The Identification Problem

If one wants to see whether one model's predictions are better than another, why not just look at the data? It turns out that this is a very difficult thing to do because economists cannot conduct experiments. Competing theories may provide differeing predictions about the relationship betweent two variables, but it is not so easy to evaulate this relationship. As Krugman points out in Metaphors and Models, models in social sciences are simplified accounts of the world: they leave a lot of things out and, in  reality, all other variables are changing as well -- for many different reasons. This is the essence of the identification problem. 
The figure below provides an illustration of the identification problem for the specific example of the relationship between national savings, investment and the interest rate. Macroeconomic theories we will study differ in the predictions about the effect of changes in the interest rate, r, on saving, S, and investment, I. In particular, some theories claim that investment is largely insensitive to changes r (i.e. the investment curve should be very steep), while others say that the curve should be quite flat. So why don't we just look at the data?The left panel of the figure below shows what we want to see: how shifts in the national saving function trace out the investment demand curve. The center panel shows what we can actually observe: just the intersections (i.e. we get to see only the interest rate and the amount of investment). Finally, as the third panel shows, we see this intersection because both curves are simultaneously shifting for many, many different reasons. Of course, statistical techniques exist that can help us estimate these saving and investment curves from the data points that we have. But the curves we get are only as good as the theoretical models and statistical techniques employed (about which there is much debate), and even then they are only probabilistic statements.
what we would like to know . . .  . . . is not what we get to observe . . .  . . . and may not be what is going on.
The Identification Problem

On Line Readings
An Apple a Day—No, Wait