How to calculate standard error and CV percent of parameter estimate using nlinfit and nlparci

Hello Everyone,

I am trying to estimate parameters from my data using nlinfit function. Using nlparci I can get the confidence intervals.

How can I get standard errors and CV% for my parameter estimates?

Thanks!

Answer 1

[Star Strider]

There is actually no reason to calculate the SE or CV values for the parameters estimated in a regression. I have never calculated them, nor have I ever seen them reported in the literature.

Neither of those have any real significance in the context of a regression. The confidence intervals have all the information you need, specifically whether the estimated parameters are significantly different from zero.

[Star Strider]

Satish’s ‘Answer’ moved here:

In field of Pharmacokinetics and Pharmacodynamics CV% or SE of parameters are reported quiet often. I agree with you that confidence interval has all the information.

I was just wondering if matlab has an output for CV% or SE.

[Star Strider]

To the best of my knowledge, MATLAB does not report that information.

You can get it from the diagonal of the covariance matrix, ‘CovB’ in the documentation. Those values are the variances of the parameters, so I would guess that the SE would be the square roots of the diagonal divided by the number of observations. The CV would be 100 times the square roots of the diagonals divided by the values of the respective parameters.

NOTE: I would defer to the literature that describes those calculations (rather than my description) if your literature provides those details. (From a physiology and clinical modeling and parameter estimation perspective, they are never calculated. Only the confidence intervals are ever calculated and reported, at least in my experience.)