By JASON LAKIN

More by this Author

I want to make some observations about a narrow but important issue that arose during the recently annulled Kenyan election and that is still relevant as we move to the fresh election in October.

The matter I have in mind is how election results are reported to the public. I am not going to say anything about the counting or transmission processes, nor will I make any claims about the results of the August election.

What I will claim is that the manner in which results were shared with the public on the Independent Electoral and Boundaries Commission (IEBC) website and on television over several days was deficient from the perspective of transparency and legitimacy.

After the election, some pundits claimed that the fact that there was a consistent gap between the two candidates in the results as reported suggested foul play.

Others responded that this is exactly what one would expect. This was a typical fight about something seemingly objective — what statistical probability would tell us about what to expect from election results, regardless of who was winning — where people’s views have been corrupted by bias (whether explicit or implicit).

The basic facts are these. The condition under which we would expect that the results from, say, 100 polling stations out of the 40,000-plus used in the election, would reflect the true result of the election is the following: these polling stations must be a random sample drawn from the total.

Randomness ensures that there is no relationship between the factors that led to selecting the 100 polling stations and any factors that affect a person’s vote.

If the reason why a polling station is in the sample is related to vote choice, then the sample will give us a biased estimate of the true result.

If the IEBC was reporting to us every few hours with updates from scores of polling stations that were randomly selected, we would expect results to come in close to the actual result throughout the period of reporting. Because of randomness, there would be no reason to expect any bias in the samples of polling stations selected.

The question is whether the results we were seeing in August were the result of randomly sampled polling stations.

Theoretically, this is certainly plausible: all polling stations were to open at the same time, close at the same time, and transmit results around the same time.

While plausible, I have no information about which polling stations were included in IEBC’s updates to test whether they were in fact random.

Practically, it seems unlikely that they were. The main factor affecting when a polling station would have been included in IEBC’s sample would be time: when a station transmitted results. What factors would determine time of transmission?

We know that polling stations that are in more remote areas or that were affected by bad weather would be unlikely to start or finish polling on time.

A substantial share of the total polling stations, and disproportionately those in more remote areas, did not have network coverage, and officers were instructed to leave the station and move to another with better coverage to report. They would have inevitably reported their results later than other stations.

So it is extremely unlikely that the selection of polling station results reported by IEBC was random. What we do not know is whether those non-random factors, such as the weather or remoteness of the station, were related to how people voted.

If more remote or marginalised areas of the country leaned toward one party, while less marginalised areas leaned toward another, then we would expect that the votes coming in earlier would be skewed toward the preferred candidates of the less marginalised areas, and the reverse would be true for results reported later.

There may have been other non-random factors affecting the time when results came in.

For example, suppose that stations with more polling agents present were more likely to fight about the results before certifying them, and suppose that there were more polling agents in more hotly contested areas.

This would mean that results would come in faster from stations with extreme votes for either party, and slower for stations where candidates were more evenly matched.

I make no claims about whether these things are true, only that they are plausible. If they were true, they would mean that a constant gap between candidates was less likely.

The best way to report results to the public is continuously, as they are received from polling stations, rather than via periodic samples.

Real time reporting is more transparent, and allows people to quickly spot anomalies where a particular polling station seems to have a vote inconsistent with past experience of how people vote in that area. We don’t need statistics when we have the actual numbers.

Jason Lakin is head of research for the International Budget Partnership. E-mail: [email protected]