Tuesday, April 7, 2015

Parochial Interests

There's a very nicely prepared dataset associated with the book The Logic of Political Survival (2003) written by Bruce B. Mesquita et al. along with code that can reproduce its statistical results. BBM is famous for The Predictioneer's Game (2009), where he detailed the use of game theory for negotations / war. It turns out he had another theory on governments, democracies, dictatorships in the earlier book. In this earlier work the claim is that it is not enough to use labels such as democracy, autocracy to describe / predict which countries to develop and others not. His main thesis is that the two most important variables in a country are W and S, the winning-coalition size and the selectorate size.

Original Dataset

My CSV Version

Winning coalition W  is # of ppl whose support a leader needs to rule, this is the group of people he needs to keep happy to stay in power, selectorate S is the # of people from among which members of W are recruited. With small W, leader only needs to keep a small group happy paying them off is easier, with large W, bribe becomes harder so leader must emphasize public services, striving to be elected on the basis of policy. S in a democracy is the entire electorate, for autocracies a much smaller group would do i.e. the military, or some sort of "guard", or a single party out of which even a smaller W would select the leader. 

BBM hypothizes there is a dependence between kleptocracy and W, S; he finds government theft is maximal when W is small and S is large, signifying a rigged electoral system where the electorate is large but their votes do not count, an inner circle decides everything. With bigger W, kleptocracy decreases.

A counterargument could be there is a hidden effect of population size on theft, more people, more revenue hence more chances to steal, or even foreign aid can tip the balance, but BBM "accounts for those variables (a statistics term)" in his calculations ,taking their effect out, keeping the focus on W and S. Kleptocracy value itself is the absolute value difference between tax revenue and expenditures in proportion to GDP. Absolute value is used because both deficits and surpluses can be a sign of theft; leader can steal surplus revenue after its reported or, overspend (on cronies, on their insane pet projects) which would result in deficit.

Another hypothesis is there is positive relationship between W and income, countries with larger winning-coalitions are richer, and this is also confirmed. With everything else the same a country going from W=0 to W=1 would add 3.0% growth to per capita income. This could seem small, but if you take per capita income $10,000 in 10 years,

R=0.03; print 10*1000 * ((1+R)**10.)

it becomes a whopping $13,439. So increasing the winning coalition to its maximum from the worst adds to per capita more than $3,000.


I ran the regressions myself, both as ordinary linear regression and as a multilevel model using region/year as the group. I was able to reproduce results close to what BBM reports, I say close because I left out some variables (the stuff that BBM ended up finding unimportant). In kleptocracy model for example R^2 is 20% (BBM gets over 40%) which is very good. All variables are significant.

Here is how to reproduce W from scratch.

There is some mad skillz displayed in accounting for "traditional democratic" effects that could be seperate from W and S. For that, BBM take W,S into regression against Polity's democracy, takes the residuals, feeds them into the main regression coupled with their variable in question. This way traditional democratic effects, or what Polity thinks as democracy is controlled for, only coalition and selectorate size effect remain. F..in A.

I used a combination of Python and R. The book uses Stata (a closed source software -nasty- but I was glad to have some software accompanying the book). 


Calculating actual W and S is the tricky business. Ideally you'd want to know the exact size, let's say country X has a military dictatorship, the military size is 100K, out of which a junta of size 10 rules the country, then S=100K and W=10. But this kind of data is hard to collect, the authors decided to use another dataset's base variables to derive W and S. This formula is educational on its own. The dataset is Polity IV, famous for capturing some base variables on countries' democratic development which also has its own democracy variable that BBM et. al did not use directly, the variables used for W derivation are RegimeType, xrcomp, xropen, parcomp.

Among these  parcomp was interesting for me, getting a 5 on that adds a 1 to your score, and many countries including Turkland has a 4, and you look up the description for that in Polity manual which says:
(4) Transitional: Any transitional arrangement from [1,2,3] patterns to fully [5] patterns [..]. Transitional arrangements are accommodative of competing, parochial interests but have not fully linked parochial with broader, general interests.
This is funny bcz the presence of "parochial interests" is a good word to describe TR at the moment.

The best score for W is 1.0. Some surprises, Singapore is rightly known for its lack of democracy and it gets a 0.4 from Polity, but its W is 0.75, a score shared by Venezuella and TR. US before Civil War scores 0.95 on democracy but 0.75 on W. In all these cases W does a better job in capturing the essence of a regime.

The research in this book appears to have evolved into The Dictator's Handbook, again by BBM.

Q&A - 21/5

Question How do you empirically prove interest rates do not cause increase or decrease in GDP growth? There is a test for that Data ,...