There is always a tradeoff between modelling a process correctly (ie without missing variables, entirely in terms of what has actually been observed and nothing else) and adapting an off the shelf model while hoping it will work out.
The reality is that the substantial effort required to model a process from scratch is not justified in nearly all cases. The few standard (statistical and physical) models that were developed from scratch since the Renaissance have been reused and extended many times, to the point that the cost of using them is merely a few years of university education. -- $E(X_t|F_s) = X_s,\quad t > s$
You're quite right that in most cases the substantial effort to make a new model from scratch is unjustified, and this gets to the heart of Bruce's issue with the institutionalist critique. We're biased by the nature and necessity of our circumstances to accept and build on the models and mistakes of others. Which means that we're likely to miss important things that don't fit the models, such as unobservable phenomenon.
However, even in making a model from scratch there is still the non-trivial issue of unobservables. This is the real problem that most economists, as well as many other social scientists, and even medical researchers, struggle to answer: "What WOULD have happened if X were true instead of Y?" -- a counterfactual, in other words. That's how causality is best inferred and how statistics is used to find the answer, but doing so is really hard work because counterfactuals are, by definition, unobservable, which means that a better theory makes all the difference.
A missing variable is a problem IF there is, in fact, a variable missing from a statistical model. The problem is that often only theory can tell you if it is missing or not -- not anything in your model itself. This means that you'll never know if it's missing if you haven't thought sufficiently about your problem.
That is why you make control experiments.
This is the real problem that most economists, as well as many other social scientists, and even medical researchers, struggle to answer: "What WOULD have happened if X were true instead of Y?" -- a counterfactual, in other words. That's how causality is best inferred and how statistics is used to find the answer, but doing so is really hard work because counterfactuals are, by definition, unobservable, which means that a better theory makes all the difference.
And this is why you do double-blind placebo-controlled clinical trials.
- Jake If you only spend 20 minutes of the rest of your life on economics, go spend them here.
