Posts Tagged ‘ training set ’

How to fail “Statistics 101”

I am blown away by anyone (let alone a “postdoc”) who would present this study as anything other than embarrassingly poor statistics.  It would be ludicrous, if it wasn’t pathetic:

http://www.theatlantic.com/science/archive/2015/10/no-scientists-have-not-found-the-gay-gene/410059/

The last line of the article is not the best (that kudo goes to “correlation does not prove causation” – but that old saw should be painfully obvious to the most casual observer); but it is almost the last line (word): “We need to change how epigenomics research is performed throughout the community.”  I respectfully offer that the problem is not epigenomics, the problem is the current understanding of statistics – regardless of the subject.