KENYA ELECTIONS: POLLING MATTERS

C O N F I D E N T I A L NAIROBI 004457 SIPDIS SIPDIS AF FOR A/S FRAZER, LONDON AND PARIS FOR AFRICA WATCHERS E.O. 12958: DECL: 11/14/2017 TAGS: KDEM PGOV PREL KE SUBJECT: KENYA ELECTIONS: POLLING MATTERS REF: NAIROBI 3969 AND PREVIOUS Classified By: Political Counselor Larry Andre for reasons 1.4 (b,d) ¶1. (C) Our election analysis (reftels) indicates a close race for the presidency. This analysis is based on voter registration data and traditional turnout rates as well as polling on presidential candidate preferences. We were concerned that widely published public opinion polls, which showed ODM's Raila Odinga well ahead of President Kibaki, did not accurately reflect the true status of the contest. Given the rising political temperature, partially due to the use of blatant ethnic appeals by both sides, we were concerned about the reaction of ODM supporters should their candidate lose in a close outcome when they were led by public opinion polls to expect a landslide victory. ¶2. (C) The Mission quietly reached out to polling firms and their clients to suggest that poll sampling distribution should be based on the regional distribution of registered voters, not on raw population (there are wide variances in voter registration rates around the country, with the high population density areas, such as Kibaki's native Central Province, enjoying higher rates). Polling on this basis would be a more accurate predictor of the outcome. When one polling firm (Consumer Insights) then started limiting its respondents on political preference polls to those who could produce an actual voter identification card, the change was dramatic: Odinga: 41 percent, Kibaki: 40.6 percent. Once the Steadman Group re-distributed their sampling according to the August voter registration figures, the Kibaki/Odinga gap shrank from 11 points to 4 points: from 39 percent Kibaki, 50 percent Odinga to 41 percent Kibaki, 45 percent Odinga. ¶3. (SBU) The Electoral Commission of Kenya has now published the final voter registration figures. The major polling firms have all committed to adjust their sampling according to these figures and limiting responses to those who at least claim to be registered voters. The results of polls based on sampling that reflects the final voter registration figures will be available in about two weeks. RANNEBERGER