Had I known before late afternoon on election day, 27 December, that “the whole reason” for the USAID-IRI exit poll was “early intelligence for the Ambassador” rather than as a tool to deter and detect potential fraud (as our consultant from UCSD and I were explicitly told some weeks before) I might have made more note of the fact that the exit poll by design generally excluded non-living voters as it was based on live interviews of people who had personally come to the polling place and cast ballots.
Admittedly if the purpose of the exit poll was to predict who the ECK would determine to be the winner, as opposed to simply how living Kenyans voted, this was a serious limitation.
One specific idiosyncrasy that afternoon that was immediately salient was the issue of release to the Ambassador of preliminary numbers reported and collated as of two hours before the polls were to generally close. I have no statistical reference or otherwise scientific and peer reviewed material to cite for this observation, but it would have been my seat-of-the-pants judgment as the “person on the ground” with some practical experience in campaigns and elections and even with “machine politics” that deceased voters have a pronounced tendency to vote last in sequence among the various voting blocs.
For those wishing to observe the voting process rather than influence it, there are two related reasons why you will not want exit poll numbers to “get out” to actors in the process before the polls close. One is that potential voters supporting the candidate who is “behind” are perceived to be subject to being discouraged. Even if this is not a big enough factor to “matter” in the primary race at issue, it has been seen to impact the outcome of other races on the same ballot. Another is that some voters who might not otherwise elect to turn out may be spurred to action by the perception that their candidate is trailing. The dead voters are one identifiable bloc that may be particularly susceptible to an appeal of this type.
At the time, I didn’t have any numbers or details to go on that would support a specific adjustment for dead voters in the exit poll. Some months later the Kreigler Commission estimated a figure of 1.2M decedents who were registered to vote on election day; in January 2010 as discussed in a previous post, Undersecretary of State Maria Otero was headlined in the Kenyan press on a visit saying “we are aware that more than two million dead people voted in 2007”.
Had these types of numbers been available to me on election day I would have understood the stakes that much better. Even though this type of voting in the United States peaked before I was born, we can easily see empirical evidence in history of a pronounced tendency of the dead voter bloc to support the party which controls the electoral mechanism, in this case in Nairobi, the ECK. With the kinds of numbers on the voting role, if ODM/Odinga had roughly six percent more live votes as reflected in the exit poll, the percentage of the deceased who needed to be inspired to cast ballots would be much lower than the overall turnout figures.
In corresponding with a diplomat from an allied country (one with which the U.S. has a mutual defense treaty) before the ECK decision I was told that his expectation was that Odinga would win by roughly five percent. I replied that this was interesting as I had decided that roughly five percent was probably the minimum threshold for a margin for Odinga that would result in him being accepted as the winner by the ECK. In hindsight I was probably “drinking the KoolAid” of democratization a bit myself.