As social scientists become more concerned with causal inference, a good thesis these days needs some sort of quasi-experimental strategy. For example, an economist wants to know the effect of military service on income but faces the problem that people who enlist are different from those who don't. So he uses the random Vietnam War lottery as a quasi-experiment with a valid control (who won the lottery and did not serve) and treatment (who lost and did serve). (More examples of these natural experiments).
As a PhD student sitting in my cubicle, I'm lost on how to identify these opportunities. Reading published work does not really help since these experimental opportunities are quite idiosyncratic. I'm not averse to going out there to find my own opportunity, but unsure about how to do this effectively. Do I read history book? Or talk to policy makers?
Since experimental opportunities are not available to all (or most) topics, I'm already paralyzed at the stage of choosing a topic (and thus can't start reading history book / talking to policy makers).
The standard advice I've got is not to pick question based on method. However, I find this quite a double standard given the concurrent push for quasi-experimental design.