Unleashed by bias and chance

When I entered research, among the first things to be taught was that when studying the effect of an intervention, e.g. a drug, bias is masking its “true” effect; to a beginner’s eye, this was picturing bias as something bad to allow to have in an experiment and that we had to eliminate it.

Later, of course I understood that not all sort of bias is equally bad. Some can be easily identified and predicted in advance. This kind of bias, such as selection bias, is important and contributing heavily to the outcome and needs to be excluded or minimized close zero during an experiment. Other forms of bias, it is alright to just have them minimized to an extent possible, such as information bias. Some form of bias while known, anticipated or expected, is unavoidable and we just need to cope with during experiments. Last it is also very likely that there is a type of bias neglected, overlooked or even not thought at all. This doesn’t stop experiments to be done or published.

Overall, to reach having truly meaningful and sound results around the effect of an action or factor, bias needs to be minimized, measured, taken under consideration, handled somehow, and this is the right thing to do during measurement and analysis of the results. But when we take an experimental effect and release it to the real-world, there is no experiment anymore. In this setting, it is widely…