The old myth is that the world sits on the back of a turtle (wouldn't it roll off, with a sphere on top of a round thing?) and the turtle is standing on a turtle. What is that turtle standing on? Well, it is turtles all the way down.
Same is true for principals and agents (alas, my latest project has finally made me drink the UCSD koolaid). Any hierarchy (government or business or union or whatever) is a series of bosses and underlings or principals and agents. The big problem is how do the bosses get their underlings to do what they want, no more and no less. The agents will always know more about what they are doing than the principals because the principals have other things to do. This information asymmetry is at the heart of the P-A problem.
Most of the work that migrated to Poli Sci, as far as I can tell, focuses on oversight--being able to monitor the agents and detect violations. The classic distinction in oversight is passive vs active--police patrols vs. fire alarms. Active means going out and inspecting, looking for violations. This serves to deter as finding a violation is more likely, but is also expensive as the inspectors have to spend much/all of their time doing oversight. The alternative is to wait to hear about a complaint--when someone pulls the fire alarm. Usually, the form this takes is someone tells the media. So, this is cheaper because inspectors don't have to be paid and so forth, but it is reactive and far more public.
The Dave and Steve book look at other aspects of the P-A problem besides oversight: selection processes that shape whether the most appropriate agents are chosen (those who think more or less like the Principal?), discretion--how much decision-making is delegated to the agent, and incentives.
Why talk about this? Because in doing this project, I keep being reminded that control of one's agents is never perfect, and so we find out that soldiers pose with dead Taliban, Secret Service agents hire prostitutes, that members of a protest movement may throw bricks onto the tracks of the metro and engage in other acts of vandalism. We have higher expectations for the US Army or the Secret Service than the student groups protesting tuition in Quebec precisely because the former organizations spend a great deal of time, thought and energy trying to manage their P-A problems through training, discipline, doctrine and all the rest.
But the systems do not always work. Size of organization obviously matters--the more agents there are, the harder it is to monitor and the harder it is to provide incentives that matter. Given how big the US Army is and how stressed its agents have been during a decade of war, it is perhaps surprising how few violations have occurred--that we know, of course. Perfection is not possible, but active oversight is the key for Army since learning about violations from the media is very costly as Abu Ghraib and other such disasters have revealed. Failures occur when the intermediate principals do not do their job. Andrew Exum, who served in Iraq and Afghanistan, always asks: where were the non-commissioned officers and the lieutenants and the captains? Obviously, it is also turtles all the way up--did the more highly ranked officers fail to instruct and discipline the lower ranking ones? I don't know in this case. In the case of Abu Ghraib, it was turtles (crappy principals) all the way up: Sanchez to Rumsfeld.
The good news: some training centers are recommending the right readings, so perhaps we can at least understand this stuff better.