bar{X}}_{d} - μ_{d})}^{'} Σ_{d}^{- 1} ({\bar{X}}_{d} - μ_{d})$ which follows closed skew chi-square distribution $C S χ_{p}^{2} (0, 0, δ_{2}, Ω)$ from Proposition 3.1. Therefore, the desired result follows by Definition 3.2.

Based on above results, we construct confidence regions for the difference of location parameter μ_d by using generalized Hotelling's T² as a pivotal statistics.

Theorem 3.2.

Assume two samples satisfying (2) with unknown Σ₁ and Σ₂ but Σ₁ = Σ₂ and known λ₁ and λ₂. Then, the $100 (1 - α) %$ confidence regions for μ_d is given by

C_{μ_{d}}^{P} (α) = \{μ_{d} : \frac{n_{1} + n_{2} - p - 1}{p (n_{1} + n_{2} - 2)} T^{2} < C S F_{p, n_{1} + n_{2} - p - 1}^{2} (1 - α)\},

where

C S F_{p, n_{1} + n_{2} - p - 1}^{2} (1 - α)

represents the 1 − αquantile of

C S F_{p, n_{1} + n_{2} - p - 1}^{2} (0, 0, δ_{2}, Ω)

The following plots present the pdf of noncentral closed skew F-distribution (see Figure 2).

Figure 2

View large Download slide

The pdfs of noncentral closed skew $F_{2, n} (5, {(5,11)}^{T}, (\begin{matrix} 5 & 11 \\ 11 & 25 \end{matrix}), (\begin{matrix} 1 & 0.5 \\ 0.5 & 1 \end{matrix}))$ distributions with n = 5, 10, 20 and 50 (black, green, red and blue)

4. Simulation study

Simulations are conducted for evaluating the performance of the proposed confidence regions for μ_d under independent multivariate SN settings using the coverage relative frequency rates. Comparisons of proposed confidence regions with those in classical independent multivariate normal distributions are given.

4.1 Coverage frequencies

To evaluate the proposed confidence regions for difference of location parameters under multivariate SN setting, Monte Carlo simulation studies (each with a number of simulation runs M = 10,000) are conducted for combinations of various values of sample sizes (n₁, n₂) = (20,25), (40, 50) and (80,100), (ρ₁, ρ₂) = ({0.1, 0.5, 0.8}, {0.1, 0.5, 0.8}), $D_{1} = (\begin{matrix} 1 & 3 \\ 2 & - 5 \end{matrix})$ and $D_{2} = (\begin{matrix} - 2 & 2 \\ 3 & 0 \end{matrix})$ ⁠.

Table 1 shows that our method provides reliable inference about the difference of location parameters with nominal confidence level (95%). To further illustrate the effectiveness of the proposed method, we confidence intervals of the coverage probability presented in the following plot.

Table 1

Coverage relative frequencies of confidence regions at confidence level α = 95% for difference of location parameters μ_d with various combinations of sample sizes, ρ₁, ρ₂ and D₁, D₂ using Hotelling's T² as the pivotal quantity when Σ₁ and Σ₂ are equal but unknown

$(n_{1}, n_{2})$	$Σ = (\begin{matrix} 1 & ρ \\ ρ & 1 \end{matrix})$	$D_{1} = (\begin{matrix} 1 & 3 \\ 2 & - 5 \end{matrix})$		$D_{2} = (\begin{matrix} - 2 & 2 \\ 3 & 0 \end{matrix})$

		F-distribution	CSF-distribution	F-distribution	CSF-distribution
$(20,25)$	ρ = 0.2	0.9268	0.9495	0.9224	0.9524
	ρ = 0.5	0.9274	0.9511	0.9135	0.9502
	ρ = 0.8	0.9206	0.9504	0.9227	0.9520
$(40,45)$	ρ = 0.2	0.9025	0.9479	0.9008	0.9534
	ρ = 0.5	0.9118	0.9451	0.9095	0.9509
	ρ = 0.8	0.9036	0.9521	0.9016	0.9524
$(80,100)$	ρ = 0.2	0.8932	0.9527	0.9080	0.9479
	ρ = 0.5	0.8917	0.9463	0.8702	0.9470
	ρ = 0.8	0.9035	0.9483	0.8933	0.9448
$(200,250)$	ρ = 0.2	0.8824	0.9503	0.8883	0.9461
	ρ = 0.5	0.88920	0.9533	0.8956	0.9559
	ρ = 0.8	0.8998	0.9492	0.8852	0.9546

