
90 Exposure-Response Modeling: Methods and Practical Implementation
problems almost surely occur during simulations, leading to a considerable
number of nonconverged runs. The calculation of (4.52) is numerically very
stable and easy to perform. The large sample bias, to gether with the small
sample property assessment, provides useful information for the evaluation of
the SM approach.
We use the combination of the power model and the Emax model to illus-
trate this approach to bias eva luation. Consider the Emax model
y
i
= E
max
/(1 + EC
50
/c
∗
i
) + ε
i
(4.53)
with c
∗
i
given by an power model. For o ur purpose, we estimate c
∗
i
from the
predicted conditional