ABSTRACT
Model specications built on the normal distribution are as common in
Bayesian statistics as they are in non-Bayesian approaches. There are sev-
eral reasons for this. First, nature seems to have an aÆnity for this form as
evidenced through empirical observation as well as from the central limit
theorem. The weakest form of the central limit theorem essentially says
that an interval measured statistic with bounded variance will eventually
be normally distributed provided suÆcient sample size. Therefore it is quite
common to see situations where quantities behave approximately normally.
Second, a huge class of posterior distributions can be modeled by combining
an assumed normal likelihood function with diering priors. Finally, when
Bayesian models were more diÆcult to estimate numerically, the normal
distribution sometimes provided an analytically tractable posterior when
other forms were less compliant.
