Abstract
The order-of-addition experiments aim at determining the optimal order of m components such that the yields are optimized. The component orthogonal array (COA) allows to economically find out the optimal order by testing some carefully selected orders from all of the m! orders. This paper proposes a new method of finding COAs of broader run sizes. As an application of the new method, some COAs with 4, 5 and 6 components are tabulated. The D-efficiencies of the COAs found by the new method are investigated under the pair-wise ordering model.
Similar content being viewed by others
References
Ding X, Matsuo K, Xu L, Yang J, Zheng L (2015) Optimized combinations of bortezomib, camptothecin, and doxorubicin show increased efficacy and reduced toxicity in treating oral cancer. Anti-Cancer Drug 26:547–554
Fuleki T, Francis FJ (1968) Quantitative methods for anthocyanins. J Food Sci 33:266–274
Land AH, Doig AG (1960) An automatic method of solving discrete programming problems. Econometrica 28(3):497–520
Lin DKJ, Peng JY (2019) Design and analysis of order of addition experiments: a review and some thoughts (with discussion). Qual Eng 31:49–59
Mee RW (2020) Order-of-addition modeling. Stat Sin 30:1543–1559
Olsen GJ, Matsuda H, Hagstrom R, Overbee R (1994) Fastdnaml: a tool for construction of phylogenetic trees of DNA sequences using maximum likelihood. Comput Appl Biosci 10(1):41–48
Peng JY, Mukerjee R, Lin DKJ (2019) Design of order-of-addition experiments. Biometrika 106(3):683–694
Stewart CA, Hart D, Berry DK, Olsen GJ, Wernert EA, Fischer W (2001) Parallel implementation and performance of fastdnaml—a program for maximum likelihood phylogenetic inference. In: ACM/IEEE 2001 conference supercomputing, pp 32–32
Van Nostrand RC (1995) Design of experiments where the order of addition is important. In: ASA proceedings, physical and engineering sciences section, pp 155–160
Voelkel JG (2019) The designs of order-of-addition experiments. J Qual Technol 51(3):230–241
Wagner JJ (1995) Sequencing of feed ingredients for ration mixing. Beef Report, Dept. of Animal Science, South Dakota State University, pp 52–54
Yang JF, Sun FS, Xu HQ (2020) A component-position model, analysis and design for order-of-addition experiments. Technometrics. https://doi.org/10.1080/00401706.2020.1764394
Zhao YN, Liu MQ, Lin DKJ (2020) Optimal designs for order-of-addition experiments. Manuscript
Acknowledgements
The authors would like to thank the associate editor, and two reviewers for their constructive comments and suggestions. This work was supported by the National Natural Science Foundation of China (Grant Nos. 11801331, 11771250, and 11771219) and Natural Science Foundation of Shandong Province (Grant No. ZR2018BA013).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflicts of interest
On behalf of all authors, the corresponding author states that there is no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Electronic supplementary material
Below is the link to the electronic supplementary material.
Appendices
Appendix I: Proofs of theorems
Proof of Theorem
1. We first prove the “only if ” part. Suppose the fractional OofA design A is a COA(n, m, 2) which consists of the \(i_1,i_2,\ldots ,i_n^{th}\) rows of \(O_m\). Then the \(i_1,i_2,\ldots ,i_n^{th}\) entries in X are 1’s with the remainder of the m! entries being 0’s. According to the definition of COA(n, m, 2), the order \(c_ic_j\) appears totally \(\lambda =\frac{n}{m(m-1)}\) times in the \(i_1,i_2,\ldots ,i_n^{th}\) rows of each 2-column subdesign \((l_p,l_q)\) of \(O_m\). This implies that there are totally \(\lambda \) 1’s in the \(i_1,i_2,\ldots ,i_n^{th}\) rows of each \(e^{l_p,l_q}_{c_ic_j}\). Therefore, \(e{^{l_p,l_q}_{c_ic_j}}'X=\lambda \) for any \(p\ne q\) and \(i\ne j\) with \(1 \le p,q,i,j\le m\). Then \(D_m'X=\lambda \mathbf{1}_{m^2(m-1)^2/2}\) follows from the structure of \(D_m\) immediately.
Now, we prove the “if ” part. Suppose A consists of the \(i_1,i_2,\ldots ,i_n^{th}\) rows of \(O_m\) and X of A is the solution of Eq. (2). Clearly, \(D_m'X=\lambda \mathbf{1}_{m^2(m-1)^2/2}\) means \(e{^{l_p,l_q}_{c_ic_j}}'X=\lambda \). This implies that the order \(c_ic_j\) appears totally \(\lambda \) times in the \(i_1,i_2,\ldots ,i_n^{th}\) rows of any 2-columns subdesign \((l_p,l_q)\) of \(O_m\) for \(p\ne q\) and \(i\ne j\) with \(1 \le p,q,i,j\le m\). Therefore A is a COA(n, m, 2). This completes the proof. \(\square \)
Proof of Theorem
2. Theorem 2 is a straightforward extension of Theorem 1. We omit the proof here. \(\square \)
Before proving Theorem 3, we first introduce some notation. For a fractional OofA design A, we use \(A({c_{i_1},c_{i_2},c_{i_3},c_{i_4}})\) to denote the resulted design after dropping the other components from A, where \(1\le {i_1}< {i_2}< {i_3}<{i_4}\le m\). Denote \(n_{i}\) as the number of the \(i^{th}\) row in \(O_4\) which appears in \( A({c_{i_1},c_{i_2},c_{i_3},c_{i_4}})\). Zhao et al. (2020) provided the necessary and sufficient conditions for constructing fractional OofA designs which have the same \(D_{PWO}\)-efficiency as the full OofA design. We introduce the results as Lemma 1 below.
Lemma 1
An n-row fractional OofA design A have the same \(D_{PWO}\)-efficiency as the full OofA design if and only if the following equations
hold for all \( A(c_{i_1},c_{i_2},c_{i_3},c_{i_4})\)’s with \(1\le {i_1}< {i_2}< {i_3}<{i_4}\le m\).
Based on Lemma 1, we give the proof of Theorem 3 below.
Proof of Theorem
3. We first prove that the equations in Lemma 1 hold for COA(n, m, 4). By the definition of COA, each of the 24 orders of the 4 components \(c_{i_1},c_{i_2},c_{i_3},c_{i_4}\) appears \(\lambda =\frac{n}{m(m-1)(m-2)(m-3)}\) times in every 4-column sundesign of a COA(n, m, 4), which means that \(A({c_{i_1},c_{i_2},c_{i_3},c_{i_4}})\) is a \(\lambda \left( {\begin{array}{c}m\\ 4\end{array}}\right) =\frac{n}{24}\) replication of \(O_4\). Clearly, for every 4-component combination \(c_{i_1},c_{i_2},c_{i_3},c_{i_4}\) the equations in Lemma 1 are valid for \(A({c_{i_1},c_{i_2},c_{i_3},c_{i_4}})\). As a result, a COA(n, m, 4) have the same \(D_{PWO}\)-efficiency as the full OofA design \(O_m\). Of course, COA(n, m, t)’s with \(4<t\le m\) have the same \(D_{PWO}\)-efficiencies as the full OofA design. This completes the proof. \(\square \)
Appendix II: Some useful design tables
Rights and permissions
About this article
Cite this article
Zhao, Y., Li, Z. & Zhao, S. A new method of finding component orthogonal arrays for order-of-addition experiments. Metrika 84, 805–824 (2021). https://doi.org/10.1007/s00184-020-00791-1
Received:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00184-020-00791-1