Professional Documents
Culture Documents
Research Article
Upper Bound of the Generalized Value for
the Population Variances of Lognormal Distributions
with Known Coefficients of Variation
Copyright 2017 Rada Somkhuean et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
This paper presents an upper bound for each of the generalized values for testing the one population variance, the difference
between two population variances, and the ratio of population variances for lognormal distribution when coefficients of variation
are known. For each of the proposed generalized values, we derive a closed form expression of the upper bound of the generalized
value. Numerical computations illustrate the theoretical results.
where 0 is a specified quantity. Let be a particular observed and the coefficient of variation for lognormal distribution are,
sample. The generalized test variable, (, , ), is required to respectively, given by
satisfy the following conditions:
2
( ) = exp ( + ),
(A1) For fixed and = (, ), the distribution (, , ) 2
is free from the nuisance parameter . (5)
var ( ) = exp (2 + 2 ) (exp (2 ) 1) ,
(A2) obs = (, , ) is free from any unknown parameter.
CV = exp (2 ) 1,
(A3) (, , ) is either stochastically increasing or
decreasing in for any given and fixed values of where CV denotes the coefficient of variation of which is
and . computed fromvar( )/( ).
It is easy to see that
Under the above conditions, if (, , ) is a stochas-
tically increasing test variable then the subset of space is = CV = exp (2 ) 1,
extreme region . For the one-sided hypothesis given above
they defined a data-based extreme region is of the form 2 = exp (2 ) 1,
(6)
= { : (, , ) 0} . (2) 2 +1= exp (2 ) ,
2 = ln (2 + 1)
Given the observed sample , the generalized p value is
defined as and the parameters of interest are
for further details and for several applications based on the = exp (21 + ln (12 + 1)) 12
generalized p value; we refer to the book by Weerahandi [11]. (7)
Moreover, Tsui and Weerahandi [10] used the generalized exp (22 + ln (22 + 1)) 22
p value () for the Behrens-Fisher problem of testing the
difference of two independent normal distribution means = 1 exp (21 + 1 ) 2 exp (22 + 2 ) ,
with possibly unequal variances. Later, Tang and Tusi [12]
extended the works of Weerahandi [11], Gamage and Weer- = 2 , = ln (2 + 1) , = 1, 2,
ahandi [14] to derive the formula of the upper bound () of
var (1 ) 1 exp (21 + 1 )
the generalized p value () which is in the form of (see also, 3 = = .
e.g., Kabaila and Lloyd [15]) var (2 ) 2 exp (22 + 2 )
For testing the null hypothesis, 01 : 1 01 vs 1 : 1 >
( () ) . (4) 01 , 02 : 2 02 vs 2 : 2 > 02 , and 03 : 3 03
vs 3 : 3 > 03 , the sufficient statistics involving these
In this paper, we also extend Tang and Tsui [12] work to problems are ( , 2 ), where
find upper bound of each of the generalized p values, (),
1
for hypotheses testing of the one population variance, the = ,
difference between two population variances, and the ratio of =1
population variances of lognormal distributions with known (8)
coefficients of variation. 1
2
2 = ( ) .
1 =1
3. Main Results for the Population Variance It is known that the probability distributions of the statistics
of Lognormal Distributions with Known below are independent:
Coefficients of Variation 2
( , ),
Let = { , , . . . , }, for ( = 1, 2, = 1, 2, . . . , ) be
random samples having lognormal distribution and let = (9)
( 1) 2
ln( ) ( , 2 ) where and 2 denote the mean and = 2 2 1 .
variance of , respectively. In particular, the mean, variance,
Journal of Probability and Statistics 3
where = 1 (1 1 )/1 (0, 1), 1 = (1 1)12 /12 2 = 1 exp (21 + 1 ) 2 exp (22 + 2 ) . (16)
21 1 .
It is easy to see that (1 , 1 , 1 ) in (12) satisfies condi- Without loss of generality, suppose 2 = 02 = 0,
tions (A1)(A3) in Section 2.
The generalized p value, (), is defined, under the null 2 = 1 exp (21 + 1 ) 2 exp (22 + 2 )
hypothesis 01 , to be
1
() = sup (1 (1 , 1 , 1 ) 1 (1 , 1 , 1 )) = 2 (1 2 ) + (ln ( ) + (1 2 )) (17)
2
01
(13)
= (1 (1 , 1 , 1 ) 1 (1 , 1 , 1 )) . = 2 (1 2 ) + 1 .
Following (13) the generalized p value for (10) can be defined Using generalized test variable for 2 is
as
() = (1 (1 , 1 , 1 ) 1 (1 , 1 , 1 ))
2 (1 , 2 , 1 , 2 , 2 ) = 2 (1 2
1
= (1 exp (2 (1 ) + 1 )
1 /1 1 1
1
01 ) = (ln (1 ) + 2 (1 ) + 1
1 /1 1 1 2 1 (2 1 ) 12 12 22 22
+ + ) + 1
12 /1 + 22 /2 1 12 2 22
1
ln (01 )) = (1
1 /1 1 1 (14)
1
(ln ( 01 ) 1 )) = (
2 1 = 2 (1 2 +
1 + 2
1 (1/2) (ln (01 /1 ) 1 ) 1
( ))
1 /1 1 1
1
1 = (1 +2 2 (((1 2 ) + )
= 2 (1 2 + 2
1 + 2 2
1 + 2 2
)) ,
((1 1) /1 ) 12 /+ ((2 1) /2 ) 22 / (1 )
(1 + 2 ) / (1 + 2 2) (20)
where 1 +2 2 () is a cdf of -distribution with 1 + 2 2
(( 1) /1 ) 12 ((2 1) /2 ) 22 degrees of freedom and () is an expectation operator with
1 + ) + 1 respect to ( Beta((1 1)/2, (2 1)/2)).
1 / (1 + 2 ) 2 / (1 + 2 )
Theorem 3. If () = ] (( + )1/(1 / + 2 /)) then for
1 +2 2 fixed 0 and 0, () is convex function for .
= 2 (1 2 +
1 + 2 2 Proof. See Appendix.
Following (25) the generalized p value for (21) can be defined 5. Conclusion
as
We proposed three new generalized p values for testing the
(2 ) = (3 (1 , 2 , 1 , 2 , 3 ) 3 (1 , 2 , 1 , 2 , 3 )) hypotheses of (1) one population variance, (2) the difference
between two population variances, and (3) the ratio of
1 +2 2 population variances of lognormal distributions when the
= (1 2 + coefficients of variation are known. We also proved new
1 + 2 2
upper bounds for our proposed generalized p values. We
note here that the results for these results for case (1),
((1 1) /1 ) 12 ((2 1) /2 ) 22 2
+ 0) case (2), and case (3) were analogous to the upper bound
1 2 of the generalized p value for the Behrens-Fisher problem
6 Journal of Probability and Statistics
Table 2: The upper bound of p value for hypothesis (15), for 1 = 1, 2 = 1, and = 0.
Table 3: The upper bound of p value for hypothesis (21), for 1 = 1, 2 = 1, and = 0.
Moreover For any < 0.5 and () < , hence, by Theorem 1, we have
((1 )) ((1 ))
1 1/2 1 3/2 1
(1 ) = ( 1 1 1 ) = 1 1 1 () = 1 ( (1 1 ))
2 1 1 4 1 1 (A.2) 1 1
> 0. 1 (A.4)
(1 1 ( ) )
1 1
Hence (1 ) > 0, and (1 ) is convex in 1 .
= (1 ) = 1 () .
1 = 1 1 (1 () + ) ,
= 1 ( ((
1 /1 1 1
(1/2) (ln (01 /1 ) 1 )
where = .
(1/2) (ln (01 /1 ) 1 ) 1 1 /1 1 /
)( )))
1 /1 1 1 1 /1
1 (A.3)
= 1 ( (( Proof of Theorem 3. Define () = (()), () =
1 1 ( )1/(1 / + 2 /), and ] is the probability density
function of -distribution of ] degrees of freedom. Hence
(1/2) (ln (01 /1 ) 1 ) 1 1
)( 1 )))
1 /1 1 1 1 ( ()) = ( ( ())) = (] ( ()) ())
(A.6)
1 2
= ( (1 1 )) , = ] ( ()) ( ()) + ( ()) () .
1 1
For 0, 0 implies that () < 0, ] (()) 0, and
(1/2) (ln (01 /1 ) 1 ) ] (()) 0.
= .
1 /1 We have
3/2
1 1 2
() = ( + ) (( + 2 ) ( ) ( 21 + ))
1 2 (1 )2
(A.7)
12 /22 + 22 /2 (1 )4 + 21 2 / (1 )3 + 21 2 /3 (1 ) + 312 22 /2 (1 )2
= ( + ) ( 5/2
) > 0.
(1 / + 2 / (1 ))
1
() = (1 +2 2 (( + ) )
1 1 / (1 1) + (1 ) 2 1 / (2 1)
8 Journal of Probability and Statistics
1/2
1 1 (1 ) 2 1
11 +2 2 ()) = ( ( + )
1 1 2 1
1 1 (1 ) 2 1 1 1 2 1 1
+ (11 +2 2 ()) ) ( )) = ( 11 +2 2 ()
1 1 2 1 1 1 2 1 4
3/2
1 1 (1 ) 2 1 1 1 (1 ) 2 1
= ( ( + (11 +2 2 ()) ( + )
1 1 2 1 1 1 2 1
2
1 1 2 1
)) , ( ) ) 0.
1 1 2 1
(A.8) (A.10)
()
(1 2 ) + 1 /2 1 + 2 2 1
= (1 +2 2 ( ( )))
12 /1 + 22 /2 ((1 1) /1 ) 12 / + ((2 1) /2 ) 22 / (1 ) 1/ (12 /1 + 22 /2 )
1 + 2 2
= (1 +2 2 (( + ) )) , (A.12)
(1 1 12 /1 ) / (12 /1 + 22 /2 ) + (2 1 22 /2 ) / (1 )
1 /2
=
12 /1 + 22 /2
1 + 2 2
= (1 +2 2 (( + ) )) .
1 1 / + (1 ) 2 1 / (1 )
For any < 0.5 and () < , hence, by Theorem 1, such that 1 + 2 2
(()) (()) ))
1 1 / + (1 ) 2 1 / (1 )
() = (1 +2 2 (( + ) 1 +2 2 ( (( + )
Journal of Probability and Statistics 9
1 + 2 2 References
))
1 1 / + (1 ) 2 1 / (1 )
[1] H. P. Singh, L. N. Upadhyaya, and U. D. Namjoshi, Estimation
of finite population variance, Current Science, vol. 57, no. 24, pp.
13311334, 1988.
= 1 +2 2 (( + )
[2] M. C. Agrawal and A. B. Sthapit, Unbiased ratio-type variance
estimation, Statistics and Probability Letters, vol. 25, no. 4, pp.
1 361364, 1995.
) [3] A. Arcos Cebrian and M. Rueda Garca, Variance estimation
1 1 / (1 1) + (1 ) 2 1 / (2 1)
using auxiliary information: an almost unbiased multivariate
ratio estimator, Metrika. International Journal for Theoretical
= 1 () .
and Applied Statistics, vol. 45, no. 2, pp. 171178, 1997.
(A.13) [4] A. Arcos, M. Rueda, M. D. Martnez, S. Gonzalez, and Y.
Roman, Incorporating the auxiliary information available in
For 0 < < 0.5, we have variance estimation, Applied Mathematics and Computation,
({ : () }) ({ : 1 () }) = () , vol. 160, no. 2, pp. 387399, 2005.
