confidence bands with ODR curve fitting

Share icon

Wed, 03/09/2011 - 05:48 am

I have some questions about prediction bands in igor...

i can´t perform confidence bands in curvefit when i use ODR=2 (i use this method because i perform fitting with error in both vaiables). on the other hand, igor generates a convariance matrix, which could help me to calculate confidence bands after fitting.

(why igor doesn´t perform confidence bands with ORD???)

in case of perform confidence bands after fitting, i use the "variance of predicted model value" as the manual says, but i found in some books that the "variance of predicted model value" is multiplied by chisquare value in order to obatain the confidence bands... how i should perform the confidence bands with ODR???

i really appreciate you help

reggards
OK- I'm back with some questions.

Are you interested in confidence bands, or confidence intervals for the fit coefficients? A confidence band is a contour about the model curve; the confidence interval for the coefficients is a +- error bar about the fit coefficient value.

If you want confidence intervals, this is possible with ODR fitting. Add this to the end of your FuncFit or CurveFit command:
/F={0.950000, 4}

This produces the 95 percent confidence interval; change 0.95 to something else to get other intervals. The 4 selects confidence intervals. Other values of the second parameter will give errors- they select confidence bands and prediction bands for the model curve.


If you really want confidence bands, it is possible to compute after a fit, but not completely straightforword, as it requires partial derivatives of the fit function with respect to the fit coefficients. Are you using a built-in fit function?
Quote:
(why igor doesn´t perform confidence bands with ORD???)

Because the code I use to perform the ODR fitting is not my own code, and I don't have ready access to the derivatives required. It would require significant further effort, and at the moment I simply don't have the time to devote to it.

John Weeks
WaveMetrics, Inc.
support@wavemetrics.com

March 10, 2011 at 04:23 pm - Permalink

hi
i'm interested in confidence bands.
i usually use some not built-in fit functions, but this functions have a analytical partial derivate respect to fitting coefficents (for example: Y=a + b/x or something like that)...

thanks

March 10, 2011 at 06:22 pm - Permalink

OK- Can you send me full details of your fitting function? Best would be send me an Igor experiment file with your fitting function and an example fit. In your message it would be most convenient if you could send me the mathematical details of your function so that I don't have to wade through code to figure out the derivatives. Send it to the address below, support@wavemetrics.com

John Weeks
WaveMetrics, Inc.
support@wavemetrics.com

March 11, 2011 at 11:56 am - Permalink

johnweeks wrote:
OK- Can you send me full details of your fitting function? Best would be send me an Igor experiment file with your fitting function and an example fit. In your message it would be most convenient if you could send me the mathematical details of your function so that I don't have to wade through code to figure out the derivatives. Send it to the address below, support@wavemetrics.com

John Weeks
WaveMetrics, Inc.
support@wavemetrics.com


I just sent the e-mail.

thanks

March 14, 2011 at 11:49 am - Permalink