With the Kenyan elections now underway, social sentiment expressed by Kenyan netizens has been analysed to determine the next president of the country.
In the week leading up to the elections (1 – 6 August), 443 000 mentions from 173 000 Kenyans discussing the election were analysed. According to the data incumbent Kenyatta is under threat to lose the presidency.
“The traditional polls have consistently put Kenyatta ahead of Odinga, but according to the opinions expressed by the public online, the majority want an Odinga victory. This is strikingly similar to what we saw with Brexit and the US Elections,” said DataEQ CEO, JP Kloppers.
Social media opinions have traditionally been very difficult to analyse in aggregate due to the scale and complexity of the conversations. The scale challenge can be solved using artificial intelligence (AI), but this approach leads to inaccurate data, as machines struggle to understand the nuances of human conversation. DataEQ solves the inaccuracy challenge by complementing the AI with trained crowds – large teams of local language speakers who review and verify the data. In the case of this election, our crowd was made up of Swahili speakers who understood the local dialects and social context.
The data confirms that the 2017 Kenyan presidential race is a two-horse contest between incumbent president, Uhuru Kenyatta, and long-time political rival, Raila Odinga. These candidates have 53% and 43% of the total election conversation respectively.
Political upstart Boniface Mwangi has received strong support over the last week – but he still occupies only 4% of the total election conversation. Mwangi is a rank outsider for the presidential seat, but his high level of support could see him gaining a seat in the Starehe Constituency.
Online negativity hurts Kenyatta
While Kenyatta and Odinga show similar levels of support online, Kenyatta has almost double the amount of negativity compared to Odinga. Central to the concerns of Kenyans are issues related to tribalism. Kenyatta has attracted much scorn for his comments that he did not need the Kamba vote to win, leading to Kamba voters feeling that Kenyatta and the Jubilee party have rejected them. This was further aggravated by the fact that he continued to campaign in the region.
Another theme of critique directed at Kenyatta has been the ongoing political intimidation, torture and murder – for which many hold Kenyatta directly responsible. The murder of Independent Electoral official Chris Msando on 31 July has only given greater impetus to this, with critics using #YouCantKillUsAll to attack the president.
Much like the US elections, fake news has been rife – leading to a lot of wariness and mistrust expressed in the public’s opinions. There have also been many predictions of violence during and post the elections. Those attacking the NASA party accuse it of using propaganda in the campaign, while anti-Jubilee members discussed “NASAgate” – claiming that Kenyatta ordered a raid on a NASA tallying center.
Whilst Odinga enjoys significantly less negative sentiment, there is a concern that he is too old to lead the country. One of the central tenets of his campaign is that he will lead Kenyans to the promised land of Canaan – a phrase that has been picked up and echoed by his supporters online. In turn, this has been met with some derision, with detractors saying he does not have the track record to justify his claims.
Odinga gains momentum
Odinga appears to have mobilized online support in the last days leading up to the election. At present his conversation volume and share of online unique authors exceeds that of Kenyatta. This, together with strong positive sentiment towards the candidate, suggests a surprise win.
Both candidates have shown strong gains in net sentiment (support minus opposition) over the last week.
Opinion mining should inform polling and political strategy
Following the global trend, Kenyan social media is increasingly representative of the electorate at large, and no longer the domain of the angry and the extreme. As such it can be used as an input into polling research and a reliable predictor of political risk.