Why aren't all econometricians Bayesians?

Andrew Gelman offers this wonderful graph:

If you have a simple model and lots of data, then it doesn't matter if you use classical or Bayesian methods. Otherwise, classical methods are more likely to generate more in the way of thorny, pointless digressions ("Hey, wait a minute! Are the data trend- or difference-stationary?") than actual answers.

Comments

[Tom Grey]

Wow, great idea in a graph.
Before looking at the Gelman data/ discussion, I should note my own a priori feeling that no other model building process will be better Bayesian for the intractable problems.

There's also an issue of data abundance, for instance on the risk of house foreclosures. If the data is by years, the last 30 years isn't so much; by months the same data becomes more, and by day it expands even more. But the usefulness of day data on house foreclosures is not 365 times greater than annual average year data, maybe not even twice as informative.

Modern data collection techniques, as well as rocket science risk management, might well disguise an intractable problem.