Tuppence writes:
Data itself cannot be biased. That is a property of the human being looking at it.
That is a significant and fundamental misunderstanding of what bias is. Bias is not the same as prejudice, which is something human. Prejudice is a human being's thinking that things must be one way and not another irrespective of evidence.
Prejudice is a kind of bias, but it is not the main or only thing we mean by data bias. Bias is a tendency. As most of you know, when experiments are done and measurements made, there is always some error in each measurement. Every measurement of the speed of light gives a value different to some degree from the actual value of the speed of light. Data bias or systematic error consists of measurements that are, for one reason or another, systematically greater or less than the actual value of the thing being measured. For example, if one were trying to measure the speed of light with a rotating-mirror type experiment with a baseline of measured length 1.00 km, but the actual length were 1.01 km (an improbably large measurement error, but this is for illustrative purposes) then the reported measured speed of light from this experiment would be systematically about one percent too small. That would be a kind of data bias having nothing to do with the prejudices of a human mind.
Finding sources of systematic error in experiments is really the essence of the experimental art. They are often subtle, and hard to detect even by those skilled in experiments. Whether apparent discrepancies between late 19th-centurements of the speed of light and contemporary measurements are the result of such data bias would require a much deeper analysis of the details of the 19th-century experiments than I am able to present, and I say nothing here to resolve that question. In this post I simply want to make clear that data bias is a real property of data independent of any human mind analyzing that data.