Saturday, April 08, 2017

Not Fake Science, Yet

(Slide presentation here)
WSJ: Biomedical research has so many problems that most published research findings are false. Follow-up studies on various "breakthroughs" could only reproduce 10% to 50% of the original claimed results.

Problems include:
  • contaminated lab samples [for example, breast-cancer experiments performed on melanoma cells);
  • cherry-picking or massaging data;
  • too-small sample sizes;
  • design flaws (for example, attributing a difference in disease rates to a drug without accounting for the role of genetics);
  • "the professional pressure to get splashy results";
  • not enough money to do experiments without cutting corners.

    The answer, so universal that it's a cliché, is transparency:
    researchers should make all of their methods and data freely available. This would allow the more rapid correction of faulty work—and would also encourage researchers to be more careful in the first place. [Virginia professor Brian] Nosek has created a free online resource called the Open Science Framework that is designed to allow scientists to make their hypotheses, methods, computer code and data freely available. For its part, Johns Hopkins University is pioneering a program that verifies exciting results from lab studies before those findings get passed along to biopharma companies.
    Results of biomedical experiments should indeed be regarded with a great deal of skepticism, but let's put this in perspective: these are flaws of hard science conducted in laboratories, the problems have been recognized, and solutions are being offered.

    Contrast the above with the even less rigorous methods of climate scientists,
  • who cannot conduct double-blind experiments (contemporaneous worlds with and without carbon dioxide concentrations),
  • who are strongly incentivized to cherry-pick and massage data,
  • who don't release the raw data on which their conclusions are based, and
  • whose models consistently predict higher global temperatures than those that actually result.

    But then again, I'm no scientist.
  • No comments: