Closed richarddmorey closed 9 years ago
Gaussian quadrature has a tendency to fail when effect sizes are very small (for instance, very small correlations). A fallback (Laplace or sampler) needs to be included for when quadrature fails.
Partially fixed this, with the t test. Large t statistics cause an approximation to be invoked.
This should be largely fixed, due to improvements in the quadrature calls.
Gaussian quadrature has a tendency to fail when effect sizes are very small (for instance, very small correlations). A fallback (Laplace or sampler) needs to be included for when quadrature fails.