pwConfidenceIntervals
ci = pwConfidenceIntervals(mode, parameterValues, nFits, criticalValue)
Determines the 95%-confidence intervals of all currently fitted
parameters.
Please note that since PW 2.1 we rather recommend using the Profile
Likelihood approach. See pwPleInit.
mode 0 Fisher-information based 95%-confidence-intervals (default)
This requires that you fitted with trustregion or marquardt
in normal parameter space.
1-3 Estimates the confidence intervals based on a fit-sequence:
1 (Data + Noise) Data is simulated based on original values plus Gaussian noise.
2 (Monte-Carlo) Data is simulated based on model values plus Gaussian noise.
3 (Bootstrap) Data points are redrawn with replacement.
This is done nFits times. The distribution of fitted parameter values
of the best 85% fits serves to estimate the confidence intervals.
The parameters are disturbed using config.optimization.strengthOfDisturbance.
4 The Hessian is approximated nummerically which requires repeated
integration of the system.
This setting is still in BETA.
parameterValues
Only for mode 4: If parameterValues are given, the Hessian is calculated at this
point in the parameter space. Else, the current parameter values of the
equalizer are taken.
Default: [] (empty)
nFits Number of fits for mode 1-3. Default: 100
criticalValue
Factor multiplied to the internally determined standard deviation
in order to calculate the confidence limits.
Default: 1.959964 corresponding to a 95% confidence limit:
qnorm((1 + 95)/2); -> 97.5 % quantile
Description
With no output argument, the confidence interval is plotted.
Else, the confidence intervals are returned for each parameter.
Since PW 2.0.44, the Profile Likelihood Estimation (pwPle) approach should
be used to determine confidence intervals in order to avoid
approximation errors for non-identifiable errors for mode 0 or 4.
See also