Network Analysis of Factors Contributing to Election Violence in Kenya: Comparing 2007-08 to 2013

Project Proposal: Peter Varnum and Mollie Zapata

This project will use networks analysis to assess election-related violence in Kenya in 2007-08, and then conduct the same analysis of 2013 when there was no election-related violence to determine if networks analysis can be used to predict violence. Additional factors in both cases will be studied separately: the networks of targeted violence in 2007, and how internal displacement in the interim period affected elections in 2013.

I. Overview: Kenya 2007-08

In December 2007, Kenya’s national elections led to two months of violent unrest across the country. In this politically motivated ethnic violence, police, youth militias, and community members killed at least 1,300 of their countrymen, which led to the displacement of an estimated 350,000 civilians. The international community was shocked. While ethnic violence was common and even expected in countries like Somalia, Rwanda, Sudan, and DR Congo, until 2007 Kenya had been seen as the beacon of stability in East Africa. 

Underlying this image of stability, however, were deep-seated ethnic, income inequality, land tenure, and other economic issues. Political parties in Kenya have always been unofficially formed along ethnic lines, with the Kikuyu-led Kenya African National Union (KANU) and Luo-led Orange Democratic Movement (ODM)  dominating. In response to what were deemed unfair elections at the end of 2007, Luo elders allegedly mobilized Luo youth militias to target Kikuyu living outside traditional Kikuyu areas. Kikuyus then retaliated, and the violence escalated across the country and continued through February 2008 when the leaders signed a power-sharing agreement.

Timeline of the Crisis:

• December 27 -- National elections begin.
• December 29 -- Opposition leader Raila Odinga of the ODM announces his victory.
• December 30 -- Incumbent President Mwai Kibaki of the Party of National Unity (PNU), a coalition of several parties formed in 2007 and dominated by the KANU, is hastily declared the winner by the Electoral Commission, despite voting irregularities and pre-election polls indicating that Odinga would prevail.
• December 30 -- Kenyan government bans public gatherings and live broadcast media (both illegal actions according to international law and Kenya’s Constitution).
• January 1  -- According to Human Rights Watch,“a mob set fire to a church where terrified Kikuyu residents were seeking refuge, soaking mattresses the victims had brought with them with petrol and stacking them against the building. At least 30 people were burned alive.” Note: This is just one example. Incidences of violence perpetrated by both sides escalated through January and continued through February. A 2010 report by the International Criminal Court noted that “crimes, such as murder, rape, mutilations, looting, destruction of property, arson and eviction seem to have occurred on the territory of the Republic of Kenya at least in the course of the events between 28/29 December 2007 and 28 February 2008.”
• February 5 -- International Criminal Court Prosecutor Luis Moreno Ocompo announces a preliminary examination of post-election violence.
• February 4 -- Government lifts ban on live broadcasts.
• February 8 -- Government lifts ban on public gatherings.
• February 28 -- President Kibaki and Raila Odinga sign a power-sharing agreement to end Kenya’s crisis.

Legacy of 2007 Elections

In the aftermath of the 2007 election, UN secretary-general Kofi Annan negotiated a power-sharing agreement between President Mwai Kibaki and Raila Odinga. The National Accord, as it was known, appointed Odinga the Prime Minister, and called for a referendum for a new Constitution of Kenya. The Constitution was passed in 2010 and divided Kenya into eight provinces -- the Rift Valley, Eastern, Northeastern, Coast, Central, Nyanza, and Western -- and 47 counties, whose size and boundaries are based on the 47 legally recognized Districts, or Sub-counties, of Kenya.

The continuation of violence for several months in early 2008 forced many people to flee their hometowns; the World Health Organization (WHO) estimates that 350,000 people were displaced. Though some internally displaced persons (IDPs) moved to IDP camps, many people -- particularly Luos -- migrated back to their ancestral homelands. As a result, the geographic division among tribes became more pronounced, with the Luos occupying the Rift Valley, Nyanza, and Western provinces and the Kikuyus occupying the others, in general. Despite this geopolitical homogeneity, little violence erupted after the 2013 elections.

Kenyan Elections 2013

The 2013 election, like the 2007 election, was controversial. Uhuru Kenyatta, the PNU incumbent president, who is Kikuyu, received 50.07% of the vote, according to the Independent Electoral and Boundaries Commission (IEBC); Raila Odinga, the ODM Prime Minister, who is Luo, received 43.28%. The two avoided a runoff, per the Kenyan Constitution (the Constitution dictates that one person must receive the majority of the vote to be elected President), by a difference of just 8,100 votes. Odinga quickly filed an appeal with the Supreme Court of Kenya, attesting that the IEBC counted votes fraudulently. The Court unanimously dismissed his case less than a month after the election.

Though the results were similar -- a close race between two rival ethnic parties, with allegations of voting irregularities from both sides -- very little violence occurred in 2013, unlike in 2007-08.

