Upload
others
View
14
Download
0
Embed Size (px)
Citation preview
Applied Mathematical Sciences, Vol. 11, 2017, no. 55, 2727 - 2766
HIKARI Ltd, www.m-hikari.com
https://doi.org/10.12988/ams.2017.79290
Comparing MINQUE and IAUE
Estimates of Variance Components
of a Random Coefficient Model
Souha K. Badr
Department of Statistics
Faculty of Science, AL-Fisaliah
King Abdulaziz University, Jeddah, Saudi Arabia
Ahmed H. Youssef
Department of Statistics
Faculty of Science
King Abdulaziz University, Jeddah, Saudi Arabia
Hanaa H. Abu-zinadah
Department of Statistics
Faculty of Science, AL-Fisaliah
King Abdulaziz University, Jeddah, Saudi Arabia
Copyright © 2017 Souha K. Badr et al. This article is distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium,
provided the original work is properly cited.
Abstract
An alternative way to estimate the coefficient variance in the Swamy’s RCR model
has been derived using Minimum Norm Quadratic Estimation (MINQUE), and the
Iteration Almost Unbiased Estimator (IAUE) methods. The estimators’
performance in the RCR model are examined in Monte Carlo study. The Monte
Carlo study provides some insight into how well the RCR model performs in small,
medium, and large samples in the case of random, mixed, and fixed coefficient
regression. We found that using MINQUE method to estimate the coefficient
variance has reduce the probability of having negative variance comparing with the
Swamy method. IAUE method was superior, since it gives zero percent of negative
2728 Souha K. Badr et al.
variance and has a low variation and a low bias in estimation coefficient parameters,
even in case of fixed coefficients.
Keywords: Mixed Model, Panel Data, Pooling Cross-section, Random Coefficient
Regression Model, Time Series Data, Variance Component Estimation, MINIQUE,
IAUE
1 Introduction
Statistical models can be characterized according to the type of data to which they
are applied. The field of survey statistics usually deals with cross-sectional data
describing each of many different individuals or units at a single point in time.
Econometrics commonly uses time series data describing a single entity. Most
researchers use methods of analysis devolved for either cross-sectional or time
series data.
There are varieties of statistical models that can be used to analyze above the cross-
sectional time-series data. Dielman (1989) gives a comprehensive review of the
statistical literature dealing with these models. One of these models is the Random
Coefficient Regression model RCR.
The random coefficient regression model (RCR) was originally proposed by C. R.
Rao (1965) and later extended by Swamy (1970, 1971, 1974). Swamy (1971) allows
for random variation in population regression coefficients over cross-sectional
units, and coefficients are viewed as fixed over time. Swamy treats both intercept
and slope as random variables that are distributed across units with the same mean
and the same variance-covariance matrix.
Negative values for estimated variances can arise in a random coefficient regression
model context, because of some parameter estimating methods. Wu (1992)
maintained that the reason for negative variance components lies in the use of the
estimation method itself, and once negative variance occurs, some other methods
should be used instead of the unsuccessful one.
Minimum norm quadratic unbiased estimator (MINQUE) is proposed as a method
of variance estimation in a series of papers starting with Rao (1970). The basic idea
for this method is to find unbiased quadratic estimators that are invariant, and to
minimize some matrix norms.
One of the major disadvantages associated with the MINQUE estimator is that
MINQUE sometimes produces negative variance components. Horn et al. (1975)
proposed an estimator that avoids the deficiencies of the MINQUE method called
the Almost Unbiased Estimator (AUE). Schaffrin (1983), and Lucas (1985)
completed the work of Horn and introduced the Iterated Almost Unbiased Estimator
(IAUE). This estimator, as Horn stated, overcomes the problem of the negative
variance and guarantees a positive value for the estimation.
In this article, we will apply these two methods of variance component estimation
on Swamy’s RCR model. A Monte Carlo simulation study is conducted to compare
the efficiency of Swamy’s RCR model when the original method of estimating
Comparing MINQUE and IAUE estimates of variance components 2729
variance is used, and also when other methods are used for the same model. The
results of the simulation are analyzed, discussed, and presented in the last part in
this article.
2 Random Coefficient Regression Model
Random Coefficient Regression Model (RCR), have a long history. Pioneering
work of Rubin, Klein, Wald and Theil in the late 1940s and early 1950s had little
practical impact and was ignored for some time. More comprehensive papers,
oriented toward practical applications, were written in the late 1960s by Rao, Fisk,
Hildreth and Houk, and Swamy. The pre-1970 literature is reviewed almost
completely by Swamy, a substantial body of theory was developed, and a number
of useful review papers appeared. The (RCR) model proposed by Swamy (1970)
has appeared in different literature and be used in a number of applications
including pooled cross-sectional and time series data. For example, Bones &
Frankfurter (1977), Mehta et al. (1978), Delicado and Romo (1999), Bhaum and
Gibbons (2001),Hobza and Morales (2011), and Cartwright and Riabko (2015).
Swamy Type Estimator
The RCR model applies to a set of N cross-sectional as a model
𝑦𝑖 = 𝑋𝑖𝛽𝑖 +∈𝑖 (1)
Where each 𝑦𝑖 represents the Tx1 vector of observations from the 𝑖𝑡ℎ cross-section,
for i=1,…N, each 𝑋𝑖 represents the TxK matrix of independent variables, 𝛽𝑖 is a
vector of unknown random parameters, and ∈i the Tx1 vector of random error
terms.
Swamy rewrites the coefficient 𝛽𝑖 as
𝛽𝑖 = �� + 𝑣𝑖 (2)
Where �� is a Kx1 vector of a fixed component, and the 𝑣𝑖 is a Kx1 vector of random
variables satisfy:
I. 𝐸(𝑣𝑖) = 0 II. 𝐸(𝑣𝑖𝑥𝑖𝑡
′ ) = 0
III. 𝐸(𝑣𝑖𝑣𝑗′) = {
∆, 𝑖𝑓 𝑖 = 𝑗0, 𝑖𝑓 𝑖 ≠ 𝑗
}
Some other assumptions on Swamy’s model are as following:
1- The number of cross-sections and the sample size of each cross- section
must be greater than the number of parameters needed to be estimated i.e
𝑁 > 𝐾 & 𝑇 > 𝐾.
2- The independent variables are non-stochastic in the sense that 𝑋𝑖 is fixed in
repeated samples on 𝑌𝑖. The rank of 𝑋𝑖 is K.
3- The ∈𝑖 is independently and identically distributed with
𝐸(∈𝑖)0 𝑎𝑛𝑑 𝐸(∈𝑖∈𝑖′) = 𝜎𝑖
2𝐼𝑇
4- The coefficient vectors 𝛽𝑖 are independently and identically distributed with
𝐸(𝛽𝑖) = �� 𝑎𝑛𝑑 𝐸(𝛽𝑖 − ��)(𝛽𝑖 − ��)′= ∆
5- The ∈𝑖and 𝛽𝑗 are independent for every i and j.
2730 Souha K. Badr et al.
Using the equations in (1) and (2), the model can be written as
𝑦𝑖 = 𝑋𝑖(�� + 𝑣𝑖) +∈𝑖= 𝑋𝑖�� + 𝑒𝑖 (3)
where 𝑒𝑖 = 𝑋𝑖𝑣𝑖 +∈𝑖. The (N) equations can be re-written as;
𝑌 = 𝑋�� + 𝑒 (4)
Where 𝑌(𝑁𝑇×1) = [
𝑌1
𝑌2
⋮𝑌𝑁
] , 𝑌𝑖(𝑇×1)= [
𝑦𝑖1
𝑦𝑖2
⋮𝑦𝑖𝑇
], 𝑋𝑁𝑇×𝑁𝑘 = [
𝑋1 0 … … 00 𝑋2 … … 0⋮0
⋮0
⋮0
⋮…
⋮𝑋𝑁
],
��(𝑘×1) =
[ ��0
��1
⋮⋮
��𝑘]
, 𝑒(𝑁𝑇×1) = [
𝑒1
𝑒2
⋮𝑒𝑁
] , 𝑒𝑖(𝑇×1)= [
𝑒𝑖1
𝑒𝑖2
⋮𝑒𝑖𝑇
]
The error vector, e, is normally distributed with zero mean and variance-covariance
matrix Ω, given by:
𝐸(𝑒𝑒′) = 𝛺𝑁𝑇×𝑁𝐾 =
[ 𝑋1∆𝑋1
′ + 𝜎12𝐼𝑇 0 … 0
0 𝑋2∆𝑋2′ + 𝜎2
2𝐼𝑇 … 0⋮0
⋮0
⋮…
⋮𝑋𝑁∆𝑋𝑁
′ + 𝜎𝑁2𝐼𝑇]
Where the zeros are all TxT null matrices and ∆ is the variance-covariance matrix
of 𝛽𝑖. If Δ and 𝜎2 are known, the best linear unbiased estimator of �� is the
Generalized Least Squares (GLS) estimator
�� = (𝑋′𝛺−1𝑋)−1𝑋′𝛺−1𝑌 (5)
If Δ and 𝜎2 are unknown, which is typically the case, we should first estimate Δ
and 𝜎2, then we estimate �� by substituting the estimated Δ and 𝜎2 into (5) to get
the Feasible Generalized Least Squares (FGLS) estimator.
�� = (𝑋′��−1𝑋)−1
𝑋′��−1𝑌 (6)
Swamy (1970) suggests the following unbiased and consistent estimators:
��𝑖2 =
∈𝑖′∈𝑖
𝑇−𝐾 (7)
And
∆=𝑆��𝑖
𝑁−1−
1
𝑁∑ ��𝑖
2(𝑋𝑖′𝑋𝑖)−1𝑁
𝑖=1 (8)
Where
𝑆��𝑖= ∑ ��𝑖��𝑖′
𝑁𝑖=1 −
1
𝑁∑ ��𝑖
𝑁𝑖=1 ∑ ��𝑖′
𝑁𝑖=1 (9)
Comparing MINQUE and IAUE estimates of variance components 2731
Also, Swamy shows that the FGLS estimator given in (6) is equivalent to
�� = {∑[∆ + ��𝑖2(𝑋𝑖′𝑋𝑖)
−1]−1
𝑁
𝑖=1
}
−1
∑[∆ + ��𝑖2(𝑋𝑖′𝑋𝑖)
−1]−1
𝑁
𝑖=1
��𝑖 (10)
Where ��𝑖 = (𝑋𝑖′𝑋𝑖)−1𝑋𝑖′𝑌𝑖 is the Ordinary Least Square (OLS) estimator of 𝛽𝑖.
Swamy showed that his estimator β is consistent as N → ∞ and T → ∞, and
asymptotically efficient as T → ∞ under certain conditions. The performance of β
in small samples was studied in Dielman (1992).The distribution of the coefficient
vector is invariant to translations along the time axis, and ∆ & ��2 are unbiased
estimators for the variance-covariance matrix ∆ and 𝜎2.
Issues with Swamy’s estimators
One particular problem with Swamy’s estimators is that the estimators of ∆,
equation (8), that he suggests may, at times, yield negative estimates of variances.
This is so because Swamy suggests an estimator for ∆ that is the difference of two
matrices. Griffiths in (1971) said that the greatest disadvantage with Swamy’s
estimator is the frequency with which it gives negative estimates. Not only because
negative values are meaningless, but because if retained they can lead to GLS
estimators of the 𝛽𝑖’s which perform extremely poorly in terms of MSE.
The gains that we earn in terms of quality of the estimate of �� justify additional
complexity in estimating ∆. There are some other alternative estimators that suggest
overcoming this problem. One of them is:
∆1= {∆ 𝑖𝑓 𝑎𝑙𝑙 𝑑𝑖𝑎𝑔𝑜𝑛𝑎𝑙 𝑒𝑙𝑒𝑚𝑒𝑛𝑡𝑠 𝑎𝑟𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒
∆𝐴 𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝑑𝑖𝑎𝑔𝑜𝑛𝑎𝑙 𝑒𝑙𝑒𝑚𝑒𝑛𝑡𝑠 𝑟𝑒𝑝𝑙𝑎𝑐𝑒𝑑 𝑏𝑦 𝑧𝑒𝑟𝑜 𝑖𝑓 𝑡ℎ𝑒𝑦 𝑎𝑟𝑒 𝑝𝑟𝑒𝑠𝑒𝑛𝑡𝑠
(11)
Griffiths (1979) mentioned that changing negative estimates to zero is not
completely satisfactory, since this implies that the corresponding coefficients are
no longer random.
3 The Minimum Norm Quadratic Estimation
The method of minimum norm quadratic estimation (MINQUE) was introduced by
Rao (1971) for regression models with heteroscedastic variances. Later, a series of
papers was published (1971, 1972, 1973) to generalize the method for variance and
covariance component models and to compare MINQUE and modified MINQUE
estimators of heteroscedastic variances with the usual sample variances in the case
of replicated data. The basic idea for this method is to find unbiased quadratic
estimators that are invariant and to minimize some matrix norms.
The principle of MINQUE estimation state:
The quadratic form 𝑌′𝐴𝑌 is said to be the MINQUE (minimum norm quadratic
unbiased estimator) of ∑ 𝜌𝑖𝜎𝑖2
𝑖 if the matrix A is determined such that
‖(𝑈′𝐴𝑈 − ∆)‖ is a minimum subject to the conditions
2732 Souha K. Badr et al.
𝐴𝑋 = 0
𝑡𝑟𝐴𝑉𝑖 = 𝜌𝑖 , 𝑖 = 1,2, … . 𝑘 (12)
MINQUE for RCR Model
Recall the random regression model equation (4),
𝑌 = 𝑋�� + 𝑒 (13)
Where Y represents NTx1 vector of observations, X represents NTxNK matrix of
independent variables, �� is a vector of unknown fixed parameters, and 𝑒 the NTx1
vector of random error terms.
From equation (13) we have that the i-th cross-section unit
𝑦𝑖 = 𝑋𝑖�� + 𝑋𝑖𝑣𝑖 + 𝜖𝑖 𝑖 = 1,2, … ,𝑁 (14)
where 𝑒𝑖 = 𝑋𝑖𝑣𝑖 + 𝜖𝑖 and the variables are defined as the same definitions and assumptions as in the
previous section.
from (13), and (14) we can rewrite the model as
𝑌 = 𝑋𝛽 + 𝑈1𝜉1 + 𝑈2𝜉2 + ⋯+ 𝑈𝑘𝜉𝑘 + 𝑈𝑘+1𝜉𝑘+1+∈ (15)
where
𝑈𝑖is 𝑁𝑇 × 𝑁𝑇 diagonal matrix whose corresponding diagonal elements of 𝑈𝑖 = 𝑋𝑖,𝑓𝑜𝑟 𝑖 = 1,2, …𝑘.
𝜉𝑖 is 𝐾 × 1 random vector has a multivariate normal distribution with mean �� and
covariance matrix ∆. Further, it is assumed that 𝜉𝑖 and 𝜉𝑗 𝑓𝑜𝑟 𝑖 ≠ 𝑗 are
uncorrelated.
Equation (15) can be expressed in a compact form as:
𝑌 = 𝑋𝛽 + 𝑈𝜉 (16)
Thus, generally we have 𝐸(𝑌) = 𝑋𝛽 and 𝐷(𝑌) = 𝛺 = ∑ ∆𝑖𝑉𝑖 + 𝜎𝑗2𝐼𝑇
𝑘𝑖=1 , 𝑓𝑜𝑟 𝑗 =
1, … . 𝑁,where 𝑉𝑖 = 𝑈𝑖𝑈𝑖′, D is called the dispersion matrix and the parameters
∆0, ∆1, … . , ∆𝑘are the unknown variance components whose values should be
estimated.
The quadratic form 𝑌′𝐴𝑌, in the observations, Y is said to be a MINOUE of ∑ 𝜌𝑖∆𝑖𝑖
if the symmetric matrix A is selected such that ‖(𝑈′𝐴𝑈 − ∆∗)‖ is a minimum
subject to the conditions
𝐴𝑋 = 0 and 𝑡𝑟𝐴𝑉𝑖 = 𝜌𝑖
For making the optimization easier, the squared Euclidean norm will be utilized.
Then we get
‖(𝑈′𝐴𝑈 − ∆∗)‖2 = 𝑡𝑟((𝑈′𝐴𝑈 − ∆∗)′(𝑈′𝐴𝑈 − ∆∗)) = 𝑡𝑟(𝐴𝑉𝐴𝑉) + ∆∗∗ (17)
where 𝑉 = ∑ 𝑉𝑖𝑘𝑖=1 = 𝑈𝑈′ and ∆∗∗ refer to a constant quantity and do not involve
A, so the second term is dropped.
Let A be a symmetric matrix and V be a symmetric and invertible matrix, then the
minimum 𝑡𝑟(𝐴𝑉𝐴𝑉), subject to invariant and unbiasedness criteria, is attained at:
𝐴 = ∑ 𝑎𝑖𝑅𝑉𝑖𝑅𝑘𝑖=1 (18)
Comparing MINQUE and IAUE estimates of variance components 2733
where
𝑎 = 𝑆−1𝜌
𝑆𝑖,𝑗 = 𝑡𝑟(𝑄′𝑉−1𝑉𝑖𝑉−1𝑄𝑉𝑗), 𝑓𝑜𝑟 𝑖 𝑎𝑛𝑑 𝑗 = 1,… , 𝑘
𝑄 = 𝐼𝑁𝑇 − 𝑋(𝑋′𝑉−1𝑋)−1𝑋′𝑉−1
𝑅 = 𝑄′𝑉−1.
Consequently, the MINQUE of 𝜌′∆ is
𝑌′𝐴𝑌 = ∑ 𝑎𝑖𝑌′𝑅𝑉𝑖𝑅
𝑘𝑖=1 𝑌 = ∑ 𝑎𝑖𝑏𝑖
∗𝑘𝑖=1 = 𝑎′𝑏∗ = 𝜌′𝑆−1𝑏∗ (19)
where
𝑏∗ = 𝑌′𝑅𝑉𝑖𝑅𝑌.
By equating (4.26) with𝜌′∆ , we get:
∆𝑀𝐼𝑁𝑄𝑈𝐸1= 𝑆−1𝑏∗ (20)
Hsiao (1972, 1974) has proposed another way to find The MINQUE estimator for
RCR model by finding The MINQUE estimator for each cross-section as follows:
Let
𝜃𝑖 = ∑ 𝑋𝑖𝑗∆𝑖𝑋𝑖𝑗′ + 𝜎2𝐼𝑇
𝐾𝑗=1 = ��𝑖𝛤 𝑓𝑜𝑟 𝑖 = 1,… ,𝑁. (21)
where
��𝑖 is 𝑇 × 𝐾 matrix whose elements are squares of the corresponding elements of
𝑍𝑖 = 𝑋𝑖, and 𝛤 is the 𝐾 × 1 column vector with elements ∆1, … , ∆𝑘, so, the
MINQUE of ∑𝜌′𝜃𝑖is the quadratic form 𝑌𝑖′𝐴𝑌𝑖, such that the 𝑡𝑟 𝐴2 is minimized
subject to the conditions:
𝐴𝑋𝑖 = 0 (22)
∑ 𝜌𝑖𝜃𝑖𝑁𝑖=1 = 𝜌′��𝑖𝛤 (23)
where
𝜌′ = (𝜌1, … , 𝜌𝑁)′ following the computational procedure suggested by Rao, we have that:
𝑌𝑖′𝐴𝑌𝑖 = 𝜌′��𝑖(��𝑖
′��𝑖��𝑖)−1��𝑖
′��𝑖 (24)
where
𝑀𝑖 = 𝐼𝑇 − 𝑍𝑖(𝑍𝑖′𝑍𝑖)
−1𝑍𝑖′
𝑟𝑖 = 𝑌𝑖 − 𝑍𝑖��𝑖
��𝑖 is the OLS estimator for each cross-section
and ��𝑖, ��𝑖 denotes the corresponding matrix and vector obtained by squaring each
element in 𝑀𝑖 , 𝑟𝑖 respectively.
Equating (24) with 𝜌′��𝑖𝛤, the unbiased estimator of 𝛤𝑖 is
𝛤𝑖∗ = (��𝑖
′��𝑖��𝑖)−1
��𝑖′��𝑖. (25)
After applying equation (25) for each cross-section unit, we can obtain the
consistent estimator 𝛤 by taking the simple average as mentioned by Hsiao (1975).
2734 Souha K. Badr et al.
4 Iteration Almost Unbiased Estimation
Three major disadvantages are associated with the MINQUE estimator:
1. MINQUE sometimes produces a negative variance component.
2. The S matrix, equation (20), is not always nonsingular.
3. MINQUE estimator needs a heavy calculation, which is time consuming.
Horn et al. (1975) proposed an estimator that avoids the deficiencies of the
MINQUE method. As Horn et al. stated, it overcomes the problem of the negative
variance and guarantees a positive value for the estimation, and exists under more
general conditions. It also needs less cost in computer time and storage than the
MINQUE estimator. They call the estimator, Almost Unbiased Estimator (AUE).
Since this estimator is proportional between a prior estimate and a true variance,
the bias introduced with the estimation is because of the failure to this proportional
factor to reach a unity, but this failure can be expected to be small. Hence, the
appellation “almost unbiased”.
Schaffrin (1983), and Lucas (1985) completed the work of Horn and introduced the
Iterated Almost Unbiased Estimator (IAUE). Like AUE, the estimation of IAUE
focuses on obtaining estimations for variance factors, which will approach unity as
prior estimates approach the true variance.
Hsu (2001) showed that the Helmert method is identical with the IAUE method. It
can easily be shown that a variance component obtained from the IAUE is always
positive provided that the resulting covariance matrix is positive definite. Hsu
(1999) indicates that it is possible for IAUE to produce some variance factors that
deviate significantly from unity due to inappropriate grouping. Egeltoft (1992),
Fotopoulos et al. (2005), Bahr et al. (2007), Amiri-Simkooei (2007), and Kall et al.
(2014) compare the former methods, which produce unbiased and invariant
estimates, but are very time-consuming, require a lot of computational resources,
and sometimes produce negative estimates, and the IAUE method, which although
is not always unbiased, does not produce negative variance component estimates,
demands fewer computational resources, and converges much more quickly than
other approaches.
IAUE for RCR Model
The working equation of IAUE as mentioned by Lucas (1985), and Hsu (1998) is
𝑓𝑖 =𝑌′𝑅∗𝑉𝑖𝑅
∗𝑌
𝑡𝑟(𝑉𝑖𝑅∗)
(26)
where
Y, and 𝑉𝑖 are as defined above.
𝑅∗ = 𝐷∗−1 − 𝐷∗−1𝑋(𝑋′𝐷∗−1𝑋)−1𝑋′𝐷∗−1
𝐷∗ = ∑ 𝜏𝑖𝑉𝑖𝐾𝑖=1 , 𝜏𝑖 is a prior estimate of ∆.
Computed 𝑓𝑖 based on 𝜏𝑖 will be updated based on the new value of 𝜏𝑖 until 𝑓𝑖′𝑠
approach one. Then ∆𝐼𝐴𝑈𝐸 can be expressed as:
∆𝐼𝐴𝑈𝐸= ∏ 𝑓𝑖𝑚−1𝑖=1 (27)
where m is the number of iterations.
Comparing MINQUE and IAUE estimates of variance components 2735
5 Simulation Study for Swamy’s RCR Estimators
A Monte Carlo simulation trailer was conducted 10,000 times to investigate the
efficiency of the estimation of the RCR model with two parameters 𝛽0 𝑎𝑛𝑑 𝛽1. The
model under study is given by:
𝑦𝑖𝑡 = 𝛽0𝑖 + 𝛽1𝑖𝑥𝑖𝑡 +∈𝑖𝑡 (28)
for 𝑖 = 1,2…… ,𝑁 ; 𝑡 = 1,2, ……𝑇.
To perform the simulation, the model in equation (28) was generated as follows:
1- The value of the independent variable, 𝑥𝑖𝑡, was generated as independent
normally distributed random variables with mean 𝜇𝑥, set equal to five, and
variance𝜎𝑥2, set equal to three.
2- The error term, ∈𝑖𝑡, were generated as independent normally distributed
random variables, independent of the 𝑥𝑖𝑡 values, with mean set equal to zero,
and standard deviation 𝜎𝜖, set equal to either 1, 5, or 10.
3- Different values of N and T were chosen to be 5, 10, 15, 20, and 50 to represent
small, medium, and large samples for the number of individuals in each cross
section and number of cross sections. Where 5, and 10 were chosen to represent
the small sample, 15, and 20 represented a medium sample size, and finally 50
represented the large sample size.
4- The parameters, 𝛽0𝑖 and 𝛽1𝑖, were set at several different values to allow study
of the estimator where there was a small or a large variety between the cross
sections unit, and under the conditions where the model was both properly and
improperly specified.
To estimate the unknown regression parameters 𝛽° and 𝛽1,different methods of
estimating variance components were used. First, the Swamy method given in
equation (8). RWZ, which replaces the negative diagonal elements of ∆ by zero if
the Swamy method for estimating the variance coefficient has failed to produce
non-negative values, given in equation (11). The MINIQUE method given in
equation (20) was used to evaluate MINIQUE1, and equation (25) to evaluate
MINIQUE2. Finally, equation (26) is used to evaluate both IAUE1 and IAUE2,
each with a different initial value. IAUE1 uses the result of solving equation (20)
as its initial value, and a vector of unity is the initial value of IAUE2 as
recommended by Lucas.
The coefficient estimator ��0 and ��1are computed using different values of the
estimated coefficient’s variance, ∆. According to the method used to estimate ∆, we
will have different values for both ��0 and ��1.
The results are recorded in table (A.1) to table (A.7). Tables consist of five panels
for the different sample sizes (5, 10, 15, 20, and 50). In addition, each panel from
these panels will have three settings for the error standard deviation (1, 5, and 10).
Each setting for each panel will provide the results for six different methods
(Swamy, RWZ, MINQUE1, MINQUE2, IAUE1, and IAUE2). Each of the tables
provides the results for a particular scheme of the regression coefficients to show
the following information:
2736 Souha K. Badr et al.
1- The coefficient mean estimators, ��0 and ��1, that are computed as in equation
(10). The values shown in the first row of each panel of each table are the
average over all 10,000 Monte Carlo trails at a particular setting.
2- The estimated variance of each coefficient, 𝑉𝑎𝑟(��𝑘) = ∆ 𝑓𝑜𝑟 𝑘 = 0,1
averaged over 10,000 trails is shown in the second row. The estimates ∆ are
computed using different methods.
3- The bias value of the coefficient mean estimators, ��0 and ��1, are computed
as
𝑏𝑖𝑎𝑠 (��) = �� − �� (29)
where �� is a vector of coefficient mean estimators and �� is the true vector of the
coefficients mean, the bias values are shown in row three of each panel.
4- The variation in estimating ��, is computed as the variance between the
estimated values of �� during the 10,000 trails, and recorded in the fourth
row of each panel.
5- The bias value of the estimate of the coefficient variances, ∆0 and ∆1, are
computed as
𝑏𝑖𝑎𝑠(∆) = ∆ − ∆ (30)
where ∆ is a vector of coefficient variances and ∆ is the true vector of the coefficient
variance, the bias values are shown in row five of each panel.
6- The variation in estimating ∆, is computed as the variance between the
estimated values of ∆ during the 10,000 trails, and recorded in the sixth row
of each panel.
7- The percentage of a negative variance estimate produced by the different
methods during the 10,000 trails were recorded in row seven in each panel.
6 Analytical Result for Simple Random Coefficient Models
In this section, we will use a Monte Carlo simulation of results to compare the
efficiency of Swamy’s RCR model when using different methods to estimate the
variance component. The behaviour of this model will be tested in small, medium,
and large samples for random, mixed, and fixed parameters. The results for these
models will be recorded in tables (A.1) to (A.7).
Table (A.1) displays the results of a simulation study when the mean and the
variance of the intercept parameter, 𝛽0, equals 10, and also the mean and variance
of the slope parameter, 𝛽1, is equal to the same value, i.e ��1 = 10 𝑎𝑛𝑑 ∆𝛽1= 10.
We will use the Swamy RCR technique to estimate the mean of the coefficient in
the model. To estimate the variance of the coefficient in the model the original
Swamy method for estimating the variance of the parameters will be used. In case
of appearing a variance with negative values during the 10,000-simulation trailer,
the alternative methods will be used for this trail.
As a guide to interpreting table (A.1), the results of using the Swamy method to
estimate the unknown parameters and their properties are recorded in the first
vertical panel of table (A.1). When 𝜎𝜀 = 1 and N=T=5, the average mean for 𝛽0and
Comparing MINQUE and IAUE estimates of variance components 2737
𝛽1over all 10,000 Monte Carlo trails is 9.779 and 10.025 respectively. Note that
the true coefficient values for the mean for both parameters are 10, the average
variances are 9.981 and 10.002 for 𝛽0and 𝛽1 respectively. To estimate these values
from the generating samples, Swamy has failed around 10 % to have a non-negative
variance value. Note that this percentage should be zero. The variation in estimating
𝛽0 is equal to 721.616 and for 𝛽1 is 21.826, while the variation in estimating ∆0
and ∆1 are 97.173 and 50.787 respectively.
The results of estimating the unknown parameters and their properties when
replacing negative variance produced by using the Swamy method to estimate the
coefficient variance by zero value during 10,000-simulation trailer, are recorded in
the second vertical panel labeled “RWZ”. When 𝜎𝜀 = 1 and N=T=5, the average
mean for 𝛽0and 𝛽1over all 10,000 Monte Carlo trails are 9.712 and 10.037
respectively. The average variances are 10.165 and 10.002 for 𝛽0and 𝛽1
respectively. By default, this method does not produce negative variances value,
since it trades with the coefficient parameters as a fixed parameter with zero
variances. The variation in estimating 𝛽0 is increased to 1091.377 and for 𝛽1 to
33.092, while the variation in estimating ∆0 is decreased to 92.922 and is still the
same as Swamy for ∆1.
The Minimum Norm Quadratic Unbiased Estimator presented in equation (20), is
used to estimate the variance component in cases where Swamy produces negative
variance values. The results are recorded in the third panel in table (A.1) and labeled
“MINQUE1”. When 𝜎𝜀 = 1 and N=T=5, the average mean for 𝛽0 over all 10,000
Monte Carlo trails is increased to 11.236 and for 𝛽1 is decreased to 9.752. The
average variances are 10.276 and 8.95 for 𝛽0and 𝛽1 respectively. The variation in
estimating the coefficient parameters are higher compared to all other methods,
equal to 8985.615, and for 𝛽1 is 356.922, while the variation in estimating ∆0 and
∆1 is 91.369 and 55.368 respectively, with 8% of negative variances, which is the
second highest percentage of negative variance after Swamy.
Another way to calculate the Minimum Norm Quadratic Unbiased Estimator
presented in equation (25) is used in the case of negative Swamy’s variance, and
recorded in the fourth panel labeled “MINQUE2”. When 𝜎𝜀 = 1 and N=T=5, the
average mean for 𝛽0 and 𝛽1 over all 10,000 Monte Carlo trails are 10.036 and
9.983 respectively, with a decrease in the estimate of the variation compared with
Swamy. The average variances are 10.221 and 8.952 for 𝛽0and 𝛽1 respectively,
with the same variation as MINQUE1, and 3% of negative variance.
The result of the variance component produced from MINQUE1 is used as an initial
value to calculate the Iteration Almost Unbiased Estimator given in equation (26),
in the case of Swamy’s negative variance, and recorded in the fifth panel labeled
“IAUE1”. When 𝜎𝜀 = 1 and N=T=5, the average mean for 𝛽0and 𝛽1over all
10,000 Monte Carlo trails is 10.069 and 9.978 respectively, with the lowest
variation in estimating those parameters compared to all other methods. The
average variances are 10.27 and 9.055 for 𝛽0and 𝛽1 respectively, with a variation
in estimating ∆0 and ∆1 equal to 90.862 and 53.591 respectively. To estimate these
values from the generating samples, IAUE1 has 0% of negative variance.
2738 Souha K. Badr et al.
Using a unity vector as an initial value to calculate the Iteration Almost Unbiased
Estimator given in equation (26), in the case of a negative Swamy variance, is
recorded in the sixth panel labeled “IAUE2”. As 𝜎𝜀 = 1 and N=T=5, the average
mean for 𝛽0 and 𝛽1 over all 10,000 Monte Carlo trails is 9.565 and 9.442
respectively. The average variances are 10.167 and 8.952 for 𝛽0and 𝛽1
respectively. The variation in estimating the coefficient parameters is equal to
57.468 and 5.793, while the variation in estimating ∆0 and ∆1 are 92.88 and 55.339
respectively, with 0% of negative variances.
As the variation in error term increases, IAUE2 has the worst estimators compared
to all other methods. The efficiency of Swamy estimators also becomes worse,
especially for the variation in estimating 𝛽 and ∆ and the percentage of negative
variance. MINQUE1 and MINQUE2 also increase the variation in estimating
parameters with the existence of negative variance less than Swamy. IAUE1 has a
zero percent of negative variance even with an increase in the error variation. In
addition, it is the least affected by an increase in the error variation in terms of
variation of estimation. RWZ is the second best method after IAUE1, but the big
issue with this method is dealing with coefficient parameters as fixed parameters
rather than random parameters.
Increasing both sample size and the number of cross sections, have improve the
estimators in all methods, except for IAUE2. For example, comparing a small
sample size, N=T=5, with a medium sample size, N=T=20, at the same standard
deviation of the error, 𝜎𝜀 = 5, reduce the negative variances produced by using the
Swamy method from 70% to 7%. The variation in estimating 𝛽1 is decreased from
7484.049 to 0.528 when using the RWZ method, and the variation in estimating ∆1
for MINQUE1 and MINQUE2 was 116.935 and 110.66 reduced to 17.255 and
17.119 respectively, and the absolute value for the average bias for 𝛽0 decreased
from 0.151 to 0.003 for IAUE1. Unfortunately, using the IAUE2 method does not
give any good results under any conditions.
Table (A.2) records the results of a simulation study when reducing the coefficient
variation of both parameters in order to study the effect of using different variance
component estimation methods on the behaviour of the RCR estimators when the
variances of both the intercept and the slope parameters are decreased from 10 to 1.
The mean of 𝛽0 and 𝛽1 are equal to 1.
Reducing coefficient variances led to producing a higher percentage of negative
variance when using the Swamy method, accompanied by a large variation in
estimating both 𝛽 and ∆. MINQUE1 and MINQUE2 have also increased in the
variation of estimating 𝛽 but with less negative variance compared to Swamy.
IAUE1 has the best results compared to other methods, for example as N=10 and
𝜎𝜀 = 1 , the absolute value of average bias for 𝛽0 and 𝛽1, when using IAUE1
method, is 0.023 and 0.005, with a variation in estimation of those parameters equal
to 7.068 and 0.358 respectively. The absolute value for the average bias of ∆0 and
∆1 is 0.026 and 0.006. IAUE1 never produces negative variances.
Table (A.3) displays the results of a simulation study when the mean of 𝛽0 and 𝛽1
is equal to 5, but the variance value for them is not equal; the variance of 𝛽0 equals
10 and the variance of 𝛽1 is 1. This model will allow us to study the effect of using
Comparing MINQUE and IAUE estimates of variance components 2739
different variance component estimation methods on the behavior of the RCR
estimators when the variances of the slope parameter are decreased.
From table (A.3) when N=T=15 and 𝜎𝜀 = 5, the absolute values for the average
bias of 𝛽0 and 𝛽1 using the Swamy method, are 0.227 and 0.055, with a variation
in estimation equal to 3046.032 and 116.873 respectively. Using the same method,
the absolute values for average bias of ∆0 and ∆1 are 0.111 and 0.002, with a
variation in estimation equal to 122.583 and 0.466 respectively. Swamy has failed
around 23% to produce positive or zero variance. MINQUE1 has the second highest
percentage of negative variance equal to 8%. The absolute value for an average bias
of 𝛽0 and 𝛽1 using the same method, reduces to 0.064 and 0.017, with a big
difference in variation of estimation compared to Swamy, and is equal to 27.226
and 1.324 respectively. IAUE1 has even smaller results, with zero percent of
negative variance, and the absolute value for average bias of 𝛽0 and 𝛽1 is 0.019
and 0.008, with a variation in estimation equal to 7.244 and 0.393 respectively.
Using the same method, the absolute value for average bias of ∆0 and ∆1 is 0.608
and 0.075, with a variation in estimation equal to 98.985 and 0.345 respectively
The results of a simulation study when the variance of the intercept parameter is
decreased are recorded and displayed in table (A.4).The mean of 𝛽0 and 𝛽1 are
equal to 1, but the variances value for them are not equal; the variance of 𝛽0 equals
1 and the variance of 𝛽1 is 5. This model will allow us to study the effect of using
different variance component estimation methods on the behavior of the RCR
estimators when the variances of 𝛽0 is decreased.
From table (A.4) for a large sample size, N=T=50, we can see that Swamy has a
zero percent of negative variances at 𝜎𝜀 = 1 , but when the error variation increases
to 𝜎𝜀 = 5 Swamy has 21% percentage of negative variances during 10,000-
simulation trails. Again, MINQUE1 has the second highest percentage of negative
variance after the Swamy method. The bias and the variation in estimating 𝛽 and ∆
is higher than Swamy for both MINQUE1 and MINQUE2. IAUE1 still does well
compared to all other methods in the sense of the bias and the variation in estimating
𝛽, the variation in estimating ∆, and the percentage of negative variances. IAUE2
is the worst.
Mixed Model
In this section, we study the efficiency of the RCR model when using different
methods of variance component estimation. The behavior of this model will be
tested in small, medium, and large samples for the model that contains both random
and fixed parameters. The Monte Carlo simulation results for these models were
present in table (A.5) and (A.6). We will study two cases within these models, the
first, when the intercept parameter is fixed while the slope parameter is random.
The second case is when the intercept parameter is random while the slope
parameter is fixed.
Table (A.5) displays the results of a simulation study when the mean for both
coefficient parameters, 𝛽0 and 𝛽1, equal 10 and the variance of 𝛽0 equals 10, while
the variance of 𝛽1 is zero. This means that the fixed parameter in this model is the
slope parameter, 𝛽1. As above, the RCR model will be estimated using the Swamy
2740 Souha K. Badr et al.
method to estimate the variance component, and in the case of a negative variance
appearing, the alternative method will be used. This model enables us to study the
efficiency of RCR estimators for Mixed RCR models, where the intercept
parameter 𝛽0 is random and the slope parameter 𝛽1 is fixed.
From table (A.5), when 𝜎𝜀 = 1 and N=T=15, the average mean for 𝛽0and 𝛽1over
all 10,000 Monte Carlo trails when we use the Swamy method to estimate the
variance component for the parameters, is 10.071 and 9.982 respectively. Note that
the true coefficient values for the mean for both parameters are 10, the average
variances are 9.944 and zero for 𝛽0and 𝛽1 respectively. To estimate these values
from the generating samples, Swamy has failed around 56 % to have a non-negative
variance value for ∆𝛽1. Note that we deal with 𝛽1 as a fixed parameter. The variation
in estimating 𝛽0 is equal to 137.701 and for 𝛽1 is 5.675, while the variation in
estimating ∆0 and ∆1 is 16.576 and zero respectively. MINQUE1 produced 22.6%
of negative variances value for ∆𝛽1, and a smaller bias and variation of 𝛽 compared
to Swamy. MINIQUE2 has almost similar results as MINQUE1 with a small
percentage of negative variance equal to 15%. RWZ, IAUE1, and IAUE2 have 0%
of negative variances, almost the same results in the bias and the variation in
estimating 𝛽 between RWZ and IAUE1 methods. IAUE1 has a high bias and
variation of ∆0 and ∆1 equal to 4.942 and 0.551 for the bias, and the variation was
equal to 28.319 and 0.231 respectively. IAUE2 has the highest values in bias and
variation for estimating both 𝛽 and ∆.
Increasing the error variation even with a large sample size, makes the Swamy
method the worst method after the IAUE2 method, with a high percentage of
negative variance for both ∆0 and ∆1, a large bias and variation in estimating 𝛽, and
a large variation in estimating ∆.
Table (A.6) displays the results of a simulation study for a mixed model where both
coefficient parameters, 𝛽0 and 𝛽1 have a mean equal to 5. The variance of 𝛽1 is 1
and since we are dealing with a mixed model, in our case where the fixed parameter,
is the intercept parameter the variance of 𝛽0 will be equal to zero. This model will
be estimated using the RCR model using the Swamy method to estimate the
variance component, and in the case of negative variance appearing, the alternative
method will be used. This model enables us to study the efficiency of RCR
estimators for Mixed RCR models where the intercept parameter, 𝛽0, is fixed and
the slope parameter, 𝛽1, is random.
Table (A.6) shows a big variation in estimating, with a large bias in estimating the
coefficient parameters, and a high percentage for negative estimated variances
when using the Swamy method, especially for small and medium sample sizes.
IAUE1 and RWZ methods have the most suitable results among all other methods,
with 0% of negative estimated variances and small bias and variation in estimating
𝛽.
Fixed Model
In this section, we study the efficiency of the RCR model when using different
methods of variance component estimation. The behavior of this model will be tested in small, medium, and large samples for the model that contains fixed parame-
Comparing MINQUE and IAUE estimates of variance components 2741
ters. The Monte Carlo simulation results for these models were present in table
(A.7).
Table (A.7) displays the results of a simulation study when the mean for
both coefficient parameters, 𝛽0 and 𝛽1, equal zero. Moreover, since we will deal
with a fixed model, then both parameters have a variance equal to zero. This model
will be estimated using the RCR model. The Swamy method will be used to
estimate the variance component, and in the case of appearing negative variance,
the alternative method will be used. This model enables us to study the efficiency
of RCR estimators for fixed models where the intercept parameter, 𝛽0, and the
slope parameter, 𝛽1, are fixed.
From table (A.7), when 𝜎𝜀 = 1 and N=T=10, the average mean for 𝛽0and 𝛽1over
all 10,000 Monte Carlo trails when we use the Swamy method to estimate the
variance component for the parameters is -2.346 and 0.446 respectively. Note that
the true coefficient values for the mean for both parameters are zero, the average
variances are 0.001 and zero for 𝛽0and 𝛽1 respectively. To estimate these values
from the generating samples, Swamy has failed around 56 % to have a non-negative
variance value for ∆𝛽0 and 55% for ∆𝛽1
. Note that we deal with 𝛽0and 𝛽1 as fixed
parameters. The variation in estimating 𝛽0 is equal to 56899.9 and for 𝛽1 is 2099.2,
while the variation in estimating ∆0 and ∆1 are 0.659 and 0.001 respectively.
MINIQUE1 and MINIQUE2 have low percentages of negative variances compared
to Swamy. The absolute bias in estimating 𝛽0and 𝛽1 for MINQUE1 is equal to
0.049 and 0.014 and for MINQUE2 is equal to 0.045 and 0.005 respectively. The
variation in estimating those parameters for MINQUE1 is 199.4 and 6.347, and for
MINQUE2 is 91.822 and 2.002. For the RWZ method the absolute bias in
estimating 𝛽0and 𝛽1 is 0.293 and 0.055, with a variation in estimating parameters
equal to 911.401 and 32.095, the absolute bias in estimating∆0 and ∆1 is 0.316 and
0.012, with a variation in estimating parameters equal to 0.306 and zero. IAUE1 as
usual has zero percent of negative variances. The absolute bias in estimating 𝛽0and
𝛽1 is 0.069 and 0.01 with a variation in estimating parameters equal to 1.426 and
78.207, the absolute bias in estimating∆0 and ∆1 are 0.614 and 0.313, with a
variation in estimating parameters equal to 0.218 and 0.306.
Graphical Analysis
For further explanation we use 2D graphical figures to show the resulting variations
in RCR estimations when using different methods to estimate the variance
component. The figures are plotted regardless of whether the RCR model is the
right model to represent this data, to compare the efficiency of the different methods
even when we use it in an improper way. The IAUE2 method was excluded from
the graph, since it gives inappropriate results and that affects the shape of the graph.
Figure (B.1) shows the absolute bias of ��°, ��1, ∆𝛽°, and ∆𝛽1
and their variation
against different sample sizes respectively. While figure (B.2) shows the absolute
bias of ��°, ��1, ∆𝛽°, and ∆𝛽1
and their variation against different standard deviation
for disturbance respectively.
2742 Souha K. Badr et al.
From figure (B.1), we can see how the absolute bias of the coefficient regression
parameters estimated by the Swamy method to estimate the variance component is
affected by sample size. As the sample size increases from 5 (small sample) to 50
(large sample), the absolute bias for both parameters rapidly decreases. For
example, when the sample size was 10 the absolute bias for ��° was close to 3, and
less than 0.5 when the sample size was 20, and almost zero when the sample size
increased to 50. Using the MINQUE1 and RWZ methods to estimate the variance
component also results in high bias values for the coefficient regression parameters,
especially when the sample size is 5, while MINQUE2 and IAUE1 have very low
bias values for the coefficient regression parameters compared with all other
methods. Using MINQUE1 as an initial value to compute IAUE1, produces
estimators with smaller bias values. Decreasing the sample size does not affect the
IAUE1 method as it does for other methods, as we can see that IAUE1 has bias for
both ��° and ��1 close to zero even with a small sample size.
The Swamy, RWZ, and MINQUE1 methods produce high variation values in
estimating the coefficient regression parameters as the sample size decrease, while
MINQU2 and IAUE1 have a very low variation, even with a small sample size
compared to other methods, with IAUE1 having the lowest variation between all
methods.
In estimating coefficient variances ∆𝛽° and ∆𝛽1
, Swamy has the lowest absolute bias
for both parameters compared to all other methods for all sample sizes. This bias is
low even with a small sample size. MINQUE1, MINQUE2, and IAUE1 produce
close values for coefficient variances estimates. Moreover, the variation in
estimating ∆𝛽° and ∆𝛽1
are very close for all methods for all sample sizes, except for
the sample size equal to five, while Swamy has a very high value compared to other
methods.
Figures (B.1.i) and (B.1.j) show that Swamy, MINQUE1, and MINQUE2 have all
produced negative variances, the percentage of which have reduced as the sample
size has increased. IAUE1 has never produced a negative variance, as it should do
over all of the different sample sizes. Note that we exclude the RWZ method from
these figures, since for this method we replace a negative variance with a non-
negative variance, i.e. zero.
From figure (B.2), we can see that as the disturbance standard deviation increased,
the absolute bias of the coefficient regression parameters and their variation
estimation increased for most methods, especially for Swamy, which has the highest
value for bias and variation in estimating β comparing it to all other methods.
IAUE1 is the method least affected by increasing the disturbance standard
deviation, with bias and variation in estimating β close to zero at different values
for disturbance standard deviation.
Again, Swamy has the lowest absolute bias for both parameters compared with all
other methods for all different error variances, while it has a high variation in
estimating ∆ which increased as error variances increased. MINQUE1, MINQUE2,
and IAUE1 have close bias and variation values. The variation for all methods are
almost the same, except for Swamy, which at 𝜎𝜀 = 10, has a very high value.
Comparing MINQUE and IAUE estimates of variance components 2743
Swamy, MINQUE1, and MINQUE2 have produced negative variances. Swamy has
produced more than 50% of negative variances in estimating ∆𝛽° when 𝜎𝜀 = 10,
while IAUE1 has 0% for all different error variances. The RWZ method was
excluded for the same reason mentioned above.
7 Conclusion
Using the Swamy method to estimate the variance component in the RCR model
produces good estimators when the RCR model is the right model to represent the
data. Of course, reducing the sample size, or increasing the disturbance standard
deviation leads to a reduction in the efficiency of the Swamy model, since more
variances that have negative values were produced. When the RCR model is not the
right model to represent the data, Swamy has a very weak attitude regarding the
percentage of negative variances and the variation in estimating both β and ∆.
When the RCR model is the right model to represent the data, MINQUE1 and
MINQUE2 produce a lower percentage of negative variance in estimating ∆𝛽° than
Swamy did, but with higher absolute bias in estimating both β and ∆. When the
RCR model is the inappropriate model to represent the data, both methods are better
than Swamy regarding absolute bias and variation in estimating β and the
percentage of negative variances. In general, we can say MINQUE2 is doing better
than MINQUE1.
IAUE1 has a zero percentage of negative variance, with the bias in estimating β and
the variation in estimating both β and ∆ being the smallest, even if the RCR model
is not the right model to represent data. The weak points in using the IAUE1 method
to estimate the variance component in the RCR model are that firstly, it has a greater
bias in estimating ∆ compared to Swamy, and secondly, it needs a heavy
computation as we use the MINQUE method as its initial value.
References
[1] A. Amiri-Simkooei, K. Moghtased-Azar, R. Tehranchi, An Alternative Method
for Non-negative Estimation of Variance Components, Journal of Geodesy, 88
(2014), 427–439. https://doi.org/10.1007/s00190-014-0693-0
[2] C. Hsiao, Statistical Inference for a Model with Both Random Cross-Sectional
and Time Effects, International Economic Review, 15 (1974), 12-30.
https://doi.org/10.2307/2526085
[3] C. Hsiao, Some Estimation Methods for a Random Coefficient Model,
Econometrica, 43 (1975), 305-325. https://doi.org/10.2307/1913588
[4] C.R. Rao, Estimation of Variance and Covariance Components in Linear
Models, Journal of the American Statistical Association, 67 (1972), 112-115.
2744 Souha K. Badr et al.
https://doi.org/10.1080/01621459.1972.10481212
[5] D.B. Duncan, S.D. Horn, R.A. Horn, Estimating Heteroscedastic Variances in
Linear Models. Journal of the American Statistical Association, 70 (1975), 380–
385. https://doi.org/10.1080/01621459.1975.10479877
[6] J.R. Lucas, A Variance Component Estimation Method for Sparse Matrix
Applications, NOAA Technical Report NOS 111 NGS 33, National Geodetic
Survey, Rockville. 1985.
[7] L. W. Johnson, Regression with Random Coefficients, Omega, 6 (1978), 71-81.
https://doi.org/10.1016/0305-0483(78)90041-5
[8] M. Aitkin, D. Anderson, J. Hinde, Statistical Modelling of Data on Teaching
Styles, Journal of the Royal Statistical Society A (General), 144 (1981), 419 -461.
https://doi.org/10.2307/2981826
[9] M. H. Pesaran, T. Yamagata, Testing Slope Homogeneity in Large Panels,
Journal of Econometrics, 142 (2008), 50-93.
https://doi.org/10.1016/j.jeconom.2007.05.010
[10] M. R. Abonazel, A. Mousa, A. H. Youssef, A Monte Carlo Study for Swamy’s
Estimate of Random Coefficient Panel Data Model, InterStat Journal, 2011 (2011),
1-12.
[11] M.S. Abdallah, Z.A. Abdel_wahed, H.A. El_leithy, On Non-negative
Estimation of Variance Components in Mixed Linear Models, Journal of Advanced
Research, 7 (2016), 59-68. https://doi.org/10.1016/j.jare.2015.02.001
[12] P. Delicado, J. Romo, Goodness of Fit Tests in Random Coefficient
Regression Models, Annals of the Institute of Statistical Mathematics, 51 (1999),
125-148. https://doi.org/10.1023/a:1003887303233
[13] P. A. Cartwright, N. Riabk, Measuring the effect of oil prices on wheat futures
price, Research in International Business and Finance, 33 (2015), 355–369.
https://doi.org/10.1016/j.ribaf.2014.04.002
[14] P. A.V. B. Swamy, Efficient inferences in a random coefficient regression
model, Econometrica, 38 (1970), 311-323. https://doi.org/10.2307/1913012
[15] P. C. Austin, V. Goel, C. V. Walraven, An Introduction to Multilevel
Regression Models, Canadian Journal of Public Health, 92 (2001), 150-154.
Comparing MINQUE and IAUE estimates of variance components 2745
[16] R. D. Gibbons, D. K. Bhaumik, Weighted Random-Effects Regression Models
with Application to Interlaboratory Calibration, Technometrics, 43 (2001), 192-
198. https://doi.org/10.1198/004017001750386305
[17] R. G. Drynan, W. E. Griffiths, S. Prakash, Bayesian Estimation of a Random
Coefficient Model, Journal of Econometrics, 10 (1979), 201-220.
https://doi.org/10.1016/0304-4076(79)90005-8
[18] R. Hsu, An Alternative Expression for the Variance Factors in Using Iterated
Almost Unbiased Estimation, Journal of Geodesy, 73 (1999), 173–179.
https://doi.org/10.1007/s001900050234
[19] R. Hsu, Helmert Method as Equivalent of Iterated Almost Unbiased
Estimation, Journal of Surveying Engineering, 127 (2001), 79–89.
https://doi.org/10.1061/(asce)0733-9453(2001)127:3(79)
[20] R. L. Carter, M. C. K. Yang, Large sample inference in random coefficient
regression models, Communications in Statistics - Theory and Methods, 15 (1986),
2507-2525. https://doi.org/10.1080/03610928608829265
[21] R. P. Leone, H. D. Oberhelman, F. J. Mulhern, Estimating Individual Cross-
Section Coefficients from the Random Coefficient Regression Model, Journal of
the Academy of Marketing Science, 21 (1993), 45-51.
https://doi.org/10.1177/0092070393211006
[22] S.R. Searle, An Overview of Variance Component Estimation, Metrika, 42
(1995), 215–230. https://doi.org/10.1007/bf01894301
[23] T. E. Dielman, Pooled Cross-Sectional and Time Series Data Analysis, Marcel
Dekker, New York, 1989.
[24] T. Hobza, D. Morales, Small Area Estimation of Poverty Proportions under
Random Regression Coefficient Models, Chapter in Modern Mathematical Tools
and Techniques in Capturing Complexity Understanding Complex System,
Springer, 2011, 315-328. https://doi.org/10.1007/978-3-642-20853-9_22
[25] T. Kall, T. Oja, K. Tanavsuu, Postglacial land uplift in Estonia based on four
precise leveling, Tectonophysics, 610 (2014), 25-38.
https://doi.org/10.1016/j.tecto.2013.10.002
[26] V. V. Anh, T. Chelliah, Estimated Generalized Least Squares for Random
Coefficient Regression Models, Scandinavian Journal of Statistics, 26 (1999), 31-
46. https://doi.org/10.1111/1467-9469.00135
2746 Souha K. Badr et al.
[27] W. E. Griffiths, R. Drynan and S. K.B. Rao, Bayesian estimation of a random
coefficient model, Journal of Econometrics, 10 (1979), 201-220
https://doi.org/10.1016/0304-4076(79)90005-8
Comparing MINQUE and IAUE estimates of variance components 2747
Table (A.1) Results of RCR Estimation When 𝛽0~𝑁(10,10) and 𝛽1~𝑁(10,10)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N
=
T
=
5
𝝈𝝐
=
1
�� 9.779 10.025 9.712 10.037 11.236 9.752 10.036 9.983 10.069 9.978 9.565 9.442
∆ 9.981 10.002 10.165 10.002 10.276 8.950 10.221 8.952 10.270 9.055 10.167 8.952
Bias of �� -0.221 0.025 -0.288 0.037 1.236 -0.248 0.036 -0.017 0.069 -0.023 -0.435 -0.558
Variation
of �� 721.616 21.826 1091.377 33.092 8985.615 356.922 72.289 4.185 55.537 3.566 57.468 5.793
Bias of∆ -0.019 0.002 0.165 0.002 0.276 -1.050 0.221 -1.048 0.270 -0.945 0.167 -1.048
Variation
of ∆ 97.173 50.787 92.922 50.787 91.367 55.368 91.849 55.339 90.862 53.591 92.880 55.339
%negative
variance 10.73% 0.01% 0% 0% 3% 5.03% 1.91% 1.2% 0% 0% 0% 0%
𝝈𝝐
=
5
�� 5.191 11.057 19.158 8.265 9.770 9.928 10.270 9.924 10.151 9.965 7.353 6.923
∆ 4.641 9.655 63.897 10.232 75.358 6.217 66.116 6.547 60.750 6.746 67.114 7.195
Bias of �� -4.809 1.057 9.158 -1.735 -0.203 -0.072 0.270 -0.076 0.151 -0.035 -2.647 -3.077
Variation �� 207135.23 7989.74 200000.9 7484.049 110613.323 4288.823 2013.824 89.111 740.288 26.614 731.565 30.840
Bias of∆ -5.359 -0.345 53.897 0.232 65.358 -3.783 56.116 -3.453 50.750 -3.254 57.114 -2.805
Variation ∆ 27771.385 133.2108 14054.02 118.865 13830.206 116.935 13311.241 110.666 13803.536 108.157 21034.601 120.739
%negative
variance 55.98% 18.85% 0% 0% 17.13% 29.69% 14.07% 5.73% 0% 0% 0% 0%
𝝈𝝐 =
10
�� 11.184 10.011 10.371 9.925 -1.116 12.078 11.373 9.650 10.015 9.978 6.842 6.856
∆ 6.235 9.825 160.731 12.997 203.175 10.593 188.908 10.816 151.854 10.187 5928.646 1001.589
Bias of �� 1.184 0.011 0.371 -0.075 -11.116 2.078 1.373 -0.350 0.015 -0.022 -3.158 -3.144
Variation
of �� 258480.478 5273.384 29222.901 987.966 8096812 236998 8271.280 508.970 890.850 29.918 810.659 31.594
Bias of∆ -3.765 -0.175 150.731 2.997 193.175 0.593 178.909 0.816 141.854 0.187 5918.646 991.589
Variation ∆ 136972.085 320.020 93834.944 308.746 136786.610 346.469 86597.467 287.221 93261.149 292.852 1472097991 38045969
%negative
variance 56.21% 35.2% 0% 0% 23.03% 26.77% 13.32% 13.95% 0.01% 0% 0% 0%
N
=
T
=
10
𝝈𝝐
=
1
�� 10.001 9.979 10.001 9.979 10.034 9.973 10.001 9.979 10.001 9.979 9.994 9.970
∆ 10.004 10.024 10.005 10.024 10.005 10.015 10.005 10.015 10.006 10.016 10.005 10.015
Bias of �� 0.001 -0.021 0.001 -0.021 0.034 -0.027 0.001 -0.021 0.001 -0.021 -0.006 -0.030
Variation �� 1.135 1.024 1.135 1.024 10.902 1.388 1.137 1.024 1.134 1.024 1.181 1.120
Bias of∆ 0.004 0.024 0.005 0.024 0.005 0.015 0.005 0.015 0.001 0.016 0.005 0.015
Variation
of ∆ 27.821 22.060 27.818 22.060 27.802 22.145 27.808 22.145 27.800 22.129 27.818 22.145
%negative
variance 0.09% 0% 0% 0% 0.01% 0.04% 0% 0.01% 0% 0% 0% 0%
𝝈𝝐
=
5
�� 12.857 9.414 9.603 10.069 9.121 10.159 9.969 10.029 9.963 9.999 31.187 15.510
∆ 9.893 9.934 14.297 9.934 23.144 6.363 21.539 6.428 14.668 6.691 24.070 7.218
Bias of �� 2.857 -0.586 -0.397 0.069 -0.879 0.159 -0.031 0.029 -0.037 -0.001 21.187 5.510
Variation
of �� 124512.354 4974.336 1018.536 35.579 2813.485 110.055 1350.637 53.270 157.880 6.066 3614.858 264.154
Bias of∆ -0.107 -0.066 4.297 -0.066 13.144 -3.637 11.539 -3.572 4.668 -3.309 14.069 -2.782
Variation
of ∆ 540.618 27.625 357.868 27.614 260.937 41.855 260.690 41.048 347.242 37.896 815.761 34.209
%negative
variance 38.12% 0.12% 0% 0% 0.19% 13.65% 0.06% 7.01% 0% 0% 0% 0%
𝝈𝝐
=
10
�� 13.421 9.355 10.148 9.952 7.556 10.487 10.306 9.920 10.128 9.960 44619.72 11911.06
∆ 9.606 9.975 32.067 10.014 79.712 6.022 71.835 6.364 32.293 6.327 27692.95 1426.831
Bias of �� 3.421 -0.645 0.148 -0.048 -2.443 0.487 0.306 -0.080 0.128 -0.040 44609.72 11901.06
Variation
of �� 52770.449 1961.913 648.642 25.480 38850.847 1602.002 3826.864 157.345 108.201 5.048 12356186592 412542364
Bias of∆ -0.394 -0.025 22.066 0.014 69.712 -3.978 61.835 -3.636 22.293 -3.673 27682.984 1416.831
Variation ∆ 5058.257 48.141 2675.037 47.292 2498.772 59.413 2268.614 55.756 2642.312 55.284 5053501756 6214327
