I saw this earlier. One claim that stood out to me was: "EFSA recommends a minimum of 10 animals per sex per group for the chronic toxicity phase, the same number that Seralini used." Now I didn't have time earlier to look at that, but I thought there was no way EFSA was going to approve 10 control animals to go with 9 other dose/treatment groups of 10 animals each. Low and behold, when reading the PDF (http://www.efsa.europa.eu/en/efsajournal/doc/3347.pdf) I discover that this claim is facile. Sure, they allow for groups of 10 but their recommendations on animal group design are much more complicated.
To start, the section talking about controls starts by taking from OECD TG 453 standards for long-term studies on chemical effects. For carcinogenicity studies, that standard recommends 50 animals per sex per dose (treatment) group with concurrent control for carcinogenicity studies. If Seralini's study had followed this recommendation and still wanted to do all 9 of their treatment groups, it sounds like they would have needed 50 * 9 = 450 control animals to go with 450 males, then the same for females. Similarly for chronic toxicity, that standard recommends 10 animals per sex per dose group ... again with concurrent control. For Seralini's study that would have been 90 male controls to go with 90 males and 90 female controls to go with 90 females. So right off the bat, we discover that Seralini's study seems to have ignored a pre-existing international recommendation on these kinds of studies! One that, I note, other less controversial long-term studies seem to be using (See http://gmopundit.blogspot.com/2012/09/the-first-long-term-2-year-study-of-gm.html).
EFSA's recommendations then go on:
EFSA emphasizes that part of the planning phase of the study should include defining clear objective(s) and a prospective power analysis in order to establish the number of animals needed for the experiment. This requires selecting a specific effect size of interest and estimating the standard deviation (in case of continuous variables). However, it is often difficult to establish an effect size of interest for every endpoint because of the multiplicity of endpoints usually assessed in toxicity testing. Further the chosen effect size should be biologically relevant (EFSA 2011b). EFSA recommends that a statistician is consulted from the start of the study planning.
There's little evidence that the researchers running Seralini's study consulted a statistician since they are very unclear about what statistics were actually being done. Certainly there's no power calculation given. But EFSA's recommendations go on with an example scenario involving two treatment groups of ten animals and explicitly says the study would need equal numbers of controls (two groups of ten). There's then a table with example numbers of animals and I really can't work out how those recommendations at all square with the group sizes chosen by the 2012 Seralini study. Just because a group of ten is permitted by the EFSA recommendation doesn't mean that it was appropriate in that case. In fact it's not at all since Seralini was looking for carcinogenic effects! So strictly speaking EFSA has endorsed group sizes of ten .. but only with concurrent controls of the same size and only if you do power calculations to make some attempt to find the right size for your hypothesis.
I'm sure the other claims by GM Watch are ridiculous, but this one is particularly ridiculous: "EFSA requires an a priori power analysis to ensure appropriate sample size, depending on the effect size that is being looked for. We've never noticed the GM industry doing one of these, resulting in experiments that are virtually guaranteed not to find anything." Of course, the way the 2012 study arranged their study groups, they also ensured that you were guaranteed to not find or find anything you wanted depending on how you over-interpret the data. The industry doing a poor job is not a good reason for an independent researcher to do a poor job!