« Tonight - Rob Corddry reports on "New Forms of Journalism" | Main | Computer woes »
March 04, 2005
A model is a set of assumptions
As you might know, my thesis work consists of evaluating current models, then using what I like about them and fixing what I don't like about them to create one of my own. It doesn't matter what sort of models I'm looking at or even what questions I want to answer about them, I'm learning from my adviser and my own experience that they all boil down to the same thing - assumptions.
For the most part, models are just short-hand for how you get from one place to another, from A to B. You start out with certain climate data and you end up with a weather prediction; you start with your personal and financial situation and wind up with an investment strategy; you start with exposure and illness information and wind up with a measure of causal association. The model you use tells the story of how all of these data relate to one another and how they come together to give you the outcome - the answer to your question . This "telling of the story" is just a description of your assumptions: this is a linear relationship, these follow a Normal distribution, this decays at this given rate.
I find I'm having a really hard time excavating assumptions from some of the models I am reviewing. Part of the problem is that modeling is a whole new field for me and I don't have much experience. Another part of it is that I hadn't been trained to identify the assumptions in the models I used through most of my schooling (mostly statistical models). But a large part of my trouble is also that many people are not accustomed to laying out their assumptions in a straight-forward manner.
So the next time someone wants you to use a model - a tool - to do some work for you or solve a problem for you, simply ask, "what are the assumptions this model is making to take me from A to B?" They you can decide for yourself if the path the model is taking is one you can accept. But if they can't answer your question, then you've got a REAL problem.
Posted by rigd0003 at March 4, 2005 11:48 AM | PhD Process