Mobile Monitoring of Elections

Researcher: Clark Gibson (UC San Diego)
Location: Kenya

Electoral fraud is a primary tool used for incumbents in developing democracies to undermine political accountability. By all accounts, Kenya’s last election was stolen, instigating ethnic clashes that resulted in 1,000 deaths and tens of thousands of displaced citizens. Electoral fraud was widely claimed, but citizens lacked the tools and organization to check it during voting and counting.

Standard election observation methods use international and domestic observers deployed to polling stations to look for and deter fraud. Despite millions of dollars spent on hundreds of such election observation missions over a generation, the technology of monitoring has evolved very little. We lack systematic and reliable data on whether they actually reduce electoral fraud.

This project aimed to redesign the way in which citizens can hold their politicians accountable through free and fair elections. Researchers refined a smartphone application to enable citizens to monitor the 2013 Kenyan national elections in real time using digital image collection and cloud-based computing. Citizens used the app to share provisional vote totals, provide pictures of the tally, and report theft of material or denial of monitor access. Data was posted on a public-facing dashboard during and after voting occurred, providing near real-time results for a national, random sample of polling stations.

Equipped with smartphones and the app, the project used a randomized controlled trial (RCT) research design to detect and suppress electoral fraud. Local researchers delivered letters to a random, nationwide sample of polling station managers. The letter announced the intention to monitor the election by taking a photo of the station’s final tally, which were publicly posted. In earlier studies in Afghanistan and Uganda, the letter significantly reduced fraud in several ways—from decreasing votes for incumbents to preventing ballot-box theftcompared to stations that did not receive a letter.

Results

Results show that the treatment letters did have significant effects on these measures. Stations treated with the letter were 25 percent less likely (from baseline) to have adjacent digits used as a measure of electoral fraud and were 16 percent more likely (from baseline) to have posted their tally.