Judea,
My (limited) understanding is that these Bayesian hierarchical models are GLMs
in which factors are nested; Bayesian methods are used to estimate model
parameters. They are thought to be useful for making estimates for very small
segments of the sample space, and are used for this reason in marketing, where
they want to know what a particular segment of a market will do in a given
situation.
This is not really explained at Chintagunta's web site:
http://gsbwww.uchicago.edu/fac/pradeep.chintagunta/research/Presentations/Movies/
In particular, on slide 8 he refers to the work of McCullogh and Rossi:
http://gsbwww.uchicago.edu/fac/pradeep.chintagunta/research/Presentations/Movies/sld008.htm
I couldn't find these references, they're not specific enough.
You should also look at
http://www.fas.harvard.edu/~stats/survey-soft/hierarchical.html
This page lists review articles.
Best regards,
Ed
On Thu, 11 Jan 2001, Judea Pearl wrote:
> Date: Thu, 11 Jan 2001 08:44:14 -0800
> From: Judea Pearl <judea@cs.ucla.edu>
> To: uai@cs.orst.edu
> Subject: Re: From the Indian press
>
> Has anyone heard about this work before?
> What is a "Hierarchical Bayes" approach?
> =========Judea
>
>
> 01/07/2001
> The Economic Times
>
> Copyright (C) 2001 The Economic Times; Source: World Reporter (TM)
>
> WANT TO know if the next planned blockbuster will hit the charts? Do you
> know that thrillers are popular in Japan and Mexico or that romantic movies
> doing well in the US, Germany, Sweden and South Africa are least likely to
> do well in other international markets? Ask Dr Pradeep Chintagunta,
> professor at the University of Chicago who has worked on a unique model
> based on the "hierarchical Bayesian" approach, a statistical form that lets
> a user input information at more than one stage in the research, that
> forecasts the future of an unreleased movie.
> According to The Economist, Chintagunta worked along with professor Ramya
> Neelamegham, on the Bayesian model using box office figures for 35 movies,
> each of which was released in 14 countries, taking into account various
> factors, including cultural preferences, genre, and the reputations of
> leading actors. Brazil and Britain were found to be the most unpredictable
> markets (43 per cent average error rate), with Japan and Germany the most
> predictable (only 21 per cent average error rate). So can Bollywood be far
> behind and would he work this model for India? Says Chintagunta: "This
> should not be a problem as long as accurate numbers are published. One of
> the issues is whether reported numbers are correct. Even if they are, will
> they be available for several periods - weeks, months - at least for a 'BAS'
> set of movies?" Meanwhile, Chintagunta has extended this model to new tech
> products like digital cameras. He claims the model can be used in
> management. "We have been contacted by a lot of folks. Unfortunately, they
> tend to have shortterm needs and do not have the data readily available to
> give us." Chintagunta says what really brought him to the US was "not a
> very challenging banking job in India." He did his BT in mechanical
> engineering from the Institute of Technology, Banaras Hindu University in
> 1984 and MS (management) from the Indian Institute of Management, Ahmedabad.
> Having got his PhD from the Northwestern University, he was an assistant
> professor of marketing, Johnson Graduate School of Management, Cornell
> University before joining the faculty at the University of Chicago graduate
> school of business where he has been since 1995. He was awarded the Hillel
> Einhorn Award for Excellence in Teaching, (Barcelona) Program, 1999. His
> journalistic career spans many publications as editor, Journal of Business;
> area editor, Marketing Science; and associate editor, Management Science,
> Journal of Business and Economic Statistics. Life moves at a cerebral pace
> at the GSB in the University of Chicago where he teaches marketing strategy
> and marketing management. He says: "Working here is terrific. Chicago is as
> close to intellectual utopia as you can get. I find it very challenging. I
> have colleagues who are super smart and I consider it a privilege to be part
> of the group of faculty here. The students, MBA and PhD are also extremely
> bright and keep you on your toes." Chintagunta is quite Indian in his food
> habits, values, music and films. "My wife is Spanish," he says, "and we need
> to get our regular `dosa fix. Family values in Spain and India are very
> similar so it works out well. We enjoy spending time with both our families
> and try to do it every year if possible." He has researched issues relating
> to brand repositioning, new brand introduction, estimation of brand use from
> household goods scanner-panel data and has done an analysis of household
> purchase behaviour. Using data on supermarket scanners, Chintagunta has
> tried to figure out how consumers make purchase decisions. Are they driven
> by price? Are they loyal? How do various brands compete with one another for
> consumer attention. Currently working on how companies interact with one
> another, he adds: "I am just finishing a paper on pharmaceutical pricing.
> Firms set prices in many geographic markets (countries) taking into account
> the local market response to price, interaction with other firms in other
> markets. "For instance, Pfizer may not want to lower prices to compete with
> Merck in UK as it knows Merck can retaliate by lowering prices in UK and the
> US. The paper attempts to decompose prices into the three components."
>
>
>
>
------- End of Forwarded Message
This archive was generated by hypermail 2b29 : Thu Jan 11 2001 - 12:01:37 PST