%negative
variance 50.17% 3.39% 0% 0% 0.22% 19.46% 0.04% 11.3% 0% 0% 0% 0%
2748 Souha K. Badr et al.
Table (A.1) Results of RCR Estimation When 𝛽0~𝑁(10,10) and 𝛽1~𝑁(10,10)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T =
15
𝝈𝝐 = 1
�� 9.996 9.998 9.996 9.998 9.996 0.998 9.996 9.998 9.996 9.998 9.996 9.998
∆ 10.057 10.052 10.057 10.052 10.057 10.052 10.057 10.052 10.057 10.052 10.057 10.052
Bias of �� -0.004 -0.002 -0.004 -0.002 -0.004 -0.002 -0.004 -0.002 -0.004 -0.002 -0.004 -0.002
Variation
of �� 0.724 0.658 0.724 0.658 0.724 0.658 0.724 0.658 0.724 0.658 0.724 0.658
Bias of∆ 0.057 0.052 0.057 0.052 0.057 0.052 0.057 0.052 0.057 0.052 0.057 0.052
Variation
of ∆ 16.242 14.411 16.242 14.411 16.242 14.411 16.242 14.411 16.242 14.411 16.242 14.411
%negative variance 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
𝝈𝝐
=
5
�� 10.097 9.987 10.009 10.005 9.311 10.139 10.368 9.914 10.051 9.996 205.535 14.914
∆ 9.979 10.007 10.856 10.007 15.588 8.073 14.785 8.112 11.053 8.249 90.339 8.699
Bias of �� 0.097 -0.013 0.009 0.005 -0.689 0.139 0.368 -0.086 0.051 -0.004 195.535 4.914
Variation �� 51.848 3.002 36.465 2.191 2659.141 93.851 1210.548 76.131 18.267 1.322 452470.041 176.275
Bias of∆ -0.021 0.007 0.856 0.007 5.588 -1.927 4.785 -1.888 1.053 -1.751 80.339 -1.300
Variation ∆ 133.243 16.403 108.928 16.403 102.602 28.795 95.684 28.196 104.798 26.099 77298.261 20.997
%negative
variance 19.66% 0% 0% 0% 0% 6.97% 0% 4.13% 0% 0% 0% 0%
𝝈𝝐
=
10
�� 9.979 9.994 9.868 10.029 9.957 10.015 10.514 9.895 9.972 10.008 353429.44 12020.45
∆ 10.250 10.076 17.968 10.076 57.615 6.361 53.530 6.548 18.136 6.669 196363.3 1294.342
Bias of �� -0.021 -0.006 -0.132 0.029 -0.042 0.015 0.514 -0.105 -0.028 0.008 353419.44 12010.45
Variation �� 2405.235 93.249 155.341 6.109 344.494 16.355 925.999 39.267 9.175 0.958 550355156962 321444969
Bias of∆ 0.250 0.076 7.968 0.076 47.615 -3.639 43.530 -3.452 8.363 -3.331 196353.261 1284.342
Variation ∆ 1038.674 23.508 605.729 23.494 1602.997 41.290 1389.807 39.231 591.012 37.441 176733180989 3739488
%negative
variance 41.16% 0.12% 0% 0% 0% 15.39% 0% 11.03% 0% 0% 0% 0%
N = T =
20
𝝈𝝐 =
1
�� 10.005 10.008 10.005 10.008 10.005 10.008 10.005 10.008 10.005 10.008 10.005 10.008
∆ 9.981 9.993 9.981 9.993 9.981 9.993 9.981 9.993 9.981 9.993 9.981 9.993
Bias of �� 0.005 0.008 0.005 0.008 0.005 0.008 0.005 0.008 0.005 0.008 0.005 0.008
Variation
of �� 0.509 0.502 0.509 0.502 0.509 0.502 0.509 0.502 0.509 0.502 0.509 0.502
Bias of∆ -0.018 -0.006 -0.502 -0.006 -0.018 -0.006 -0.018 -0.006 -0.018 -0.006 -0.018 -0.006
Variation
of ∆ 11.491 10.459 11.491 10.459 11.491 10.459 11.491 10.459 11.491 10.459 11.491 10.459
%negative
variance 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 10.040 9.997 10.023 10.0003 10.046 9.996 10.045 9.996 10.025 10.00004 374.861 11.647
∆ 9.806 9.986 9.951 9.986 11.722 9.275 11.571 9.282 10.024 9.342 148.571 9.491
Bias of �� 0.040 -0.002 0.023 0.0003 0.046 -0.003 0.045 -0.003 0.026 3.507𝑒−5 364.861 1.647
Variation
of �� 3.063 0.593 1.178 0.528 2.913 0.591 5.555 0.683 1.159 0.527 3.163𝑒6 50.843
Bias of∆ -0.193 -0.013 -0.048 -0.013 1.722 -0.724 1.571 -0.717 0.024 -0.657 138.571 -0.508
Variation
of ∆ 54.972 11.441 51.577 11.441 57.621 17.255 55.321 17.119 50.202 16.075 468530.181 13.980
%negative
variance 7.242% 0% 0% 0% 0% 2.628% 0% 2.057% 0% 0% 0% 0%
𝝈𝝐 =
10 �� 10.543 9.902 10.006 10.0008 10.031 9.998 10.134 9.977 10.015 9.999 69159.116 7238.752
∆ 9.683 10.038 14.650 10.038 51.628 6.510 47.778 6.674 15.026 6.833 36404.810 733.570
Bias of �� 0.543 -0.097 0.006 0.0008 0.031 -0.001 0.134 -0.022 0.015 -0.0008 69149.116 7228.752
Variation
of �� 1125.0004 36.129 5.366 0.657 68.548 2.926 27.748 1.650 3.685 0.613 15858467574 110798072
Bias of∆ -0.316 0.038 4.650 0.038 41.628 -3.489 37.778 -3.325 5.026 -3.166 36394.810 723.570
Variation
of ∆ 571.773 15.428 351.915 15.428 1630.149 34.565 1361.342 32.649 341.128 30.523 4604695739 1113180
%negative
variance 37.62 0% 0% 0% 0% 15.42% 0% 10.04% 0% 0% 0% 0%
Comparing MINQUE and IAUE estimates of variance components 2749
Table (A.1) Results of RCR Estimation When 𝛽0~𝑁(10,10) and 𝛽1~𝑁(10,10)
Table (A.2) Results of RCR Estimation When 𝛽0~𝑁(1,1) and 𝛽1~𝑁(1,1)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T =
50
𝝈𝝐 = 1
�� 10.016 9.983 10.016 9.983 10.016 9.983 10.016 9.983 10.016 9.983 10.016 9.983
∆ 9.971 10.110 9.971 10.110 9.971 10.110 9.971 10.110 9.971 10.110 9.971 10.110
Bias of �� 0.016 -0.016 0.016 -0.016 0.016 -0.016 0.016 -0.016 0.016 -0.016 0.016 -0.016
Variation
of �� 0.201 0.209 0.201 0.209 0.201 0.209 0.201 0.209 0.201 0.209 0.201 0.209
Bias of∆ 0.028 0.110 -0.028 0.110 -0.028 0.110 -0.028 0.110 -0.028 0.110 -0.028 0.110
Variation ∆ 5.611 5.740 5.611 5.740 5.611 5.740 5.611 5.740 5.611 5.740 5.611 5.740
%negative variance 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 10.0007 9.977 10.0007 9.977 9.977 10.0007 10.0007 9.977 10.0007 9.977 10.0007 9.977
∆ 10.035 10.082 10.035 10.082 10.035 10.082 10.035 10.082 10.035 10.082 10.035 10.082
Bias of �� 0.0007 0.022 0.0007 -0.022 0.0007 -0.022 0.0007 -0.022 0.0007 -0.022 0.0007 -0.022
Variation
of �� 0.306 0.215 0.306 0.215 0.306 0.215 0.306 0.215 0.306 0.215 0.306 0.215
Bias of∆ 0.035 0.082 0.035 0.082 0.035 0.082 0.035 0.082 0.035 0.082 0.035 0.082
Variation ∆ 6.315 6.124 6.315 6.124 6.315 6.124 6.315 6.124 6.315 6.124 6.315 6.124
%negative
variance 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
𝝈𝝐
=
10
�� 10.014 9.966 10.014 9.966 10.016 9.966 10.168 9.936 9.966 10.014 204253.519 1662.048
∆ 9.478 10.103 9.627 10.103 15.850 9.500 15.728 9.506 9.689 9.564 89434.122 164.859
Bias of �� 0.014 0.014 0.014 -0.033 0.016 -0.033 0.168 -0.063 0.014 -0.033 204243.519 1652.048
Variation
of �� 0.556 0.235 0.555 0.235 0.589 0.241 3.979 0.358 0.555 0.235 735805679721 43549521
Bias of∆ -0.521 0.103 -0.372 0.103 5.850 -0.499 5.728 -0.493 -0.310 -0.435 89424.122 154.859
Variation ∆ 8.159 6.535 8.106 6.535 48.687 6.187 47.370 6.188 8.105 6.186 9.870𝑒9 2.635𝑒4
%negative
variance 6.140% 0% 0% 0% 0% 3.728% 0% 3.070% 0% 0% 0% 0%
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T = 5
𝝈𝝐 = 1
�� 0.929 0.999 0.704 1.056 -0.163 1.178 0.811 1.033 0.919 1.016 0.631 0.581
∆ 0.924 0.999 1.455 1.001 1.803 0.617 1.674 0.620 1.859 1.023 1.446 0.617
Bias of �� -0.071 -0.000 -0.296 0.056 -1.163 0.178 -0.189 0.033 -0.081 0.016 -0.369 -0.419
Variation �� 481.235 12.628 516.487 16.585 6575.22 391.109 113.669 4.442 78.304 2.853 78.131 3.033
Bias of∆ -7.643e-02
-2.887e-
05 0.455 0.001 0.803 -0.383 0.674 -0.379 0.859 0.023 0.446 -0.383
Variation ∆ 6.842 0.595 4.519 0.594 4.797 0.623 4.018 0.617 3.538 0.365 4.479 0.621
%neg var 41.2% 1.71% 0% 0% 11.76% 17.37% 3.55% 5.1% 0% 0% 0% 0%
𝝈𝝐
=
5
�� 1.101 0.946 22.339 -2.377 19.937 -2.019 0.542 1.040 23.374 0.932 1.575 0.936
∆ 0.389 0.969 24.527 1.518 34.381 1.426 31.062 1.491 23.578 1.832 24.772 1.872
Bias of �� 0.101 -0.054 21.339 -3.377 18.937 -3.019 -0.458 0.040 0.374 -0.068 0.575 -0.064
Variation �� 13927.761 415.395 3333078.
7 76446.3 2501743.58 59000.95 7572.619 272.480 3633.998
118.7
85 3642.418 121.024
Bias of∆ -0.611 -0.031 23.527 0.518 33.381 0.426 30.062 0.491 22.578 0.832 23.772 0.872
Variation ∆ 4232.460 7.579 2080.306 5.069 2835.699 5.683 1945.490 4.644 2052.794 3.816 2179.569 6.734
%neg var 56.5% 41.96% 0% 0% 20.62% 25.08% 14.69% 11.77% 0% 0% 0% 0%
𝝈𝝐
=
1
0
�� -0.388 1.444 -0.387 1.271 -0.541 1.322 1.319 0.805 -0.573 1.334 57.408 78.552
∆ -7.124 0.873 496.091 23.788 529.115 24.847 523.316 25.036 492.571
24.20
8 1064.881 169.581
Bias of �� -1.388 0.444 -1.387 0.271 -1.541 0.322 0.319 -0.195 -1.573 0.334 56.408 77.552
Variation �� 380125.77 10780.8
6
138294.2
53
5587.72
2 35959.875 1478.565
51304.04
9 3259.662
18478.28
0
739.3
13 2022708 1727922
Bias of∆ -8.124 -0.127 495.091 22.788 528.115 23.847 522.316 24.036 491.571
23.20
8 1063.881 168.581
Variation ∆ 2011408.0
67
4558.04
3
1074521.
619
2525.35
6
1066316.95
8 2530.221
1050651.
576 2471.044
1076851.
668
2503.
343 97585076 5347288
%neg var 59.52% 58.88% 0% 0% 19.71% 28.29% 15.83% 12.47% 0% 0% 0% 0%
2750 Souha K. Badr et al.
Table (A.2) Results of RCR Estimation When 𝛽0~𝑁(1,1) and 𝛽1~𝑁(1,1)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T =
10
𝝈𝝐 = 1
�� 1.099 0.977 0.959 1.009 1.015 0.997 0.563 1.094 0.977 1.005 0.832 0.788
∆ 0.971 0.993 1.052 0.993 1.248 0.794 1.188 0.797 1.262 0.994 1.053 0.794
Bias of �� 0.099 -0.023 -0.041 0.009 0.015 -0.003 -0.437 0.094 -0.023 0.005 -0.168 -0.212
Variation
of �� 937.431 36.979 15.299 0.647 3121.383 90.432 1573.326 73.376 7.068 0.358 7.111 0.517
Bias of∆ -0.029 -0.007 0.052 -0.007 0.248 -0.206 0.188 -0.203 0.026 -0.006 0.053 -0.206
Variation ∆ 1.366 0.244 1.152 0.244 0.919 0.358 0.962 0.354 0.879 0.197 1.150 0.359
%negative variance 20.65% 0.02% 0% 0% 0.18% 7.96% 0.03% 2.99% 0% 0% 0% 0%
𝝈𝝐
=
5
�� -1.311 1.344 0.436 1.082 0.156 1.127 0.713 1.007 0.165 1.129 32.629 3.051
∆ 0.727 0.991 6.264 1.021 19.464 0.679 17.018 0.791 6.703 1.182 150.979 3.044
Bias of �� -2.311 0.344 -0.564 0.082 -0.844 0.127 -0.287 0.007 -0.835 0.129 31.629 2.051
Variation
of �� 69662.826 1905.495 7247.355 192.481 8095.771 236.789 10389.923 357.215 6748.363 175.765 15194.225 190.176
Bias of∆ -0.273 -0.009 5.264 0.021 18.464 -0.321 16.018 -0.209 5.703 0.182 149.979 2.044
Variation ∆ 250.031 0.889 132.386 0.817 138.971 0.884 124.052 0.787 125.194 0.453 171328.453 12.655
%negative
variance 55.63% 12.41% 0.13% 20.7% 0% 11.48% 0% 0% 0% 0%
𝝈𝝐
=
10
�� 0.327 1.125 0.799 1.019 0.729 1.056 1.452 0.891 1.029 0.986 3431.953 989.384
∆ 0.966 1.039 23.278 1.596 76.744 1.612 67.027 1.987 22.683 1.904 21612.99 1183.096
Bias of �� -0.673 0.125 -0.200 0.019 -0.270 0.056 0.452 -0.109 0.029 -0.014 3430.953 988.384
Variation
of �� 42361.539 1546.976 1169.258 35.151 379.898 17.508 1639.011 82.949 26.289 0.802 67067069 2583673
Bias of∆ -0.034 0.039 22.278 0.596 75.744 0.612 66.027 0.987 21.683 0.904 21611.989 1182.096
Variation ∆ 3449.968 7.454 1613.779 4.824 1894.981 5.106 1671.923 4.453 1591.793 3.657 2431396941 3441040
%negative
variance 54.88% 39.66% 0% 0% 0.25% 21.01% 0.16% 11.93% 0% 0% 0% 0%
N = T =
15
𝝈𝝐 =
1
�� 0.996 1.003 0.995 1.003 0.994 1.004 0.842 1.033 0.995 1.003 0.962 0.956
∆ 0.998 1.005 1.004 1.005 1.049 0.959 1.042 0.959 1.050 1.003 1.004 0.959
Bias of �� -0.004 0.003 -0.005 0.003 -0.006 0.004 -0.158 0.033 -0.005 0.003 0.038 -0.044
Variation
of �� 0.254 0.073 0.149 0.069 5.084 0.246 94.914 3.455 0.149 0.069 0.168 0.113
Bias of∆ -0.002 0.005 0.004 0.005 0.049 -0.041 0.042 -0.040 0.050 0.003 0.004 -0.041
Variation
of ∆ 0.489 0.153 0.474 0.153 0.428 0.189 0.431 0.189 0.426 0.146 0.474 0.189
%negative
variance 4.543% 0% 0% 0% 0% 1.757% 0% 1.043% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 0.458 1.096 1.524 0.906 0.973 1.003 1.389 0.918 0.991 1.001 30.003 1.188
∆ 1.210 1.019 3.707 1.020 15.066 0.646 13.239 0.721 4.181 1.081 130.310 1.261
Bias of �� -0.542 0.096 0.524 -0.094 -0.027 0.003 0.389 -0.019 -0.009 0.001 29.003 0.188
Variation
of �� 609.309 19.162 2144.166 66.882 83.493 2.777 2198.858 93.732 2.083 0.136 4499.030 0.551
Bias of∆ 0.210 0.019 2.707 0.020 14.066 -0.354 12239 -0.279 3.181 0.081 129.310 0.261
Variation
of ∆ 63.454 0.369 32.476 0.365 96.834 0.524 81.829 0.463 29.039 0.214 8.03e+04 0.497
%negative
variance 48.7% 1.7% 0% 0% 0% 15.371% 0% 10.286% 0% 0% 0% 0%
𝝈𝝐 =
10 �� -0.137 1.199 3.433 0.569 1.353 0.926 16.923 -1.910 1.071 0.987 11655.933 836.337
∆ 0.245 0.968 13.286 1.119 69.051 0.912 62.033 1.191 13.539 1.373 66679.069 875.238
Bias of �� -1.137 0.199 2.433 -0.430 0.353 -0.074 15.923 -2.910 0.071 -0.013 11654.933 835.337
Variation
of �� 2953.459 88.138 14568.094 458.842 384.358 15.457 1700533.60 56276.33 11.439 0.398 445311044.5 962632.1
Bias of∆ -0.755 -0.032 12.286 0.119 68.051 -0.088 61.033 0.191 12.539 0.373 66678.069 874.238
Variation
of ∆ 1152.025 1.933 538.839 1.475 1604.081 1.645 1385.088 1.555 525.064 0.907 1.41e+10 9.728e+05
%negative
variance 55.657% 25.057% 0% 0% 0% 22.4% 0% 15.2% 0% 0% 0% 0%
Comparing MINQUE and IAUE estimates of variance components 2751
Table (A.2) Results of RCR Estimation When 𝛽0~𝑁(1,1) and 𝛽1~𝑁(1,1)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T =
20
𝝈𝝐 = 1
�� 1.0003 1.002 1.0004 1.002 1.0004 1.002 1.001 1.002 1.0004 1.002 0.997 0.998
∆ 1.0005 0.995 1.0008 0.995 1.004 0.991 1.004 0.991 1.004 0.995 1.0008 0.991
Bias of �� 0.0003 0.002 0.0004 0.002 0.0004 0.002 0.001 0.002 0.0004 0.002 -0.002 -0.001
Variation
of �� 0.076 0.051 0.075 0.051 0.078 0.051 0.086 0.051 0.075 0.051 0.077 0.055
Bias of∆ 0.0005 -0.004 0.0008 -0.004 0.004 -0.008 0.004 -0.008 0.004 -0.004 0.0008 -0.0008
Variation ∆ 0.257 0.108 0.256 0.108 0.252 0.112 0.253 0.112 0.252 0.108 0.256 0.112
%negative variance 0.414% 0% 0% 0% 0% 0.171% 0% 0.085 0% 0% 0% 0%
𝝈𝝐
=
5
�� 0.919 1.018 1.015 1.001 0.970 1.009 0.905 1.024 1.004 1.003 42.096 1.573
∆ 0.967 0.999 2.370 1.00006 13.466 0.615 12.304 0.668 2.823 1.038 162.916 1.551
Bias of �� -0.080 0.018 0.015 0.001 -0.029 0.009 -0.094 0.024 0.004 0.003 41.096 0.573
Variation
of �� 48.444 1.582 2.301 2.301 24.562 1.101 30.772 1.301 0.744 0.075 5505.083 1.028
Bias of∆ -3.220 -7.714 1.370 0.00006 12.466 -0.384 11.304 -0.331 1.823 0.038 161.916 0.551
Variation ∆ 22.650 0.230 11.639 0.230 112.0007 0.418 95.518 0.373 9.732 0.139 7.960 7.393
%negative
variance 45.642% 0.228% 0% 0% 0% 16.814% 0% 11.471% 0% 0% 0% 0%
𝝈𝝐
=
10
�� 1.871 0.855 0.998 1.001 -0.925 1.382 0.839 1.026 1.0100 0.999 129928.716 1230.514
∆ 1.059 1.009 6.726 1.030 58.002 0.749 54.142 0.930 7.191 1.179 688799.498 1235.629
Bias of �� 0.871 -0.144 -0.001 0.001 -1.925 0.382 -0.160 0.026 0.010 -0.0004 129927.716 1229.514
Variation
of �� 1722.087 43.184 12.176 0.460 11608.102 452.030 256.889 10.842 2.368 0.127 40307097265 1962165
Bias of∆ 0.059 0.009 5.726 0.030 57.002 -0.250 53.142 -0.069 6.191 0.179 688798.498 1234.629
Variation ∆ 252.876 0.659 118.668 0.607 1861.528 0.771 1687.344 0.803 112.001 0.336 1.079 1.861
%negative
variance 52.1% 8.885% 0% 0% 0% 19.028% 0% 14.857% 0% 0% 0% 0%
N = T =
50
𝝈𝝐 =
1
�� 1.002 0.997 1.002 0.997 1.002 0.997 1.002 0.997 1.002 0.997 1.002 0.997
∆ 1.002 1.0005 1.002 1.0005 1.002 1.0005 1.002 1.0005 1.002 1.0005 1.002 1.0005
Bias of �� 0.002 -0.002 0.002 -0.002 0.002 -0.002 0.002 -0.002 0.002 -0.002 0.002 -0.002
Variation
of �� 0.025 0.020 0.025 0.020 0.025 0.020 0.025 0.020 0.025 0.020 0.025 0.020
Bias of∆ 0.002 0.0005 0.002 0.0005 0.002 0.0005 0.002 0.0005 0.002 0.0005 0.002 0.0005
Variation ∆ 0.190 0.186 0.190 0.186 0.190 0.186 0.190 0.186 0.190 0.186 0.190 0.186
%negative
variance 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 1.010 0.995 1.010 0.995 1.071 0.983 1.002 0.999 1.010 0.995 1.010 0.995
∆ 1.065 0.994 1.167 0.994 6.244 0.801 6.101 0.808 1.369 1.002 1.369 1.002
Bias of �� 0.010 -0.004 0.010 -0.004 0.071 -0.016 0.002 -0.0002 0.010 -0.004 0.010 -0.004
Variation
of �� 0.123 0.024 0.122 0.024 4.882 0.217 41.276 1.597 0.121 0.024 0.121 0.024
Bias of∆ 0.065 -0.005 0.167 -0.005 5.244 -0.198 5.101 -0.191 0.369 0.002 0.369 0.002
Variation ∆ 0.461 0.164 0.435 0.164 23.734 0.139 22.631 0.139 0.455 0.164 0.455 0.164
%negative
variance 20.104% 0% 0% 0% 0% 10.833% 0% 7.395% 0% 0% 0% 0%
𝝈𝝐 =
10 �� 1.017 0.994 1.008 0.995 1.008 0.994 0.880 1.020 1.007 0.995 1.007 0.995
∆ 1.137 1.003 2.267 1.003 43.434 0.638 42.136 0.703 2.680 1.045 2.680 1.045
Bias of �� 0.017 -0.005 0.008 -0.004 0.008 -0.005 -0.119 0.020 0.007 -0.004 0.007 -0.004
Variation
of �� 0.444 0.034 0.394 0.032 7.763 0.289 18.478 0.806 0.392 0.032 0.392 0.032
Bias of∆ 0.137 0.003 1.267 0.003 42.434 -0.361 41.136 -0.296 1.680 0.045 1.680 0.045
Variation ∆ 4.572 0.220 3.117 0.220 922.670 0.169 874.185 0.189 3.092 0.226 3.092 0.226
