On the cleverness scale, it’s hard to beat Ray Fisman and Ted Miguel’s study of national corruption norms by analyzing the parking behavior of United Nations officials in Manhattan (Times coverage here). The only challenge is, what good does such a corruption index do for combating the problem?
On the clever+useful scale, however, my nomination goes to Columbia political science grad students Bernd Beber and Alex Scacco. In a new working paper, they use known psychological biases in number patterns to statistically detect election fraud in Nigeria.
When asked to invent lists of random numbers, lab experiments show that people tend to disproportionately select small numbers. Also, people tend to underestimate the likelihood of digit repetition–a 4 following a 4 in the list, or even a 24 following a 44.
So, when we look at election tally sheets–of voters registered or votes cast for a candidate, for instance–we should tend to see a disproportionate number of small last digits, and fewer repeated last digits than normal.
The base case: Swedes.
Here is the distribution of last digits in vote returns for the Social Democratic Party (SAP) and the Moderate Party (MSP) as well as the tallies of registered voters:
The digits that come up are evenly and uniformly distributed, as we’d expect of random last digits. No one digit is more likely to appear than the other.
Now we turn to Nigeria, and analysis of handwritten vote return sheets from one of its states, where the 2003 election was widely believed to be manipulated in favor of the ruling PDP. Here is the last digit distribution of PDP returns, total vote counts, and number of registered voters:
The black bars represent the significant, nonrandom deviations. You can see the paper for a discussion of why this is likely fraud rather than error or laziness.
Thus we have a useful tool for election monitors on the ground. It doesn’t guard against all sorts of fraud–ballot box stuffing would not be apparent in this case, for instance–but it can guard against fraudulent filling out of tally sheets.
The mischievous question of interest to Floridians: do hanging chads and electronic voting machines have psychological biases in number invention?
For a more personal take on Nigerian corruption, it’s hard to beat Teju Cole’s chronicle of his trvel home to Nigeria after a decade abroad.
6 Responses
It is an interesting proposition. We can use this inside an electronic voting system to check for reasonability of the results.
oh expletive! This is new?
19th century and Chicago elections likely have some similar patterns.
your reference to the fisman/miguel paper reminds me of a recent incident involving my local MP here in Dar es Salaam, Rita Mlaki. She committed a parking offence no less! Darmites are as concerned by bad parking as New Yorkers. Swahili readers click through here
http://www.globalpublisherstz.com/2008/05/14/ritta_mlaki_afanyiwa_kitu_mbaya.html
nice comment from anon from 5:00 pm
which is to say, as of today, the finding is no longer useful for monitoring future elections….
with a $100 laptop and a random number generator from a free Ubuntu spreadsheet anybody can bypass this statistical control.
One step forward, one step back, thus is the nature of corruption in Nigeria
What a clever idea! It would be a lot more convincing if the excess number wasn’t zero. If I’m manipulating election results, I’m probably going to avoid numbers that end in zero, because it just looks suspiciously round. I suspect if they ignore the zeroes and remean the results for 1 through 9, the other dark bars will lose significance.