Theme 32: Use of Information and Analytical Systems

Description

Paired-paddock comparisons are a common way of evaluating new grassland practices at a scale more relevant to farmers. They are also being used to replace or augment designed experiments and can be the only source of information available on a practice. However, it is often uncertain if the differences between paddocks are significant. Importantly, a current trend among funding organisations is to support paddock comparisons. The need for valid procedures to compare unreplicated treatments is increasingly urgent. It is suggested that a range of tools be used to infer statistical significance from using typical error values from related studies or subsampling, through to multivariate techniques to follow trends. Local ‘rules of thumb’ could be developed and data evaluated with calibrated models. A final judgement on treatment effects would need to be based upon the use of several criteria to achieve a ‘balance of probabilities’. Consideration of these problems suggests that paired-paddocks should only be used to evaluate contrasting treatments where large effects are expected and not small variations within a practice.

Share

COinS
 

The Analysis of Results from Paired Paddock Comparisons

Paired-paddock comparisons are a common way of evaluating new grassland practices at a scale more relevant to farmers. They are also being used to replace or augment designed experiments and can be the only source of information available on a practice. However, it is often uncertain if the differences between paddocks are significant. Importantly, a current trend among funding organisations is to support paddock comparisons. The need for valid procedures to compare unreplicated treatments is increasingly urgent. It is suggested that a range of tools be used to infer statistical significance from using typical error values from related studies or subsampling, through to multivariate techniques to follow trends. Local ‘rules of thumb’ could be developed and data evaluated with calibrated models. A final judgement on treatment effects would need to be based upon the use of several criteria to achieve a ‘balance of probabilities’. Consideration of these problems suggests that paired-paddocks should only be used to evaluate contrasting treatments where large effects are expected and not small variations within a practice.