%negative
variance 4.572% 0% 0% 0% 0% 21.048% 0% 16.044% 0% 0% 0% 0%
2752 Souha K. Badr et al.
Table (A.3) Results of RCR Estimation When 𝛽0~𝑁(5,10) and 𝛽1~𝑁(5,1)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T = 5
𝝈𝝐 = 1
�� 4.846 5.028 4.747 5.046 5.133 4.989 5.285 4.908 4.884 5.020 3.776 2.947
∆ 9.999 0.982 12.442 1.046 11.599 0.840 11.589 0.840 11.766 1.231 11.382 0.849
Bias of �� -0.154 0.028 -0.253 0.046 0.133 -0.010 0.285 -0.092 -0.116 0.020 -1.224 -2.053
Variation
of �� 686.401 27.656 167.997 6.907 7065.117 334.422 1411.358 103.495 89.136 3.793 89.729 10.128
Bias of∆ -0.000 0.018 2.442 0.046 1.599 -0.159 0.1589 -0.159 1.766 0.231 1.382 -0.151
Variation ∆ 375.861 1.492 289.019 1.322 285.223 1.347 283.756 1.343 279.659 0.930 288.149 1.329
%negative variance 32.41% 18.3% 0% 0% 13.6% 18.03% 9.68% 8.24% 0% 0% 0% 0%
𝝈𝝐
=
5
�� 4.958 4.928 5.107 4.992 2.222 5.544 5.106 5.023 5.525 4.965 5.969 4.355
∆ 0.834 0.670 53.154 2.179 68.921 1.951 56.915 2.290 51.308 2.603 51.893 2.495
Bias of �� -0.042 -0.072 0.207 -0.008 52.821 0.829 56.926 2.291 51.584 2.604 52.834 2.911
Variation
of �� 65549.822 1699.397 49026.125 1104.163 99403.568 2870.227 25173.098 504.122 21818.178 416.498 21911.188 440.965
Bias of∆ -9.166 -0.329 43.154 1.179 58.921 0.951 46.915 1.290 41.308 1.603 41.893 1.495
Variation ∆ 22716.709 28.612 12003.865 16.836 17932.175 23.880 11631.592 15.943 12018.537 14.738 12099.407 16.327
%negative
variance 58.43% 52.02% 0% 0% 23.59 34.16 17.97% 7.42% 0% 0% 0% 0%
𝝈𝝐
=
10
�� 4.756 5.117 2.169 5.539 8.456 4.399 4.920 5.009 4.601 5.081 11950.182 2117.064
∆ -5.114 0.493 110.711 3.779 162.094 3.967 136.802 4.698 105.703 4.122 21005.81 680.833
Bias of �� -0.244 0.117 -2.831 0.539 3.456 -0.601 -0.079 0.009 -0.399 0.081 11945.182 2112.064
Variation
of �� 66180.821 1667.105 73405.722 2399.038 192667.681 5756.227 4227.460 162.986 1404.661 52.319 349400889598 266405627
Bias of∆ -15.114 -0.507 100.711 2.779 152.094 2.967 126.802 3.698 95.702 3.122 20995.808 679.833
Variation ∆ 92615.276 89.361 44360.843 44.647 58261.801 58.079 40842.794 39.599 44614.617 40.817 999140174892 28399851
%negative
variance 57.71% 54.76% 0% 0% 17.87% 29.03% 10.5% 8.14% 0% 0% 0% 0%
N = T =
10
𝝈𝝐 =
1
�� 4.998 4.995 4.998 4.995 4.999 4.994 4.999 4.994 4.998 4.995 4.997 4.993
∆ 10.005 1.008 10.005 1.008 10.005 1.007 10.005 1.007 10.005 1.007 10.005 1.007
Bias of �� -0.002 -0.005 -0.002 -0.005 -0.001 -0.006 -0.000 -0.006 -0.002 -0.005 -0.003 -0.007
Variation
of �� 1.126 0.106 1.126 0.106 1.139 0.107 1.156 0.108 1.126 0.106 1.129 0.118
Bias of∆ 0.005 0.008 0.005 0.008 0.005 0.007 0.005 0.007 0.005 0.007 0.005 0.007
Variation ∆ 27.540 0.245 27.539 0.245 27.532 0.245 27.532 0.245 27.532 0.245 27.539 0.245
%negative
variance 0.04% 0% 0% 0% 0% 0.01% 0% 0.01% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 7.036 4.593 4.904 5.014 4.544 5.082 4.905 5.015 4.760 5.042 40.364 9.359
∆ 9.639 0.982 13.597 1.036 22.982 0.842 21.277 0.913 13.122 1.228 44.348 2.067
Bias of �� 2.036 -0.407 -0.096 0.014 -0.456 0.082 -0.095 0.015 -0.239 0.042 35.364 4.359
Variation
of �� 56797.744 2247.801 1450.777 55.129 482.679 18.692 1498.749 57.756 384.994 14.842 9115.353 119.157
Bias of∆ -0.361 -0.018 3.597 0.035 12.982 -0.158 11.277 -0.087 3.122 0.228 34.348 1.067
Variation ∆ 489.029 1.124 334.525 0.987 246.004 1.052 241.109 0.953 324.375 0.631 6736.187 4.151
%negative
variance 37.58% 16.95% 0% 0% 0.21% 16.68% 0% 9.19% 0% 0% 0% 0%
𝝈𝝐 =
10
�� 2.019 5.527 4.278 5.123 -1.281 6.389 5.477 4.892 4.909 5.003 41506.528 2615.932
∆ 9.319 0.986 30.135 1.521 80.142 1.514 57.755 2.399 28.416 1.837 53391.14 602.671
Bias of �� -2.981 0.527 -0.722 0.123 -6.281 1.389 0.477 -0.108 -0.091 0.003 41501.528
2610.932
Variation
of �� 42098.527 1449.086 8012.731 280.192 358948.87 18144.04 2155.271 78.972 217.859 5.158 23170464851 19584387
Bias of∆ -0.681 -0.014 20.135 0.521 70.142 0.514 47.755 1.399 18.416 0.837 53381.136 601.671
Variation ∆ 4394.206 6.668 2311.897 4.252 2301.606 4.749 2019.998 3.407 2295.829 3.164 40594028775 1107469
%negative
variance 49.71% 39.46% 0% 0% 0.7% 23.05% 0.24% 5.8% 0% 0% 0% 0%
Comparing MINQUE and IAUE estimates of variance components 2753
Table (A.3) Results of RCR Estimation When 𝛽0~𝑁(5,10) and 𝛽1~𝑁(5,1)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T =
15
𝝈𝝐 = 1
�� 5.004 4.999 5.004 4.999 5.004 4.999 5.004 4.999 5.004 4.999 5.004 4.999
∆ 10.006 0.999 10.006 0.999 10.006 0.999 10.006 0.999 10.006 0.999 10.006 0.999
Bias of �� 0.004 -0.001 0.004 -0.001 0.004 -0.001 0.004 -0.001 0.004 -0.001 0.004 -0.001
Variation
of �� 0.727 0.067 0.727 0.067 0.727 0.067 0.727 0.067 0.727 0.067 0.727 0.067
Bias of∆ 0.006 -0.000 0.006 -0.000 0.006 -0.000 0.006 -0.000 0.006 -0.000 0.006 -0.000
Variation ∆ 16.421 0.149 16.421 0.149 16.421 0.149 16.421 0.149 16.421 0.149 16.421 0.149
%negative variance 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 4.773 5.055 4.996 5.006 4.936 5.017 4.925 5.022 4.981 5.008 100.479 9.782
∆ 9.889 0.998 10.661 1.005 15.647 0.879 15.002 0.912 10.608 1.075 84.661 1.955
Bias of �� -0.227 0.055 -0.005 0.006 -0.064 0.017 -0.075 0.022 -0.019 0.008 95.479 4.782
Variation
of �� 3046.032 116.873 40.165 1.395 27.226 1.324 38.566 2.365 7.244 0.393 83855.459 138.117
Bias of∆ -0.111 -0.002 0.661 0.005 5.647 -0.121 5.002 -0.089 0.608 0.075 74.661 0.955
Variation ∆ 122.583 0.466 101.723 0.451 100.406 0.533 93.712 0.493 98.985 0.345 51408.168 4.852
%negative
variance 18.64% 4.52% 0% 0% 0% 8.28% 0% 5.54% 0% 0% 0% 0%
𝝈𝝐
=
10
�� 5.007 4.999 4.793 5.037 4.509 5.092 4.956 5.007 4.982 5.001 84805.38 5839.10
∆ 10.149 1.014 17.969 1.168 63.640 1.053 57.648 1.323 17.264 1.418 94449.02 1244.769
Bias of �� 0.007 -0.001 -0.207 0.037 -0.491 0.092 -0.044 0.007 -0.018 0.001 84800.38 5834.10
Variation
of �� 3912.271 138.359 704.283 25.123 575.278 22.368 20.156 0.909 5.639 0.242 26655708488 58088742
Bias of∆ 0.149 0.014 7.969 0.168 53.640 0.053 47.648 0.323 7.264 0.418 94439.017 1243.769
Variation ∆ 915.515 1.907 567.209 1.536 1815.486 1.625 1526.727 1.613 554.138 1.152 30681549209 2501753
%negative
variance 42.039% 25.375% 0% 0% 0% 18.383% 0% 11.799% 0% 0% 0% 0%
N = T =
20
𝝈𝝐 =
1
�� 5.007 5.0009 5.007 5.0009 5.007 5.0009 5.007 5.0009 5.007 5.0009 5.007 5.0009
∆ 9.942 1.003 9.942 1.003 9.942 1.003 9.942 1.003 9.942 1.003 9.942 1.003
Bias of �� 0.007 0.0009 0.007 0.0009 0.007 0.0009 0.007 0.0009 0.007 0.0009 0.007 0.0009
Variation
of �� 0.507 0.052 0.507 0.052 0.507 0.052 0.507 0.052 0.507 0.052 0.507 0.052
Bias of∆ -0.057 0.003 -0.057 0.003 -0.057 0.003 -0.057 0.003 -0.057 0.003 -0.057 0.003
Variation ∆ 11.378 0.107 11.378 0.107 11.378 0.107 11.378 0.107 11.378 0.107 11.378 0.107
%negative
variance 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 5.141 4.973 5.078 4.984 5.216 4.955 5.027 4.993 5.019 4.995 29.768 5.925
∆ 10.014 1.003 10.204 1.003 12.273 0.939 12.094 0.948 10.270 1.018 29.194 1.186
Bias of �� 0.141 -0.026 0.078 -0.015 0.216 -0.044 0.027 -0.006 0.019 -0.004 24.768 0.925
Variation
of �� 56.642 1.800 12.752 0.500 251.964 10.286 1.564 0.089 1.241 0.076 11783.019 13.825
Bias of∆ 0.014 0.003 0.204 0.003 2.273 -0.060 2.094 -0.051 0.270 0.018 19.194 0.186
Variation ∆ 60.629 0.232 56.077 0.232 62.426 0.288 59.245 0.275 54.624 0.211 6859.306 0.642
%negative
variance 8.114% 0.371% 0% 0% 0% 3.457% 0% 2.385% 0% 0% 0% 0%
𝝈𝝐 =
10 �� 5.196 4.959 5.163 4.970 5.523 4.903 4.695 5.070 5.023 4.998 221966.45 5139.51
∆ 10.193 1.020 14.329 1.071 53.169 0.909 49.580 1.074 14.177 1.250 231207.123 1035.703
Bias of �� 0.196 -0.040 0.163 -0.029 0.523 -0.096 -0.304 0.070 0.023 -0.001 221961.45 5134.51
Variation
of �� 1161.808 49.628 55.619 2.111 938.508 35.310 541.203 24.080 3.161 0.145 156552817279 50276565
Bias of∆ 0.193 0.020 4.329 0.071 43.169 -0.090 39.580 0.074 4.177 0.250 231197.123 1034.703
Variation ∆ 489.722 1.027 314.927 0.892 1623.762 1.037 1398.915 0.981 308.220 0.584 177216288388 2053979
