UPDATE 4PM: RESPONSE TO COMMENTS (SEE END OF POST) Step 1: You're right, almost all of the biggest growth success stories are autocracies! Step 2: Wait a second, all of the worst growth failures are also autocracies! Step 3: Solving mystery: autocracies have much higher variance of growth rates, so they have both best and worst growth rates
Wonky Moral of the story: focusing on success only in Step 1 created a selection bias that led to the erroneous conclusion that autocracy was good for growth.
Plain English Moral of the Story: autocracy is extremely risky: it could result in high growth, but it could just as well result in a growth collapse -- for every Lee Kuan Yew, there is a Jean-Bédel Bokassa.
Extra credit question: why would arguing that the autocrats under Step 1 are benevolent while Step 2 autocrats are malevolent be logically fallacious?
RESPONSE TO COMMENTS 4PM To answer some questions, the growth rate is the geometric average 1960-2008 of per capita growth per annum. The source is WDI.
The source for the democracy data is Polity IV, which has some problems, but is enough for the kind of illustration here.
The point about causality is well taken, I am just making a point about how what for these data is actually a POSITIVE and SIGNIFICANT correlation between democracy and growth is turned into an apparent NEGATIVE association in Step 1, which is where the "benevolent autocrat" discussion usually stops.
(FOR WONKS ONLY) When I write this up more fully in an eventual paper, I will explain also some exploration of different functional forms for transforming the original POLITY index from -10 to 10, which is an arbitrary scale (autocracy being the negative direction). To illustrate the strongest possible POSITIVE correlation, I chose from 3 alternatives the one with the strongest statistical significance , which was the following function: POLITY/(11-POLITY). I would normally NOT like this kind of "data mining" among several different functions, but again the point of the exercise is to show the fallacy by which a STRONG POSITIVE association appears to be a STRONG NEGATIVE ASSOCIATION.