Kenyan pollsters eat humble pie

No one in the mainstream Kenyan media, at least not yet, wants to talk about the failure of opinion polls to predict the outcome of last Monday’s election (For some thoughts on the election check out my post on the monkey cage blog here).

A week to the election the three main polls showed the race to be neck and neck between Messrs Raila Odinga and Uhuru Kenyatta, with a slight advantage to Mr. Odinga, on average.

But in the end it was not even close. Mr. Kenyatta handily beat Mr. Odinga by almost 7 percentage points (50.07% – 43.28%).

So what went wrong?

In my view, the pollsters missed the mark both by not taking turnout into account (despite my unwarranted advice!) and perhaps poorly weighting the results from the different regions of the country (Kenyans largely vote along ethnic lines for various instrumental reasons; ethnic groups are geographically concentrated, with variance in size and population density).

The polling firms ought to have done a better job of basing their results on likely voters as opposed to self-declared registered voters. Kenyatta’s strongholds registered voters at higher rates, and based on past elsctions, were likely to register higher turnout rates than  Odinga’s strongholds- and they did (88.6% to 84%).

My own pre-election predictions a month before the election highlighted this fact. In my estimation the polls consistently overestimated Mr. Odinga’s support.

I eagerly await the polling firms’ own rationalization of what happened. Hopefully the misses this time won’t permanently damage the public’s perception of opinion polls.

Despite the difficulties in forecasting political outcomes, it is better to have polls than fly blind into an election.