%negative
variance 35.728% 14.328% 0% 0% 0% 15.657% 0% 11.242% 0% 0% 0% 0%
2754 Souha K. Badr et al.
Table (A.3) Results of RCR Estimation When 𝛽0~𝑁(5,10) and 𝛽1~𝑁(5,1)
Table (A.4) Results of RCR Estimation When 𝛽0~𝑁(1,1) and 𝛽1~𝑁(1,5)
SWAMY R W Z
MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T = 5
𝝈𝝐 = 1
�� 0.784 1.067 0.798 1.056 17.419 -1.681 0.188 1.103 0.232 1.149 -0.023 0.709
∆ 0.996 4.959 1.708 4.959 2.037 2.820 1.945 2.822 2.140 3.244 1.721 2.827
Bias of �� -0.216 0.067 -0.202 0.056 16.419 -2.681 -0.812 0.103 -0.768 0.149 -1.023 -0.291
Variation
of �� 6978.418 265.869 8072.326 293.388 2176060.20 52456.47 6831.249 265.171 5772.118 233.951 5771.543 233.910
Bias of∆ -0.004 -0.040 0.708 -0.040 1.037 -2.179 0.945 -2.178 1.140 -1.756 0.721 -2.173
Variation ∆ 10.403 12.788 6.689 12.787 7.169 13.289 6.189 13.278 5.458 11.146 6.645 13.253
%negative variance 43.65% 0.1% 0% 0% 14.35% 17.28% 9.88% 8% 0% 0% 0% 0%
𝝈𝝐
=
5
�� -8.532 2.681 -0.767 1.324 8.478 -0.494 0.602 1.086 0.872 1.033 2.311 1.354
∆ -8.851 4.701 60.336 5.889 71.451 4.310 63.746 4.578 57l899 4.814 65.173 5.152
Bias of �� -7.532 1.681 -1.767 0.324 7.478 -1.494 -0.398 0.086 -0.128 0.033 1.331 0.354
Variation
of �� 427016.31 13359.49 6409.330 231.502 382936.08 16025.21 374.403 15.368 149.516 6.519 792.656 22.613
Bias of∆ -9.851 -0.299 59.336 0.889 70.451 -0.689 62.747 -0.422 565.899 -0.186 64.173 0.152
Variation ∆ 29987.926 81.406 14307.072 61.987 14317.955 62.806 13651.791 58.811 14170.992 56.796 23942.568 62.322
%negative
variance 59.69% 33.2% 0% 0% 19.09% 30.99% 14.01% 5.75% 0% 0% 0% 0%
𝝈𝝐 =
10
�� 1.695 0.886 -33.672 7.225 -7.292 2.522 1.495 0.913 1.211 0.968 94.038 70.330
∆ -28.973 3.815 195.154 9.457 262.075 8.321 214.986 9.649 190.717 9.051 1017.729 118.670
Bias of �� 0.695 -0.114 -34.672 6.225 -8.292 1.522 0.495 -0.087 0.211 -0.032 93.038 69.330
Variation
of �� 18112.328 506.164 10813135.5 356181.5 314150.06 10065.42 3475.557 105.124 399.126 10.616 2996315.6 678493.6
Bias of∆ -29.973 -1.185 194.154 4.457 261.075 3.321 213.986 4.649 189.717 4.051 1016.729 113.670
Variation ∆ 352317.988 479.053 178591.585 292.279 275702.967 407.389 172626.330 279.014 179022.366 286.344 152629107 1409920
%negative
variance 61.6% 49.54% 0% 0% 24.01% 34.3% 18.05% 7.18% 0.06% 0% 0% 0%
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T =
50
𝝈𝝐 = 1
�� 5.0002 4.996 5.0002 4.996 5.0002 4.996 5.0002 4.996 5.0002 4.996 5.0002 4.996
∆ 9.986 1.004 9.986 1.004 9.986 1.004 9.986 1.004 9.986 1.004 9.986 1.004
Bias of �� 0.0002 -0.003 0.0002 -0.003 0.0002 -0.003 0.0002 -0.003 0.0002 -0.003 0.0002 -0.003
Variation
of �� 0.214 0.020 0.214 0.020 0.214 0.020 0.214 0.020 0.214 0.020 0.214 0.020
Bias of∆ -0.013 0.004 -0.013 0.004 -0.013 0.004 -0.013 0.004 -0.013 0.004 -0.013 0.004
Variation ∆ 16.644 0.168 16.644 0.168 16.644 0.168 16.644 0.168 16.644 0.168 16.644 0.168
%negative variance 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 5.017 5.0003 5.017 5.0003 5.017 5.0003 5.017 5.0003 5.017 5.0003 5.017 5.0003
∆ 10.292 1.024 10.292 1.024 10.292 1.024 10.292 1.024 10.292 1.024 10.292 1.024
Bias of �� 0.017 0.0003 0.017 0.0003 0.017 0.0003 0.017 0.0003 0.017 0.0003 0.017 0.0003
Variation
of �� 0.301 0.022 0.301 0.022 0.301 0.022 0.301 0.022 0.301 0.022 0.301 0.022
Bias of∆ 0.292 0.024 0.292 0.024 0.292 0.024 0.292 0.024 0.292 0.024 0.292 0.024
Variation ∆ 7.193 0.067 7.193 0.067 7.193 0.067 7.193 0.067 7.193 0.067 7.193 0.067
%negative
variance 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
𝝈𝝐
=
10
�� 5.060 4.985 5.059 4.985 5.057 4.986 5.206 4.974 5.058 4.985 66460.380 496.900
∆ 9.973 0.993 10.086 0.993 13.788 0.962 13.748 0.964 10.123 1.0001 57519.522 93.926
Bias of �� 0.060 -0.014 0.059 -0.014 0.057 -0.013 0.206 -0.025 0.058 -0.014 66455.380 491.900
Variation
of �� 0.603 0.035 0.602 0.035 0.600 0.035 12.355 0.237 0.601 0.035 129413933150 6504362
Bias of∆ -0.026 -0.006 0.086 -0.006 3.788 -0.037 3.748 -0.035 0.123 0.0001 57509.522 92.926
Variation ∆ 6.437 0.052 6.409 0.052 24.873 0.051 24.664 0.051 6.409 0.053 5.175𝑒9 1.200𝑒4
%negative
variance 3.693% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
Comparing MINQUE and IAUE estimates of variance components 2755
Table (A.4) Results of RCR Estimation When 𝛽0~𝑁(1,1) and 𝛽1~𝑁(1,5)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T =
10
𝝈𝝐 = 1
�� 1.009 0.985 1.096 0.970 1.139 0.951 2.162 0.726 1.001 0.986 0.894 0.829
∆ 0.997 4.989 1.042 4.988 1.189 4.238 1.160 4.240 1.196 4.386 1.043 4.239
Bias of �� 0.009 -0.015 0.096 -0.029 0.139 -0.049 1.162 -0.274 0.001 -0.014 -0.106 -0.171
Variation
of �� 4.171 0.597 100.147 3.623 500.345 28.105 15416.010 768.742 0.579 0.518 0.621 0.577
Bias of∆ -0.003 -0.012 0.042 -0.012 0.189 -0.761 0.160 -0.758 0.196 -0.614 0.043 -0.761
Variation ∆ 1.080 5.628 0.967 5.628 0.799 8.038 0.816 8.027 0.778 6.909 0.966 8.040
%negative variance 15.18% 0% 0% 0% 0.06% 5.82% 0% 3.01% 0% 0% 0% 0%
𝝈𝝐
=
5
�� 0.008 1.188 0.467 1.093 0.807 1.039 0.736 1.041 0.904 1.010 6.575 1.841
∆ 0.827 5.025 8739 5.029 21.573 2.646 19.341 2.732 9.256 3.124 34.511 3.931
Bias of �� -0.992 0.288 -0.533 0.093 -0.193 0.039 -0.264 0.041 -0.096 0.010 5.575 0.841
Variation
of �� 13510.282 519.757 2746.641 108.491 1005.785 38.101 2254.118 80.582 26.198 1.365 301.015 6.954
Bias of∆ -0.173 0.025 7.739 0.029 20.573 -2.354 18.341 -2.268 8.256 -1.876 33.511 -1.069
Variation ∆ 464.134 9.769 225.210 9.733 195.770 12.681 175.587 12.238 215.741 10.368 3709.283 10.127
%negative
variance 54.45% 0.97% 0% 0% 0.59% 21.6% 0.15% 12.56% 0% 0% 0% 0%
𝝈𝝐 =
10
�� -8.209 2.652 -0.422 1.287 0.307 1.122 2.269 0.745 1.095 0.969 6516.748 1335.972
∆ -0.684 4.926 27.549 5.082 81.585 3.294 72.509 3.677 27.508 3.634 41582.47 1628.939
Bias of �� -9.209 1.652 -1.422 0.287 -0.693 0.122 1.269 -0.255 0.095 -0.031 6515.748 1334.972
Variation
of �� 357987.84 11851.81 15554.002 695.089 1025.717 38.081 10213.152 384.109 378.546 11.966 324371655 6123064
Bias of∆ -1.684 -0.074 26.549 0.082 80.585 -1.706 71.509 -1.323 26.508 -1.366 41581.471 1623.939
Variation ∆ 5215.377 22.009 2506.748 20.143 2409.734 22.207 2116.426 20.101 2463.606 19.463 8954740995 6073156
%negative
variance 57.25% 12.53% 0% 0% 0.35% 22.47% 0.03% 12.25% 0% 0% 0% 0%
N = T =
15
𝝈𝝐 =
1
�� 0.995 0.998 0.996 0.997 0.991 0.996 0.995 0.997 0.996 0.997 0.972 0.963
∆ 0.998 5.015 1.002 5.015 1.034 4.846 1.028 4.847 1.036 4.879 1.002 4.846
Bias of �� -0.005 -0.002 -0.004 -0.003 -0.009 -0.004 -0.005 -0.003 -0.004 -0.003 -0.028 -0.037
Variation
of �� 0.149 0.335 0.123 0.334 11.163 0.665 0.227 0.338 0.122 0.334 0.136 0.358
Bias of∆ -0.002 0.015 0.002 0.015 0.034 -0.154 0.028 -0.153 0.036 -0.121 0.002 -0.154
Variation ∆ 0.437 3.622 0.427 3.622 0.394 4.296 0.397 4.293 0.392 4.014 0.427 4.297
%negative
variance 3.371% 0% 0% 0% 0% 1.086% 0% 0.586% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 1.272 0.953 1.144 0.976 1.980 0.798 0.792 1.052 1.117 0.981 100.363 2.864
∆ 0.880 5.029 3.675 5.029 15.755 2.601 14.284 2.677 4.185 3.057 382.601 4.763
Bias of �� 0.272 -0.047 0.144 -0.024 0.980 -0.202 -0.208 0.052 0.117 -0.019 99.363 1.864
Variation
of �� 283.105 10.163 52.427 1.722 1610.709 73.327 855.091 42.525 47.531 1.547 50986.224 13.223
Bias of∆ -0.119 0.029 2.675 0.029 14.755 -2.399 13.284 -2.323 3.185 -1.943 381.601 -0.237
Variation ∆ 51.413 8.469 28.904 8.469 123.248 7.935 104.762 7.759 27.193 6.927 4.537967e+05 8.901121e+00
%negative
variance 50.912% 0% 0% 0% 0% 17.830 0% 12.892% 0% 0% 0% 0%
𝝈𝝐 =
10 �� -0.532 1.298 1.891 0.826 1.112 0.964 0.578 1.099 0.909 1.022 33791.06 2417.51
∆ 0.693 4.965 12.209 4.980 64.526 2.883 58.057 3.213 12.692 3.276 178926.357 2637.287
Bias of �� -1.532 0.298 0.891 -0.174 0.112 -0.036 -0.422 0.099 -0.091 0.022 33790.06 2416.51
Variation
of �� 5881.109 219.431 10631.766 423.454 678.102 38.688 690.866 34.804 148.043 5.407 4629160902 12014951
Bias of∆ -0.307 -0.035 11.209 -0.019 63.526 -2.117 57.057 -1.787 11.692 -1.724 178925.357 2632.287
Variation ∆ 910.095 9.447 407.998 9.316 1664.037 13.008 1399.474 11.682 395.059 10.699 102755061124 10281354
%negative
variance 54% 2.657% 0% 0% 0% 20.886% 0% 13.614% 0% 0% 0% 0%
2756 Souha K. Badr et al.
Table (A.4) Results of RCR Estimation When 𝛽0~𝑁(1,1) and 𝛽1~𝑁(1,5)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T =
20
𝝈𝝐 = 1
�� 1.003 1.009 1.003 1.009 1.005 1.008 0.961 1.018 1.003 1.009 1.0008 1.0005
∆ 0.994 4.991 0.994 4.991 0.998 4.972 0.997 4.972 0.998 4.975 0.994 4.972
Bias of �� 0.003 0.009 0.003 0.009 0.005 0.008 -0.038 0.018 0.003 0.009 0.0008 0.005
Variation
of �� 0.075 0.248 0.075 0.248 0.087 0.249 9.583 0.707 0.075 0.248 0.077 0.251
Bias of∆ -0.005 -0.005 -0.005 -0.008 -0.001 -0.027 -0.002 -0.027 -0.001 -0.024 -0.005 -0.027
Variation ∆ 0.241 2.620 0.240 2.620 0.237 2.701 0.237 2.701 0.237 2.669 0.240 2.701
%negative variance 0.371 % 0 % 0% 0% 0 % 0.114 % 0 % 0.085 % 0 % 0 % 0 % 0 %
𝝈𝝐
=
5
�� 1.130 0.985 0.997 1.012 -31.092 7.579 0.669 1.061 0.997 1.012 38.285 1.669
∆ 0.866 4.973 2.429 4.973 14.061 2.684 12.898 2.736 2.905 3.133 145.204 3.726
Bias of �� 0.130 -0.0145 -0.002 0.012 -32.092 6.579 -0.330 0.061 -0.002 0.012 37.285 0.669
Variation
of �� 138.036 4.948 0.794 0.276 7096232.4 298019.5 1163.453 32.641 0.758 0.275 4227.075 1.790
Bias of∆ -0.133 -0.026 1.429 -0.026 13.061 -2.315 11.898 -2.263 1.905 -1.866 144.204 -1.273
Variation ∆ 26.662 3.074 13.868 3.074 113.286 8.053 96.516 7.793 11.804 5.888 52573.214 4.104
%negative
variance 47.642 % 0 % 0% 0% 0 % 18.785 0 % 13.4 % 0 % 0 % 0 % 0 %
𝝈𝝐
=
10
�� 3.501 0.546 0.995 0.995 1.557 0.883 0.913 1.010 1.014 0.991 21388.233 1460.862
∆ 0.906 5.015 7.913 5.015 59.313 2.688 54.608 2.914 8.437 3.132 110669.983 1478.983
Bias of �� 2.5018 -0.453 -0.004 -0.004 0.557 -0.116 -0.086 0.010 0.014 -0.008 21387.233 1459.862
Variation
of �� 45139.172 1439.135 4.097 0.399 1242.303 50.201 110.307 4.049 2.747 0.346 1203281340 3555518
Bias of∆ -0.093 0.0150 6.913 0.015 58.313 -2.311 53.608 -2.085 7.437 -1.867 110668.983 1473.983
Variation ∆ 363.200 5.042 165.201 5.041 1856.567 9.903 1612.585 9.010 157.164 7.682 27732470542 2730336
%negative
variance 52.557 % 0.0428 % 0% 0% 0 % 20.7 % 0 % 14.485 % 0 % 0 % 0 % 0%
N = T =
50
𝝈𝝐 =
1
�� 1.003 0.992 1.003 0.992 1.003 0.992 1.003 0.992 1.003 0.992 1.003 0.992
∆ 1.006 5.000 1.006 5.000 1.006 5.000 1.006 5.000 1.006 5.000 1.006 5.000
Bias of �� 0.003 -0.007 0.003 -0.007 0.003 -0.007 0.003 -0.007 0.003 -0.007 0.003 -0.007
Variation
of �� 0.024 0.103 0.024 0.103 0.024 0.103 0.024 0.103 0.024 0.103 0.024 0.103
Bias of∆ 0.006 0.000 0.006 0.000 0.006 0.000 0.006 0.000 0.006 0.000 0.006 0.000
Variation ∆ 0.057 0.056 0.057 1.056 0.057 1.056 0.057 1.056 0.057 1.056 0.057 1.056
%negative
variance 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 1.001 0.994 1.001 0.994 -1.106 1.388 1.426 0.912 1.001 0.995 1.001 0.995
∆ 1.000 5.028 1.133 5.028 6.392 3.970 6.249 3.978 1.343 4.179 1.343 4.179
Bias of �� 0.001 -0.005 0.001 -0.005 -2.106 0.388 0.426 -0.087 0.001 -0.004 0.001 -0.004
Variation
of �� 0.110 0.101 0.109 0.101 4201.232 146.093 76.427 2.771 0.109 0.101 0.109 0.101
Bias of∆ 2.171𝑒−5 0.028 0.133 0.028 5.392 -1.029 5.249 -1.021 0.343 -0.820 0.343 -0.820
Variation ∆ 1.570 1.148 1.158 1.148 93.763 5.093 89.497 5.036 0.847 3.601 0.847 3.601
%negative
variance 21% 0% 0% 0% 0% 9.6% 0% 8% 0% 0% 0% 0%
𝝈𝝐 =
10 �� 1.047 0.988 1.047 0.988 1.121 0.975 0.117 1.200 1.046 0.988 1.046 0.988
∆ 1.002 5.052 2.242 5.052 43.811 3.024 42.410 3.096 2.659 3.440 2.659 3.440
Bias of �� 0.047 -0.011 0.047 -0.011 0.121 -0.024 -0.882 0.200 0.046 -0.011 0.046 -0.011
Variation
of �� 0.423 0.111 0.398 0.111 7.766 0.364 1163.976 47.642 0.397 0.111 0.397 0.111
Bias of∆ 0.002 0.052 1.242 0.052 42.811 -1.975 41.410 -1.903 1.659 -1.559 1.659 -1.559
Variation ∆ 18.580 1.333 9.020 1.333 22.82.727 7.240 2178.037 6.941 7.396 4.959 7.396 4.959
%negative
variance 41.4% 0% 0% 0% 0% 21.7% 0% 16.8% 0% 0% 0% 0%
Comparing MINQUE and IAUE estimates of variance components 2757
Table (A.5) Results of RCR Estimation When 𝛽0~𝑁(10,10) and 𝛽1~𝑁(10,0)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T = 5
𝝈𝝐 = 1
�� 7.615 10.560 9.588 10.092 10.114 9.999 10.408 9.931 10.212 9.979 6.309 3.601
∆ 9.599 -0.011 10.231 0.069 6.482 0.078 6.318 0.082 6.652 0.681 6.052 0.085
Bias of �� -2.385 0.560 -0.412 0.092 0.114 -0.000 0.408 -0.69 0.212 -0.021 -3.691 -6.399
Variation
of �� 23747.446 1282.242 4300.622 135.109 1462.088 44.905 453.421 12.860 285.539 2.239 295.664 26.698
Bias of∆ -0.401 -0.011 0.232 0.069 -3.518 0.078 -3.682 0.082 -3.348 0.681 -3.948 0.085
Variation ∆ 141.207 0.041 125.148 0.019 118.554 0.020 118.709 0.018 114.657 0.171 121.665 0.017
%negative variance 19.6% 59.55% 0% 0% 21.55% 22.71% 12.54% 7.1% 0% 0% 0% 0%
𝝈𝝐
=
5
�� 13.856 9.365 11.914 9.705 12.197 9.737 10.894 9.912 10.705 9.960 32.724 20.059
∆ 0.044 -0.284 55.204 1.727 64.302 1.991 60.057 2.085 53.472 2.312 65.683 3.459
Bias of �� 3.856 -0.635 1.914 -0.295 2.197 -0.263 0.894 -0.088 0.705 -0.039 22.724 10.059
Variation
of �� 84877.103 2390.529 39278.504 1320.559 48653.911 1102.417 6313.202 63.894 5592.819 20.966 41342.409 4142.664
Bias of∆ -9.956 -0.284 45.204 1.727 54.302 1.992 50.057 2.085 43.472 2.312 55.683 3.459
Variation ∆ 23183.807 25.487 11815.255 12.048 12037.706 12.801 11322.968 11.056 11857.737 10.204 24972.939 66.329
%negative
variance 57.48% 59.6% 0% 0% 22.28% 23.18% 13.31% 7.42% 0% 0% 0% 0%
𝝈𝝐
=
10
�� 32.545 5.968 16.325 8.876 13.782 9.510 10.754 9.959 10.419 10.037 14290.438 8111.947
∆ -28.786 -1.089 202.966 6.939 244.385 8.033 227.719 8.404 199.257 7.425 13111.43 1376.92
Bias of �� 22.545 -4.032 6.325 -1.124 3.782 -0.489 0.754 -0.040 0.419 0.037 14280.438 8101.947
Variation
of �� 5980799.3 260540.2 305368.16 11442.11 168788.528 4908.427 22430.02 1312.31 20073.156 1179.716 13570036160 1825606951
Bias of∆ -38.786 -1.089 192.966 6.939 234.385 8.033 217.719 8.404 189.257 7.425 13101.43 1376.92
Variation ∆ 353051.347 409.610 171699.425 194.263 175649.570 206.222 163696.456 178.343 172419.601 186.857 15359361643 68279481
%negative
variance 59.81% 59.55% 0% 0% 22.24% 23.1% 13% 7.29% 0% 0% 0% 0%
N = T =
10
𝝈𝝐 =
1
�� 9.868 10.024 10.041 9.992 9.988 10.001 9.964 10.007 9.992 10.001 6.363 4.143
∆ 10.062 -3.631e-05 10.062 0.012 5.302 0.014 5.222 0.017 5.345 0.561 4.789 0.017
Bias of �� -0.132 0.024 0.041 0.008 -0.012 0.001 -0.036 0.006 -0.008 0.001 -3.637 -5.857
Variation
of �� 842.753 33.142 19.540 0.578 42.989 1.582 15.266 0.417 7.605 0.126 17.598 27.326
Bias of∆ 0.061 3.631e-05 0.062 0.012 -4.698 0.014 -4.778 0.017 -4.655 0.561 -5.210 0.017
Variation ∆ 29.189 9.35e-04 29.174 4.344e-04 38.26 .4 80e-04 38.964 0.000 37.814 0.228 42.876 5.008e-04
%negative
variance 0.16% 55.76% 0% 0% 0.28% 21.39% 0.2% 12.76% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 9.333 10.091 10.545 9.901 9.477 10.108 9.738 10.056 9.25 10.039 34.404 16.465
∆ 9.932 -0.001 14.665 0.297 25.844 0.346 23.669 0.434 12.521 0.876 23.569 1.495
Bias of �� -0.667 0.091 0.545 -0.099 -0.523 0.108 -0.262 0.056 -0.175 0.039 24.404 6.465
Variation
of �� 7666.998 335.578 4595.949 165.068 633.662 25.851 379.376 15.596 264.809 11.278 3072.953 256.880
Bias of∆ -0.068 -0.001 4.665 0.297 15.844 0.346 13.669 0.434 2.521 0.876 13.569 1.495
Variation ∆ 578.375 0.584 373.991 0.272 236.612 0.304 245.706 0.0264 374.037 0.170 726.736 2.572
%negative
variance 38.17% 55.75% 0% 0% 0.3% 23.03% 0.2% 13.73% 0% 0% 0% 0%
𝝈𝝐 =
10
�� 5.953 10.800 13.293 9.314 9.649 10.052 10.209 9.937 10.383 0.903 19248.268 9351.227
∆ 9.643 -0.004 37.028 1.187 90.929 1.397 82.077 1.754 34.920 1.744 12091.39 1117.366
Bias of �� 4.047 0.800 3.293 -0.686 -0.351 0.052 0.209 -0.063 0.384 -0.097 19238.268 9341.227
Variation
of �� 131627.599 4573.942 29128.773 1173.811 1841.826 77.354 807.742 34.735 328.339 16.612 1565714833 219644006
Bias of∆ -0.357 -0.004 27.028 2.287 80.929 1.397 72.077 1.754 24.920 1.744 12081.394 1117.366
Variation ∆ 6967.775 9.351 3542.718 4.345 2607.445 4.904 2538.349 4.238 3569.485 3.243 665060845 3340543
