Beware of correlations!

1 Apr 2016

March is the month of the Cambridge Science Festival and I attended a few lectures. The programme was less plant-orientated this year and one of the major themes was big data.

The lecture I attended on big data and medical science was superb. The lecturer was brilliantly modest, ensuring that his points were understood by the wide-spectrum of knowledge in the audience. Put it this way, even I understood it!

First of all, he undermined some of the bunkum that goes with the term big data. As he said, it is data analysis. The issue is that so much data is now being generated in the medical world that it is becoming ever more challenging to analyse it and extract meaningful messages.

He warned that correlations in data do not mean causation. The well known example he quoted was the close link between ice cream sales and shark attacks in Australia. This is simply explained by the fact that there are more people in the sea on the hot days when ice cream sales are high. Hence, stopping ice cream sales will not prevent shark attacks.

Another example I can remember was when someone found a correlation between the fall in sparrow numbers in the UK and the rise in lead-free petrol sales. This amazingly lead to suggestions that there should be an investigation into the environmental impact of the additives in lead-free petrol but that proposal soon ran out of gas (pun intended).

In the medical world, there is a real danger of correlations leading to false conclusions that may impact on the lives of many people and so much of the lecture was based on this issue. There are some similarities in arable agriculture.

During the 1970s it was fashionable to say that high wheat yields were correlated with crops that had a high above-ground biomass at harvest. When you think about it, it is stating the b******* obvious. At harvest much of the biomass is in the grain and the rest is in the remaining above ground plant matter, notably straw. The two are related. Varieties tend to have a consistent harvest index, which is the proportion of total above-ground dry matter that is in the grain. Hence, inevitably high yields must be associated with high levels of biomass.

The correlation between wheat grain yield and high biomass crops was the basis of the Schleswig-Holstein system. This attempted to have a high biomass throughout the life of the crop, starting with high seed-rates and continuing with encouraging high growth rates from early spring onwards. Also in the 1970s there was another approach called the Laloux system. This stemmed from research by the then Professor of Agronomy at the University of Gembloux in Belgium. This was based on modest seed rates (roughly what we use now) and a more measured programme of feeding and protecting the crop in the spring and early summer. I did trials comparing the two approaches and the Laloux system consistently outyielded the Schleswig-Holstein system. The striking observation of the trials was that by harvest the two systems gave similar looking crops. Obviously the weather intervened and influenced final tiller numbers and biomass.

In the event, the then best UK practice was equal to or superior to the Laloux system. This was partly due to the fact that the Laloux system relied on late nitrogen applications. These are relevant to Belgium because of the regular incidence of thunderstorms in late May and June but not to the UK where it may be as dry as toast at that time of year.

Hence, whilst there is an obvious correlation between wheat biomass at harvest and wheat yield, it may not necessarily be pointing a way towards growing higher yielding crops. Obviously, things would be different if there was a correlation between yield and wheat biomass at a far earlier growth stage.

Last year proved that good standard UK practice can result in very high levels of biomass and yields provided that the weather is with us. Therefore, the approach must surely be to ensure that crops are able to take advantage of such conditions in order to achieve high levels of biomass at harvest without spending a shed load of money.

By the way, Gembloux is a very small town. I have visited the superb Department of Agronomy of the Agricultural University a few times and on occasions my wife came along too. The first time I suggested she looked around the town and we would meet up for lunch. I think it took her less than an hour to ‘do’ the town but the salads in the restaurant at the gates of the University made up for that.