The reasoning was that since there's hundreds of millions of people sleeping in America every night it's statistically guaranteed that someof those people are going to have dreams that come true. What do you think of that? Do you think that's good reasoning? I don't. How do you statistically determine how many dreams will come true? The ironic thing is that this was in a book on critical thinking. Sheesh.

Well, if you have a null hypothesis of the form "no one in the US had a dream last year that predicted a future event" all you'd have to do to reject it is find one exception. It'd be much more difficult to prove effectively the general rule that there's always a guarantee that someone will have a prophetic dream but you could probably operationalize it as a dream of the form that "X will die tomorrow," conduct a poll of people to see if anyone had such a dream on day Y and then call back on day Y+1 to see if any of the predictions came true. If you called a few thousand people I think you'd be likely to get one or two hits. And that's just one type of prediction. Of course I could be all wet, and you'd call 30,000 people and never get a single hit... but it's at least a testable hypothesis. You probably couldn't call everyone in the US on every night and record every kind of potential prophesy, but you wouldn't have to... to make the point.

So, with the caveat that the way you've expressed it to us isn't very rigorous you could probably do something like it without too much trouble. You could compare the number of successful prophecies with the percentage of people who knew someone who died to see if there were a deviation that was statistically significant allowing you to reject the null hypothesis that the "prophesies" were just random blindfolded dart throws.

Also, I think Juan's critique of Paddy's statistics misses the mark, if I'm reading him right. True, Paddy is drawing from a small sample who fit a similar demographic, but I think it's enough to determine if people who stick with M are happy or not. What IS missing from Paddy's statistics are the people who DIDN'T stick with M, or are questioning their involvement.

Well, that's one aspect of the selection bias but he's also omitted (by virtue of the process of selection) most of those premies who aren't "happy" (meaning living normal relatively financially secure lives). And since he doesn't state or even imply an hypothesis it's not clear what the heck he's trying to prove.

What is THEIR happiness quotient? Also, what's the percentage of people who stuck with chubby compared to those who split? Paddy just gives stats for people he knows that stuck with M. That is far from the complete story.

Well, I think you've captured the essence of selection bias.

For instance, I have a friend who was doing a study to determine what it was about survivors of the holocaust in Poland that ensured their survival. She had some good hypotheses:

They spoke Polish fairly well.
They had friends in the Polish Catholic community who could give them critical help.
They had access to official papers and documents.