My (all-soy) beef with econometrics (plus: a rambling argument for fieldwork in economics)

Only my economist friends will understand this angst, but this has really been bugging me for a while.

How can econometrics adjudicate between different data-generating processes?

We have lots of techniques for estimating coefficients - OLS, 2SLS, GMM, systems estimation, but that only picks out the model that appears to fit the data best from a restricted class of possible DGPs. And the class of DGPs - to noneconomists, this is the "mechanism" that "generates the data," i.e. a probabilistic model that we suppose (hope?) characterises the real world - is simply assumed a priori, often chosen for convenience, not for plausibility on economic grounds.

I find it troubling that empirical work hardly pays any attention to the theoretical implications of the functional forms it uses. If we estimate a linear wage equation, what does that implicitly say about firms' marginal revenue product curves (if we want to believe that labour markets are competitive)? If we presume semilogarithmic household demand functions, what are we saying about intra-household bargaining and the utility function of the household (or, by extension, its constituents)?

I know there are lots of nonparametric estimation techniques out there, which I guess I feel are more honest. Obviously, I'm aware that they're way more data-intensive than parametric techniques - the curse of dimensionality and all that.

I wonder, though - and Rob, Sean, Alex, Dean (sorry about the technical terms) and Simon, help me out here, since you're really the intended audience for this post - if there shouldn't be more to the scientific process (as it relates to economics, at least) than the estimation of conditional mean functions? OK, so we get the partial derivatives of some function which, as far as we can tell, represents reality. Those are our "effects," our betas. We've dealt with selection bias, attrition, and endogeneity.

Now what? Interesting though it is, can you think of anything else that we as economists should do to explore the validity of our ideas? If we were molecular biologists or solid-state physicists, we could go screw around in a lab and hope to notice related things that might give us clues about our theories. But how can economists do anything of that nature? We come to the data with an idea in our heads that we want to test. But our ability to let the real world speak to us in return - to poke around and see something that we weren't quite looking for - is severely limited. This is the anxiety that makes me so enthusiastic about fieldwork - actually going out there to root around in the dirt to see what's going on out there.