How precise are the sciences, Dieter Ebert?
Text: Dieter Ebert
The humanities and natural sciences differ fundamentally on the issue of how to depict the world accurately. They also deal with the problem of imprecise findings in different ways.
The natural sciences – sometimes described as the “exact sciences” – are often associated with the uncovering of facts and laws. Even at school, people learn physics, chemistry and biology in the form of clear concepts and established facts. That is what distinguishes the natural sciences from the humanities. However, in modern biology – although not only there – this is becoming less and less true, with statements of fact increasingly being replaced by probabilities.
The sort of statements we are familiar with from the weather forecast – for instance, that there is a 50% chance of rain – also feature in biology and medicine. Vaccines reduce your likelihood of coming down with a particular infection, specific gene variants increase the risk of breast cancer, and there is a certain probability that plant seeds will germinate the following spring. The reasons for this lack of precision are different in each case, and are often impossible to check for. If I have only one seed, I cannot say whether it will germinate. If I have 100 seeds, however, it is possible to state what percentage of them is likely to germinate. Thus, to express findings we need statistics – indications of the frequency with which a statement holds true.
In order to place statements within a meaningful framework, scientists compare probabilities. The effect of one treatment (a drug, a fertilizer, a toxin or radiation) is compared with that of another (no active ingredient or a placebo), for example. If the difference between the objects that have been exposed to different treatments is big, we say that there has been a significant treatment effect. The question of whether the difference is great enough to constitute a significant effect is determined by a convention, which can be explained like this: If you conduct an experiment 20 times using an ineffective treatment, it may produce a significant difference on one occasion by chance. In other words, now and again you find a significant effect that is not really there.
The converse is also true: By chance, an effective treatment may fail to produce a significant effect. By increasing the outlay, it is possible to reduce the incidence of such false conclusions, but they can never be wholly eliminated. The gold standard is to repeat an experiment that has demonstrated an effect. If it has revealed a true correlation, it will probably do so again. Unfortunately, experiments are often not repeated when they yield positive results, but rather only when the results are negative. Unfortunately, this even reduces the probability of finding the real correlation.
Such practices, and others, are partly responsible for the so-called replication crisis, which has been rocking the life sciences for some time now. According to various studies, a high proportion of biological and medical findings are not replicable. Can biology still be trusted? I think that it can – that is quite clear from the advances that the field is making overall. However, progress could be faster, as every negative result is a backwards step.
As a scientist, I have developed a healthy mistrust when evaluating new findings. For example, I evaluate new scientific findings within the framework of what we think we already know. If something is totally new and cannot be compared with any previous findings, I am automatically more skeptical. Another factor is the quality of the scientific approach employed. Besides inspiration and motivation, good science requires the use of considered and careful strategies, which is unfortunately rare. We need to teach such strategies at universities. Good scientific approaches and an understanding of quantitative and statistical relationships are necessary basic tools of modern science.
More articles in the current issue of UNI NOVA.