When philosophers formulate hypotheses, it’s obvious that it’s not so easy to test them empirically, says the epistemologist Jan Sprenger. | Photo: Bea De Giacomo

Jan Sprenger, you deal with the philosophical aspects of objectivity and subjectivity in research. If someone has been personally involved or affected by the topic of their own research, to what should they pay particular attention?

In the philosophy of science, we distinguish between two dimensions. When you’re generating a hypothesis, personal experiences or perspectives are allowed to play a role. But you have to evaluate your hypothesis according to the evidence you produce, and you have to work in accordance with the standards that are the norm in your discipline. What’s more, you can’t use your own experiences as evidence to confirm specific theories. And you naturally have to watch out that your personal experience doesn’t cloud your judgement.

So whether you’re allowed to let your personal experiences come into play depends on your position in the knowledge production process?

Yes. That also applies to incorporating your own perspectives.

What about you – how objectively can you ponder the topic of objectivity?

Well, when we philosophers formulate our hypotheses, it’s naturally difficult to test them empirically. Instead, we rather try to explain concepts such as ‘objectivity’ by describing the function they assume in the scientific discourse. Then we explain why one approach is more promising than another.

How does objectivity function in the social sciences and the natural sciences respectively?

The natural sciences have an easier time of it. This is because they have a greater number of quantitative theories that are also more successful, enabling researchers to make precise predictions and test them accurately. There’s a specific example that you often find in physics textbooks: deviations in the orbit of the planet Uranus allowed researchers to predict with success the position of Neptune. In cases of doubt, the natural sciences also have laboratories at their disposal where you can conduct experiments without any relevant external influences – experiments into thermodynamics, for example. The situation in the social sciences is much trickier because they’re dealing with complex systems and a whole network of causal influences that can’t be screened off so easily.

Can you give us an example?

How does racism come about? This is a question that’s hard to test in the lab. And there are many factors that impinge on it. Of course people also conduct ‘lab experiments’ in the social sciences to address specific questions such as: How much do people trust each other in matters of money? But it remains open as to how such idealised scenarios might afterwards be applied to society at large.

So in concrete terms, how can you establish objectivity in research?

It’s complicated. But there are two basic ideas that dominate here. The one deals with the actual product of research. For example, the findings from an experiment might be deemed ‘objective’ if they can be replicated independently by different researchers at different times in different places. Then there is the process by which personal bias can be kept out of empirical research. When we talk about objectivity in scholarship without specifying it further, we’re usually referring to an area in which these two ideas of objectivity overlap. But I find it important to distinguish between the objectivity of the product and of the process.

That’s pretty abstract.

It is. But what interests me most about the concept of objectivity is how useful it can be in everyday research practice. Let’s remain with the process aspect for the moment. When it’s impossible to keep personal values out of their scientific arguments or their data analysis, researchers have to make this fact transparent. It might concern their selection of the hypotheses to be tested, or the probabilities assigned to them. The researchers in question could also reveal whether the conclusions from an experiment would change, were they to base it on different assumptions.

In light of this, can we avoid subjectivity at all?

Well, it’s impossible to analyse a dataset without making certain assumptions. These could be about what factors influence people’s trust in each other, and what factors do not. In such cases, our subjective judgement is essential. This is why, in so-called Bayesian statistics, these subjective elements are an explicit component of the scientific reasoning. Hypotheses are assigned subjective probabilities whose value can change in light of the experimental findings. In classical statistics, however, no one speaks of the probability of a hypothesis. All that matters is how well or not the null hypothesis can explain the data – a null hypothesis meaning that no causal relationship exists, which is the standard assumption that we make without having conducted any investigation. Any and every subjective judgement will be dismissed as non-scientific.

So where’s the problem?

Apart from anything else, this form of standardisation means that statistical research ultimately only publishes so-called significant results – results that correspond to a specific p-value (the probability value). However, these figures only allow us to state that the data are unlikely to have been obtained by chance. I can explain this with an example. If there is a large experiment involving 100,000 patients in which a drug performs even minimally better than a placebo, then the very large sample size almost certainly makes this finding statistically significant. However, this says nothing about whether the drug in question actually contributes to combating the disease. This is why, for several years now, there’s been a movement in science that is urging us to move away from an exclusive reliance on p-values. But that used to be the dominant approach for several decades and caused a lot of damage.

Can objectivity be a good thing at all?

It’s thanks to objectivity that people trust science, and it’s objectivity that makes science relevant in societal discourse. If you take the objectivity of a process as your starting point – in other words, if your primary endeavour is to exclude personal bias as much as possible – then the idea of objectivity isn’t associated with overly high expectations. But if you understand objectivity as conforming to reality, that’s something that science can’t always guarantee. It’s far too complex for that.

Would such an understanding add a new dimension of objectivity, alongside the objectivity of the process and the product?

The idea of objectivity conforming with reality is naturally an old one. It is also partly implicit in the notion of the objectivity of the product. But this conformity is difficult to measure directly.

So to sum up, what would have to be fulfilled to achieve a practical combination of product and process objectivity?

It would have to be based on experimental work; there would have to be a high degree of reproducibility; and it would have to entail an assessment of theories that is guided more by evidence than by personal convictions.

That would fit well in the natural and social sciences. But what about the humanities? A historian can quite deliberately choose to consider a source from the perspective of a specific, marginalised group. In such a case, the ‘bias’ would itself comprise the method.

Such an approach can be perfectly sensible when it can allow you to develop a previously unknown perspective. But it’s important to state clearly that you’ve adopted this approach. And it’s important that there are sufficient other approaches that interpret the source from other perspectives. Scholarly progress in the humanities arises precisely from such confrontation.

What do you mean by that?

Natural scientists accept a paradigm by working on it. In the humanities, people instead tend to try and deepen their understanding of a particular phenomenon by using contrasting interpretative approaches. In such cases, a scholarly consensus would represent an objective finding. Unlike Newton’s laws in the field of physics, the humanities have hardly any general theories to guide researchers. They are much more dependent on studying specific, individual cases.