“…In other words, this is a decent approximation of SE for sufficiently long tests only. In fact, the inverse Fisher information tends to overestimate the variance of
and
for short tests especially at the extremes (e.g., Magis,
2014), so exact SEs and confidence intervals have been proposed in recent literature (e.g., Doebler, Doebler, & Holling,
2013; Ippel & Magis,
2020; Liu et al.,
2018; Magis,
2016). Furthermore, the SE of boundary estimates are not well understood, because the asymptotic properties of ML and WL only hold under certain regularity conditions, one of which is that the estimated parameter must be an interior point of the parameter space.…”