http://www.deloitte.com/dtt/cda/doc/content/Facility.pdfhttp://faculty.washington.edu/krumme/location/saturn.htmlSaturn: A Bayesian Example
(
http://faculty.washington.edu/krumme/location/saturn.html)
Supporting & Related Pages:
Bayes
Decisions and Uncertainty
--------------------------------------------------------------------------------
In 1985 Roger Smith, GMs CEO, announced the secret Saturn project in order to "leap-frog" the Japanese car makers. Production started in October 1990 The project resulted in a $ 3.5 billion investment from GM and in an all new factory in Spring Hills, Tennessee From the space-frame Saturn car an almost self-contained production was launched with an "empowered" organization Nowadays the Saturn corporation is manufacturing about 320.000 cars a year and employs 10.000 people (Source)
--------------------------------------------------------------------------------
General Motors (GM) is launching a new car (Saturn) and is looking for a location and production site. After GM's location and real estate department has found its preferred site, it will assign a 'subjective' probability of .9 to this site suggesting that "we are 90% certain that Tennessee is superior to its closest competitors Michigan and New Jersey."
However, GM management is still sufficiently confident that this assessment is correct since it has relatively little prior knowledge of the state. It also does not want to give the impression of merely imitating Nissan's prior decision for Smyrna, relying too much on Nissan's experience or yielding to the Governor's persuasive abilities.
GM engages the FANTUS corporation, a large (formerly independent) location and corporate real estate consulting firm located in New Jersey (and acquired in 1996 by Deloitte & Touche), to carry out a quick survey and evaluation to confirm or reject its initial assessment. Fantus President Yaseen (of course a geographer by training) tells GM that the re-evaluation will be only 80% reliable because of time and measurement constraints and the limits GM has imposed on consulting fees.
GM now wants to know what its own, revised pro-Tennessee probability will be after Fantus's collection of additional independent information and unbiased reevaluation:
In terms of Bayes' Theorem:
E1 = event "Tennessee is superior over Michigan or N.J.
E2 = event "Tennessee is inferior"
P(E1) = .9
} = prior probabilities
P(E2) = .1
F = "Location evaluation by Fantus inicates Tennessee is superior"; then the conditional probabilities ("likelihoods") are:
P (F|E1) = .8 ( = reliability of new information
or the likelihood that new information supports the original hypothesis)
P (F|E2) = .2
Bayes' Theorem:
Posterior Probability = P (E1|F) = P(E1) P(F|E1)
---------------------
P(E1) P(F|E1) + P(E2) P(F|E2) = .9 x .8
-----------
.9 x .8 + .1 x .2 = .97
If the Fantus location survey indicates that Tennessee is superior, GM's new (posterior) probablity (confidence) should be .97
If, however, Fantus says: "Michigan or N.J. is a better location, what is GM's new probability (confidence) that Tennessee is still better?
1 - posterior probability that
Michigan or N.J. is superior = 1 - .1 x .8
--------------
.1 x .8 + .9 x.2 = .69 = posterior probability for Tennessee.
Are you still surprised that GM did not come to Washington State?
Criteria all spelled out by the company "planners"
http://www.deloitte.com/dtt/cda/doc/content/Facility.pdf