Reconciling subjectivity and objectivity in science
What’s the difference between the Trump team’s ‘alternative facts’ and scientific truth? Objective methods of scientific inference is the standard response. But while there is an – objective – difference between scientific facts and falsehoods, the idea that sound science is free of personal values or subjective assumptions can lead to dangerous biases. The Objectivity (Making Scientific Inferences More Objective) project, which received funding from the European Research Council, is rethinking the concept of objectivity in scientific inference. Outdated ideas of scientific objectivity as being free of any subjective element often hinder the promotion of scientific progress, remarks Jan Sprenger, principal investigator of the project and professor of the Philosophy of Science from the Department of Philosophy, University of Turin. “Unfortunately, many scientists and journal editors tend to sweep these elements under the carpet.” According to Sprenger, this practice has contributed significantly to the ongoing replication crisis, which sees researchers struggle to reproduce the results of previous experiments. “We have argued that an explicitly subjective stance on scientific inference increases the transparency of scientific reasoning. Thus, it also facilitates the verification of scientific claims and contributes to a higher degree of reliability of the conclusions.” So how do we translate this approach into better science? The project team developed practical tools for improving statistical, causal and explanatory inference reconciling subjective choice with the aim of objective knowledge.
Subjectivity built-in
One example of statistical malpractice contributing to the replication crisis is p-hacking, where researchers select the analysis or data that best fits the desired conclusion. The Objectivity team highlights the promise of Bayesianism, which uses the subjective interpretation of probability, for improving statistical inference: their work shows that experiments designed and analysed using this method led to more accurate estimates compared to the conventional method. Causal inference is the process through which causes are inferred from data. In medicine, for instance, randomised controlled trials aim to measure causal strength to study the effectiveness of a new treatment. The researchers argue for a specific measure of causal strength: the difference that interventions on the cause make for the probability of the effect. This probability can be interpreted objectively (as frequencies, propensities, etc.) or as subjective degrees of belief, dependent on the context. Explanatory inference is the process of choosing the hypothesis that best explains the data at hand. This concept has been notoriously vague, notes Sprenger: “What is a ‘good’ explanation? The gut feeling of a scientist? In our work, we have provided a rigorous foundation of this mode of inference via the construction and comparison of various measures of explanatory power.” The team identified a close relationship between prior beliefs and explanatory power. The quality of an explanation, and the inference of the ‘best explanation’, is hence not a purely objective matter, but entangled with subjective beliefs. “Procedures for evaluating experiments and their statistical analysis should be adapted: we need to lose our fear of subjective elements in inference,” Sprenger concludes. “Science is superior to superstition not because it does not allow for subjective elements, but because its conclusions are rather resistant to variation in subjective input, and because it allows for rational criticism of the assumptions it makes.”
Keywords
Objectivity, subjectivity, scientific inference, statistical inference, causal inference, explanatory inference, Bayesianism, bias