| T O P I C R E V I E W |
| roka |
Posted - 08/06/2013 : 11:07:31 AM
Hello,
I've fitted one dataset with two different models, one made by me (model 1; userdefined stretched exponential) and the other is a double exponential decay (model 2; expdec2). I did not use weights for fitting. Visually, the data are much better described by model 2(as you can see on the graph below), but if I use the model comparison tool, the answer is that model 1 is more likely to be correct. I don't understand this and I hope that someone can help me.
Best wishes,
Romee
http://www.originlab.com/ftp/forum_and_kbase/Images/copy for forum.pdf |
| 4 L A T E S T R E P L I E S (Newest First) |
| roka |
Posted - 08/07/2013 : 08:54:13 AM
Hi all,
Thank you very much for your input. I think indeed it has to do with the log scale and also with the amount of datapoints. The degree of freedom of the first model is bigger than that of the second model due to the number of parameters, so that will also explain the answer of the model comparison.
To answer the first question: the AIC gives a preference to model 1; the F-test is inconclusive.
I am reconsidering how to present my data and whether I really need the fitting. Thanks you again for your help!
Romee |
| Hideo Fujii |
Posted - 08/06/2013 : 11:54:32 AM
Hi Romee,
Your plot's Y has log10 scale, therefore the contribution to evaluate the fit is much more significant at higher Y values (X is around in around 10^-1 -- 10^2 area) than at lower Y region (X around 10^3 -- 10^5 area). If you change the Y scale to linear, it may display more understandably. I suggest to convert your Y data itself to log10 values pre-processed, if you want to evaluate the fitting "fairly" within log10 scale post-processed. (Of course, you need to modify the fit function to cope with that conversion.)
--Hideo Fujii
OriginLab |
| greg |
Posted - 08/06/2013 : 11:43:24 AM
A picture may be worth a thousand words, but for an Origin analysis we really need to see an OPJ!
tech@originlab.com
|
| meili_yang |
Posted - 08/06/2013 : 11:42:03 AM
Hi Roka,
There are two models for comparing fitting functinos, F-test and AIC:
http://www.originlab.com/www/helponline/Origin/en/UserGuide/Comparing_Two_Fitting_Functions.html
Do these two methods give you the same results?
Meili
OriginLab Tech Support |
|
|
