The Square Array Design
R. A. Bailey, L. M. Haines · 2024
Read The Square Array Design by R. A. Bailey, L. M. Haines (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-09166.txt.
Introduction
Rowland Bailey and Linda Haines set out a family of layouts for field trials where each test line gets one plot, in a 2024 paper. Such trials still need fair checks against known control lines. Their square array design grows from a helper block design: a balanced set of blocks, each holding the same few treatments. Laid out as a square, the helper becomes rows and columns with a few controls in each.
The form extends the old Youden square trick for single-plot trials. The paper probes when such designs stay connected, so every contrast can be read, and how to randomize them fairly. It hunts the arrays that shrink the error on two prized contrasts: test line against control, and test line against test line. Cyclic helper designs give a workable, bendy family. Non-cyclic helpers earn their own treatment for cases the ring cannot serve.
Catalog record
| CREATOR | R. A. Bailey, L. M. Haines |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 51 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-09166
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
The Square Array Design R. A. Bailey, L. M. Haines Abstract This paper is about the construction of augmented row-column designs for unreplicated trials. The method uses the representation of a k × t k\times t equireplicate incomplete-block design with t t treatments in t t blocks of size k k , termed an auxiliary block design, as a t × t t\times t square array design with k k controls, where k < t k<t . This can be regarded as an extension of the representation of a Youden square as a partial latin square for unreplicated trials. Properties of the designs, in particular in relation to connectedness and randomization, are explored. Particular attention is given to square array designs which minimize the average variances of the estimates of paired comparisons between test lines and controls and between test-line and test-line effects. The use of equireplicate cyclic designs as auxiliary block designs is highlighted. These provide a flexible and workable family of augmented row-column square array designs. Designs whose auxiliary block designs are not cyclic are also covered. Keywords: augmented row-column design, auxiliary block design, square array design, cyclic design, A-optimal design 1 Introduction The construction of designs for plant breeding experiments is challenging because the quantity of seed available for each test line is usually sufficient only for a single plot. Augmented designs which incorporate controls and thereby facilitate comparisons between the unreplicated test lines are well-researched and widely used. The experiments are most commonly implemented as block designs, with either complete or incomplete blocks, so that the blocks accommodate one-way heterogeneity in the field ( Piepho and Williams, 2016 ) . Augmented row-column designs which allow for two-way heterogeneity are not as straightforward to construct and have received limited attention in the literature. The early papers of Federer and Raghavarao (1975) , Federer, Nair and Raghavarao (1975) , Lin and Poushinsky (1983) and Williams and John (2003) provide somewhat restrictive augmented row-column designs, while the more recent papers by Piepho and Williams (2016) and Vo-Thanh and Piepho (2020) are primarily concerned with the introduction of an additional blocking structure in order to accommodate regional heterogeneity. The present study is concerned with a family of augmented row-column designs for unreplicated trials which are theoretically tractable and practically appealing. The paper is organised as follows. The designs of interest, that is the t\times t square array designs with k controls, where 2<k<t , and the attendant fixed-effects model adopted for analysis, are introduced in Section 2 . The main results relating to the construction and randomization of augmented row-column square array designs developed from auxiliary block designs are presented formally in Section 3 . Section 4 provides a comprehensive account of the use of equireplicate cyclic designs as auxiliary block designs. Section 5 extends this to auxiliary block designs which are not cyclic and provide augmented square array designs with low values of the average variance metrics. A brief discussion of the paper and some pointers for future research are given in Section 6 . An R package for constructing auxiliary block designs and square array designs and R programs to reproduce the examples in the paper are available in the Supplementary Material. 