Currently, we take the curvature at the ML value for x_i | theta_i, with no average. This could possibly create some troubles, if the noise in x_i | theta_i has a strong effect.
Also verify exactly if this is smaller/larger than the average FI we used before, which matched nicely between all ways of estimating it.
Currently, we take the curvature at the ML value for x_i | theta_i, with no average. This could possibly create some troubles, if the noise in x_i | theta_i has a strong effect.
Also verify exactly if this is smaller/larger than the average FI we used before, which matched nicely between all ways of estimating it.