In PROC HPLOGISTIC output, the 'Convergence Status' indicates 'Convergence criterion (GCONV=1E-8) satisfied.' What does this mean?
Select an answer to reveal the explanation.
Short Explanation
Here's the deal — when the convergence status says the gradient criterion is satisfied, the normalized gradient is tiny enough that the optimizer has found a stationary point of the log-likelihood. That means you've obtained the maximum likelihood estimates.
Full Explanation
The option saying the gradient of the log-likelihood is sufficiently close to zero, indicating the algorithm found the maximum likelihood estimates, is correct because GCONV is a gradient-based convergence criterion in PROC HPLOGISTIC. When normalized gradient length falls below GCONV=1E-8, the optimizer has found a stationary point of the log-likelihood, so maximum likelihood estimates have been obtained. The maximum-iterations-without-converging option describes non-convergence. The poor-fit-because-gradient-small option is wrong because satisfied convergence is positive, not poor fit. The increase-GCONV-to-1E-6 option is wrong because increasing GCONV makes the criterion less stringent, and current convergence is satisfactory.