Rouxléne van der Merwe
Senior lecturer at Plant Breeding, University of the Free State

Plant breeding trials, or essentially field experiments, are used to test genotypes (such as sunflower hybrids, wheat lines) or any other type of treatment and estimate their potential for a range of variables (e.g. yield, quality characteristics or disease resistance) at all stages of the breeding process. Just as a plant breeder chooses a specific breeding technique that is applicable to the crop being genetically improved or more specifically the plant characteristics (or variables) targeted during selection, the trial design used should also be appropriate in order to statistically analyse the data collected and to derive valid conclusions from the results obtained. Thus, to ensure that only the best genotypes are kept for further selection in the breeding process, those variables used as a basis for selection need to be accurately estimated. Accurate estimations of variables can only be obtained from statistically designed trials and data analysis. The principle for experimental design is to keep the design as simple as possible while satisfying the required level of scientific soundness. According to Michael Casler1, experimental design is based on four pillars, i.e. replication, randomisation, blocking and experimental units (plots). These individually need to be considered properly and require a conscious decision regarding number, size, scale, shape or form as they have a profound impact on the experimental design, data analysis and conclusions that result from the experimental conduct. Many design types have been developed for field experiments. Any of several designs may be possible for a particular experiment, but each design has its own advantages and disadvantages. For example, the completely randomised design (Fig. 1) is used to compare genotypes when environmental conditions are fairly uniform and the plots that receive the same genotype (or other treatment) are uniform. If this were not the case, then it would be difficult to detect real differences between genotypes. By increasing the number of replications per genotype, the precision can be improved; however, a larger number of replications will result in a larger trial site required and finally environmental conditions will not be uniform anymore. To increase precision of genotype comparisons, the principle of blocking should be used. A randomised complete block design (Fig. 2) is most often used in field experimentation if systematic variation in soil conditions (e.g. gradual increase in pH, soil fertility or water capacity) across the trial site is suspected. With this design, the genotypes are randomly assigned to plots within a block such that each block contains a complete set of genotypes. In this case a complete block is representative of a replication. By forming blocks of similar plots, experimental error is reduced and the precision of genotype comparisons is increased. This is done by blocking the effects of site variability where blocks are arranged at right angles to the gradient. This design should, however, not be used when a large number of genotypes (e.g. more than 21) are being tested since larger complete blocks will not be uniform and true differences between genotypes will not be detected anymore. One solution to this is to use the incomplete block design. With an incomplete block design, the blocks are subdivided into smaller, incomplete blocks, which are more homogeneous. Thus, the replicate is divided into incomplete blocks that contain a fraction of the total number of genotypes. Genotypes are distributed among the blocks so that pairs occur in the same incomplete block in nearly equal frequency. The most commonly used incomplete block design is the alpha-lattice for replicated trials. In general, the alpha-lattice design (Fig. 3) is a risk free way to increase the precision of large field trials of more than 21 genotypes where soil heterogeneity plays an important role. However, this design does not allow for the analysis of a factorial type relationship among treatments, for example testing for the relationship between two different fertilizer treatments on several genotypes. In addition, constraints to the use of these designs include a lack of an appropriate language-independent software interface for entering and transforming data files, difficulty of handling information in a common format and lack of integration with other analytical tools. Keeping in mind that field experimentation and planting of breeding trials are expensive, it is of utmost importance that satisfactory results are obtained and that correct inferences about genotype differences are developed. Finally, it is advisable for plant breeders to become proficient in experimental trial design and data analysis or alternatively to befriend someone who has the necessary experience.

1Casler MD (2015) Fundamentals of experimental design: guidelines for designing successful experiments. Agronomy Journal 107:692-705

Fig. 1 Completely randomised design

A1 B1 C1 A2
D1 A3 D2 C2
B2 D3 C3 B3
C4 A4 B4 D4

Four treatments (A-D) with four replications (1-4) are assigned to each plot completely at random.

Fig. 2 Randomised complete block design

Block 1 Block 2 Block 3 Block 4

Four treatments (A-D) are randomised and each occurs once in each block. Each block represents a replication.

Fig. 3 Alpha-lattice (incomplete block design)

Block 1 S 1 1, 2, 3, 4
S 2 5, 6, 7, 8
S 3  9, 10, 11, 12
S 4 13, 14, 15, 16
S 5 17, 18, 19, 20
S 6 21, 22, 23, 24
Block 2 S 1 10, 15, 19, 23
S 2 1, 5, 14, 20
S 3 2, 6, 12, 22
S 4 3, 9, 17, 21
S 5 4, 8, 18, 24
S 6 7, 11, 13, 16

Twenty-four treatments (1-24) are divided into six incomplete blocks (S1-6) with four treatments per incomplete block. Treatments are randomised across and within the incomplete blocks and this design needs a minimum of two blocks since trial areas can become large.