$(n_{1}, n_{2})$	$Σ = (\begin{matrix} 1 & ρ \\ ρ & 1 \end{matrix})$	$D_{1} = (\begin{matrix} 1 & 3 \\ 2 & - 5 \end{matrix})$		$D_{2} = (\begin{matrix} - 2 & 2 \\ 3 & 0 \end{matrix})$

		F-distribution	CSF-distribution	F-distribution	CSF-distribution
$(20,25)$	ρ = 0.2	0.9268	0.9495	0.9224	0.9524
	ρ = 0.5	0.9274	0.9511	0.9135	0.9502
	ρ = 0.8	0.9206	0.9504	0.9227	0.9520
$(40,45)$	ρ = 0.2	0.9025	0.9479	0.9008	0.9534
	ρ = 0.5	0.9118	0.9451	0.9095	0.9509
	ρ = 0.8	0.9036	0.9521	0.9016	0.9524
$(80,100)$	ρ = 0.2	0.8932	0.9527	0.9080	0.9479
	ρ = 0.5	0.8917	0.9463	0.8702	0.9470
	ρ = 0.8	0.9035	0.9483	0.8933	0.9448
$(200,250)$	ρ = 0.2	0.8824	0.9503	0.8883	0.9461
	ρ = 0.5	0.88920	0.9533	0.8956	0.9559
	ρ = 0.8	0.8998	0.9492	0.8852	0.9546

From Figure 3, we can see clearly that the pivotal quantity-based closed skew F-distribution produce more robust confidence region than that based on F-distribution. The coverage relative frequencies, based on the SN model, are close to the nominal confidence level 95% consistently for the combination of different sample sizes, scale parameter and skewness parameters. But the coverage relative frequencies, based on the normal model, are lower than the nominal confidence level.

Figure 3

View large Download slide

The confidence intervals of coverage relative frequencies at confidence level α = 0.95, red ones based on SN model and blue ones based on normal model with sample size (n₁, n₂) = (20, 25), (40, 50), (80, 100) and (200, 250) (from left to right in each figure), ρ = 0.2, 0.5 and 0.8 (in each row), and D₁ and D₂ (from left to right), respectively

5. Real data example

In this section, we illustrate the effectiveness and applicability of the proposed methods by applying them to Australian Institute of Sport (AIS) data (Cook and Weisberg, 2009). We explore the difference of body mass index (BMI) and lean body mass (LMB) between males and females athletes in AIS data.

The point estimates of the parameters for AIS data are reported in Table 2.

Table 2

Point estimates of SN parameters for the males and females AIS data, respectively

	Males	Females
${\hat{μ}}^{'}$	$(24.47, 79.55)$	$(23.18, 61.42)$
$\hat{Σ}$	$(\begin{matrix} 4.81 & 18.95 \\ 18.95 & 114.00 \end{matrix})$	$(\begin{matrix} 8.32 & 21.30 \\ 21.30 & 89.96 \end{matrix})$
${\hat{λ}}^{'}$	$(- 0.20, - 0.80)$	$(0.88, - 2.63)$

In Figure 4, the scatter plots and contour plots of fitted bivariate SN distributions are presented. Based on our previous work (Azzalini and Valle, 1996), this data set prefers multivariate SN model. So we adopt multivariate SN model as well to explore the difference of location parameters. Using point estimates listed in Table 2 and applying Theorem 2.1, the differences of sample mean has closed SN distribution, ${\bar{X}}_{d} \sim C S N_{2,2} (μ, Σ, D, Δ)$ with estimated parameters

\hat{μ} = {(1.29, 18.13)}^{'}, \hat{Σ} = (\begin{matrix} 0.13 & 0.41 \\ 0.41 & 2.09 \end{matrix})

\hat{D} = (\begin{matrix} 3.20 & - 4.83 \\ - 29.80 & 16.82 \end{matrix}), \hat{Δ} = (\begin{matrix} 28.90 & 101.01 \\ 101.01 & 473.42 \end{matrix}) .

