Postprint version. Published in International Journal of Approximate Reasoning, Volume 50, Issue 3, March 1, 2009, pages 416-428.
The definitive version is available at http://dx.doi.org/10.1016/j.ijar.2008.05.005.
This paper describes a process by which anthropologists, computer scientists, and social welfare case managers collaborated to build a stochastic model of welfare advising in Kentucky. In the process of collaboration,the research team rethought the Bayesian network model of Markov decision processes and designed a new knowledge elicitation format. We expect that this model will have wide applicability in other domains.