2 Preliminaries The designs for unreplicated trials introduced here are augmented row-column designs based on square arrays. Specifically, consider a row-column design with plots arranged in t rows and t columns and with k controls (with k<t ) located once in each of the rows and once in each of the columns. There are thus a total of tk plots for the controls and t(t-k) plots for the test lines. The designs are implemented as follows. Initially, the k controls are randomly allocated to the specified sets of plots occurring once in each row and once in each column, and the test lines are randomly allocated to the remaining plots. Then, following Bailey (2008, p. 108) , the rows and columns of the square array design are randomly permuted. A fixed effects model for the t^{2} yields of the design and a total of v=k+t(t-k) treatments, that is controls and test lines, is adopted. Specifically, the t^{2}\times 1 vector of yields, denoted \mathbf{y} , is expressed as \mathbf{y}=\mathbf{X}\bm{\tau}+\mathbf{Z}_{r}\bm{\rho}+\mathbf{Z}_{c}\bm{\gamma}+\mathbf{e}, where \bm{\tau} is a v\times 1 vector of fixed treatment effects with a t^{2}\times v design matrix \mathbf{X} , and \bm{\rho} and \bm{\gamma} are vectors representing fixed effects for the t rows and t columns with attendant t^{2}\times t design matrices \mathbf{Z}_{r} and \mathbf{Z}_{c} , respectively. The error term \mathbf{e} is assumed to be distributed as N(0,\sigma^{2}\mathbf{I}) , where \mathbf{I} is the identity matrix, and the variance \sigma^{2} is taken, without loss of generality, to be one. The treatment information matrix associated with the model follows immediately and is given by \mathbf{C}=\mathbf{X}^{\top}(\mathbf{I}-\mathbf{Z}(\mathbf{Z}^{\top}\mathbf{Z})^{-}\mathbf{Z}^{\top})\mathbf{X} , where \mathbf{Z}=[\mathbf{Z}_{r}~\mathbf{Z}_{c}] and (\mathbf{Z}^{\top}\mathbf{Z})^{-} is a generalized inverse of \mathbf{Z}^{\top}\mathbf{Z} . Then, if \mathbf{C} has rank v-1 , the design is connected and all treatment contrasts are estimable. The precision with which contrasts of the treatment effects are estimated is assessed by the average variance of estimates of the pairwise differences between the test-line effects, denoted A_{tt} , between the control and test-line effects, denoted A_{ct} , and between the control effects, denoted A_{cc} . These metrics can be invoked as criteria for comparing candidate designs, with A_{tt} deemed to be of primary interest and A_{ct} of secondary interest. The metric A_{cc} is introduced here for completeness. Thus, t\times t square array designs for which A_{tt} is a minimum and A_{ct} is small, and possibly a minimum, are sought. Section 3.2 shows that these two metrics achieve their minima together. Kempton (1984) suggested a range of 20 to 25 % for the percentage of plots occupied by the controls in unreplicated trials as a balance between theory and practice. This rule-of-thumb remains in general use today and is broadly followed in the examples in the present paper. In addition, the number of error degrees of freedom in a square array design is equal to tk-2(t-1)-(k-1)-1=(t-1)(k-2), which is zero if k=2 . The assumption that k\geq 3 is therefore made throughout the text. An early example of an augmented row-column square array design was given by Federer and Raghavarao (1975, pp. 33–34) . They transformed a 3\times 7 Youden square by interchanging rows and letters into a 7\times 7 augmented row-column design with three controls; the construction is clearly illustrated in their paper. It seems that the Youden square has been interpreted as the key component in this construction and that, as a consequence, the approach has not been pursued. Indeed, Vo-Thanh and Piepho (2020) state that “ \ldots the dimensions are restrictive since they [sic] used Youden designs.” This example forms the basis for a much broader strategy of construction for augmented row-column square array designs from equireplicate incomplete-block designs which is developed in the present study. 3 Main results 3.1 Auxiliary block designs The interpretation of the Youden square as a balanced incomplete-block design (BIBD) and, more broadly, as an equireplicate, incomplete-block design in the construction of Federer and Raghavarao (1975) can be used to great advantage by noting the following theorem in Bailey (2008, p. 235) . Theorem 1 Suppose that \Gamma(k,t) is