10. Use the data from Gelfand et al. (1990) relating to growth for n = 30 rats at five ages and add a DPP as in West et al. (1994, p. 373) and Escobar and West (1998, p. 16). See also the birats example on the WINBUGS site. Thus the bivariate normal model for varying intercepts and slopes is replaced by a DPP that allows clustering of intercepts and slopes.

Specifically one could retain as G0 the bivariate normal with a precision matrix distributed as Wishart(C, 2), where

and take M = 20. The Dirichlet parameter can be assigned either a Ga(1,1) prior or a Ga(0.01,0.01) prior as in Escobar and West (1998). Both studies applying a DPP to these data found multimodal posteriors for the predictive distribution of the slopes.

c = [100001]