4. Application: modelling the UK income distribution
Since the late 1970s, real income has increased substantially in the UK, but the gap between the poorest and the richest has also increased faster than in any other comparable industrial countries. We will provide a decomposition analysis of this increase in inequality using the Family Expenditure Survey (FES).
4.2. Model selection
We provide prior information that serves both for inference and for model selection. We have a strong prior for a mixture with three members because we want to interpret each member as representative of one of the social classes: poor, middle, and rich. But for model selection purposes, we consider the possibility of more members. We selected four quantiles of the density of log incomes 0.25, 0.50, 0.75, 0.95 and also 0.99 which is needed in one case. These values will serve to define prior expectations for the μj. We chose n0=1 as prior precision. For the prior on σ2, we selected a prior mean of 0.50, common for all the samples as σ2 is scale free in the log-normal distribution. We chose a common value of 5 for ν0. For γ0, we selected the value of 5 for all the members of the mixture, which is a rather soft prior when compared to the sample size. Table 2 summarizes the different sample quantiles and means of the logs.