%negative
variance 50.58% 55.76% 0% 0% 0.3% 23.53% 0.21% 13.99% 0% 0% 0% 0%
2758 Souha K. Badr et al.
Table (A.5) Results of RCR Estimation When 𝛽0~𝑁(10,10) and 𝛽1~𝑁(10,0)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T =
15
𝝈𝝐 = 1
�� 10.071 9.982 9.989 10.001 9.966 10.006 9.990 10.002 9.987 10.002 5.979 4.149
∆ 9.944 -0.000 9.944 0.005 5.040 0.006 4.985 0.009 5.057 0.551 4.499 0.006
Bias of �� 0.071 -0.018 -0.010 0.001 -0.034 0.006 -0.009 0.002 -0.013 0.002 -4.021 -5.851
Variation
of �� 137.107 5.675 2.365 0.076 8.928 0.355 4.155 0.131 1.114 0.022 13.413 27.126
Bias of∆ 0.056 -0.000 -0.056 0.005 -4.959 0.006 -5.015 0.009 -4.942 0.551 -5.501 0.006
Variation ∆ 16.576 1.986e-04 16.576 9.10e-05 28.486 1.123e-04 28.943 1.077e-04 28.319 0.231 33.096 1.007
%negative variance 0% 55.84 0% 0% 0% 22.61% 0% 14.54% 0% 0% 0% 0%
𝝈𝝐
=
5
�� 10.496 9.876 10.667 9.846 10.453 9.885 10.630 9.853 10.561 9.865 66.794 18.963
∆ 9.715 -0.005 10.788 0.134 21.652 0.157 20.208 0.221 8.072 0.701 31.276 1.539
Bias of �� 0.496 -0.124 0.667 -0.154 0.453 -0.115 0.630 -0.147 0.561 -0.135 56.794 8.963
Variation
of �� 9221.823 442.783 5530.883 278.438 5478.876 276.591 5455.286 275.594 5367.169 272.546 12032.041 462.230
Bias of∆ -0.285 -0.005 0.788 0.134 11.652 0.157 10.208 0.221 -1.928 0.701 21.276 1.539
Variation ∆ 149.988 0.124 119.788 0.057 79.295 0.071 77.164 0.068 123.629 0.142 1009.874 2.187
%negative
variance 22.31% 55.84% 0% 0% 0% 23.64% 0% 15.3% 0% 0% 0% 0%
𝝈𝝐
=
10
�� 10.596 9.899 8.834 10.218 9.839 10.025 10.176 9.965 10.023 9.992 45572.36 13920.18
∆ 9.209 -0.019 19.717 0.535 76.195 0.627 70.102 0.896 17.805 1.109 25275.93 1489.561
Bias of �� 0.596 -0.101 -1.166 0.218 -0.160 0.025 0.176 -0.035 0.023 -0.008 45562.36 13910.18
Variation
of �� 6052.616 233.660 9665.184 347.718 497.735 17.907 404.664 14.357 34.549 1.576 4053009339 223964037
Bias of∆ -0.792 -0.019 9.717 0.535 66.195 0.627 60.102 0.896 7.805 1.109 25265.926 1489.561
Variation ∆ 1490.563 1.987 823.832 0.910 1331.139 1.151 1167.574 1.102 827.039 0.534 1324714240 2650099
%negative
variance 45.15% 55.84% 0% 0% 0% 24.67% 0% 15.79% 0% 0% 0% 0%
N = T =
20
𝝈𝝐 =
1
�� 9.990 10.001 9.993 10.0004 10.029 9.993 9.982 10.003 9.993 10.0004 5.775 4.220
∆ 10.05 1.524𝑒−5 10.053 0.002 5.100 0.003 5.055 0.005 5.105 0.546 4.550 0.002
Bias of �� 0.009 0.001 -0.006 0.0004 0.029 -0.006 -0.017 0.003 -0.006 0.0004 -4.224 -5.779
Variation
of �� 6.789 0.273 0.546 0.001 2.555 0.078 7.649 0.313 0.544 0.001 14.478 26.792
Bias of∆ 0.053 0.00001 0.053 0.002 -4.899 0.003 -4.944 0.005 -4.894 0.546 -5.449 0.002
Variation ∆ 12.087 5.613𝑒−5 12.087 2.492𝑒−5 26.599 3.498𝑒−5 26.976 4.416𝑒−5 26.547 0.235 31.349 2.519𝑒−5
%negative
variance 0% 55.5% 0% 0% 0% 24.471% 0% 16.557% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 9.685 10.062 9.978 10.003 10.140 9.972 9.849 10.031 9.993 10.0005 136.814 19.158
∆ 10.131 0.0003 10.357 0.073 20.331 0.083 19.195 0.134 6.977 0.623 56.042 1.458
Bias of �� -0.314 0.062 -0.021 0.003 0.140 -0.027 -0.150 0.031 -0.006 0.0005 126.814 9.158
Variation
of �� 717.974 28.753 41.381 1.541 40.630 1.555 50.623 2.296 1.420 0.033 28103.589 139.783
Bias of∆ 0.131 0.0003 0.357 0.073 10.331 0.083 9.195 0.134 -3.022 0.623 46.042 1.458
Variation ∆ 69.300 0.035 63.742 0.015 61.570 0.021 57.207 0.027 74.578 0.173 3956.074 1.924
%negative
variance 8.871% 55.5% 0% 0% 0% 24.814% 0% 16.757% 0% 0% 0% 0%
𝝈𝝐 =
10
�� 9.802 10.044 10.011 9.997 10.119 9.977 9.699 10.063 9.978 10.003 101729.75 14621.23
∆ 10.280 0.001 14.967 0.294 70.544 0.332 65.822 0.543 12.899 0.862 53068.826 1481.897
Bias of �� -0.197 0.044 0.011 -0.002 0.119 -0.022 -0.300 0.063 -0.021 0.003 101719.75 14611.23
Variation
of �� 5667.787 224.860 16.039 0.561 225.126 8.775 185.417 8.411 3.583 0.111 16533359114 210848071
Bias of∆ 0.280 0.001 4.967 0.294 60.544 0.332 55.822 0.543 2.899 0.862 53058.826 1481.897
Variation ∆ 575.550 0.561 364.800 0.249 1452.747 0.357 1276.926 0.453 370.941 0.154 4769889275 2203940
%negative
variance 36.385% 55.5% 0% 0% 0% 26% 0% 17.642% 0% 0% 0% 0%
Comparing MINQUE and IAUE estimates of variance components 2759
Table (A.5) Results of RCR Estimation When 𝛽0~𝑁(10,10) and 𝛽1~𝑁(10,0)
Table (A.6) Results of RCR Estimation When 𝛽0~𝑁(5,0) and 𝛽1~𝑁(5,1)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T = 5
𝝈𝝐 = 1
�� 5.079 4.977 1.149 5.832 5.492 4.916 5.347 4.884 4.985 5.004 3.063 1.863
∆ -0.397 0.976 1.965 0.982 2.356 0.472 2.193 0.476 2.523 1.066 1.932 0.479
Bias of �� 0.079 -0.023 -3.851 0.832 0.492 -0.084 0.347 -0.116 -0.015 0.004 -1.937 -3.137
Variation
of �� 613.956 21.066 217081.88 9123.07 8178.796 258.791 2256.190 198.561 109.696 3.394 111.916 9.515
Bias of∆ -0.397 -0.024 1.965 -0.018 2.356 -0.528 2.193 -0.524 2.523 0.066 1.932 -0.521
Variation ∆ 34.569 0.729 16.464 0.717 16.516 0.651 15.321 0.646 14.046 0.338 16.069 0.643
%negative variance 60.38% 5.71% 0% 0% %21.32 22.24 %12.57 %7.02 %0 %0 %0 %0
𝝈𝝐
=
5
�� 11.610 4.064 8.735 4.191 7.088 4.659 5.740 4.819 5.299 4.958 16.219 10.026
∆ -10.207 0.687 48.931 2.353 58.641 2.424 54.381 2.519 47.729 2.749 59.818 3.896
Bias of �� 6.610 -0.936 3.735 -0.809 2.088 -0.341 0.740 -0.181 0.299 -0.042 11.219 5.026
Variation
of �� 355983.736 9690.219 197638.796 8030.697 48939.471 1305.639 3653.827 286.838 482.052 15.080 9030.502 1044.520
Bias of∆ -10.207 -0.313 48.931 1.353 58.641 1.424 54.381 1.519 47.729 1.749 59.818 2.896
Variation ∆ 21496.707 30.748 10193.414 17.176 10481.720 17.827 9716.825 15.988 10181.562 14.927 23174.759 69.618
%negative
variance 60.44% 51.96% 0% 0% 22.55% 23.41% 13.55% 7.42% 0% 0% 0% 0%
𝝈𝝐
=
10
�� 4.274 5.566 -6.222 7.115 9.196 4.363 5.835 4.863 5.305 5.023 7099.283 4066.495
∆ -39.504 -0.146 196.583 7.515 238.113 8.463 221.381 8.841 192.757 7.866 13145.19 1380.224
Bias of �� -0.726 0.566 -11.222 2.115 4.196 -0.637 0.835 -0.137 0.305 0.023 7094.283 4061.495
Variation
of �� 40957.38 2879.85 808381.54 27746.52 160557.846 4128.049 8826.749 480.950 3529.257 68.224 3078169239 454146401
Bias of∆ -39.504 -1.146 196.583 6.515 238.113 7.463 221.381 7.841 192.757 6.866 13145.189 1379.244
Variation ∆ 345745.105 428.146 164687.750 211.972 168937.061 223.825 156870.383 195.600 165269.923 204.898 15345841886 68173474
%negative
variance 60.37% 57.55% 0% 0% 22.26% 23.25% 13.02% 7.43% 0% 0% 0% 0%
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T =
50
𝝈𝝐 = 1
�� 10.014 10.0004 10.014 10.0003 10.016 10.00006 9.835 10.0377 10.015 10.0001 0.031 4.464
∆ 9.751 -0.00008 9.751 0.0005 5.076 0.0006 5.060 0.001 5.077 0.535 4.537 0.0005
Bias of �� 0.014 0.0004 0.014 0.0003 0.016 5.575𝑒−5 -0.164 0.037 0.015 0.0001 -4.757 5.535
Variation
of �� 0.209 0.0001 0.208 0.0001 0.210 0.0001 20.150 0.880 0.208 0.0001 19.426 25.927
Bias of∆ -0.248 -0.00008 -0.248 -0.0005 -4.923 0.0006 -4.939 0.001 -4.922 0.535 -5.462 0.0005
Variation ∆ 5.797 1.275𝑒−7 5.797 6.355𝑒−8 2.839 1.436𝑒−7 2.838 7.491𝑒−7 2.839 0.031 2.825 6.355𝑒−8
%negative variance 0% 54.233% 0% 0% 0% 24.713% 0% 20.137% 0% 0% 0% 0%
𝝈𝝐
=
5
�� 10.012 10.002 10.023 10.0003 10.044 9.993 9.276 10.158 10.027 9.999 2832.960 20.414
∆ 9.464 -0.002 9.464 0.012 18.359 0.015 17.941 0.037 5.341 0.549 828.504 1.481
Bias of �� 0.012 0.002 0.023 0.0003 0.044 -0.006 -0.723 0.158 0.027 -0.0004 2822.960 10.414
Variation
of �� 0.358 0.005 0.304 0.003 0.554 0.023 538.872 23.574 0.308 0.003 8729769.997 109.230
Bias of∆ -0.535 -0.002 -0.535 0.012 8.359 0.015 7.941 0.037 -4.658 0.549 818.504 1.481
Variation ∆ 5.251 6.910𝑒−5 5.251 3.373𝑒−5 20.91 7.910𝑒−5 19.94 4.279𝑒−4 2.995 0.028 7.860𝑒4 0.221
%negative
variance 0% 54.475% 0% 0% 0% 25.575% 0% 20.204% 0% 0% 0% 0%
𝝈𝝐 =
10
�� 10.035 9.996 10.027 9.998 10.027 9.996 10.189 9.982 10.031 9.997 10.031 9.997
∆ 9.845 0.0003 9.912 0.057 59.440 0.061 57.817 0.146 6.688 0.587 6.688 0.587
Bias of �� 0.035 -0.003 0.027 -0.001 0.027 -0.003 0.189 -0.017 0.031 -0.002 0.031 -0.002
Variation
of �� 0.632 0.015 0.057 0.014 1.253 0.058 2069.557 89.195 0.600 0.014 0.600 0.014
Bias of∆ -0.154 0.0003 -0.087 0.057 49.440 0.061 47.817 0.146 -3.311 0.587 -3.311 0.587
Variation ∆ 24.862 0.004 24.755 0.002 1.021𝑒3 5.483𝑒−3 971.849 0.035 19.071 0.102 19.071 0.102
%negative
variance 3.934% 53.156% 0% 0% 0% 26.532% 0% 21.043% 0% 0% 0% 0%
2760 Souha K. Badr et al.
Table (A.6) Results of RCR Estimation When 𝛽0~𝑁(5,0) and 𝛽1~𝑁(5,1)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T =
10
𝝈𝝐 = 1
�� 5.016 4.999 6.527 4.717 5.073 4.983 4.838 5.033 5.009 5.002 3.186 2.056
∆ -0.003 1.009 0.309 1.009 0.829 0.461 0.747 0.464 0.872 1.011 0.314 0.464
Bias of �� 0.016 -0.001 1.527 -0.283 0.073 -0.017 -0.162 0.033 0.009 0.002 -1.814 -2.944
Variation
of �� 107.367 3.909 22448.893 786.046 1078.146 47.441 362.494 14.623 24.783 0.830 27.404 7.558
Bias of∆ -0.003 0.009 0.309 0.009 0.829 -0.539 0.747 -0.536 0.872 0.011 0.314 -0.536
Variation ∆ 0.634 0.248 0.289 0.248 0.258 0.384 0.239 0.381 0.189 0.119 0.286 0.382
%negative variance 56.1% 0.01% 0% 0% 0.28% 21.47% 0.21% 12.73% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 10.318 4.089 25.563 1.106 9.423 4.193 9.341 4.211 9.576 4.167 21.609 7.341
∆ -0.066 1.011 7.724 1.109 21.065 0.802 18.936 0.888 8.094 1.316 18.926 1.928
Bias of �� 5.318 -0.910 20.563 -3.894 4.423 -0.807 4.341 -0.789 4.576 -0.833 16.609 2.341
Variation
of �� 221544.680 7303.003 2251268.39 80001.79 211644.706 7062.807 210890.063 7029.873 210532.962 7015.654 211125.454 7082.889
Bias of∆ -0.066 0.011 7.724 0.096 21.065 -0.198 18.936 -0.112 8.094 0.316 18.926 0.928
Variation ∆ 396.195 1.475 180.957 1.240 158.091 1.331 146.911 1.214 172.123 0.741 623.601 2.602
%negative
variance 56.09% 20.54% 0% 0% 0.29% 22.55% 0.22% 13.29% 0% 0% 0% 0%
𝝈𝝐
=
10
�� -2.026 6.255 5.251 4.943 4.535 5.085 4.633 5.065 4.963 5.002 9501.623 4633.263
∆ -0.277 1.010 30.886 1.865 85.905 1.874 77.126 2.228 30.302 2.213 11926.7 1107.536
Bias of �� -7.026 1.255 0.251 -0.057 -0.465 0.085 -0.367 0.065 -0.037 0.002 9496.623 4628.263
Variation
of �� 496876.6 17539.5 1416.989 54.699 1955.485 78.465 750.000 28.335 142.503 5.020 383190940 54779790
Bias of∆ -0.277 0.010 30.886 0.865 85.905 0.874 77.126 1.228 30.302 1.213 11926.699 1106.536
Variation ∆ 6338.535 12.319 2895.139 7.412 2446.048 7.881 2292.884 6.897 2877.509 5.923 647998985 3310072
%negative
variance 56.1% 44.12% 0% 0% 0.3% 23.35% 0.23% 13.83% 0% 0% 0% 0%
N = T =
15
𝝈𝝐 =
1
�� 5.333 4.936 4.996 4.999 4.954 5.008 4.879 5.023 4.997 4.999 3.017 2.108
∆ -0.005 0.999 0.136 0.999 0.672 0.452 0.617 0.455 0.688 0.990 0.137 0.451
Bias of �� 0.333 -0.064 -0.004 -0.001 -0.046 0.008 -0.121 0.023 -0.003 -0.001 -1.983 -2.892
Variation
of �� 1089.503 38.026 0.371 0.078 403.579 16.305 87.698 3.256 0.091 0.068 3.248 6.819
Bias of∆ -0.005 -0.001 0.136 -0.001 0.672 -0.548 0.617 -0.545 0.688 -0.009 0.137 -0.549
Variation ∆ 0.126 0.154 0.055 0.154 0.167 0.319 0.142 0.318 0.152 0.072 0.055 0.320
%negative
variance 55.18% 0% 0% 0% 0% 22.15% 0% 14.32% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 0.399 5.848 5.235 4.952 4.902 5.017 4.943 5.011 5.010 4.996 32.042 9.427
∆ -0.123 1.001 3.411 1.009 16.939 0.603 15.538 0.665 3.943 1.122 26.233 1.938
Bias of �� -4.601 0.848 0.235 -0.048 -0.098 0.017 -0.057 0.011 0.010 -0.004 27.042 4.427
Variation
of �� 179706.656 6169.358 404.529 15.905 457.071 19.425 83.583 3.251 3.887 0.197 1674.441 47.264
Bias of∆ -0.123 0.001 3.411 0.009 16.939 -0.397 15.538 -0.335 3.943 0.122 26.233 0.938
Variation ∆ 78.461 0.514 34.382 0.496 104.023 0.644 88.531 0.587 30.831 0.263 1091.579 1.609
%negative
variance 55.18% 4.88% 0% 0% 0% 22.43% 0% 14.5% 0% 0% 0% 0%
𝝈𝝐 =
10
�� -2.431 6.362 5.039 4.983 4.742 5.043 4.995 5.000 4.954 5.004 22002.368 6779.398
∆ -0.657 0.993 13.576 1.251 70.147 1.075 64.236 1.337 13.729 1.539 24391.63 1452.068
Bias of �� -7.431 1.362 0.039 -0.017 -0.258 0.043 -0.005 0.000 -0.046 0.004 21997.368 6774.398
Variation
of �� 177217.189 7069.555 342.657 13.229 828.592 34.454 285.781 10.473 6.221 0.286 984401503 56427535
Bias of∆ -0.657 -0.007 13.576 0.251 70.147 0.075 64.236 0.337 13.729 0.539 24391.627 1451.068
Variation ∆ 1091.849 2.833 499.329 2.114 1955.753 2.300 1658.239 2.126 487.421 1.571 1172761594 2549327
%negative
variance 55.444% 31.199% 0% 0% 0% 23.622% 0% 15.156% 0% 0% 0% 0%
Comparing MINQUE and IAUE estimates of variance components 2761
Table (A.6) Results of RCR Estimation When 𝛽0~𝑁(5,0) and 𝛽1~𝑁(5,1)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T =
20
𝝈𝝐 = 1
�� 5.001 4.998 4.997 4.999 5.080 4.9833 5.230 4.952 4.997 4.999 2.914 2.145
∆ 0.001 1.003 0.081 1.003 0.625 0.456 0.581 0.458 4.997 0.991 0.081 0.455
Bias of �� 0.001 -0.001 -0.002 -0.0007 0.080 -0.016 0.230 -0.047 -0.002 -0.0007 -2.085 -2.854
Variation
of �� 7.115 0.271 0.050 0.052 72.912 2.847 411.837 16.795 0.048 0.052 3.599 6.736
Bias of∆ 0.001 0.003 0.081 0.003 0.625 -0.543 0.581 -0.541 0.629 -0.008 0.081 -0.544
Variation ∆ 0.043 0.113 0.019 0.113 0.186 0.302 0.162 0.300 0.177 0.053 0.019 0.303
%negative variance 54.8% 0% 0% 0% 0% 24.171% 0% 16.557% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 5.600 4.890 5.019 4.995 5.178 4.964 5.170 4.965 4.997 4.999 67.386 9.513
∆ 0.028 1.0005 2.042 1.0008 15.645 0.528 14.546 0.577 2.588 1.054 50.810 1.878
Bias of �� 0.600 -0.109 0.019 -0.004 0.178 -0.035 0.170 -0.034 -0.002 -0.0004 62.386 4.513
Variation
of �� 3275.317 109.839 3.829 0.179 101.369 3.943 348.718 14.471 0.895 0.081 7064.270 35.208
Bias of∆ 0.028 0.0005 2.042 0.0008 15.645 -0.471 14.546 -0.422 2.588 0.054 50.810 0.878
Variation ∆ 27.084 0.250 12.121 0.249 116.448 0.442 101.705 0.404 10.128 0.128 4311.114 1.161
%negative
variance 54.8% 0.8% 0% 0% 0% 24.214% 0% 16.6% 0% 0% 0% 0%
𝝈𝝐
=
10
�� 5.045 4.886 5.088 4.532 4.941 5.292 4.985 5.073 17.042 4.988 7053.785 49116.405
∆ 0.997 0.114 1.059 8.171 0.763 64.013 0.967 59.472 1.274 8.604 1429.831 51203.923
Bias of �� -0.113 0.045 -0.467 0.088 0.292 -0.058 0.073 -0.014 -0.011 0.001 49111.405 7048.785
Variation
of �� 548.834 19.126 931.221 34.401 207.648 8.087 309.286 13.336 3.060 0.157 4130473558 53314982
Bias of∆ 0.114 -0.002 8.171 0.059 64.013 -0.236 59.472 -0.032 8.604 0.274 51203.923 1428.831
Variation ∆ 433.351 1.084 193.945 0.919 1834.232 1.143 1603.609 1.057 185.390 0.516 4736818021 2221692
%negative
variance 54.8% 16.171% 0% 0% 0% 25.014% 0% 17.042% 0% 0% 0% 0%
N = T =
50
𝝈𝝐 =
1
�� 5.003 5.001 5.002 5.001 4.959 5.014 5.994 4.798 5.003 5.001 5.003 5.001
∆ -0.0002 0.995 0.0161 0.995 0.540 0.477 0.523 0.478 0.540 0.997 0.540 0.997
Bias of �� 0.003 0.001 0.002 0.001 -0.040 0.014 0.994 -0.201 0.003 0.001 0.003 0.001
Variation
of �� 0.004 0.021 0.003 0.021 2.207 0.195 2052.725 84.557 0.003 0.021 0.003 0.021
Bias of∆ -0.0002 -0.004 0.0161 -0.004 0.540 -0.522 0.523 -0.521 0.540 -0.002 0.540 -0.002
Variation ∆ 0.0002 0.123 0.0001 0.123 0.067 0.064
0.063 0.064 0.067 0.121 0.067 0.121
%negative
variance 52.387% 0% 0% 0% 0% 27.496% 0% 19.247% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 5.003 4.999 5.009 4.997 4.963 5.010 3.231 5.395 5.009 4.997 5.009 4.997
∆ -0.007 1.004 0.402 1.004 13.744 0.484 13.336 0.505 0.936 1.013 0.936 1.013
Bias of �� 0.0003 -0.0008 0.009 -0.002 -0.036 0.010 -1.768 0.395 0.009 -0.002 0.009 -0.002
Variation
of �� 0.158 0.026 0.096 0.024 1.413 0.119 8953.042 402.322 0.095 0.024 0.095 0.024
Bias of∆ -0.007 0.004 0.402 0.004 13.744 -0.515 13.336 -0.494 0.936 0.013 0.936 0.013
Variation ∆ 0.152 0.136 0.080 0.136 46.061 0.073 43.726 0.074 0.144 0.135 0.144 0.135
%negative
variance 53.395% 0% 0% 0% 0% 27.163% 0% 19.707% 0% 0% 0% 0%
𝝈𝝐 =
10
�� 5.020 5.0002 5.009 5.001 4.978 5.009 4.414 5.161 5.010 5.001 5.010 5.001
∆ 0.032 1.004 1.614 1.004 53.578 0.538 51.820 0.629 2.134 1.050 2.134 1.050
Bias of �� 0.020 0.0002 0.009 0.001 -0.021 0.009 -0.585 0.161 0.010 0.001 0.010 0.001
Variation
of �� 0.992 0.050 0.387 0.033 1.029 0.066 8206.602 362.642 0.384 0.033 0.384 0.033
Bias of∆ 0.032 0.004 1.614 0.004 53.578 -0.461 51.820 -0.370 2.134 0.050 2.134 0.050
Variation ∆ 3.935 0.216 2.028 0.216 1122.120 0.147 1059.365 0.174 2.037 0.221 2.037 0.221
