From: Charles Geyer Date: Wed, Jan 14, 2015 at 8:48 AM Subject: Re: seen on hacker news To: Ruth Shaw , Aaron Rendahl https://www.youtube.com/watch?v=lrm8iEMQZNw (found through reddit) is an amazing talk about what was wrong statistically with those Duke trials covered in the Cancer Letter article I sent you yesterday. There is also an Annals of Applied Statistics paper (cited in the talk at the end, the speaker is one of the authors) that was the paper that made all the problems public. It turns out that all of the problems were simple data errors. Mislabeling data. Having headers on CSV data when the Matlab code expected no headers. And a lot of other stuff of the same sort. It was not using theoretically bogus statistics or anything that elaborate. Because the "high tech", "big data" methodology was thousands of pairwise comparisons using two-sample t-tests and they did not actually screw that up. The talk makes a strong push that everybody doing this has to do reproducible research (like I have been doing since 2005). All the data and all the computer code must be publically available. The speaker in the YouTube video has switched all of the bioinformatics at M D Anderson Cancer Center (in Houston) to using Sweave for all analysis. So that's another point. I know Evolution requires data. But it should require all the computer code too. On Tue, Jan 13, 2015 at 8:26 AM, Charles Geyer wrote: http://www.cancerletter.com/articles/20150109_1 A very interesting article about the research scandal at Duke. The article is about a 3rd year medical student who blew the whistle but was squelched by Duke and whose very existence was denied to later investigators and this will be important evidence in the lawsuit. But more interesting to us is just the details of research fraud that is entirely bogus statistics. I admit I only read half way through the five page article, but it would be fascinating to know all the details of what they were doing wrong. Aaron: you are teaching about ethics in the statistical consulting course? This would be a good read for that. Ruth: just thought you might like to see this too. And you might pass it along to whoever is mentoring in your dept too.