| T O P I C R E V I E W |
| tlkurth |
Posted - 05/03/2004 : 4:35:28 PM
I have a data set that I would like to fit in two ways. The two ways consist of exchanging the x and y axes. When I fit the same function to each plot I get two different solutions. This is due to the fact that the deviation in one axis is different than the deviation in the other. Is there a simple way in Origin to modify the error function so that the same result is obtained regardless of which axis is independent? i.e. the error in x and y are both considered or the error in x instead of y. Simply fitting to both plots simultaneously will not work as the ultimate goal is to fit a more complex function in only the x-y exchanged form. |
| 3 L A T E S T R E P L I E S (Newest First) |
| easwar |
Posted - 05/04/2004 : 2:56:49 PM
Hi Todd,
Our fitting/minimization routines currently only support error in y. We will consider supporting error in x in a future version. Thank you for reporting this issue.
Easwar
OriginLab
|
| Hideo Fujii |
Posted - 05/03/2004 : 6:25:10 PM
I think that the factor analysis assumes the linear relationship. So, obviously it is NOT what you are looking for. Not sure, but programming may help??
Hideo Fujii
OriginLab Corp.
|
| tlkurth |
Posted - 05/03/2004 : 5:55:22 PM
I'm not sure. I'm familiar with deconvolution methods utilizing factor analysis but my problem is much simpler.
I'm trying to fit multiple data sets with a Langmuir equation: y=kx/(1+kx,) which is easy. The FFG model is an extension of this model: y=kx*exp(by)/(1+kx*exp(by)), also easy. This can be rearranged to get x=f(y). To obtain consistent fits utilizing each of the two models for different data sets it is important to have a consistent error function (chi^2). However, the error function for the fit to x=f(y) depends on the error on the x-axis. The chi^2 for y=f(x) depends on the error on the y-axis. Nearly horizontal and vertical trends in the data make the difference between calculated errors, i.e. the error surfaces to be minimized, quite large.
If there were an error function (2-dimensional chi^2) that included the x-axis and y-axis deviation the resulting fit could be identical.
Thanks,
Todd
quote:
If my understanding is correct, I think you're looking for the "factor analysis", which Origin corrently doesn't have. The factor analysis tries to reduce dimensionality - in your case from 2 to 1. The method is often used in social sciences, but is it useful in your field, too?
Hideo Fujii
OriginLab Corp.
|
|
|
