
168 Bayesian Methods in Health Economics
possible to use the posterior distribution of β =(β
0
,...,β
L
), which is derived
from the data in D
pos
, to compute the predictive distribution
μ
∗
i
= β
0
+
L
l=1
β
l
z
∗
il
,
where z
∗
il
is the value of the l−th covariate measured on the i−th individual
in D
null
.
ThiscaninturnbeusedtoestimatethecostforthesubjectsinD
null
in
terms of a mixture of the two components as
c
∗
i
= π
i
× exp
μ
∗
i
+
σ
2
2
+(1−π
i
) × 0. (5.7)
Of course, if different specifications are used for the cost distribution, it is
necessary to modify (5.7) to rescale the costs on the natural scale.
In addition, the different models can be formally assessed in terms of their
predictive ...