Skip to content

guide

Running an audience wave on TI

Audience survey software for Kenyan media: how one wave goes from the outlet list to weighted figures, step by step.

Takwimu editorial team, Editorial team8 min read

Audience survey software takes one wave of research from the first question to the final report: the questionnaire, the sample, the interviews, the weighting and the figures. On Takwimu Intelligence (TI) a wave runs on the Media Audience pack, and most of the decisions that protect the figures are made before the first interview.

Key points:

  • A wave is one round of fieldwork, and its worth grows when the next round can be read against it.
  • The prompted outlet list has to be rewritten for your market before fielding.
  • The sampling design is recorded on the study, and it decides what the figures may claim.
  • Weighting uses KNBS census totals at the level where they exist, in practice the county.
  • Replies from an on-air announcement are self-selected, so they never measure reach.

What is an audience wave?

An audience wave is one round of fieldwork, with one questionnaire, in one period. One reading says little. The second wave is where decisions start, since it shows what moved.

That only works if the instrument holds still. Change the outlet list, the time bands or the sampling design between waves and the trend is gone. Treat the first wave as the template for every wave after it.

What should audience survey software do?

It should keep the decisions that make a figure defensible in writing, next to the figure. A general form tool will collect answers. It will not carry the outlet list, the sample, the weights and the limits through to the page an advertiser reads.

For an outlet, four things have to stay attached to every number: who was asked, how they were chosen, how the answers were weighted, and which question produced it. On TI each of those sits on the study itself. The pack allows fourteen days to answer a request for a methodology note, and that time goes on review, with nothing left to assemble.

Which outlets go on the prompted list?

Start here, because share depends on it. The questionnaire comes with a list of stations, channels and titles in alphabetical order, and alphabetical order is no guide to size. Replace it with the outlets broadcasting into your market before fielding.

For the question "Which one station did you listen to most yesterday?", put your station and your main competitor in the first two answers, and take them out of the list that follows. Station share is read from this question. A write-in question catches any station the list missed.

Comparators earn their place. An outlet's reach can rise while the whole market's rises, and only the other names on the list show which happened. Our guide to prompted and open recall explains why the prompt matters so much.

Why ask about yesterday?

Yesterday is specific, recent and easy to check against memory. Asking how often someone "usually" listens gets a claim. The pack rests on a plain rule: claimed frequency is not behaviour, and open recall measures which outlets are famous.

Day-after recall asks about radio, television and news reading in one pass, prompted at every point. The opening line defines yesterday as the whole day, from waking to sleeping, at home, at work, in a matatu or in a shop. Any listening counts, however short, and whether or not the person chose the station.

Everybody is asked everything. Where something does not apply, there is an answer that says so, and it stays out of the base.

How are the dayparts asked?

Radio has seven bands: before 6am, 6am to 10am, 10am to 1pm, 1pm to 4pm, 4pm to 7pm, 7pm to 10pm and after 10pm. Television has five: before noon, noon to 5pm, 5pm to 8pm, 8pm to 11pm and after 11pm.

Each band is its own question. A grid looks tidier on paper, but every channel TI collects on asks one question per screen, and a grid arrives on all of them as an empty text box. Answering seven short questions takes seconds.

The same bands feed each daypart in radio and TV you report later. Keep them fixed across waves.

Who gets interviewed, and how many?

This decision sets the claim. Record the sampling design on the study before fieldwork, because the reports quote it. A probability design can speak for a population. A convenience or quota design describes the people who answered, and the reports say so in words.

Age comes in the bands the census publishes, so the sample can be reconciled and quota'd without recoding. Somebody under 18 is recorded and stays outside every figure. The county is asked directly, so figures break down by the answer the respondent gave.

An invented example shows why sample size follows the question you want answered. Take 800 interviews across two counties, 400 in each, on a simple random sample at 95% confidence. A figure near 50% carries about 3.5 points either way across both counties and about 4.9 points in one county. A group of 100 people inside that county carries about 9.8 points.

Plan the sample around the smallest group you intend to report. Clustering adds to these margins, so leave room.

How do replies come in?

Sampled interviews go through field interviewers, in the mode your budget and market allow.

A study can also be announced on air. Replies come by USSD, SMS, voice or web through the outlet's own gateway account, and a code per station shows which station brought them in. A group running several stations can see which one sent people.

Those replies are self-selected. They never measure reach. Keep them for what they do well, such as hearing from engaged listeners and testing a question, and read why on-air responses do not measure reach before anyone is tempted to quote them.

How are the answers weighted?

Answers are weighted to KNBS census totals at the level where those totals exist, which in practice is the county. Weighting corrects for who was easier to reach. It cannot repair a sample that left a group out.

Who answered shapes the figure. The Communications Authority of Kenya's April 2026 report finds radio listening higher among men than women and rising with age, so a sample that leans young will understate radio unless it is corrected.

An invented example: if women aged 18 to 24 are 10% of your interviews and 15% of the county's adults, each of their answers counts 1.5 times. The report then shows how the people who answered compare with the population, so a reader can judge the correction. See weighting an audience survey for the method.

Each figure shows its margin of error where the design supports one. On a convenience or quota design the figures carry none, and the report says that plainly.

How do respondents stay protected?

The questionnaire opens by saying it wants no name and nothing that identifies the person. Respondents' identities stay apart from analysis, and an outlet owns its own data. Plan the consent wording and retention with data protection for audience research in view.

One rule helps here. If a question is not going to change a decision, drop it.

What do you read first when the results arrive?

Read the sample before the figures. Check interviews achieved, then how the people who answered compare with the population.

Then read the warnings the study raises. They flag advert fatigue, a signal problem, a presenter problem, a schedule people cannot place, and a breakfast show smaller than the afternoon. Each has a declared rule, so nobody has to argue about whether 22% is high.

After that, move to the audience wave report. Reach says how big a problem is. The reasons say which department owns it.

What goes wrong in an audience wave?

Most failures are made before fieldwork. These are the ones to watch.

  • The default outlet list goes into the field unchanged, so the main competitor is missing or the market is wrong.
  • The client's own staff or fans are recruited as respondents, and the sample stops describing anyone else.
  • Replies from an on-air promotion are mixed into the sample to lift the count.
  • The questionnaire is edited between waves, and the comparison is lost.
  • Figures are quoted to an advertiser before anyone has read the sampling design.

Fix those five and the rest of the wave becomes routine.

Where TI fits

A wave is decided before fieldwork. TI holds the questionnaire, the outlet list, the sampling design and the weighting on one study, then carries them through to every report.

To see it on your own station or title, book a walk-through.

Takwimu Intelligence (TI) is audience measurement software for Kenyan media.

Sources

  1. 1

    Audience Measurement and Industry Trends Report, abridged version, April 2026 · Communications Authority of Kenya, April 2026Seven-day and daily media use to Q3 2025/26, radio genre and language, advertising window, industry challenges. Opened 5 November 2026.

More from the library