Gibbs sampling and data augmentation have revolutionised Bayesian inference, particularly in extensions of the normal-linear model. The separate works collected here present applications of these ideas to data on an emerging milk market in the Ethiopian highlands. Fundamental to this collection is the notion of statistical robustness.
The concept that statistical measures remain robust to a diverse set of alternative model formulations is important for policy. This collection has summarised the results of a search for three measures of particular relevance to market participation policy, namely the levels of three essential inputs in the milk production and selling exercise. In this regard, this collection has highlighted the need for policymakers, administrators, extension specialists etc. to focus attentions on the additions of about 2–3 crossbred cows, 7–8 local breed cows or 9–10 visits by extension agents as paramount in effecting participation among representative non-participants. That these quantities have been discovered under so many alternative specifications suggests an important conclusion from the exercise. These estimates are robust.
Although this search for robustness was the main objective of this exercise—an objective shared also in the genesis of data collection (Nicholson C.F., personal communication)—we discovered, along the way, another important fact. This fact is that routine applications of Markov chain Monte Carlo methods—Gibbs sampling and data augmentation in particular—provide important measures for market-development policy. Indeed, in view of their worth, it is surprising that these methods have received so little attention by agricultural economists and development economists to date. Agricultural economists, we believe, particularly those with empirical interests, should aim to focus more attention on the method; a series of fundamental contributions (Chib 1992, 1995, 1996); their application (Dorfman 1996); and, we hope, their profitable exploitation in policy formation.
