Blog

Technical Pieces Friday 27th October 2017

New Problems - New Designs

In 1919, nearly 100 years ago, R.A. Fisher started work at Rothamsted Experimental Station. His experience there led to him publishing in 1925 the book Statistical Methods for Research Workers. It has rightly been highly influential. Much of our scientific knowledge about agriculture and many other subjects was discovered with the methods of experimental design and analysis that Fisher developed and that students still learn.

The conceptual frame behind the methods is straight forward. There is a treatment effect on a response that we want to understand, so we aim to get a precise and unbiased estimate of it. Variation around the effect is 'error' so we aim to reduce it as much as possible, then replicate and average to minimize its impact. As a consequence, researchers used a randomized block design, or some elaboration of it. The concept can be summarized by Figure A below.

Figure A: ‘Fisher-type’ randomised experiment

Sometimes agricultural researchers expect the results to depend on an environmental factor – such as soil type – and do a multi-environmental trial, repeating the experiment in a few different types of soil, as in figure B:

Figure B: A Multi-environment trial

But the world of research and development has moved on, at least for research focusing on the problems of smallholder farmers. One example is the way that farms and farm livelihoods are now seen as integrated systems that need understanding as such. Another is that farms and farmers are recognised to be endlessly variable – one is never a replicate of another. Two consequences of these are:

• First, people are involved.  Social and biophysical components interact strongly.
• Second, the influence diagrams have become much more complicated. We recognise that many social, ecological and other dimensions of environment influence what is happening.

What this means is that the old types of research design, as developed by Fisher, are no longer sufficient.

Figure C: The influence diagram gets more complex

Practitioners bring people in through the use of participatory methods, and try to cope with the multiple factors influencing outcome by working in many environments and with many people at the same time – so called 'Large-N trials'. The designs often look messy and complex, and the resulting data can be confusing to analyse with no single method that can be applied universally. But they are generating important insights and results that help solve real practical problems.

Maybe what we need now is a modern-day Fisher who can put the array of ideas and methods being developed and used into a coherent and sound framework that will satisfy the needs of researchers for the next 100 years. Not an easy task!

So, how would you start to develop such a framework? Do you think it's something that can be done? Leave a comment with your thoughts, and we'll continue the discussion in future blog posts.