Significance
Here is one small example (indirectly via Patri), of a multitude I could choose, of why I'm working on what I'm working on right now: in this paper the authors check how damaged the hearts of marathon runners are compared to relatively sedentary people. Surprise! Their hearts tend to actually be in *worse* shape -- i.e. they show more scarring, which raises your odds of dying of a heart attack enormously. Oops!
You'd think such a counter-intuitive result would be considered highly remarkable, but the catch is that the result only cleared the 0.92 level of statistical significance. Oh, well then. Beneath the standard 0.95, so safe to ignore it, right? Nevermind that the empirical techniques they used are as definitive as you can get, short of an autopsy. Nevermind that we have good reasons to entertain the notion that the inflammation and tissue stress caused by stupid activities like marathon running would do more harm than good. (Here, as everywhere, there are diminishing returns: short, intense bursts of exercise are closer to optimum.) Meh!
The chapter on how to navigate the morass of statistical results will draw heavily on one of my favorite papers: Gelman & Stern's "The Difference Between 'Significant' and Not Significant' is not Itself Statistically Significant". Gotta love the title -- almost as good as Ioannidis' "Why Most Published Research Findings Are False", which will also get a shout-out.
You'd think such a counter-intuitive result would be considered highly remarkable, but the catch is that the result only cleared the 0.92 level of statistical significance. Oh, well then. Beneath the standard 0.95, so safe to ignore it, right? Nevermind that the empirical techniques they used are as definitive as you can get, short of an autopsy. Nevermind that we have good reasons to entertain the notion that the inflammation and tissue stress caused by stupid activities like marathon running would do more harm than good. (Here, as everywhere, there are diminishing returns: short, intense bursts of exercise are closer to optimum.) Meh!
The chapter on how to navigate the morass of statistical results will draw heavily on one of my favorite papers: Gelman & Stern's "The Difference Between 'Significant' and Not Significant' is not Itself Statistically Significant". Gotta love the title -- almost as good as Ioannidis' "Why Most Published Research Findings Are False", which will also get a shout-out.