Six Mistakes People Make Reading Kenyan Election Numbers

Election numbers are not hard. Reading them carelessly is easy.

The same handful of errors accounts for most of the confidently wrong conclusions that circulate in any Kenyan election week, and each one has a fix that takes less than a minute.

Dividing by the wrong population

Turnout is votes cast over registered voters. Not over adults, not over the census population, not over the number of people somebody thinks live there.

Suppose a ward has 10,000 registered voters and 6,000 vote. Turnout is 60 per cent. Divide the same 6,000 by an estimated adult population and you get a different, smaller number that answers a different question. Both are useful. Only one is turnout. Related: how the fraction is built.

Comparing turnout across two different offices

Kenyans mark six ballots on one day and each contest is counted separately.

A presidential turnout figure and an MCA turnout figure for the same ward are two different measurements. Comparing them across counties or across years without checking the office is the most common invisible error in Kenyan election commentary, because both numbers look identical in a headline. Related: what a long-run turnout comparison looks like when it is done properly.

Treating a raw vote count as a trend

A candidate whose votes rose can have lost ground, and a candidate whose votes fell can have gained it.

Suppose a ward casts 6,000 votes at one election and 7,200 at the next, and a candidate goes from 3,000 votes to 3,300. That is 300 more voters and roughly four percentage points less of the ward. The honest sentence has both halves in it. Related: how to compare a ward with itself.

Reading a partial tally as a result

Stations do not report in a representative order, so an incomplete tally is an exact count of a non-random group.

Numbers moving through the night is what an incomplete count does, and it is not by itself evidence of anything at all. The things that would be evidence are specific and checkable: votes exceeding a station's register, an unsigned form, a published figure that does not match the signed copy an agent holds. Related: why the tally moves.

Ignoring the line for rejected ballots

A rejected ballot was cast and then set aside at the count. A spoilt ballot was replaced before it ever reached the box. Only the first appears as a counted line on the results form.

That line matters because it is often larger than the winning margin. When it is, the result stands but it cannot be independently checked without knowing how the pile was handled, and by then it is sealed. Related: the difference between the two, and what makes a margin close.

The sixth mistake, and the habit that fixes all six

The sixth is reading a by-election as a national poll.

A by-election is one seat, contested within 90 days of a vacancy, with a national party machine concentrated on an area that will never receive that much attention again. It is excellent evidence about organisation and no evidence about national vote share. Related: what it does and does not show.

Every one of the six is the same error wearing different clothes. A number was separated from the thing it measures, and then compared with a number measuring something else.

The habit that fixes all of them is to write four labels beside every figure before you use it: which office, which level, which population it is divided by, and which document it came from. If you cannot fill in all four, you do not have a result. You have a claim, and you should say so when you repeat it. Related: where the documents actually live.

Start practising on old numbers now rather than on live ones. Take any result you can find for your own ward, label it with those four fields, and notice how often one of them turns out to be missing from how the figure was originally presented to you. That is the whole discipline, and by 10 August 2027 it will be automatic.