The old idea of p>.01 being the gold standard falls down when you are checking thousands of parameters. Perhaps p>.01/Nparameters needs to be the standard. But you would need a lot of data to get that, and the more data the more parameters you probably have to check.
