Monday, May 15, 2017

Striking example of how an apparently well-replicated finding can still fail to pan out

From among a list of links at Slate Star Codex, Links 5/17: Rip Van Linkle.
I mentioned the debate over 5-HTTLPR, a gene supposedly linked to various mental health outcomes, in my review of pharmacogenomics. Now a very complete meta-analysis finds that a lot of the hype around it isn’t true. This is pretty impressive since there are dozens of papers claiming otherwise, and maybe the most striking example yet of how apparently well-replicated a finding can be and still fail to pan out.
We really like to treat science as authoritative but it is always contingent. Whatever we think we know may be usefully true, but it is always only contingently true. Till that point in time when we discover a different way of looking at the phenomena, measuring it, defining it, understanding it. The Structure of Scientific Revolutions by Thomas Kuhn and all that.

Always be careful of anyone who claims that the science is settled or tries to resolve an argument by referencing the percentage of scientists who support a particular interpretation. They are ideological salesmen with a convenient, but not accurate, understanding of science.

No comments:

Post a Comment