[5] C. Kadilar and H. Cingi, Ratio estimators for the population
by Theorem 4, variance in simple and stratified random sampling, Applied
() max { (0) , (1)} Mathematics and Computation, vol. 173, no. 2, pp. 10471059,
2006.
1 [6] V. Cojbasic and A. Tomovic, Nonparametric confidence inter-
= max { (1 +2 2 (( + ) ) ) ,
1 1 / (1 1) vals for population variance of one sample and the difference
of variances of two samples, Computational Statistics & Data
1 Analysis, vol. 51, no. 12, pp. 55625578, 2007.
(1 +2 2 (( + ) ) )}
2 1 / (2 1) [7] V. Cojbasic and D. Loncar, One-sided confidence intervals
for population variances of skewed distributions, Journal of
1 1 Statistical Planning and Inference, vol. 141, no. 5, pp. 16671672,
= max { (
1 1 / (1 1) 1 1 / (1 1) 2011.
[8] V. C. Rajic, J. Kocovic, D. Loncar, and T. R. Antic, Testing
1
+ 11 +2 2 ()) , ( population variance in case of one sample and the difference of
2 1 / (2 1) variances in case of two samples: example of wage and pension
(A.14) data sets in Serbia, Economic Modelling, vol. 29, no. 3, pp. 610
1
+ 11 +2 2 ())} 613, 2012.
2 1 / (2 1)
[9] R. Singh and S. Malik, Improved estimation of population
1 variance using information on auxiliary attribute in simple
= max { (1 1 ( + 11 +2 2 ())) , random sampling, Applied Mathematics and Computation, vol.
1 1 / (1 1)
235, pp. 4349, 2014.
1 [10] K.-W. Tsui and S. Weerahandi, Generalized p-values in sig-
(2 1 ( + 11 +2 2 ()))} nificance testing of hypotheses in the presence of nuisance
2 1 / (2 1)
parameters, Journal of the American Statistical Association, vol.
1 84, no. 406, pp. 602607, 1989.
max {1 1 ( ( ) + 11 +2 2 ()) ,
1 1 / (1 1) [11] S. Weerahandi, Exact Statistical Methods for Data Analysis,
Springer, New York, NY, USA, 1995.
1 [12] S. Tang and K.-W. Tsui, Distributional properties for the
2 1 ( ( ) + 11 +2 2 ())}
2 1 / (2 1) generalized p-value for the Behrens-Fisher problem, Statistics
& Probability Letters, vol. 77, no. 1, pp. 18, 2007.
1 /2
= min{1 1,2 1} (11 +2 2 () ) , = . [13] R. Somkhuean, S. Niwitpong, and S.-A. Niwitpong, Upper
12 /1 + 22 /2 bounds of generalized p-values for testing the coefficients of
variation of lognormal distributions, Chiang Mai Journal of
Science, vol. 43, pp. 671681, 2016.
[14] J. Gamage and S. Weerahandi, Size performance of some tests
Competing Interests in one-way ANOVA, Communications in StatisticsSimulation
and Computation, vol. 27, no. 3, pp. 625640, 1998.
The authors declare that there is no conflict of interests [15] P. Kabaila and C. J. Lloyd, Tight upper confidence limits from
regarding the publication of this paper. discrete data, The Australian Journal of Statistics, vol. 39, no. 2,
pp. 193204, 1997.
Acknowledgments [16] The R Development Core Team, An Introduction to R, R
Foundation for Statistical Computing, Vienna, Austria, 2010,
The second author is grateful to Grant no. KMUTNB-GOV- http://www.R-project.org.
59-36 from King Mongkuts University of Technology North
Bangkok.
Advances in Advances in Journal of Journal of
Operations Research
Hindawi Publishing Corporation
Decision Sciences
Hindawi Publishing Corporation
Applied Mathematics
Hindawi Publishing Corporation
Algebra
Hindawi Publishing Corporation
Probability and Statistics
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014
International
Journal of Journal of
Mathematics and
Mathematical
Discrete Mathematics
Sciences