PolitIQ Kenya Log in Request access
Reference

Analytic Standards

PolitIQ Kenya · Nairobi · Updated 2026-08-19

This note sets out how we reach a judgement and how we expect it to be read. It applies to everything we publish and to the client reporting behind our access wall. We publish it because a judgement whose method is hidden cannot be checked, and an intelligence product that cannot be checked is an opinion with better typesetting.

What we monitor

Kenyan political discourse as it appears in openly available sources: national and regional press, broadcast and radio, public video commentary, public social posts, and official records including gazette notices and procurement disclosures. Coverage is national, with county-level attribution wherever the source supports it.

We monitor discourse and events. We do not monitor private individuals, we do not acquire private communications, and we do not buy access to accounts.

Separating fact from judgement

Our reports keep three things visually distinct, because collapsing them is the most common way an analysis misleads without stating anything false.

LayerWhat it is
ObservationWhat the sources show. Countable, and traceable to the item it came from.
AssessmentWhat we think it means. Carries a probability term and a confidence level.
ImplicationWhat follows for the reader, stated as conditional on the assessment holding.

Probability language

Terms of likelihood in our reporting carry defined ranges. They are estimates of probability, not expressions of preference, and they are used consistently across every product we issue.

TermRange
Almost certainly95–99%
Very likely80–95%
Likely55–80%
Roughly even chance45–55%
Unlikely20–45%
Very unlikely5–20%
Almost certainly not1–5%

Confidence is stated separately from probability. A judgement can be very likely and still be held with low confidence when the sourcing behind it is thin.

Confidence levels

LevelBasis
HighMultiple independent sources, consistent over time, with a method validated against known outcomes.
ModerateCredible sourcing with gaps, or a method sound in principle but not yet tested against this class of question.
LowFragmentary or single-stream sourcing. Published when the question matters enough that silence would be the greater distortion.

Alternative hypotheses

Every assessment is written against at least one competing explanation, and that explanation is stated in the report rather than dismissed in private. Where the evidence does not separate two readings, we say so instead of choosing the more interesting one.

Indicators and warnings

Each assessment carries the observations that would change it, in both directions. An analysis that nothing the future might show could falsify is not an analysis.

Sourcing

Volume is not corroboration. A claim carried by twenty outlets running the same wire copy is one source, and we count it as one. For every item we record the outlet, its reach, whether the account is primary or downstream, and the date of first appearance.

We distinguish organic discourse from coordinated amplification. Where a surge shows the signatures of coordination, that is reported as a finding about the campaign, not as evidence about its subject.

What the method cannot see

Ethics and the law

We operate under the Kenya Data Protection Act 2019. Identifiers in our pipeline are pseudonymised before storage. Public reporting is aggregate: we do not publish material that identifies a private individual, and we do not republish hate speech even as illustration — findings on ethnic dangerous speech are issued as counts and categories, never as quotable text.

We report on people who hold or seek public office, in their public capacity. That is the whole of our subject matter.

Corrections

Reports carry an information cutoff. Where a published judgement is later contradicted, we amend the report, mark the change, and leave the original assessment visible. Quietly editing an assessment after the fact destroys the only thing a track record is worth.