From help-octave-request at bevo dot che dot wisc dot edu Mon Nov 24 13:45:41 2003 Subject: Re: axes labels on errorbar plots From: Przemek Klosowski To: help-octave at bevo dot che dot wisc dot edu Date: Mon, 24 Nov 2003 14:45:34 -0500 (EST) przeme >numerical and measurement error should be treated the same; I think przeme >the fundamental principle is that there are two sources of error: Yeah, I'd say that the three are distinct: * error due to inexact representation floating-point numbers. * error due to the finite resolution of any real-world measurement ( i.e., significant figures, & error range ). * error due to calculation via an inexact (vs. exact) method. why do you think that inexact representation has to be treated differently than measurement error? They both can be described by confidence intervals. True, they have different probability distribution: hardbox like for truncation (x IS between x_with_truncated_LSB and x_with_set_LSB) and gaussian for your typical error, with 68.26% probability that x is within x-sigma and x+sigma. However, if you start playing those games, why not distinguish between different statistical distributions of measurement error: what if you have an experimental variable with Poisson statistics, instead of Gaussian? I am comfortable with describing all errors in the same way, as confidence intervals. If you need a confidence higher than 68%, you can widen your interval (+- 2 sigma will give you 95% confidence, etc). The Central Limit Theorem says that almost always you quickly get into Gaussian distribution region, anyway, so if you start with certain confidence level, your confidence intervals proceed to be at this level. ------------------------------------------------------------- Octave is freely available under the terms of the GNU GPL. Octave's home on the web: http://www.octave.org How to fund new projects: http://www.octave.org/funding.html Subscription information: http://www.octave.org/archive.html -------------------------------------------------------------