Plant breeder, biology instructor, and computer geek. Clemson, SC
Ben hits the nail on the head, identifying yet another example from CRIIGEN of how not to use statistics.
This analysis by Seralini of Monsanto's report on the safety of NK603 gives some insight into Seralini's experiment design methods: [Referring to 2007 Committee for Independent Research and Information on Genetic Engineering Report on NK 603 GM maize produced by Monsanto company June 2007 Controversial effects on health reported after subchronic toxicity test: 90-day study feeding rats] http://bit.ly/RKFFNq@ TwitLonger — When you talk too much for Twitter:
He seems to think Monsanto included too many controls, and that ANOVA is inadequate for the statistical analysis - he would prefer PCA or something else, again, showing his preference for more exotic statistical methods.
His philosophy appears to violate the prime directive for statistical analysis - that you design your experiment to answer certain specific hypotheses, and you make sure you have enough power that the results can show significance, if there is any. And you limit yourself to the specific questions you designed the experiment to answer in the first place.
He seems more inclined to prefer exotic techniques to find anything of significance he can, as someone said earlier, a statistical fishing trip. What he seems to fail to understand, it doesn't matter how "detailed and extensive" your experiment, if it suffers from poor design, the results are meaningless, and a waste of time, money, and mice.
No comments:
Post a Comment