Figure 4

View large Download slide

The scatter plots and contour plots for the AIS data of males and females

Then, we apply Theorem 3.2 to construct the confidence region of difference of location parameters μ_d. In Figure 5, the confidence regions for the difference of location parameters are given below at 95% confidence level.

Figure 5

View large Download slide

The confidence region of difference of location parameter μ_d at confidence level 95% for AIS data between males and females, where the black triangle represents the ${\hat{μ}}_{d}$ , and the blue dot represents the $μ_{d}^{0}$

6. Conclusion

In this work, the difference for location parameters between two independent samples under multivariate SN setting is studied. The construction of confidence region procedure is developed. From the results of simulation studies, the confidence region based on SN model has better performance than normal model in term of relative coverage frequencies to capture the true value when the data are generated from skewed distribution.

The authors would like to thank reviewers for their valuable suggestions and comments, which improved the manuscript.

References

Adcock

,

C.

and

Azzalini

,

A.

(

2020

), “

A selective overview of skew-elliptical and related distributions and of their applications

”,

Symmetry

, Vol.

12

No.

1

, p.

118

.

Google Scholar

Crossref

Arellano-Valle

,

R.

,

Bolfarine

,

H.

and

Lachos

,

V.

(

2005

), “

Skew-normal linear mixed models

”,

Journal of Data Science

, Vol.

3

No.

4

, pp.

415

-

438

.

Google Scholar

Crossref

Azzalini

,

A.

(

1985

), “

A class of distributions which included the normal ones

”,

Scandinavian Journal of Statistics

, Vol.

12

No.

2

, pp.

171

-

178

.

Google Scholar

Azzalini

,

A.

and

Capitanio

,

A.

(

1999

), “

Statistical applications of the multivariate skew normal distribution

”,

Journal of the Royal Statistical Society. Series B (Statistical Methodology)

, Vol.

61

No.

3

, pp.

579

-

602

.

Google Scholar

Crossref

Azzalini

,

A.

and

Valle

,

A.D.

(

1996

), “

The multivariate skew-normal distribution

”,

Biometrika

, Vol.

83

No.

4

, pp.

715

-

726

.

Google Scholar

Crossref

Carmichael

,

B.

and

Coën

,

A.

(

2013

), “

Asset pricing with skewed-normal return

”,

Finance Research Letters

, Vol.

10

No.

2

, pp.

50

-

57

.

Google Scholar

Crossref

Chen

,

J.T.

and

Gupta

,

A.K.

(

2005

), “

Matrix variate skew normal distributions

”,

Statistics

, Vol.

39

No.

3

, pp.

247

-

253

.

Google Scholar

Crossref

Cook

,

R.D.

and

Weisberg

,

S.

(

2009

),

An Introduction to Regression Graphics

,

John Wiley & Sons

,

New York

, Vol.

405

.

Google Scholar

Gonzalez-Farias

,

G.

,

Dominguez-Molina

,

A.

and

Gupta

,

A.K.

(

2004

), “

Additive properties of skew normal random vectors

”,

Journal of Statistical Planning and Inference

, Vol.

126

No.

2

, pp.

521

-

534

.

Google Scholar

Crossref

Li

,

B.

,

Tian

,

W.

and

Wang

,

T.

(

2018

), “

Remarks for the singular multivariate skew-normal distribution and its quadratic forms

”,

Statistics and Probability Letters

, Vol.

137

, pp.

105

-

112

.

Google Scholar

Crossref

Ma

,

Z.

,

Chen

,

Y.-J.

,

Wang

,

T.

and

Peng

,

W.

(

2019

), “The inference on the location parameters under multivariate skew normal settings”, in

Kreinovich

,

V.

,

Trung

,

N.

and

Thach

,

N.

(Eds),

Beyond Traditional Probabilistic Methods in Economics

,

Springer Nature

, pp.

146

-

162

.

Google Scholar

Crossref

Ma

,

Z.

,

Zhu

,

X.

,

Wang

,

T.

and

Autchariyapanitkul

,

K.

(

2018

), “

Joint plausibility regions for parameters of skew normal family

”, in

Krennovich

,

V.

,

Sriboonchitta

,

S.

and

Chakpitak

,

N.

(Eds),

Predictive Econometrics and Big Data

,

Springer-Verlag

,

New York

, pp.

233

-

245

.

Google Scholar

Crossref

Mardia

,

K.V.

,

Kent

,

J.T.

and

Bibby

,

J.M.

(

1980

),

Multivariate Analysis (Probability and Mathematical Statistics)

.

Google Scholar

Wang

,

T.

,

Li

,

B.

and

Gupta

,

A.K.

(

2009

), “

Distribution of quadratic forms under skew normal settings

”,

Journal of Multivariate Analysis

, Vol.

100

No.

3

, pp.

533

-

545

.

Google Scholar

Crossref

Wei

,

Z.

,

Zhu

,

X.

and

Wang

,

T.

(

2021

), “

The extended skew-normal-based stochastic frontier model with a solution to ‘wrong skewness’ problem

”,

Statistics

, pp.

1

-

20

, doi:

https://doi.org/10.1080/02331888.2021.2004142

.

Google Scholar

Ye

,

R.

,

Wang

,

T.

and

Gupta

,

A.K.

(

2014

), “

Distribution of matrix quadratic forms under skew-normal settings

”,

Journal of Multivariate Analysis

, Vol.

131

, pp.

229

-

239

, 00010.

Google Scholar

Crossref

Young

,

P.D.

,

Harvill

,

J.L.

and

Young

,

D.M.

(

2016

), “

A derivation of the multivariate singular skew-normal density function

”,

Statistics and Probability Letters

, Vol.

117

, pp.

40

-

45

.

Google Scholar

Crossref

Zhu

,

X.

,

Li

,

B.

,

Wu

,

M.

and

Wang

,

T.

(

2018

), “

Plausibility regions on parameters of the skew normal distribution based on inferential models

”, in

Krennovich

,

V.

,

Sriboonchitta

,

S.

and

Chakpitak

,

N.

(Eds),

Predictive Econometrics and Big Data

,

Springer-Verlag

,

New York

, pp.

287

-

302

.

Google Scholar

Zhu

,

X.

,

Li

,

B.

,

Wang

,

T.

and

Gupta

,

A.K.

(

2019

), “

Sampling distributions of skew normal populations associated with closed skew normal distributions

”,

Random Operators and Stochastic Equations

, Vol.

27

No.

2

, pp.

75

-

87

.

Google Scholar

Crossref

Zhu

,

X.

,

Ma

,

Z.

,

Wang

,

T.

and

Teetranont

,

T.

(

2017

), “Plausibility regions on the skewness parameter of skew normal distributions based on inferential models”, in

Krennovich

,

V.

,

Sriboonchitta

,

S.

and

Huynh

,

V.

(Eds),

Robustness in Econometrics

,

Springer-Verlag

,

New York

, pp.

267

-

286

.

Google Scholar

Crossref

2022

Ziwei Ma, Tonghui Wang, Zheng Wei and Xiaonan Zhu

Published in Asian Journal of Economics and Banking. Published by Emerald Publishing Limited. This article is published under the Creative Commons Attribution (CC BY 4.0) licence. Anyone may reproduce, distribute, translate and create derivative works of this article (for both commercial and non-commercial purposes), subject to full attribution to the original publication and authors. The full terms of this licence may be seen at http://creativecommons.org/licences/by/4.0/legalcode

Inferences on location parameters based on independent multivariate skew normal distributions

1. Introduction

2. Matrix variate SN distributions and sampling distributions

3. Inference on difference of location parameters

3.1 Some related distributions

3.2 Confidence region of μ_d

4. Simulation study

4.1 Coverage frequencies

5. Real data example

6. Conclusion

References

New and popular articles

Email Alerts

Cited By

Inferences on location parameters based on independent multivariate skew normal distributions

1. Introduction

2. Matrix variate SN distributions and sampling distributions

3. Inference on difference of location parameters

3.1 Some related distributions

3.2 Confidence region of μd

4. Simulation study

4.1 Coverage frequencies

5. Real data example

6. Conclusion

References

New and popular articles

Email Alerts

Related Chapters

Recommended for you

Cited By

Sharing Unavailable

3.2 Confidence region of μ_d