New Politrack poll puts Ruto ahead: the questions behind the headline numbers

A presidential-preference survey reported on 1 October places William Ruto ahead of the other named candidates. Its most useful contribution to the election debate will depend on whether readers can examine how respondents were selected, what they were asked and how their answers were weighted.

The Star’s account attributes the poll to the Politrack Research Team and reports Ruto at 45.1 per cent, Edwin Sifuna at 19.6 per cent and Kalonzo Musyoka at 10.3 per cent. Fred Matiang’i and Rigathi Gachagua are reported at 7.2 and 6 per cent respectively, with 8.4 per cent undecided.

Those figures describe the reported answers to a survey, not votes cast in an election. This article has not independently audited the underlying dataset. The account says fieldwork ran from 15 to 25 September, involving 20,419 participants invited through a text-based platform. It reports the pollster’s claimed margin of error as three percentage points at a 95 per cent confidence level.

Start with the people the survey was designed to represent

A national poll needs a clear definition of its target population. Adults, registered voters and people likely to vote are overlapping but different groups. A percentage can mean something different depending on which of those groups the study intends to describe.

The American Association for Public Opinion Research’s disclosure standards call for researchers to explain their population, sampling approach, fieldwork and measures of precision. These are useful questions for Kenyan readers too; citing the standards does not imply that the association assessed or endorsed this poll.

The immediate request to the pollster should be for the complete methodology and questionnaire. The summary available to readers identifies several important features, but it does not enable an independent reconstruction of how the published result was produced.

That limitation does not, by itself, prove the figures are wrong. It sets the boundary around what can responsibly be concluded from them. Reporting a poll and validating its representativeness are separate tasks.

A large sample answers only one part of the quality question

Twenty thousand responses can sound decisive. But the number of participants cannot, on its own, show whether people with different characteristics had a fair opportunity to be included. The process for reaching and recruiting them matters as well.

Imagine an illustrative survey whose invitations reached one kind of phone user much more frequently than another. Collecting additional replies from the same source could increase the response count without fully correcting that imbalance. This example is not a finding about Politrack’s recruitment method.

AAPOR’s best-practice guidance explains why recruitment, survey mode, weighting and the questions themselves need disclosure alongside sample size. It also warns that missing particular kinds of respondents can affect estimates, even when many others participate.

A text-based platform is therefore neither automatic proof of quality nor automatic grounds for dismissal. Readers need to know how users were selected, how duplicate responses were handled and whether the final sample was adjusted to match a defined population.

What a margin of error can—and cannot—tell you

A stated margin of error needs a methodological explanation. It should not be read as a blanket promise that every possible source of error is contained within that interval, or that a future election result must fall within it.

Sampling uncertainty differs from problems caused by who declines to respond, how a question is understood or whether the people reached match the intended population. Those issues are not resolved simply by displaying a confidence percentage next to the result.

AAPOR’s guidance on credibility intervals distinguishes conventional sampling-error estimates from model-dependent precision measures used with some non-probability samples. The appropriate interpretation depends on the design and assumptions.

The available report does not provide enough detail to classify every part of this survey’s design. It would therefore be premature either to validate the three-point claim independently or to substitute a different interval calculated from the response count alone.

The fieldwork dates are part of the result

The reported interviews ended on 25 September. Political statements, endorsements and events after that date cannot be treated as developments that respondents necessarily considered. A publication date is not a substitute for the dates when answers were collected.

Nor does one poll establish momentum. To describe a candidate as gaining or losing support requires an appropriate earlier comparison, with attention to changes in wording, candidate lists, sampling and fieldwork. Comparing two unlike surveys can create an apparent trend that reflects method as well as opinion.

The same caution applies to coalition arithmetic. Adding every opposition candidate’s reported percentage does not establish how supporters would vote if those candidates endorsed one person. Some could change preference, remain undecided or choose not to vote.

What would make the findings easier to assess?

Useful accompanying material would include the complete questions and answer options, the recruitment procedure, weighting variables, treatment of undecided respondents and the basis for the precision claim. Information about sponsorship would also help readers understand who commissioned the exercise.

Political Headache’s report on the Judiciary’s election-year budget request examines another aspect of 2027 preparations. Both debates benefit when important claims can be checked against identifiable evidence, although a preference survey and the institutions handling election disputes serve different purposes.

For now, the defensible conclusion is narrow: the published account reports a lead for Ruto in a September survey. It is an item of political information to examine, not a declaration of the 2027 winner. The next useful contribution is fuller methodological disclosure rather than more confident predictions from the same headline figures.