From TomWiki
Jump to navigationJump to search


Compute parameter statistics for least squares problems.

Calling Syntax

LS = StatLS(x_k, r_k, J_k);


Input Description
x_k Optimal parameter vector, length n.
r_k Residual vector, length m.
J_k Jacobian matrix, length m by n.


Structure LS with fields:

Output Description
SSQ Sum of squares: r'k * rk
covar Covariance matrix: Inverse of J' * diag(1./(r'k * rk )) * J
sigma2 Estimate squared standard deviation of problem, SSQ / Degrees of freedom, i.e. SSQ/(m-n)
Corr Correlation matrix: Normalized Covariance matrix
Cov./(CovDiag * CovDiag'), where CovDiag = sqrt(diag(Cov))
StdDev Estimated standard deviation in parameters: CovDiag * sqrt(sigma2)
x =x_k, the input x
ConfLim 95 % Confidence limit (roughly) assuming normal distribution of errors ConfLim = 2 * LS.StdDev
CoeffVar The coefficients of variation of estimates: StdDev./xk