Dear Users,

1. The value of ConfidenceInterval by Fit and FitPlot was changed. It has been Sqrt[Cii], where Cii is the diagonal component of the covariant matrix, but it may not be right. Now it is changed to Sqrt[ DeltaChisq Cii], where DeltaChisq is the solution of

GammaRegularized[nvar/2, DeltaChisq/2 ] == Erfc[Sqrt[1/2]] .

Dear Users,

1. On top of the previous change, to obtain the covariant matrix was improved to enable analytic differentiation even when the equation to fit contains Part whose index is the independent variable. This kind of change will be also applied for similar errors with Fit or Plot in future.

2. Derivative of Erf was wrong by a factor of Sqrt[Pi].



> Dear Users,
>
> 1. The value of ConfidenceInterval by Fit and FitPlot was changed. It has been Sqrt[Cii], where Cii is the diagonal component of the covariant matrix, but it may not be right. Now it is changed to Sqrt[ DeltaChisq Cii], where DeltaChisq is the solution of
>
> GammaRegularized[nvar/2, DeltaChisq/2 ] == Erfc[Sqrt[1/2]] .

Dear Users,

1. FitPlot was modified to avoid unnecessary error messages when Part is used in the fit function.

Dear Users,

1. Now Fit has a new option Cutoff to specify the saturation amplitude to calculate chi-square:

chi-suare = Sum_i ( Min[Cutoff, Max[-Cutoff, (d_i - f_ i) / sigma_i]] ) ,

which is a sort of robust M-Estimates. By Cutoff, the fit tends to ignore tail data which are beyond Cutoff.
If Cutoff is zero (default), it is ignored.

There was a mistake in the expression:

Dear Users,

1. Now Fit has a new option Cutoff to specify the saturation amplitude to calculate chi-square:

chi-suare = Sum_i ( Min[Cutoff, Max[-Cutoff, (d_i - f_ i) / sigma_i]]^2 ) ,

which is a sort of robust M-Estimates. By Cutoff, the fit tends to ignore tail data which are beyond Cutoff.
If Cutoff is zero (default), it is ignored.

> Dear Users,
>
> 1. Now Fit has a new option Cutoff to specify the saturation amplitude to calculate chi-square:
>
> chi-suare = Sum_i ( Min[Cutoff, Max[-Cutoff, (d_i - f_ i) / sigma_i]] ) ,
>
> which is a sort of robust M-Estimates. By Cutoff, the fit tends to ignore tail data which are beyond Cutoff.
> If Cutoff is zero (default), it is ignored.