Dear all,
THX to Antigoni discovering a defect in the Univariate 95% confidence interval (WNL Linear Model 502) and CI of fixed effects in Linear Mixed Effects. The internally used critical value of the inverse t is wrong (especially for small degrees of freedom). Confirmed in PHX/WNL 6.3, 6.4, 7.0, and 8.0.
Until the defect is corrected, use one of these workarounds.
Hi Helmut,
To clarify the issue, the CI’s are accurate (were shown to match other tools) for NDF >= 5, using Antigoni’s example with a couple more data points. For NDF <= 4, the accuracy of the estimate declines as NDF → 1, where NDF is the degrees of freedom (loosely speaking, the number of data points minus the number of parameters).
This has been a known issue in our tracking system since WNL 5.1, but has not been considered a priority since it is only about the accuracy of an estimate in the case of very little data, where generally any estimation process may perform poorly. The issue was logged for the LinMix tool, but since the recent report, it is also being tracked as an issue for WNL Classic Modeling.
Thanks for your post,
Linda Hughes
Hi Linda,
THX for the clarification! To check whether (pseudo-) steady state is reached it is common to assess three pre-dose concentrations. With a linear model we end up with only one DF. My workaround is clumsy.