How to select the most promising varieties in my trials?
Trials are composed of a set of varieties that we want to rank, according to their results measured on several variables. These varieties are replicated a given number of times, possibly in different locations or other experimental conditions.
According to the possible number of replications (depending on the quantity of seeds and/or available space) and to the preliminary knowledge of heterogeneities in the field, we choose a more or less restricting design. Restricting designs, such as lattices or alpha designs, are harder to set but they provide, in return, more powerful results. In particular, these advanced designs enable to adjust the mean of each variety, thus removing possible field gradients.
It often happens that the data contains missing values. We must look into details to understand their origin. Knowing that, it can be decided which treatment we apply to them: either keep them into the analysis, by estimating their value with the available data, or remove the involved variety or replication.
A complete dataset can now be analyzed, first of all by doing an ANOVA. It helps us to validate the model - especially by studying the residuals, which can lead us to point out suspect values or unexpected field effects - as well as to find the factors having a significant effect on the results of the dataset. In addition to replication or location effects, we mainly look for a possible variety effect.
The presence of a global variety effect shows that all varieties cannot be considered as having the same level of performance. To find out which ones are the best and how we can group them, we then need to study the variety means (possibility of adjusted means when allowed by the design) on which we do multiple comparison tests. According to the sought aim, we choose a more or less conservative procedure (for example Tukey, Newman-Keuls or LSD) to construct the variety groups. Other specific methods exist to answer more precise questions, such as the Dunnett’s test which enables to select varieties significantly better than the checks.
“Elaborate statistics are not substitute for meticulous experimentation.” G.W. Snedecor
Being aware that this methodology is a standard procedure for the choice of the most promising varieties, Doriane has developed tools to facilitate and automate this treatment. Thus, the standard software RnDExperience™ permits to:
- Directly use the trial data stored into the software to make all the statistical analysis
- Generate standard outputs that can be freely customized
- Assist the user in the choice and definition of the experimental designs
Throughout the whole testing process, RnDExperience™ includes all the tools needed to prepare the campaign, manage the experiment, edit the culture units, analyze the results, share the conclusions and help in decision-making.
Doriane, your R&D IT partner
About DorianeThe expertise of Doriane brings the three players of R&D the resources to organize, share and valorize their research projects:
RESEARCH DEPARTMENTS: Fast implementation of business processes ; Liberation of researchers creativity
IT DEPARTMENTS: Real-time centralization of multilocation data ; Security and integrity of data
MANAGEMENT: Maximization of research investments ; Research departments monitoring ; Intellectual assets valorization and sustainability