MathGroup Archive 2011

[Date Index] [Thread Index] [Author Index]

Search the Archive


On Jul 17, 3:06 am, maria giovanna dainotti
<mariagiovannadaino... at> wrote:
> Dear Mathematica Group,
> I am fitting a function of 3 parameters with the NonlinearModelFit
> and with the Marquardt Levemberg algorithm.  Since  this method gives
> the interval of the Parameters of the best fit considering the parameters
> not correlated, I was looking for a tool that takes into account the
> fact that the parameters are correlated and the  parameters interval
> should respect the following rule:
> The confidence interval for a given parameter should be computed by
> varying the parameter value until the chi^2 increases by a particular
> amount above the minimum, or best-fit value.  The amount that the chi
> square is allowed to increase (also referred to as the critical
> delta_chi^2) depends on the confidence level one requires, and on
> the number of parameters whose confidence space is being calculated.
> The critical delta_chi^2 for common cases are given in Avni, 1976:
>  For example in the case I am using for 3 parameters delta_chi^2=3.50.
> I would be very grateful if someone of you is acquainted with a such
> a tool in Mathematica or some of you has done a code to sort this
> problem out.  Thanks a lot in advance for your help.
> Best regards,
> Maria
discusses a similar problem. Here is the code I used.

color = {"Blue","Brown","Green","Orange","Red","Yellow"};
{k,n} = {Length@#,Tr@#}&[f = {127, 63, 122, 147, 93, 76}];
e = Array[x,k];
init = Transpose@{e,f-1,f+1};
b[\[Alpha]_] = Tr[f^2/e] == n + 2 InverseGammaRegularized[(k-1)/2,
  \[Alpha]] && And@@Thread[e > 0] && Tr@e == n;

"         95% Limits\nColor   Lower  Upper\n" <>

         95% Limits
Color   Lower  Upper
Blue    0.154  0.261
Brown   0.067  0.147
Green   0.147  0.252
Orange  0.183  0.295
Red     0.107  0.201
Yellow  0.084  0.171

  • Prev by Date: Re: Solve never calls Equal?
  • Next by Date: Re: Solve never calls Equal?
  • Previous by thread: I: NONLINEARMODELFIT
  • Next by thread: infection model of numeric coercion in Mathematica [Was: Re: Numerical accuracy/precision - this is a bug or]