Harnessing complexity may strangle it: what is a well-controlled experiment?

This article discusses a core ideal of biological experimentation, the goal of controlling as strictly as possible. We present the epistemological argument underlying the ideal of a perfectly controlled experiment. While most scientists agree that the ideal is unattainable in practice, they still uphold the pragmatic goal that one should strive to approximate the ideal as much as possible in all cases. We argue that this desire for strict control is problematic and outline an alternative understanding of control in experimentation. First, we distinguish four different ways in which real-world biological experiments can fall short of the ideal of perfect control. Then, we argue that strict control does not always yield better experimental results. Certain core features of biological systems actually suggest that strict control can be counterproductive for learning from experiments. We thus argue for 'controlling well', for striking a balance between holding variables fixed while allowing some degrees of freedom. What it means to control well cannot be specified in general terms; it is context- and purpose-dependent and needs to be determined relative to a particular experimental project.