Comments on: Kepler and the Art of Astrophysical Inference

By: vlk

vlk — Fri, 18 Apr 2008 03:55:28 +0000

Good point, Aneta. I was going to respond, using the original example, that now we know with certainty that the Earth rotates, so that is an example of a prior that has become a delta function. But of course it is conditional on Newtonian mechanics and Einsteinian relativity, so strictly speaking it isn’t a prior! Oh well, like they say, if you push an analogy, it can fall off the edge of the world!

By: aneta

aneta — Fri, 18 Apr 2008 01:55:50 +0000

I like the VLK image of the priors becoming a delta function! this is then certainty! do we have such things in astronomy?

By: vlk

vlk — Thu, 17 Apr 2008 20:26:36 +0000

Oh I definitely did not mean to imply that hypothesis tests and principled statistical calculations and the like are useless. Without a rigorous assessment of whatever-it-is, there is no science, and everything becomes an opinion. The point that Owen was making is simply that astronomers (and scientists in general) tend to make progress based on a coherent “story” without waiting for a formal Kuhnian revolution, and usually when a final proof of a worldview does arrive, it is rather anti-climactic (he refers to Foucalt’s pendulum as an example). Think of it as continuously updating your priors to make it narrower and narrower until at long last it becomes a delta function. So you can see how a marginal result along the way has little effect on the believability of the model.

By: hlee

hlee — Thu, 17 Apr 2008 19:42:44 +0000

Replying to the last sentence, not to exaggerate, but to amuse, astronomers care best fits and error bars more than statisticians and the statement seems to imply all the efforts to get best fits and errors are unrealistic. The idea of proving is interpreted in different ways, I guess. What I go against is using a statistic to prove empirical laws like the law of equal areas without comprehending the fundamentals of the statistic, which explains astronomers’ suspicion if we change the direction of our view points. In my opinion, it is not a fault of statistics but the people who use it inappropriately. Speaking of best fits and error bars, from the tree growing point of view, the method of growing trees receives more attention than estimating errors. Statisticians do care finding laws (best ways to grow a tree) that represent the truth of the nature. (Note: tree = classification tree or regression). Not knowing detailed history (haven’t listened Prof. Gingerich’s talk but I will soon), I’m very opinionated but one thing I’m sure is that in migration, adaption happens but not necessarily it reflects its origin. Selective mutation is what I’m afraid of.