This is a guest-post from Fredrik M Sjoberg, currently a Postdoctoral Scholar at NYU. The paper on which this post is based is here.
In recent years, non-democratic rulers have surprisingly begun to embrace fraud-reducing technologies, like web cameras or transparent ballot boxes. Most recently Russia had webcams installed across the vast territory (see this earlier post at The Monkey Cage). Ukraine and Albania are also supposedly considering it. Little is known about the effects it has on reducing fraud. Yes, earlier studies have shown that turnout is reduced in the presence of webcams, but that the ruling party vote share is not necessarily affected (e.g., see here). However, relying solely on proxy measures like turnout and vote shares is problematic since both clearly depend on many different things, some of which might not be related to fraud at all. In any case we would need to account for historical turnout and vote share.
With the help of new fraud identification techniques, I argue that the installation of web cameras in polling stations changes how fraud is conducted. First, we need to recognize that election fraud comes in different forms. Ballot box stuffing, multiple voting, and fabrication of vote are not necessarily affected equally by a fraud-reducing technology like webcams.
Micro-level evidence from Azerbaijan, the first country to install web cameras at the precinct level, suggests that ballot box stuffing is reduced, as measures by turnout, but that authorities instead focus on fabricating the vote totals. Miscounting of the ballots can be detected by digit-based tests where the distribution of the last digit is compared to a uniform distribution (the green line in the graph). As Beber & Scacco show in a recent article, absent outright fabrication of vote totals we expect a uniform distribution on the last digit (Political Analysis, 2012, What the Numbers Say: A Digit-Based Test for Election Fraud). In Azerbaijan the distribution on the last digit in webcam and non-webcam precincts is given by the graph:
* Note: the data is here matched (CEM) based on size, district, historical turnout and competitiveness to account for biases in how the cameras were allocated.
In precincts with no webcam there is no significant deviation, as indicated by a p value of .6088 (chi square goodness of fit test). In the monitored group, on the other hand, there is a deviation from the uniform, significant at the 5-percent level. This suggests that the officially reported vote totals were fabricated in precincts with a webcam.
By replacing one form of fraud with another incumbents are able to prevent vote share losses, while contributing a veneer of legitimacy by self-initiating anti-fraud measures. It therefore seems like a win-win for the autocrat (and for the company that managed to sell 100,000 webcams to Putin)
