@Kodiologist but doesn't it depend on which field? (RE: xkcd.com/435 )

I mean, it's still OK for a physicist to fit a power-law exponent "by eye" on a log-log plot, right?

I noticed a couple recent question from pop musicians? James Brown? Frank Zappa?

I thought we were strictly down with the cool jazz cats? (like Charlie Parker)

@GeoMatt22 I don't know any physics, but it's hard to see the sense in eyeballing what you could compute.

@Kodiologist I don't know much physics either. But commonly physicists want a particular scaling exponent (based on theory), so sometimes they just put the "reference line" on their log-log data plot, as a visual guide. (This is usually theorists I think, not experimentalists)

Ok, let's take that CI is random and parameter is random for the fixed credible interval (despite I just watched Bayesian made simple explaining how credible is built on observation of heads, which means that true probability of heads is unknown and may be different from evidence-based interval estimation). Anyway. Interval is random in one case and parameter is random in the other. In which case I can say that parameter lies within the interval with probability and in which I cannot?

Is it only me who currently has technical issues posting answers/updates?

To follow up - the culprit seems to be this code I want to include in an answer:

The offending part seems to have been # comments after R code. When I post this stats.stackexchange.com/questions/234958/… it works, but not when I for example write y <- rnorm(n) with "# this is the dependent variable"?!

@ChristophHanck, I added the comment for you. I didn't have any problems; I don't know what the issue is.

I also made some formatting tweaks that I thought might make it a little easier to read. I hope you don't mind. If you do, roll it back with my apologies. I can always re-add just the comment.

@gung: thanks for the useful edits - now I can insert comments, too - could the tabs I had included to make the comments align below each other be an issue (not that I'd have any idea why - it was copy & paste from RStudio)?

@ChristophHanck, no idea I'm afraid

I have read this answer many times - "surfing" the site really is a Markov process, where you land over and over on some "classics." Anyway, I understand that there is a ton of advanced math in manifolds applied to probabilities, but I'd like to get a bit of an idea for the advantages manifolds provide, etc. Do you know of any accessible (intuition style) posts or references?

@ChristophHanck I could not reproduce the problem, either.

@AntoniParellada Do Carmo is a classic text. I think it's still used for undergraduate courses on differential geometry. I learned differential geometry from Spivak's Comprehensive Introduction... (volumes II through V), so my language and approach likely reflect that. The answer at math.stackexchange.com/a/46523/1489 looks like it contains a good rundown of more recent textbooks at the advanced undergraduate/early graduate level.

@whuber Thank you. I like practical applications because they allow me to deal with my lack of formal training using R as a crutch, but what I'm interested in in the long run (wishful thinking) is the math.

@AntoniParellada for an exposition of differential geometry with interesting visualizations there are some nice tutorials from the computer graphics community. (emphasis 2D-manifolds in 3D euclidean space). The "Differential Geometry Primer" here was pretty good (hope I remembered the right one): inf.usi.ch/hormann/parameterization

The lecture "Nonlinear Methods" (right after the primer at my link, lect. 3 & 4) has perhaps my favorite diff-geometry picture ever (on slide 2): Essentially when the author's physics professor said "assume a spherical cow", she must have been like "Yes, but what is the metric distortion? we can do better!".

@GeoMatt22 Thank you. I ordered Do Carmo's book, but I'd like to get a SE post, or some youtube tutorial that frames the intuition behind the need or advantages to using manifolds in probability, so as to have a bit of an idea of why it is relevant.

Hello everyone, I have a question in mind which is a little abstract, and I'm not sure it actually constitutes a real CV question, so I want to ask it here first. The two concepts, 1) all models being wrong and 2) bias/variance dilemma. Are they related, or even the same thing? Is 2) the reason of 1)? Can we say that no model can be exact because we either keep it too simple and introduce a high bias or make it too complex and obtain unwanted variance, or are there other reasons for 1)?

@AntoniParellada this video about manifolds is not exactly what you asked for (context = nonlinear dimension reduction for visualizing high-D data in 2D, t-SNE et al.), but may help: youtube.com/…

@GeoMatt22 Thank you very much.

@AntoniParellada Although I have only seen the table of contents, the Murray & Rice book, Differential Geometry and Statistics, looks like it begins at a relatively elementary level and is dedicated to applying geometrical thinking to statistics.

A quick overview, freely available on JSTOR, is Barndorff-Nielsen, O. E., Cox D. R., and Reid N. "The Role of Differential Geometry in Statistical Theory." International Statistical Review / Revue Internationale De Statistique 54, no. 1 (1986): 83-96.

@whuber Thank you. Both recommendations are very promising. I have the article on JSTOR ready to read on the "shelf". For now I guess I'll hold off on the book for now, because it's $140, and it would pay to do some resarch perhaps with Do Carmo's book, which I already ordered. I see I'm not the only one with this question.

This sounds like literally homework for a class stats.stackexchange.com/questions/235237/… ... I have not been in Uni for a while: Are teaching assistants / office hours no longer a thing?