Author |
Message |
Benoit Igne (benoit)
Member Username: benoit
Post Number: 12 Registered: 11-2006
| Posted on Friday, October 23, 2009 - 2:24 pm: | |
Howard, Thank you for your answer. While assumptions might be difficult to meet with NIR data, the chi^2 approach might be the easiest way to get a first estimate of the error. Benoit |
Howard Mark (hlmark)
Senior Member Username: hlmark
Post Number: 279 Registered: 9-2001
| Posted on Friday, October 23, 2009 - 2:03 pm: | |
Benoit - in principle, if the errors are really random, independent and Normally distributed, and the system is stationary (i.e., it's underlying behavior doesn't change with time - something very unusual in NIR!) then under those ideal conditions, the square of the SEP (and those other measures of error) will be distributed as chi^2, and there are statistical tables giving the confidence limits for the chi^2 distribution corresponding to different numbers of degrees of freedom and confidence levels. Then you could simply look up the confidence limits, multiply them by your SEP^2 and take their square root to get the confidence limits for your SEP. I have to point out, though, that that is an awful lot of conditions for actual NIR data to meet. Don't forget, for example, that nonlinearity in the response introduces non-randomness and non-Normality, as does a simple skew of the model. \o/ /_\ |
Benoit Igne (benoit)
Member Username: benoit
Post Number: 11 Registered: 11-2006
| Posted on Friday, October 23, 2009 - 1:25 pm: | |
Hello all, I am trying to calculate an uncertainty value for my SEP but I would like to avoid having to perform a bootstrap type method. Would you know of any simpler method to calculate confidence limits on an SEP or SEC or SECV? Thank you for your help, Ben |
|