## Revision 1360

1001 1001
is that the prior on the random effects acts as a penalty function.
1002 1002

1003 1003
We can now return to the role of the initial values specified for the random
1004
effects in the \texttt{.pin} file. Each time step~1 above is performed, these
1005
values are used---unless you use the command line option \texttt{-noinit}, in
1006
which case the previous optimum is used as the starting value.
1004
effects in the \texttt{.pin} file. Unless you use the command line option \texttt{-noinit}
1005
these values are not used (but dummy values must nevertheless be provided
1006
in the \texttt{.pin} file).  Each time step~1 above is performed
1007
the maximization of the penalized likelihood is always initiated at \$x=0\$.
1008
If the command line option \texttt{-noinit} is used the value
1009
from the \texttt{.pin} file is used the first time step~1 above is performed.
1010
The remaining times the value from the previous optimum is used as the starting value.
1007 1011

1008 1012
\paragraph{Empirical Bayes} is commonly used to refer to Bayesian estimates of
1009 1013
the random effects, with the hyper-parameters fixed at their maximum likelihood

Also available in: Unified diff