2010, Volume 18, Paper 8
ISSN: 1883-5675

Old model, new problem: when should you update a model and what happens when you do?

G.R. Griffith
L.R. Malcolm
S.W. Mounter
H. Slattery

Introduction

A couple of months ago the phone rings:

BM: G’day mate…(then follows light-hearted bagging of various cricket teams, football teams, state bureaucrats, university administrators, etc; catching up on mates and family; red wine finds; etc)…(then a bit on AAR editorial issues)…(then) listen mate I’ve got this final year student Henry looking for a project and I thought we might be able to do something extra with Amy’s survey on that pig stuff.
GG: do you mean the willingness to pay for low cholesterol pork?
BM: yeah.
GG: well, we could use her results as inputs into that pig edm model we published in the journal a few years back – Stuart Mounter’s model – and look at the industry wide benefits.
BM: sounds good, what do we have to do?
GG: that’s a tricky one. I’d better come down and meet Henry and see what he wants to do.
BM: good, how about early May?
GG: I’m available then. Dinner at Jimmy’s?
BM: all clear.

So a “quick” phone call generates a good idea, which at first blush seems a simple thing to do – apply an existing model to a new problem and then write it up. But the question “What do we have to do?” is a good one, and at second blush things are not so simple. In this short piece we talk about some of the considerations involved in applying an existing model to a new problem, in particular in deciding whether to update or not, and some of the issues involved in interpreting the output from the new application.

Download full document here