6. Exercise 5 6.odc contains a 10% sample (n = 441) of the 4406 observations on J = 6 count responses relating to health care use; these data are considered by Chib andWinkelmann

(2001). The responses are y1 = visits to physician in an office setting, y2 = visits to a nonphysician in office setting, y3 = visits to physician in hospital outpatient setting, y4 = visits to nonphysician in hospital outpatient setting, y5 = visits to an emergency room, y6 = number of hospital stays. Correlations between the ui j as in the model set out in (5.7) might in this instance represent substitution effects between different forms of health demand. One possibility for the prior on the precision matrix is

−1 ∼ Wishart

(J, j ˆ), where ˆ is a prior estimate of ; Chib and Winkelmann (2001) assume

−1 ∼

Wishart(6, I). Compare the model in (5.8) with one that assumes scale mixing and may better accommodate outlier subjects; thus

is equivalent to assuming ui follows a multivariate Student t with ν degrees of freedom.
In particular, compare inferences under the two models on the correlations r56, and r26;
the latter may be taken as representing the association between serious and less serious morbidity.

~ Yijlui, Bj Polojj) log(0)=xj;+ Uij (uil, uiuij)~N(0, /ki) K; Ga(0.5v, 0.5v)