I have a set of 260 sets of measurements (for each set of measurements there is an amplitude measured as a function of 8 radii). Since I do not get measurement errors and I am interested in the properties of the distribution of measurements, I am using jacknifing to determine the errors for the distribution, and bootstrapping in order to estimate the 8x8 covariance matrix. The problem is that even at one million bootstrap samples the determinant is very close to 0 (10^-21).

Would you suggest alternative ways of estimating the covariance matrix? I need it for finding the minimum reduced chi square for a set of models that I am fitting to the data.

  • $\begingroup$ how many amplitude-radius pairs per set? $\endgroup$ – Memming Jun 10 '16 at 10:39
  • $\begingroup$ I've edited to include that information. There are 8 measurements per set. $\endgroup$ – mannaroth Jun 10 '16 at 10:47

Since I had enough measurements (273 sets), I could use the data itself to directly create the covariance matrix. This has fixed the determinant as well, it hovers around 0.5.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.