II. Project Plan

This project will determine what networks can tell us about election-related violence and whether networks analysis can be used as a tool to predict this type of violence. While extensive studies have been done on the causes and drivers of Kenya’s 2007 election violence, no one has yet used network analysis tools to assess how political and geographic organization contributed to unrest, nor compared that data to 2013, when the election results were similar but the aftermaths were completely different.

There will be two comparative analyses:

1. We plan to assess if and how the formation of political party networks contributed to election-related violence in 2007-08, and then, similarly, assess conditions in the 2013 elections to see if and how the networks changed.

2. We will also look at the networks of Kenyan communities based on election results in 2007-08 and 2013, with communities and the parties they support as the nodes and ethnicity as their primary attribute. We will see if patterns emerge and whether SNA could be used to indicate causality and possibly predict election-related violence.

Additionally, we will assess related factors for both election processes to gather a more complete picture of the situation.

2007-08: We will analyze who directed violence against whom over time to determine if the violence escalated due to a cycle of reprisals.

2013: We will assess if and how the internal displacement that was caused by the 2007-08 violence affected the way communities voted in 2013.

III. Objectives, Research Questions, and SNA Methods

Comparing 2007-08 to 2013:

Objective 1: To conduct a network analysis of inter-political party alliances in the 2007 elections to set a baseline to compare to a network analysis of inter-party alliances in 2013.

• Research Question: How did the network of party alliances change from 2007 to 2013?
• SNA Methods: We will analyze a one-mode dataset of political parties. This will be a dichotomized dataset with a “1” indicating that the parties are allied, and a “0” indicating that they were not.
• Attributes: 
• Number of members of each political party
• Ethnicity
• Geography of party base

Objective 2: To conduct a network analysis of how communities connected with (i.e. voted for) political parties.

• Research Questions: 
• Did the way county voted affect level of violence? 
• Were more homogenous counties less likely to experience violence? Alternately, were more diverse counties more likely to experience violence? Homogeneity will be assessed by voting records -- we can infer that those voting for a party are of the same ethnicity as that party, because political parties are overwhelmingly aligned with ethnicity.
• Did this change from 2007-2013? If so, how? 
• SNA Methods: We will start by analyzing bimodal data (counties and political parties), assessing the strength of ties based on percentage of votes (stronger ties = higher percentage of votes). This will be visualized by the width of ties between county nodes. We can infer a person’s ethnicity from the way they voted -- for example: a county voting 100% for the PNU can be assumed (for the purposes of our analysis) to be 100% Kikuyu. This will be a valued dataset, with the values corresponding with the percentage of votes going to each party. For example, if a county voted 20% for the PNU and 80% for the ODM, those values would be reflected in the weight of the ties.
• Attributes:  
• Incidences of violence reported in each community. (Node size will be larger for more incidences of violence.) 
• If possible, we will also consider attributes relating to other stresses present in the communities: Police force-perpetrated violence, land tenure issues, previous election-related violence, income inequalities, perception of misrepresentation in government.

Year-Specific Objectives:

Objective 3 (2007-08)*: To use networks analysis to assess patterns of violence in communities over time.

• Research Question: Who committed violence against whom? In what order? Can the networks tell us anything about why the violence escalated so quickly and continued for so long? 
• SNA Methods: We will select the counties that experienced the most violence in 2007-08 and use data of incidences of violence from Ushahidi and news reports to determine, over time, who committed violence against whom. This will be a bimodal dataset, with the nodes being: violence Kikuyu against Luo; violence Luo against Kikuyu; reciprocal violence; and no violence.  We will look at the directionality of ties. By doing a series of network analyses over time, we will assess whether the reason the violence escalated so quickly and completely was due to a cycle of reprisals that the government was unable to stop. 
• Attribute: 
• County

*Due to an inability to accurately determine who instigated violence against whom, it is unlikely that we (or anyone) will be able to conduct such an analysis for this project.

Objective 4 (2013): To use networks analysis to determine if and how internal displacement changed election results in districts from 2007-2013.

• Research Question: Could heterogeneity of communities cause election-related violence? If communities were not homogeneous in 2007-08 and they experienced violence -- and they were homogeneous in 2013 and did not (due to displacement/migration) -- then that could be an indicator that violence was caused by tribal heterogeneity within communities.
• SNA Methods: This will involve two datasets, combined. Each dataset will have ethnicity and counties as their axes. The first will answer the question, “Was there violence?” and yield either a 1, or a 0 (1 indicating there was; 0 indicating there was not). The second will answer the question, “Was there an influx of IDPs?” and yield a 10 or a 0 (10 indicating there were IDPs; 0 indicating there were not). We will then add the datasets together, resulting in a 0, 1, 10, or 11 meaning:
• 0: There was neither violence nor IDPs
• 1: There was violence, but no IDPs
• 10: There was no violence, but there were IDPs
• 11: There was both violence and IDPs.
• Attributes:
• Who perpetrated the violence -- civilians or police?
• How the county voted in the 2007 election.