blind analysis

Blind analysis in science, is carrying out analysis in such a way that the scientists carrying it out cannot tell whether values produced by the analysis are near previously determined or presumed results. For example, carrying out a new measurement of a physical constant in such a way that you do not know the real data. The purpose is to avoid confirmation bias, a tendency to accept results that match what you expect. The method has become common in some branches of physics, e.g., particle physics and cosmology, after some review revealed that confirmation bias affects those fields (even though one might imagine "hard science" fields are immune). There is current interest in making use of it in other fields: many fields of science and research have shown evidence of a bias toward a result that is expected and/or would be considered "successful".

I expect there is some creativity in developing a means of carrying out effective blind analysis for an experiment (or some particular type of experiment). An example method, in principle, might be having some of the input data multiplied by a number you don't know, such as to produce a result off by some multiple which can be revealed after your analysis is done. Among other things, you give up the simple sanity tests throughout the experiment and analysis of comparing your results-so-far to what is already known (or presumed). Such simultaneous testing helps avoid wasted time after errors in calculation or analysis, but also encourages biases that blind analysis attempts to avoid.