Nice article,
@ggingues. This bit is useful for those interested in these issues:
article said:
data dredging comes in two basic forms, known as p-hacking and HARKing.
P-hacking is when you run multiple tests on a data until you find one that's significant, then report only that. Its a problem because the more tests you run, the greater your likelihood of getting a positive results by chance (we presume the chance of getting a significant result when no difference exists out there in the real world is about 1 in 20, so if you do 20 tests, at least 1 is bound to be significant by chance alone).
HARKing stands for "hypothesizing after the results are known." If the initial hypothesis is not confirmed, you hunt through the data retrospectively to find something positive, then you pretend that's what you hypothesised all along. This increases the chances of finding false positives and makes results seem stronger than they really are.
