[ADMB Users] sdreport
mmaunder at iattc.org
Sun Feb 19 13:04:36 PST 2012
Try doing a profile likelihood on one of the parameters/quantities of interest and see if that gives better coverage. The CIs may be asymmetric or you may have a parameter close to a bound.
From: users-bounces at admb-project.org [mailto:users-bounces at admb-project.org] On Behalf Of Mollie Brooks
Sent: Saturday, February 18, 2012 5:13 PM
To: users at admb-project.org Users
Subject: [ADMB Users] sdreport
I'm looking for information on how sdreport vectors are calculated.
I'm doing a multivariate normal model and I use the solve function to calculate my objective function from the predictions and the variance-covariance matrix.
I'm interested in the uncertainty of my predictions. So I put them into an sdreport vector. I've looked at CIs calculated as (prediction +/- 2*std) and they are way too narrow (almost never include the true value). I'm wondering if I should try calculating the standard deviations from scratch or if ADMB does exactly what I would do with the variance-covariance matrix. The other possibility is that my measurement error only model is bad and I need to include process error.
NSF IGERT Fellow
University of Florida
mbrooks at ufl.edu
Users mailing list
Users at admb-project.org
More information about the Users