# Drawing Blanks

Premature Optimization is a Prerequisite for Success

## Russian elections: 2011 vs 2012

I managed to merge the 2011 and 2012 election data and I can now look at all changes that happened at precinct level.

As we saw here https://bbzippo.wordpress.com/2012/03/27/russian-presidential-election-2012, turnout increased, and the leader support increased even more.

What’s the relationship between the change in turnout and the change in leader support? Following Steven Coleman, we can use a nice additive measure – entropy – to compare those.

Below, on the horizontal axis is the change in turnout entropy. On the vertical axis is the change in choice entropy normalized to the number of candidates. The straight line is a regression with zero intercept (which is Coleman’s theoretical prediction, if I’m interpreting his theory correctly). The other line is the actual relationship (mean choice entropy vs. turnout entropy in 50 bins):

Even though overall the change in choice follows the change in turnout as predicted, there is a very significant drop in the entropy of choice in the central region of the graph where the majority of precincts are located.

That indicates that in 2012 (presidential) elections the opposition was much weaker that in 2011 (parliamentary).

Below, on the horizontal axis is the change in turnout percent, and on the vertical – the change in the votes for the leader as a share of all registered voters. Note that this measure is different from the votes as a share of all cast ballots. I use the former because unlike the latter it behaves linearly when we add ballots for one candidate. The color indicates the level of support for the leader in 2011.

The most dense cluster here lies in the north-north-east sector of the graph. Those are precincts that showed low support for the leader in 2011, but in 2012 increased turnout and leader support. And the majority of points being located above the diagonal, means that the increased leader support cannot be fully explained by the increase in turnout – some people who had voted for the opposition in 2011 must have voted for the leader in 2012.

The opposite cluster shows that the situation with the precincts who strongly supported the leader in 2011 is fully symmetric: turnout dropped and the leader support dropped even more. This may be interpreted as a decrease in fraud.

And the right side of the graph being almost empty below the diagonal, means that the opposition did not attract many new voters.

Below are pictures broken down by the 3 categories of precincts: ethic outskirts, smaller precincts (presumably countryside) and larger precincts (presumably cities). They have no qualitative differences: