| Author |
 Topic  |
|
|
dblasing
4 Posts |
 Posted - 03/25/2017 : 11:56:18 AM
|
Origin Ver. and Service Release (Select Help-->About Origin): 2016
Operating System: Windows 10
Hi All,
I have a question that would seem to be of broad importance (and have searched for the answer for sometime). I have some data with uncertainties that I'm fitting too. Origin reports some fitting parameters and I was imagining that the "standard error" that origin reports is the uncertainty in the extracted fit parameter. The weird thing is that when I artificially "inflate" the data uncertainty, the "standard error" of the fit parameter does not increase. I noticed that it does increase if I uncheck "scale error with square root of reduced chi squared."
So my question is, if I want to report the uncertainty of fitting parameters, do I want to check "scale error with square root of reduced chi squared" or not. It would seem to me that I want it unchecked, since only then does it appear to be sensitive to the size of the error bars that I feed the fit.
Thanks very much,
-David |
|
|
AmandaLu
439 Posts |
Posted - 03/27/2017 : 06:35:01 AM
|
Hi,
Parameter Standard Error provides a way to consider the precise of the parameter. Please refer to our document:
http://www.originlab.com/doc/Origin-Help/NLFit-Algorithm#Parameter_Standard_Errors
From the formula in the document, when you check the “scale error with square root of reduced chi squared”,
If uncheck, the formula becomes
The Scale Error with sqrt(Reduced Chi-Sqr) box is checked by default to keep parameter's standard error and related results compatible with other software. We have a quick help for a similar question:
http://www.originlab.com/doc/Quick-Help/Chi-Se-Remain
Thanks,
Amanda
OriginLab Technical Service
|
Edited by - AmandaLu on 03/27/2017 06:37:55 AM |
 |
|
|
dblasing
4 Posts |
Posted - 03/27/2017 : 08:26:30 AM
|
Hi Amanda,
I appreciate your reply. I am, though, still somewhat confused - so let me be a little more clear with my question and then perhaps you can help me better. Let's say I want to report an experimental frequency, with an appropriate error. Suppose I do 15 data runs of a sine wave, each of which contains 30 points.
The first way I know of is to fit all 15 data runs individually and extract the mean frequency. Then I report the mean and standard error of the mean of those 15 extracted frequencies as the error.
Now suppose alternatively, I instead average all of the 30 data points firs, and include, in a column set as the yerr, the standard error of the mean of each individual data point. In this second example, I have now one data set of 30 points with error. I can fit and extract one frequency as the experimental average frequency. My question is, wether or not "standard error" of the extracted frequency is the uncertainty analogous to the standard error of the mean in the above first way? Do I want to click or unclick "The Scale Error with sqrt(Reduced Chi-Sqr) box" if my goal is to report the uncertainty of the average mean frequency?
Thanks very much,
-David
quote:
Originally posted by AmandaLu
Hi,
Parameter Standard Error provides a way to consider the precise of the parameter. Please refer to our document:
http://www.originlab.com/doc/Origin-Help/NLFit-Algorithm#Parameter_Standard_Errors
From the formula in the document, when you check the “scale error with square root of reduced chi squared”,
If uncheck, the formula becomes
The Scale Error with sqrt(Reduced Chi-Sqr) box is checked by default to keep parameter's standard error and related results compatible with other software. We have a quick help for a similar question:
http://www.originlab.com/doc/Quick-Help/Chi-Se-Remain
Thanks,
Amanda
OriginLab Technical Service
|
|
|
|
dblasing
4 Posts |
Posted - 03/29/2017 : 3:28:00 PM
|
Hi Amanda,
I think this is an important question and I'd really appreciate a reply.
Thanks,
-David |
|
|
|
easwar
USA
1964 Posts |
Posted - 03/29/2017 : 5:34:06 PM
|
Hi David,
Regarding your first post where the data uncertainties are inflated:
If there is an error column included in the fit, the weighting method is by default set as instrumental. Then note that in our help page:
http://www.originlab.com/doc/Origin-Help/NLFit-Dialog-SettingsTab#Advanced
in the Fit Control section where this check box is described, we state:
"This option is checked by default to keep parameter's standard error and related results compatible with other software. It is recommended to uncheck this option when fitting data with instrumental weight, so that parameter's standard error can reflect the magnitude of weight."
Regarding the second scenario you described where you have multiple sets of data (each with no weight) and your question was whether fitting each dataset individually and averaging the parameter value, or first computing the data average and mean and then performing the fit:
The above two procedures will most likely result in different parameter values themselves, and not just different standard errors on the parameter. If indeed you have multiple measurements, the best way to fit the data is to use the Concatenate mode in Origin. Origin then internally combines ALL of the data points and performs the fit. In this case there is no measurement uncertainty to deal with, and the "scale error..." check box is set as checked, and cannot be unchecked.
See this page for description of the Concatenate Fit mode:
http://www.originlab.com/doc/Origin-Help/NLFit-Dialog-SettingsTab#Data_Selection
Hope this helps.
Easwar
OriginLab |
|
|
|
dblasing
4 Posts |
Posted - 04/12/2017 : 3:55:23 PM
|
Hi,
Can you flush out a little bit by what you mean by "best" in reference to using concatenate fit? Is the fit somehow more accurate and thus best or perhaps it may take less time to accomplish?
Thanks,
-David |
|
|
| |
Topic |
|
|
|
Parameter uncertainties and error bars