Abstract
In this paper, Resolution III saturated , factorial designs and specially the cases , , are studied, in order to obtain D-optimal plans.
AMS Subject Classifications:
Public interest statement
An issue of interesting in experimental designs is the saturated designs. An experimental design is called saturated if all the degrees of freedom are consumed by the estimation of the parameters without leaving degrees of freedom for error variance estimation. The saturated factorial designs, where the interest is to estimate the general mean and the main effects while all higher order interactions are negligible (resolution III plans), are commonly used in screening experiments. In recent years, there has been a considerable interest in optimal saturated main effect designs. Most researchers have dealt with the case where two or three factors are involved in the experiment on two levels. The problem is different and becomes more difficult when three or four factors are involved in the experiment on three or more levels.
1. Introduction
Saturated factorial plans is a very interesting issue in theory of exeprimental designs, since the reduced number of observations is very usefull in practise especially in screening experiments, where are used to determine which of many factors affects the measure of pertinent quality characteristics. In saturated designs the number of observation is equal to the number of parameters, so all degrees of freedom are consumed by the estimation of parameters, leaving no degrees of freedom for error variance estimation. The purpose of this paper is to give saturated resolution III designs, minimizing the generalized variance of the main effects and the general mean, that is, D-optimal designs. In recent years, there has been a considerable interest in optimal saturated main effect designs with two or three factors. Mukerjee et al. (Citation1986) and Kraft (Citation1990) showed all two-factor designs are equivalent with respect to D-optimality criterion. Later Mukerjee and Sinha (Citation1990) considered, for the two-factor case, the optimality results on almost saturated main effect designs. Pesotan and Raktoe (Citation1988) worked also in the special case for factorials and a subclass of factorials.
Chatterjee and Mukerjee (Citation1993) were the first who attempted to extend the two factor results to three factors. They consider , , factorial to derive D-optimal saturated main effect designs. Later Chatterjee and Narasimhan (Citation2002), using techniques from Graph Theory and Combinatorics, claimed about the upper bound of the determinant of the saturated , , factorials when is odd. Chatzopoulos and Kolyva-Machera (Citation2006) extend the results concerning D-optimal saturated main effect designs for to factorials, when and . Karagiannis and Moyssiadis (Citation2005) and Karagiannis and Moyssiadis (Citation2008) extend the Graph theoretic approach of Chatterjee and Narasimhan (Citation2002) and the results of Chatzopoulos and Kolyva-Machera (Citation2006), and give the D-optimal saturated , , designs. In this paper, we study the D-optimality for saturated factorials. Moreover, we give the upper bound of the determinant for the , , , saturated designs and the corresponding design, which attains this bound. The paper is organized as follows. Some notations and preliminaries are first presented in Section 2. Section 3 deals with the main results of this paper.
2. Notations and preliminaries
In this paper, we follow the same notations as in Chatzopoulos and Kolyva-Machera (Citation2006) adapted for four factors. Let us consider the setup of an , saturated factorial experiment, involving four factors , , and appearing at , , and levels, respectively, with runs. For let the levels of be denoted by and coded as . Our interest is to find D-optimal resolution III designs. There are altogether treatment combinations denoted by , that will hereafter be assumed to be lexicographically ordered.
Let, for , be the vector with each element unity, the identity matrix of order , denotes the Kronecker product of matrices and be an matrix such that is orthogonal ( denotes the transpose of matrix A). The usual fixed effect model under the absence of interactions is , where Y is the response vector of the experiment, is the vector of uncorrelated random errors with zero mean and the same variance and is the vector of unknown parameters, is consider. In our case , where is the unknown general mean and the elements of the vectors are unknown parameters representing a full set of mutually orthogonal contrasts belonging to the main effects and , where , , and . It is easy to see that the D-optimal design does not depend on the choice of , .
Following Mukerjee and Sinha (Citation1990) let , where , , and . We denote , the matrices obtained by deleting the first column of , . Consider the matrix U, which is a submatrix of given by , which has full column rank. The u rows of matrix U like those of W, correspond to the lexicographically ordered treatment combinations. Moreover the columns of U span those of and hence those of W, which also has full column rank.
