Here was an interesting paragraph from a recent Simon Wren-Lewis says, there are many models, and the key questions are about their applicability.
Problem 1: Ex Ante vs. Ex Post.
I'd say this is no good. Telling just-so stories about past events - what some people cynically call "explanation without prediction" - is useless unless it gives us some insight into the future. If we have a grab-bag of models, and we simply pull one out to "explain" every event after it happens, and then sit there in smug satisfaction, well, we're not really adding much value to the human species.
For situationalism to be a useful approach, you need to have some way of telling which model to use ex ante. That brings us to the second hurdle...
Problem 2: Judgment vs. Scope Conditions
In natural sciences, models usually come with some set of scope conditions. For example, if you're modeling a hockey puck, you probably want to ignore friction and use a two-dimensional model. But if you're modeling a cannonball, you probably want to use a 3-dimensional model with air resistance. If you're modeling the progression of a disease in a person with low T-cell count, you probably want to use a different model than if the person has lots of T-cells, and so on. These situations have indicators that give you guidance about which model to use - the coefficient of friction, the T-cell count, or whatever. You can observe these indicators before you choose your model.
In other cases, expert judgment matters more. A surgeon may look at a bunch of technical indicators when deciding which kind of surgery to perform, but there are no hard-and-fast rules. Instead, it's up to her own expert judgment. These two kinds of situationalism fall on a continuum - there's still a little judgment involved when making an approximation in physics, and there are still quantitative indicators that help a surgeon make an expert decision.
In econ, people don't seem to think very hard about how much to rely on judgment vs. evidence for model selection (hopefully Rodrik's book will cover this). Central bankers seem to rely mostly on expert judgment. Some theories come with quantitative indicators telling you which theory to use - for example, New Keynesian models behave differently when interest rates are near zero, and some models explicitly separate short-run from long-run effects.
But in general, economic models are hobbled by the fact that the underlying theory itself (the "microfoundation") is in dispute. Physics models are usually just approximations of a single underlying theory. Econ models actually contain different assumptions about the nature of human and institutional behavior.
So maybe we're stuck with expert judgment. But how good is expert judgment in economic policymaking? That's a very hard thing to assess, especially in macro but really in all areas of policymaking. So I think we're kind of in the dark over how well judgment-driven situationalism, of the kind advocated by Rodrik and Wren-Lewis and others, actually works.
Problem 3: Model Combination
There's an alternative approach to situationalism: Why not just combine models into one big fat super-model? Some people at central banks have suggested something like this (see here, here, and here, for example).
That's fine in principle, if you have one big overarching theory, and different models are just different approximations of that theory. In that case, you're basically assigning probabilities to different assumptions about which approximation is best in this case.
In other words, many economic models only work if all other models totally fail to work. This is obviously a problem for the "model combination" approach.
So I think situationalism in econ faces some big hurdles. The question is how to tell ex ante which model to use. The danger is that by ignoring this question, economists can use situationalism as an excuse to "explain" everything but predict nothing at all.