This paper posted to the arxiv the other day provides a very readable, practical discussion of error analysis and curve fitting. While I have not had the time to go through this in detail, at a quick read it looks like it is the sort of thing every physics student should know and use as a refresher.
The point is, far too many people who really should know better never learn the right way to think about uncertainties or how to properly fit data. For example, in the world of economics, some people actually think that the curve on this graph actually has some statistical significance. (That's not a political statement - I'm laughing at their innumeracy.)