Wednesday, April 18, 2018

Junk Science

Headline of op ed in Tuesday's Wall St Journal, "How bad is the Government's Science?"  It speaks to the reproducibility crisis in science, where a large number of published scientific papers simply cannot be reproduced by other workers.  Which says that the published paper was just plain wrong.  A 2015 study estimated that $28 billion a year was spent on wrong science.  Which is a terrible waste of both money and the time of scarce and hard to train scientists. 
   I ran into the reproducibility problem myself back when I was developing a portable heart monitor.  I needed a way to compress the sampled EKG so that the device could store more EKG data in its limited memory.  I researched the literature, and bingo, I found a paper discussing compression of EKG and offering a method that claimed much higher performance than the standard technique.  I read the article thru, and then programmed the new algorithm into our prototype monitor.  It worked, it did compress the data, and the decompressed EKG was of good quality, but, I could only obtain one half the amount of compression that the author claimed.  I troubleshot and debugged and finally telephoned the author to ask for help.  The author rather sheepishly, admitted that he had left out a key factor in his paper, and that yes, the compression obtained would be only half of what he had claimed.  I managed not to express my dismay over the waste of two weeks of the project's time. 
   One thing legislators could do about this.  Require that all government financed researchers publish all their raw data.  Right now, a lot of researchers keep their data private, hoping to either use it for another publication, or to prevent skeptics from going over it looking for faults.  Far as I am concerned, if the taxpayers are paying the freight,  the taxpayers own the results.  This policy would go far to squelch the likes of leftie greenie "climate scientist" Michael Moore, inventor of the global warming hockey stick. 
  Another thing, someone ought to keep score.  Any scientist who publishes unreproducible results should be barred from future government research grant money.  That will make them a bit more careful. 

No comments: