| T O P I C R E V I E W |
| jcarifin |
Posted - 04/07/2023 : 12:19:49 AM
Origin Ver. and Service Release (Select Help-->About Origin): 2016 (64 bits)
Operating System: Windows
Hello,
I have several datasets (histograms). Each dataset contains 2 overlapped distributions (G1 and G2). Using "Multiple Peak Fit", I have successfully fitted the histograms, and found the peaks and their distribution.
But is it possible to separate or remove one distribution from the datasets, so that only G1 or G2 is left on the histograms?
Basically what I want to do is compare G1 and G2 from one dataset, with G1 and G2 from other datasets (their mean/peak and statistical significance difference). By seeing the distribution, the hypothesis is only G2 changes. G1 doesn't change.
I have tried:
1. Analysis -> Fitting -> Compare Datasets. It has a problem since it always compares G1 and G2 together, so I could not prove that G1 doesn't change.
2. Analysis -> Data Manipulation -> Substract Reference Data. It requires a specific dataset. Since G1 and G2 have overlapped area, I am not sure how to generate such G1 or G2 datasets based on the fitted distribution.
3. "Compare Datasets and Fit Parameters" app. It gives a similar problem to the 1st case. And comparing the fitted curves would always give a result that the datasets are statistically different because the F value is tremendously big, probably because of the df.
So, can I do it with OriginLab?
Thank you very much.
Kind regards,
jcarifin |
| 4 L A T E S T R E P L I E S (Newest First) |
| jcarifin |
Posted - 04/07/2023 : 08:03:43 AM
Dear Amanda,
This is the result of comparing two fitted G1 the with the app (my 3rd method, your 2nd method)
Result:
Graphic of the two distribution:
The F-test gives the result that these two distribution, or parameters are statistically different (0.05 significance level). Even if I were to compare only 1 parameter (xc), it would still give me statistically different, with F-value in the order of ~e30.
Is the result really correct? I mean, if I see the fitted parameters, intuitively the two distributions are not significantly different. At least, the "Prob > F" won't be = 0.
I am confused about the result.
Thank you.
|
| AmandaLu |
Posted - 04/07/2023 : 06:27:48 AM
Hi,
Yes, this is your 3rd. When you use "Compare Datasets and Fit Parameters" app to compare fit parameter “a”, all other parameters will be treated as the same between two datasets. If your model G1 and G2 do not share same parameters, the result will be “statistically different”.
Thanks,
Amanda
OriginLab Technical Service
|
| jcarifin |
Posted - 04/07/2023 : 03:50:25 AM
Dear Amanda,
Thank you for the quick response.
I am not clear about the 2nd method. When you said "compare the G1 fitted data of dataset 1", did you mean the curve drawn from "nlfitpeaksCurve", the dataset we got from "Multiple Peak Fit"?
Because if yes, that's my 3rd attempt, and the statistical test would always give a "statistically different" result as the F-value didn't make sense (in the order of E39). The df is > 1000.
Did I misunderstand the 2nd method? |
| AmandaLu |
Posted - 04/07/2023 : 02:47:32 AM
Hi,
Two ways to compare G1 and G2.
1. Use "Compare Datasets and Fit Parameters" app. You will need to define the fitting function as the sum of two distributions: f = f(G1) + f(G2).
This way you will need to give good initial parameters to get a reliable result.
Please note that when you compare fit parameter “a” using "Compare Datasets and Fit Parameters" app, all other parameters will be treated as the same between two datasets.
2. If you separate G1 and G2 using "Multiple Peak Fit", you can get fitted data of two datasets, respectively. You can then use "Compare Datasets and Fit Parameters" app to compare G1 fitted data of dataset 1 and G1 fitted data of dataset 2. Same for G2.
Thanks,
Amanda
OriginLab Technical Service
|
|
|
