| T O P I C R E V I E W |
| giordandue |
Posted - 10/20/2010 : 07:11:34 AM
Dear friends,
by Origin definition, the prediction band for the desired confidence level is the interval within which x% of all the experimental points in a series of repeated measurements are expected to fall.
In alternative, is it possible to get the bands based on the standard error s of the regression? In other word, the the bands which includes for example 1s, 2s, 3s, i.e. the 68%, 95%, 99% of the samples?
Thank you in advance |
| 4 L A T E S T R E P L I E S (Newest First) |
| giordandue |
Posted - 10/22/2010 : 03:05:11 AM
Hi Larry, thank you again. Maybe I am not able to explain better, but what I would to do is the following: once derived the fitting curve, to trace (if it is possible) the parallel bands that contain a certain percent of the data point (let's say 68%, 95% etc.) In my opinion these curves should not be the Confidence or the Prediction bands, which represent the bands to be confident that your best fitting and sampled data could be representative of the real population. Do you agree?  |
| larry_lan |
Posted - 10/22/2010 : 02:28:28 AM
Regarding your question, "trace automatically of the bands", I guess you want the XY value of these bands? For every fits, Origin generates two worksheets, one is the Report Worksheet, while the other one is data worksheet for the curves. You can get the bands' values in the second worksheet.
Maybe you want the X value of the bands exactly equal to your source data? If so, you need to select "Same as Input" before doing the fit. If you are using the NLFit dialog, you can find this option in Settings tab, Fitted Curves page, X Data Type drop-down.
Thanks
Larry |
| giordandue |
Posted - 10/21/2010 : 03:38:56 AM
Hi Larry-lan, thank you for you reply. In reality, I think that the Confidence bands should represent, with the desired confidence, the possible variation of the best fitting if sampling would be repeated. In other words, if I am not wrong, the Confidence bands should be equivalent to the confidence range for the estimate of the mean in statistical theory, i.e. [X+-zc*standard deviation/(n)^0.5], where X=mean, zc=confidence coefficient, n=number of tests. However, the bands representing n% of data should be in the case related only to the standard deviation (i.e. for normal distribution, +-1 standard deviation contain 68.27% of data, etc.)and my question focus on this: is it possible to trace automatically these bands?
Giordano |
| larry_lan |
Posted - 10/20/2010 : 11:56:02 PM
I think you can output the Confidence Band? It shows n% of the sample data points will fall within the band.
Larry |
|
|
