不是统计方法，而是样本太小得出的结论没有代表性。

就像前面俺提到的抛掷硬币的次数太少一样。

另外， 请注意Nature commentary 中下面一段话：

Some scientists, however, have long questioned whether such feeding studies are appropriate for testing the safety of whole foods, says Peter Kearns, head of food safety, nanosafety and chemical accidents for the Organisation for Economic Co-operation and Development in Paris. They were designed for testing chemicals where precise doses of purified and well-characterized compounds can be administered, whereas compounds in foods are heterogeneous, and doses are difficult to control. Regulators rely mainly on more robust tests that compare the toxicological and nutritional profiles of GM foods with their non-GM counterparts to screen for potential concerns.