Hence, one may obtain , where matrix H is a nonsingular matrix of order . For any design d in the class of the saturated resolution III designs with runs, the design matrix is , where is a square matrix of order such that for if the i-th run in d is given by the treatment combination then the j-th row of is the row of U corresponding to the treatment combination . A design d is said to be D-optimal in the class , if it maximizes the quantity . Since matrix H is nonsingular a design is D-optimal if it maximizes the quantity , where:(1) (1)
The matrices , and are obtained from the matrices and in a similar way, as is obtained from U.
Definition 2.1
For , if the i-th factor enters the experiment at level 0 then the corresponding row of the matrix is a row vector with elements zero. On the other hand if the i-th factor enters the experiment at level p, , then the corresponding row of the matrix equals the p-th row of the identity matrix of order . Similarly, if the fourth factor enters the experiment at level p, , then the corresponding row of the matrix equals to the -th row of the identity matrix . Let , , denote the number of these rows. It holds that
Definition 2.2
For and , , let , denote the number of runs where the i-th factor appears at level p and the j-th factor appears at level q. It holds that
Definition 2.3
For and , , let , denote the number of runs where the i-th factor appears at p level, the j-th factor appears at level q and the k-th factor appears at level r. It holds that
Remark 2.1
It holds that , , , since the design matrix of a saturated design has full column rank.
Remark 2.2
By the choice of the labels for the levels one can always assume, without loss of generality (w.l.g), that .
The following lemmas are crucial for the main results of our paper and can be founded in Chatterjee and Mukerjee (Citation1993) and Chatzopoulos and Kolyva-Machera (Citation2006).
Lemma 2.1
Consider the saturated design e, , with runs and corresponding matrix . It holds that(2) (2)
Proof
See Chatterjee and Mukerjee (Citation1993).
Lemma 2.2
Consider the saturated designs , , with runs and corresponding matrix . It holds that(3) (3)
Proof
See Chatterjee and Mukerjee (Citation1993).
Lemma 2.3
Consider the saturated design d. If and for some then , where is a saturated design.
Proof
See Chatzopoulos and Kolyva-Machera (Citation2006), lemma 2.1.
Corollary 2.1
Consider the saturated , design d with runs. If , then using the pigeonhole principle we can easily verify that , for some at least w times. Applying, w times, lemma 2.3, we get , where is a saturated design.
Remark 2.3
For we have to study only the cases where , that is the cases , and , .
Lemma 2.4
Let d be a saturated , design. If and , , then d is D-optimal.
Proof
See Chatzopoulos and Kolyva-Machera (Citation2006), theorem 2.1.
3. Main results
Lemma 3.1
The determinant of the matrix , given in (Equation1(1) (1) ), which corresponds to a , saturated factorial design d is left invariant by interchanging the levels of the factors.
Proof
For the fourth factor, we can interchange the columns which correspond to two levels and the proof is obvious. Similarly, for the nonzero levels of the first, second and the third factor, interchanging the columns p and q, the levels and are interchanged. Moreover, for the first (or second or third) factor, adding all the columns of matrix (or or ) to the column which corresponds to level p, subtracting the sum of all columns of matrix and multiplying the resulting column by the levels 0 and p are interchanged.
Lemma 3.2
Let d be a , saturated factorial design with corresponding matrix as given in (Equation1(1) (1) ). Let for some and some , , . Then
where , , and is , , , , saturated factorial design, respectively.
Proof
Let as assume that , , , , which means that the saturated design contains the runs , . By subtracting the row corresponding to run from the other rows which correspond to runs , adding the columns corresponding to levels to the column corresponding to level and expanding along the rows which contain levels , we get , where is saturated design. The proof is similar for , and .
3.1. D-optimality of saturated designs
Lemma 3.3
Consider the saturated design d with runs and corresponding matrix as given in (Equation1(1) (1) ). For the D-optimal design it holds that:
Proof
Expanding along its first column, we have that , where , are saturated designs with corresponding matrices . Let max . Then(4) (4)
From lemma 3.1, by interchanging the levels 0 and 1 of the first factor, it also holds(5) (5)
From (Equation4(4) (4) )-(Equation5(5) (5) ), we get . According to lemma 2.4, in order to find the D-optimal design , it must hold that , , which implies , , , , , . Expanding along its second column, and following the same procedure we get .
