On Sun, May 12, 2013 at 4:37 PM, Jed Rothwell <jedrothw...@gmail.com> wrote:
> > How many replications does it take to ensure an effect is real? > > Everyone will have a different answer. A knowledgeable person will want to > look at the papers, and evaluate the skills of the researchers, the choice > of instruments, the signal to noise ratios and so on. For an experimental > finding as surprising as cold fusion, I think most people will demand 5 or > 10 "quality" replications from professional labs. What constitutes > "quality" is partly matter of opinion. > > Exactly. And the prevailing opinion is that there are *no* quality replications. When P&F claim 140 W output with 40 W input with one type of calorimetry, and McKubre gets 1 W out with 10 W input, that's not a quality replication. I would say that by 1990 there were so many replications of heat and > tritium that any continued doubts were irrational. > > Right, so Gell-Mann, Lederman, Glashow, Huizenga, Koonin, Lewis, etc etc can all be dismissed with a simple statement from a computer scientist as being irrational. > A Bayesian analysis sheds some light on this. > No. It really doesn't. A big pile of marginal results makes it look more pathological, not less.