| Author |
 Topic  |
|
|
xjtuthermo
Germany
1 Posts |
 Posted - 04/07/2012 : 05:48:19 AM
|
Origin Ver. and Service Release (Select Help-->About Origin):
Operating System:
Dear all,
Please explain me the way to evaluate the standard errors of parameters form origin 8.0. I check the help file offered by origin8.0,and the errors in the parameters correspond to the product fo the meam residual variance and the square roots of the diagonal terms in the error matrix. But from the books, the standard errors of parameters for nonlinear fitting only correspond to the square roots of the diagonal terms in the error matrix. And i coded in C++ using L-M algorithm in such books and found that the standard errors of the parameters fitted is much larger than that fitted from origin 8.0. Please explain me why? |
|
|
Sam Fang
293 Posts |
Posted - 04/10/2012 : 03:09:04 AM
|
By default Origin uses the same method as most software to calculate the standard errors of parameters, i.e. including the mean residual variance. If weight is specified, you can clear Use Reduced Chi-Sqr check box in the Fit Control branch of Settings: Advanced page, and the standard error will only correspond to the square roots of the diagonal terms in the error matrix as you said.
But it may be more reasonable to include the mean residual variance, e.g. if y data is scaled by 10, the amplitude parameter is also scaled by 10, its standard error will not change if the mean residual variance is excluded in the calculation of standard error, and it will be also scaled by 10 if it is included.
Sam
OriginLab Technical Services |
 |
|
| |
Topic |
|
|
|
Standard errors of the parameters for NL fitting