Notes‎ > ‎

An Agent-based Bayesian Forecasting Model for the University

posted 2 Apr 2012, 03:59 by David Sherlock

Here are some unfinished notes I jotted down last night. I have to thank Bruce Carelton for  this. I have lots of background reading to do but lots of work to do, so these are very early thoughts that may never get looked at again.


I wonder what an early warning system for the University would look like. While researching models of management that could incorporate such a system I find myself going in circles reading themes around Beers work in the 70s but really failing to really grasp the core concepts. Although I have been unable to invest the time I would like into his work I have found Bruce Carlton’s Viable System Agent a way into understanding the VSM and how some of its core components. Recently Bruce has introduced me to Cyberfilter (Beer, 1979). From what I gather Cyberfilter is a tool to measure performance in the VSM. To implement Cyberfilter I gather that Bruce is creating an alerting mechanism based upon the variables created by his Viable System Agent.

I am particularly interested in the method that Bruce is using to create the alerting system. He is using the Harris-Stevens Model of Bayesian theory based upon the agent’s capability variable. Bruce pointed out to me that Beer himself was confident in that the techniques would detect any insatiability in the organisation before Human’s would. Bruce pointed me towards Beer references.

-Harrison, P. J., and Stevens, C. F., 'A Bayesian Approach to Short-term Forecasting', Operational Research Quarterly, 22, No. 4. December 1971.

-Harrison, P. J., and Stevens, C. F., 'Bayesian Forecasting', Journal of the Royal Statistical Society, Series B (Methodological), 38, No.3, 1976.

I am interested in forecast models. I find it very depressing that we have lots of educational data but not are not sure what to do with other then make pretty pictures.  I am concerned with what is considered creditable when we talk about measuring or visualising this data. Measuring pre-determined outputs set by the people with the money rings my alarm bells.My thought process seems to be.

Lots of ways to collect data -> Explore Forecast models we can use with that data


Espinosa, A. & Jackson M.C. (2002). A systemic look at educational development programs: Two perspectives on a recent Colombian experience, Kybernetes, November