Lemma 3.4
Consider the saturated design d with runs and corresponding matrix as given in (Equation1(1) (1) ). For the D-optimal design it holds that:
Proof
Suppose, w.l.g, that for some . Then from the pigeonhole principle, we get for some q, . From lemmas 3.1 and 3.2, we may assume that in a design d the following treatment combinations exist: , , , . Consider now matrix , which corresponds to the design d, as given in (Equation1(1) (1) ). Subtract the row corresponding to the treatment combination from the row corresponding to the treatment combination . The row corresponding to the treatment combination is now , where the second ace is at the -th column of r, that is at the j-th column of . Then, subtract the row r from the row corresponding to the treatment combination . Continue by adding the column of , which corresponds to to the column of which corresponds to . Consequently, treatment combination is now in the position of treatment combination . Add the row corresponding to the treatment combination to row r. The resulting row corresponds to the treatment combination . Hence, the design d contains the treatment combinations , , , , which implies . Then, proceeding as in lemma 3.2, we have that , where is saturated design.
Corollary 3.1
For the D-optimal saturated design d with runs, if there exists the treatment combination () then there exists the treatment combination ().
Lemma 3.5
Consider the saturated design d with runs and corresponding matrix as given in (Equation1(1) (1) ). For the D-optimal design it holds that:
Proof
The proof is similar as in lemma 3.4.
Corollary 3.2
Let us now consider the saturated design d with runs. Then, from lemma 3.4 and using the pigeonhole principle, we get:
Corollary 3.3
For the D-optimal saturated design d with runs it holds that:
Lemma 3.6
Consider the saturated design d with runs and corresponding matrix as given in (Equation1(1) (1) ). For the D-optimal design it holds that:(6) (6)
Proof
From lemma 3.3, we have , and from lemma 3.2 and corollary 3.2 the D-optimal design includes one of the following sets of treatment combinations: (, , ) or (, , ) or (, , ) or (, , ). Applying lemma 3.1, we can always choose, w.l.g. the treatment combination , , . This choice, using pigeonhole principle, implies the existence of the treatment combinations , , and the proof of (Equation6(6) (6) ) is obvious.
Theorem 3.1
Consider the saturated design d with runs and corresponding matrix as given in (Equation1(1) (1) ). It holds that:(7) (7)
Proof
Let . So, from corollary 3.3, we have if mod 2 or if mod 2. Moreover, from lemmas 3.1-3.6 and corollaries 3.1-3.2, w.l.g., the D-optimal saturated design can be written as:
Let be a vector with all elements equal to zero, be the identify matrix of order k. For , it can be easily seen that:(8) (8)
Matrix as given in (Equation1(1) (1) ), can be written as:
Using relation (Equation8(8) (8) ), and after permutation of columns matrix can be written as:
Now subtract the (2u)-th column from the -th column and add the last columns to the -th column. Matrix , as given in (9), can be written as:
Matrix is the design matrix of the saturated design with runs. Matrix is not design matrix as its first column is , but , where matrix is the design matrix of the saturated design with runs. Hence, . From (Equation3(3) (3) ), we get . Recalling that if mod 2, or if mod 2, we get that (Equation7(7) (7) ) holds.
Theorem 3.2
Let if mod 2, or if mod 2. The saturated , design with the following treatment combinations 00ii (), , (), 101u, (), , 01ii (), () is a D-optimal design in the class of all , saturated designs.
Proof
The proof is obvious from lemma 2.3, lemmas 3.1-3.6 and theorem 3.1.
3.2. D-optimality of saturated designs
Lemma 3.7
Consider the saturated design d with runs and corresponding matrix as given in (Equation1(1) (1) ). For the D-optimal design, it holds that:
Proof
The proof is similar as lemma 3.3.
Corollary 3.4
For the saturated design d with runs, from lemma 3.2 and using the pigeonhole principle we get:
Lemma 3.8
Consider the saturated design d with runs and corresponding matrix as given in (Equation1(1) (1) ). For the D-optimal design, it holds that:(9) (9) (10) (10)
Proof
For the proof of relation (10) see lemma 3.4. The proof of relation (11) is obvious, since or .
