There are many choices to make when statistically analyzing the data, and it is not terribly hard to play around with choices until you get a favorable result.

There are many choices to make when statistically analyzing the data, and it is not terribly hard to play around with choices until you get a favorable result. For example, one common problem is truth inflation: clinical trials are conducted with too-small sample sizes, so they can't detect the result they're looking for. If you run enough of these, the only statistically significant results you will get will significantly exaggerate the true effects of the drug. Some explanation and citations here, along with many other statistical tricks: http://www.refsmmat.com/articles/statistics.html#truth-inflation http://www.refsmmat.com/articles/statistics.html#truth-infla... There's also a well-known problem called "vibration of effects" where you can play with controlling different variables, using different analyses methods, and so on, resulting in an estimate of a drug's effects that oscillates by a factor of two. Look for John Ioannidis' paper "Why Most Discovered True Associations Are Inflated".