%negative
variance 51.967% 0% 0% 0% 0% 26.184% 0% 19.116% 0% 0% 0% 0%
2762 Souha K. Badr et al.
Table (A.7) Results of RCR Estimation When 𝛽0~𝑁(0,0) and 𝛽1~𝑁(0,0)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N
=
T
=
5
𝝈𝝐
=
1
�� -0.333 0.049 -0.849 0.118 -0.294 0.056 0.022 0.003 -0.071 0.011 -0.071 0.011
∆ -0.468 -0.013 1.913 0.068 2.330 0.079 2.173 0.083 2.521 0.701 1.901 0.086
Bias of �� -0.333 0.049 -0.849 0.118 -0.294 0.056 0.022 0.003 -0.071 0.011 -0.071 0.011
Variation
of ��
619.942 22.299 5523.281 109.694 570.983 21.123 101.693 2.479 54.838 1.273 54.481 1.259
Bias of∆ -0.468 -0.013 1.913 0.068 2.330 0.079 2.173 0.083 2.521 0.701 1.901 0.086
Variation ∆ 34.586 0.041 16.515 0.020 16.993 0.021 15.747 0.018 14.410 0.164 16.512 0.018
%negativ
e variance
61.03% 59.98% 0% 0% 22.43% 23.17% 12.65% 7.18% 0% 0% 0% 0%
𝝈𝝐
=
5
�� -1.474 0.220 -4.394 0.619 -1.489 0.285 0.124 0.011 -0.369 0.059 -0.290 0.067
∆ -11.834 -0.326 47.679 1.686 58.016 1.979 54.065 2.064 47.510 2.289 59.772 3.458
Bias of �� -1.474 0.220 -4.394 0.619 -1.489 0.285 0.124 0.011 -0.369 0.059 -0.290 0.067
Variation
of ��
15159.38
1
551.565 138032.4
06 2740.094 14257.348 527.267 2515.730 60.848 1353.533 31.008 1584.113 32.436
Bias of∆ -11.834 -0.326 47.679 1.686 58.016 1.979 54.065 2.064 47.510 2.289 59.772 3.458
Variation ∆ 21640.51 25.901 10371.49
9 12.528 10678.721 13.229 9904.022 11.547
10357.05
4 10.709 22080.311 60.377
%negativ
e variance
61.07% 60.06% 0% 0% 22.53% 23.17% 13.16% 7.12% 0% 0% 0% 0%
𝝈𝝐
=
1
0
�� -3.328 0.486 -8.488 1.181 -2.942 0.562 0.218 0.027 -0.719 0.114 78.886 9.487
∆ -46.706 -1.272 191.331 6.774 233.109 7.931 217.399 8.269 188.806 7.285 13339.49 1381.005
Bias of �� -3.328 0.486 -8.488 1.181 -2.942 0.562 0.218 0.027 -0.719 0.114 78.886 9.487
Variation
of ��
61994.19
8
2229.949 552328.1 10969.4 57098.255 2112.271 10169.359 247.976 5486.307 127.348 112255008 1888581
Bias of∆ -46.706 -1.272 191.331 6.774 233.109 7.931 217.399 8.269 188.806 7.285 13339.487 1381.005
Variation ∆ 345946.2 414.999 165204.3
2 200.223 169979.762 211.582 157523.903 184.643
165680.5
3 192.885 14445024787 57729448
%negativ
e variance
61.03% 59.98% 0% 0% 22.43% 23.17% 12.65% 7.18% 0% 0% 0% 0%
N
=
T
=
1
0
𝝈𝝐
=
1
�� -2.346 0.446 -0.293 0.055 0.049 -0.014 -0.045 0.005 -0.069 0.010 -0.066 0.009
∆ 0.001 0.000 0.316 0.012 0.868 0.015 0.779 0.018 0.921 0.614 0.313 0.017
Bias of �� -2.346 0.446 -0.293 0.055 0.049 -0.014 -0.045 0.005 -0.069 0.010 -0.066 0.009
Variation
of ��
56899.91
0
2099.181 911.401 32.095 199.411 6.347 91.822 2.002 78.274 1.426 78.207 1.424
Bias of∆ 0.001 0.000 0.316 0.012 0.868 0.015 0.779 0.018 0.921 0.614 0.313 0.017
Variation ∆ 0.659 0.001 0.306 0.000 0.247 0.001 0.234 0.000 0.171 0.218 0.306 0.001
%negativ
e variance
56.07% 55.41% 0% 0% 0.45% 22.96% 0.25% 13.45% 0% 0% 0% 0%
𝝈𝝐
=
5
�� -11.737 2.233 -1.465 0.273 0.243 -0.068 -0.223 0.026 -0.343 0.050 -0.349 0.051
∆ 0.033 0.009 7.906 0.305 21.709 0.364 19.476 0.453 8.291 0.897 18.966 1.497
Bias of �� -11.737 2.233 -1.465 0.273 0.243 -0.068 -0.223 0.026 -0.343 0.050 -0.349 0.051
Variation
of ��
1422497.
26
52479.49 22785.02
9 802.371 4985.278 158.667 2295.539 50.052 1957.071 35.666 1963.282 35.724
Bias of∆ 0.033 0.009 7.906 0.305 21.709 0.364 19.476 0.453 8.291 0.897 18.966 1.497
Variation ∆ 411.675 0.601 191.399 0.284 154.525 0.315 145.987 0.271 184.014 0.162 577.803 2.506
%negativ
e variance
56.07% 55.42% 0% 0% 0.45% 22.97% 0.24% 13.45% 0% 0% 0% 0%
𝝈𝝐
=
1
0
�� -23.463 4.464 -2.931 0.547 0.486 -0.137 -0.447 0.053 -0.686 0.100 0.105 0.800
∆ 0.135 0.035 31.623 1.218 86.824 1.454 77.894 1.813 31.322 1.783 11486.04 1085.071
Bias of �� -23.463 4.464 -2.931 0.547 0.486 -0.137 -0.447 0.053 -0.686 0.100 0.105 0.800
Variation
of ��
5689991.
0
209918.1 91140.10
7 3209.484 19941.109 634.669 9182.155 200.207 7828.250 142.659 1012998.54 83348.46
Bias of∆ 0.135 0.035 31.623 1.218 86.824 1.454 77.894 1.813 31.322 1.783 11486.041 1085.071
Variation ∆ 6586.211 9.613 3061.993 4.537 2473.295 5.042 2336.589 4.341 3053.393 3.393 586801868 3245085
%negativ
e variance
56.07% 55.41% 0% 0% 0.45% 22.96% 0.25% 13.45% 0% 0% 0% 0%
Comparing MINQUE and IAUE estimates of variance components 2763
Table (A.7) Results of RCR Estimation When 𝛽0~𝑁(0,0) and 𝛽1~𝑁(0,0)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T =
15
𝝈𝝐 = 1
�� 0.246 -0.068 -0.025 0.005 0.018 -0.033 -0.059 0.013 0.000 0.000 -0.001 0.000
∆ 0.002 0.000 0.141 0.006 0.733 0.007 0.672 0.009 0.752 0.606 0.137 0.006
Bias of �� 0.246 -0.069 -0.025 0.005 0.018 -0.033 -0.059 0.013 0.000 0.000 -0.001 0.000
Variation
of �� 693.584 35.516 3.724 0.130 3.344 0.126 11.325 0.484 1.447 0.047 1.417 0.046
Bias of∆ 0.002 0.000 0.142 0.006 0.733 0.007 0.672 0.009 0.752 0.606 0.137 0.006
Variation ∆ 0.129 0.000 0.006 9.516e-05 0.146 0.000 0.126 0.000 0.129 0.221 0.059 0.000
%negative variance 55.0.43% 55.371% 0% 0% 0% 25.529% 0% 15.943% 0% 0% 0% 0%
𝝈𝝐
=
5
�� 1.232 -0.344 -0.123 0.024 0.089 -0.017 -0.295 0.064 0.003 0.001 0.04 -3.95e-05
∆ 0.054 0.004 3.543 0.139 18.313 0.163 16.805 0.229 4.019 0.736 28.035 1.613
Bias of �� 1.232 -0.344 -0.123 0.024 0.089 -0.017 -0.295 0.064 0.003 0.001 0.04 -3.95e-05
Variation
of �� 17339.608 887.893 93.092 3.252 83.595 3.154 283.117 12.097 36.238 1.173 47.949 1.203
Bias of∆ 0.054 0.004 3.543 0.139 18.313 0.163 16.805 0.229 4.019 0.736 28.035 1.613
Variation ∆ 81.072 0.127 36.989 0.059 91.523 0.074 78.532 0.069 33.480 0.126 1057.620 2.181
%negative
variance 55.043% 55.371% 0% 0% 0% 25.529% 0% 15.943% 0% 0% 0% 0%
𝝈𝝐
=
10
�� 2.465 -0.687 -0.247 0.047 0.179 -0.034 -0.590 0.128 0.006 0.001 19.393 -0.898
∆ 0.216 0.017 14.173 0.559 73.252 0.065 67.219 0.919 14.230 1.140 25233.726 1504.641
Bias of �� 2.465 -0.687 -0.247 0.047 0.179 -0.034 -0.590 0.128 0.006 0.001 19.393 -0.898
Variation
of �� 69358.43 3551.57 327.366 13.007 334.379 12.614 1132.470 48.387 144.929 4.692 2686886.23 48315.47
Bias of∆ 0.216 0.017 14.173 0.559 73.252 0.651 67.219 0.919 14.230 1.140 25233.726 1504.641
Variation ∆ 1297.158 2.024 591.826 0.952 1464.372 1.191 1256.510 1.108 582.429 0.542 1245036621 2657338
%negative
variance 55.043% 55.371% 0% 0% 0% 25.529% 0% 15.943% 0% 0% 0% 0%
N = T =
20
𝝈𝝐 =
1
�� 0.049 -0.013 -0.017 0.003 -0.003 0.0006 0.058 -0.010 -0.006 0.001 -0.007 0.001
∆ 1.996𝑒−4 2.562𝑒−5 0.081 0.002 0.677 0.003 0.628 0.005 0.685 0.597 0.078 0.002
Bias of �� 0.049 -0.013 -0.017 0.003 -0.003 0.0006 0.058 -0.010 -0.006 0.001 -0.007 0.001
Variation
of �� 63.855 2.414 1.215 0.032 10.129 0.397 48.294 1.802 0.718 0.013 0.702 0.013
Bias of∆ 1.996𝑒−4 2.562𝑒−5 0.081 0.002 0.677 0.003 0.628 0.005 0.685 0.597 0.078 0.002
Variation ∆ 0.042 5.523𝑒−5 0.018 2.428𝑒−5 0.169 3.561𝑒−5 0.148 4.462𝑒−5 0.161 0.227 0.018 2.481𝑒−5
%negative
variance 54.185% 54.957% 0% 0% 0% 26.771% 0% 17.828% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 0.247 -0.065 -0.086 0.017 -0.015 0.003 0.292 -0.054 -0.033 0.007 0.029 0.006
∆ 0.004 0.0006 2.039 0.073 16.928 0.084 15.715 0.138 2.562 0.665 54.074 1.548
Bias of �� 0.247 -0.065 -0.086 0.017 -0.015 0.003 0.292 -0.054 -0.033 0.007 0.029 0.006
Variation
of �� 1596.384 60.361 30.383 0.811 253.243 9.930 1207.373 45.065 17.963 0.345 51.457 0.360
Bias of∆ 0.004 0.0006 2.039 0.073 16.928 0.084 15.715 0.138 2.562 0.665 54.075 1.548
Variation ∆ 26.470 0.034 11.460 0.015 106.157 0.022 92.814 0.027 9.483 0.161 3859.753 1.813
%negative
variance 54.185% 54.957% 0% 0% 0% 26.771% 0% 17.828% 0% 0% 0% 0%
𝝈𝝐 =
10
�� 0.494 -0.130 -0.173 0.035 -0.030 0.006 0.584 -0.109 -0.065 0.014 27.433 -1.927
∆ 0.019 0.002 8.159 0.293 67.712 0.336 62.863 0.554 8.427 0.877 53377.486 1513.367
Bias of �� 0.494 -0.130 -0.173 0.035 -0.030 0.006 0.584 -0.109 -0.065 0.014 27.433 -1.927
Variation
of �� 6385.538 241.447 121.533 3.246 1012.974 39.723 4829.494 180.261 71.868 1.380 9788718.25 29223.41
Bias of∆ 0.019 0.002 8.159 0.293 67.712 0.336 62.863 0.554 8.427 0.877 53377.486 1513.367
Variation ∆ 423.530 0.552 183.367 0.242 1698.522 0.356 1485.037 0.446 176.642 0.139 4247977718 2104146
%negative
variance 54.185% 54.957% 0% 0% 0% 26.771% 0% 17.828% 0% 0% 0% 0%
2764 Souha K. Badr et al.
Table (A.7) Results of RCR Estimation When 𝛽0~𝑁(0,0) and 𝛽1~𝑁(0,0)
SWAMY R W Z MINQUE1 MINQUE2 IAUE1 IAUE2
𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏 𝜷𝟎 𝜷𝟏
N = T =
50
𝝈𝝐 = 1
�� 0.002 -0.000 -0.001 0.000 0.021 -0.005 -0.003 0.000 -0.001 0.000 -0.001 0.000
∆ -1.580𝑒−3 -4.687𝑒−5 0.015 0.000 0.632 0.000 0.611 0.001 0.634 0.615 0.634 0.615
Bias of �� 0.002 0.000 -0.001 0.000 0.021 -0.005 -0.003 0.000 -0.001 0.000 -0.001 0.000
Variation
of �� 0.016 0.000 0.003 0.000 3.836 0.169 0.121 0.005 0.003 0.000 0.003 0.000
Bias of∆ -1.580𝑒−3 -4.687𝑒−5 0.015 0.000 0.632 0.000 0.611 0.001 0.634 0.615 0.634 0.615
Variation ∆ 3.548𝑒−4 4.614𝑒−7 1.749𝑒−4 2.288𝑒−7 0.117 6.370𝑒−7 0.111 3.873𝑒−6 0.118 0.117 0.118 0.117
%negative variance 54.864% 53.693% 0% 0% 0% 29.279% 0% 21.891% 0% 0% 0% 0%
𝝈𝝐 =
5
�� 0.010 -0.001 -0.011 0.002 0.117 -0.028 -0.006 0.001 -0.012 0.002 -0.012 0.002
∆ -0.032 -0.000 0.013 0.383 15.692 0.015 15.176 0.041 0.978 0.622 0.978 0.622
Bias of �� 0.010 -0.001 -0.011 0.002 0.117 -0.028 -0.006 0.001 -0.012 0.002 -0.012 0.002
Variation
of �� 0.450 0.018 0.096 0.003 105.933 4.673 2.786 0.116 0.095 0.003 0.095 0.003
Bias of∆ -0.032 -0.000 0.383 0.013 15.692 0.015 15.176 0.041 0.978 0.622 0.978 0.622
Variation ∆ 0.201 0.000 0.100 0.000 67.21 3.579𝑒−4 63.411 0.002 0.189 0.106 0.189 0.106
%negative
variance 54.427% 53.034% 0% 0% 0% 29.054% 0% 21.592% 0% 0% 0% 0%
𝝈𝝐 =
10
�� 0.032 -0.005 -0.026 0.005 0.267 -0.064 0.079 -0.015 -0.027 0.006 -0.027 0.006
∆ -0.186 -0.005 1.521 0.055 63.025 0.062 60.838 0.174 2.061 0.662 2.061 0.662
Bias of �� 0.032 -0.005 -0.026 0.005 0.267 -0.064 0.079 -0.015 -0.027 0.006 -0.027 0.006
Variation
of �� 2.079 0.084 0.386 0.013 513.889 22.681 6.077 0.243 0.385 0.013 0.385 0.013
Bias of∆ -0.186 -0.005 1.521 0.055 63.025 0.062 60.838 0.174 2.061 0.662 2.061 0.662
Variation ∆ 2.699 0.003 1.336 0.001 911.3 4.750𝑒−3 856.001 0.028 1.359 0.090 1.359 0.090
%negative
variance 55.380% 53.083% 0% 0% 0% 28.657% 0% 20.798% 0% 0% 0% 0%
Comparing MINQUE and IAUE estimates of variance components 2765
Fig (B.1) 2D graph comparing different methods of estimating variance
component applying on RCR model according to sample size
a) Comparing the bias in estimating 𝛽°
between 5 methods
according to the sample size
b) Comparing the bias in estimating 𝛽1
between 5 methods
according to the sample size
c) Comparing the variation in estimating 𝛽°
between 5 methods according to the sample size
d) Comparing the variation in estimating 𝛽1
between 5
methods according to the sample size
e) Comparing the bias in estimating ∆° between 5 methods
according to the sample size
f) Comparing the bias in estimating ∆1
between 5 methods according to the sample size
g) Comparing the variation in estimating ∆°
between 5 methods according to the sample size
h) Comparing the variation in estimating ∆1
between 5 methods according to the sample size
i) Comparing the negative value in estimating ∆°
between 4 methods according to the sample size
j) Comparing the negative value in estimating ∆1
between 4 methods according to the sample size
2766 Souha K. Badr et al.
Fig (B.2) 2D graph comparing different methods of estimating variance
component applying on RCR model according to error variance
Received: October 10, 2017; Published: October 30, 2017
a) Comparing the bias in estimating 𝛽°
between 5 methods
according to the error variance
b) Comparing the bias in estimating 𝛽1
between 5 methods
according to the error variance
c) Comparing the variation in estimating 𝛽°
between 5 methods
according to the error variance
d) Comparing the variation in estimating 𝛽1
between 5
methods according to the error variance
e) Comparing the bias in estimating ∆°
between 5 methods according to the error variance
f) Comparing the bias in estimating ∆1
between 5 methods
according to the error variance
g) Comparing the variation in estimating ∆°
between 5 methods according to the error variance
h) Comparing the variation in estimating ∆1
between 5
methods according to the error variance
i) Comparing the negative value in estimating ∆°
between 4
methods according to the error variance
j) Comparing the negative value in estimating ∆1
between
4 methods according to the error variance