Corollary 3.5
For the D-optimal saturated design d with runs it holds that:
Theorem 3.3
Consider the saturated design d with runs and corresponding matrix as given in (Equation1(1) (1) ). It holds that:(11) (11)
Proof
If mod 2, then and , while if mod 2, then and . From lemmas 3.7-3.8 and corollaries 3.4-3.5, the D-optimal saturated design, w.l.g., can be written as:
By interchanging the levels 0 and of the 4-th factor, according to lemma 3.1, the determinant of the matrix is left invariant. Hence the D-optimal saturated design, w.l.g., can be written as:
Matrix as given in (Equation1(1) (1) ), can be written as:
Using relation (Equation8(8) (8) ), and after a suitable permutation of columns of the matrix , in order to make the left bottom block of the matrix a zero matrix, matrix can be written as:
Analytically, the permutation is: the first u columns of matrix are the first u columns of matrix , columns of matrix are the first u columns of matrix , -th column of matrix is matrix , -th column of matrix is matrix , while the remaining columns of matrix are the rest columns of matrices and .
Now subtract the -th column from the -th column. Then, matrix is a block triangular matrix. It holds that , where matrix is a matrix, which does not correspond to any design and matrix is the design matrix of the saturated design with runs. From (Equation3(3) (3) ), we get . Moreover, expanding along its last column we get that , where, after some manipulations, correspond to two factor saturated designs , with, according to lemma 2.1, . Hence, . Recalling that if mod 2, or if mod 2, we get that (Equation11(11) (11) ) holds.
Theorem 3.4
Let if mod 2, or if mod 2. The saturated , design with the following runs , , (), 10u0, 01ii (1 1), 01u(1), 110u, 11ii (), 11()0, 0000, 00()i (), is a D-optimal design in the class of all , saturated designs.
Proof
From lemma 2.3, lemmas 3.1, 3.2, 3.7, 3.8 and theorem 3.3, the proof is obvious.
Acknowledgements
We would like to thank the referees and the journal editorial team for providing valuable advice that improved the quality of the original manuscript.
Additional information
Funding
Notes on contributors
St. A. Chatzopoulos
The problem of finding optimal designs under different types of criteria preoccupies many researchers the last decades. Most of the work on constructing optimal designs for the estimation of parameters in fractional factorials is concentrated on factors at two levels. Chatzopoulos, Kolyva-Machera, and Chatterjee (2009), studied the optimality of designs which are obtained by adding p runs to an orthogonal array for experiments involving m factors each at s levels. Chatterjee, Kolyva-Machera, and Chatzopoulos (2011), considered the issue of optimality of fractional factorial experiments involving m factors each at two levels. Pericleous, Chatzopoulos, Kolyva-Machera and Kounias, study the problem of estimating the standardized linear and quadratic contrasts in fractional factorials with k factors, each at 3 levels, when the number of runs or assemblies is N = 3 and introduced a different notion of Balanced Arrays. Chatzopoulos and Kolyva-Machera (2005), studied the saturated m1 × m2 × m3 designs and Chatzopoulos & Kolyva-Machera (2008), considered the problem of finding D-optimal saturated 4 × m2 × m3 designs.
References
- Chatterjee, K., & Mukerjee, R. (1993). D-optimal saturated main effect plans for 2 × s2 × s3 factorials. Journal of Combinatorics, Information & System Sciences, 18, 116–122.
- Chatterjee, K., & Narasimhan, G. (2002). Graph theoretic techniques in D-optimal design problems. Journal of Statistical Planning and Inference, 102, 377–387.
- Chatzopoulos, S. A., & Kolyva-Machera, F. (2006). Some D-optimal saturated designs for 3 × m2 × m3 factorials. Journal of Statistical Planning and Inference, 136, 2820–2830.
- Karagiannis, V., & Moyssiadis, C. (2005). Construction of D-optimal s1 × s2 × s3 factorial designs using graph theory. Metrika, 62, 283–307.
- Karagiannis, V., & Moyssiadis, C. (2008). A graphical construction of the D-optimal saturated, 3 × s2 main effect, factorial design. Journal of Statistical Planning and Inference, 138, 1679–1696.
- Kraft, O. (1990). Some matrix representations occurring in two factor models. In R. R. Bahadur (Ed.), Probability, statistics and design of experiments (pp. 461–470). New Delhi: Wiley Eastern.
- Mukerjee, R., Chatterjee, K., & Sen, M. (1986). D-optimality of a class of saturated main effect plans and allied results. Statistics, 17(3), 349–355.
- Mukerjee, R., & Sinha, B. K. (1990). Almost saturated D-optimal main effect plans and allied results. Metrika, 37, 301–307.
- Pesotan, H., & Raktoe, B. L. (1988). On invariance and randomization in factorial designs with applications to D-optimal main effect designs of the symmetrical factorial. Journal of Statistical Planning and Inference, 19, 283–298.