Map of life expectancy at birth from Global Education Project.

Sunday, July 11, 2021

On the bias of science: Part three

Okay, we have established that universities, as a class, do not have any agenda or preference for that the outcome of scientific investigations will be. The exceptions, obviously, are institutions like Liberty University, Bob Jones, and Oral Roberts, that deny evolution and cosmology a priori. But at colleges and universities that are not run by religious fanatics, faculty investigators are independent and they pursue their own lines of research, conduct investigations of their design, and interpret the results as they understand them. We have also established that the government does not have any agenda or preference for what the outcome of the research it funds may be. Well, of course people who conduct biomedical research are hoping they'll find the cure for cancer or Alzheimer's or whatever, but if they don't, they don't. 


Can they try to fake it? Sure, not everyone is ethically scrupulous, and there have been plenty of examples of scientific fraud. People want fame, tenure, promotions, and sometimes they think they need a big discovery to get it. There are basically two reasons why frauds usually don't survive for long, however, at least not in areas that are consequential. (Obviously a finding that nobody much cares about won't attract a lot of attention. We'll get to that.) By the way, just so we're clear, fraud is not all that common.


The scientific publication process is the first screen. It isn't actually very effective at detecting sophisticated fraud, but you do need to get past it. First, a journal editor will read your submission and decide if the work appears to be methodologically sound, and of sufficient interest to the journal's readers to merit consideration. If it's a go, the paper is sent out to peer reviewers, usually three, who are volunteer scientists with relevant expertise. They normally won't be able to tell if you just made up your numbers, but they can assure that the methodology is thoroughly described and appropriate, and the results are plausible. If the numbers don't add up, you're gonzo. 


Say you did make up the numbers, but you were able to get them to add up and seem plausible, and the paper gets past peer review and is published. If the results are important to the field, and suggest new directions of research that build on them, other scientists will try to do so, and if it doesn't work they'll find out soon enough. 


But frauds that don't get in the middle of other people's programs of research may not be noticed for a long time. On the other hand, they don't do much harm. There are various ways that fraudsters get caught eventually, though no doubt some never do. But again, they have little consequence for the broad enterprise of science. If you're interested in learning more about this, you might want to check out Retraction Watch.


However, the more important or at least systemic problems don't have to do with fraud, per se. I'll get to that next time.


No comments: