Bayesian Statistics
1.7 Families with random parameters
In this section we are interested to take a model family and treat the parameter as a random variable, which will be denoted by a capital letter . We
think of first sampling the value of and then (using whatever value we obtain) taking a sample from . The resulting distribution is sometimes known as a compound or
mixture distribution. We will have a detailed discussion of how this idea becomes useful in Section 2.1. For now
let us note that it increases the range of models that we have available.
To make sense of the idea, let us state it more precisely. We want random variables and such that . In this section we show that a pair
with this property is given by the distribution
where is a family of distributions with range , as defined in Section 1.3, and is a probability
density function with range . This is a type of random variable you may not have seen before. We will shortly show that the part is a continuous random variable, but the part
might be discrete or continuous, depending on .
Our notation strongly suggests that we expect to be the (marginal) probability density function of , and we can confirm this by setting , in which case equation (1.13) becomes . We can also find the marginal
distribution of , by setting , giving
but that formula doesn’t really explain what is going on here. The relationship that we are interested in comes from the following lemma.
Proof: We give a sketch proof to illustrate the idea, in similar style to Lemma 1.6.1. From Lemma 1.5.1, for
we have
The second equality follows from (1.13) with , for the
numerator, and from the fact that is the p.d.f. of , for the denominator. Using continuity, from the statement of the lemma and from Assumption 1.3.2, for we can approximate and . This gives
Letting we have , so by Definition 1.2.1 and (1.9) we have that is well defined and
. ∎