How does Prism compute the normalized covariance matrix?
When you check an option on the diagnostics tab, Prism reports the normalized covariance matrix. Each value ranges from -1 to 1. For details on how Prism computes the covariance matrix, look at page 9 and 10 of this article. The covariance matrix is the inverse of the Hessian (Gaussian merit”) matrix shown in Equation 12. Prism reports the normalized matrix, which is obtained by multiplying each value by: Syx2 / (SE(i) * SE(j) ) How to convert to the nonnormalized matrix Some other programs report the actual (not normalized) variance-covariance matrix. Compute the actual covariance — cov(i,j) — of any two parameters (so i does not equal j) from the normalized matrix Prism reports — NormCov(i,j) — using this equation: Cov(i, j) = NormCov(i, j) * SE(i) * SE(j) / Syx2 That equation calculates the covariance from the normalized covariance, the standard errors (SE) of the two parameters and the standard deviation of the residuals (Syx). How to compute the variance of a parameter? Prism
