“Widespread but minor frauds occurred” in the August 8 election and appear to have benefited both Uhuru Kenyatta and Raila Odinga, according to an American statistics professor.
Professor Walter Mebane of the University of Michigan was called in by an unnamed NASA official on August 14 to look at the Kenyan election data. Mebane is a specialist in the forensic analysis of elections around the world.
Initially, he and his team said they had found patterns that showed widespread manipulation.
“It was unlike any data set I had ever seen,” he said. “Every single indicator came up signaling anomalies. It’s a huge red flag that something weird is going on,” he told the New York Times.
But Mebane has now moderated his position in a draft paper on Anomalies and Frauds (?) in the Kenya 2017 Presidential Election.
“Election forensics analysis suggests that frauds may have been widespread throughout Kenya in the August 8 presidential election but that the magnitude of frauds was small. Only a few per cent of votes, at most, are likely to have been produced by frauds,” he writes.
However, Mebane qualifies his analysis by saying “Kenya presents a very hard case for election forensics analysis. As will become clear, voting in Kenya is extremely polarised”. He says this can â€œdistort election forensics results”, especially when combined with strategic voting.
Mebane analysed the data from 40,818 polling stations which he scraped from the IEBC website and matched with the Kenya Gazette.
He found warning signs on the numbers for both Uhuru (Jubilee Party) and Raila (National Super Alliance).
“Every indicator of anomalies is triggered for the votes for Kenyatta or Odinga,” he says.
“Such excesses may be interpreted as symptoms of excesses occurring due to coordination problems among agents manipulating votes, for example through vote buying”, he says.
The most problematic counties were Kisumu, Migori, Tana River, Nyandarua and Lamu.
Mebane said Uhuru got 444,102 votes from polling stations with “spikes” but that “this model asserts that proportions of votes for Kenyatta in these polling stations is excessive, not that all the votes in the polling stations are fraudulent.”
His analysis indicated no particular problem with Form 34B.
“Spike frauds are not generally more frequent in polling stations included in problematic Forms 34B than they are in polling stations that do not have problematic Forms 34B,” he states.
He said that academic forensic analysis of elections does not yet have models where multiple candidates benefit from fraud so he analysed each county twice.
He found â€œ18 times when Kenyatta is specified to benefit from any frauds and 11 times when Odinga is specified to benefit from any frauds”.
“The total number of votes due to incremental frauds across the counties, for which estimates could be produced, is 36,907 for Kenyatta and 25,093 for Odinga.”
However, Mebane says the turnout figures look plausible across the country.
“Appearance of turnout not being manipulated persists when examined separately for each county,” he writes.