Takwimu
Takwimu is Kiswahili for data.
In TIIt is also the first half of the name Takwimu Intelligence (TI).
- Related
- Takwimu Intelligence, Data intelligence
- Read more
- More on this · On its own page
Glossary
This data and research glossary explains, in plain words, the terms you meet in surveys, opinion polls, public participation and data protection in Kenya, each with a real situation where it matters.
What Takwimu Intelligence is called, and the words its screens use.
Takwimu is Kiswahili for data.
In TIIt is also the first half of the name Takwimu Intelligence (TI).
Takwimu Intelligence (TI) is a data intelligence platform that organisations in Kenya use to collect their own data and make decisions with it.
Think of a county budget officer in January, weeks before the fiscal strategy paper is due. She uses TI to hear what residents in each ward want funded, and when the assembly later asks what became of a view, the answer is already written down beside it.
In TIThe platform covers the five stages of a study: collect, analyse, strategy, message and follow up.
TI Academy is the training side of Takwimu Intelligence, preparing people for the paid jobs research needs, with the courses taught by training institutions on the TI platform.
Its ladder starts with the jobs people walk into, Data Entry Clerk, Field Interviewer and Media Monitor, then climbs through research data manager and Field Supervisor to Quantitative Analyst, Qualitative Analyst and Study Designer, then M&E Officer, Campaign Data Officer and Communication Officer, and at the top, Research Manager. Every role ends in a practical piece of the job, marked on the platform.
In TILearners practise each session in a training workspace on simulated data, and each of the role courses ends in a practical the platform marks.
A workspace is your organisation's own space on TI, holding its studies, its people and its data.
Each person you invite sees only what their role allows, so a data clerk can key forms without ever seeing the director's reports.
In TIYou set up your workspace once, then invite your team to it.
A study is one piece of research: one question, the data collected to answer it, and what that data shows.
A tracker poll that runs every month for a year is still one study, with twelve rounds.
In TIYou create a study in your workspace and start by saying what it is trying to decide.
A sector pack is a ready-made starting point for one kind of work, with the questions, indicators and reports that sector already uses.
A county starting its budget consultation opens the public participation pack and finds the legal calendar and a place for a decision on every view already set up.
In TIEach sector pack starts a study with its sector's questions and reports already in place.
Data intelligence is turning the data an organisation collects into decisions, and then checking whether those decisions worked.
A county health team that surveys mothers before choosing where to open an evening clinic, then counts visits six months later to see whether it chose well, is doing exactly that.
In TIThe data intelligence platform joins the five stages, from collecting the data to measuring what a decision moved.
An insight is a finding that tells you what to do differently.
That 58 per cent of voters back a candidate is a figure. That support among women under 35 in two wards has fallen to a third over school fees is an insight: it tells the campaign where to go and what to talk about.
In TICoded answers become findings, each linked to the passages behind it.
An informed decision is one made with the evidence in hand and its limits understood.
A health officer choosing between two wards for an evening clinic needs more than which ward's figure is higher. If each figure rests on 40 interviews, the margin is about 15 points either way, and the two wards may not differ at all.
In TIA study opens by stating what it is trying to decide, and its results are read against that question.
How answers are gathered, by whom, and how they are checked.
A survey asks a chosen group of people the same questions in the same way, so their answers can be counted and compared.
A county that wants to know how residents rate the water service cannot ask everyone, so it might interview 800 households spread across its wards and read the results ward by ward.
In TIOne survey can collect on several of the nine collection channels at once, with every answer landing in the same study.
A questionnaire is the fixed set of questions, answer choices and instructions that every interview follows.
Its wording matters more than people expect. Ask whether residents support the new market fee that will pay for repairs and you have told them the answer; ask whether they support or oppose the new market fee and you will hear what they think.
In TIYou build a questionnaire on screen, or bring one in from KoboToolbox, ODK or SurveyCTO as XLSForm.
A census counts everyone in a population, where a survey asks a sample.
Kenya's 2019 census counted 47,564,296 people. Because it counts everyone, survey results are weighted to its county and sub-county figures.
In TIEach study shows the census figures it can weight to for its area.
A sample is the smaller group you actually interview, chosen so their answers can stand for everyone else.
A national poll of 1,000 adults picked at random can speak for Kenya's adults. It cannot speak for Lamu, where a poll that size would find three or four of them.
In TIThe platform can work out how many interviews you need for the precision you want in each area.
A sampling frame is the list or map you draw your sample from.
If a phone survey draws its numbers from sign-in sheets at chiefs' barazas, everyone who never attends a baraza is missing before the first call is made, and more interviews will not bring them back.
In TIField areas are cut from the map and handed out as visit lists, so every interviewer knows which households are theirs.
A quota fixes how many interviews must come from each group, such as women aged 18 to 34 in a ward.
Interviewers working in the daytime find more older people at home. A quota stops them filling the sample with whoever opens the door first, though it does not make the sample random.
In TITargets for each group are set when you set quotas on a questionnaire.
An enumerator is the person who goes out to ask a survey's questions and record the answers, a job research firms also call field interviewer.
Most of the work happens at someone's door. An enumerator on a household survey in Kakamega might reach a dozen homes in a day, explain who the study is for, read the consent statement in full and ask every question as it is written, including the awkward one about income. The supervisor will phone some of those households afterwards, so a rushed or invented interview does not stay hidden for long.
In TIThe Field Interviewer course trains enumerators on the real screens and ends in a practical of three households the platform marks.
A field supervisor leads a team of enumerators and answers for whether their interviews are real and right.
Her evenings go on the day's quality flags: an interview finished in four minutes, five households with identical answers. She phones a few respondents back, decides which interviews stand, and sets out tomorrow's areas.
In TISupervisors check interviewers' work from each day's quality flags, and the Field Supervisor course trains them for it.
A back-check is a second call or visit to a respondent to confirm that the interview happened and was recorded as they said.
A supervisor rings a household from yesterday's list and asks two of the same questions again. If the woman who answers says nobody came, every interview by that enumerator is looked at again.
In TIBack-checks are drawn, recorded and compared with the interview when you settle back-checks.
A pilot is a small trial run of a questionnaire before the real fieldwork.
Twenty interviews in an Embu market can save a study. They show the income question nobody will answer, or the Kiswahili word that means something else in that county, while there is still time to fix it.
In TIA questionnaire is frozen for a pilot, and every pilot interview records the version it answered.
CAPI, computer-assisted personal interviewing, is a face-to-face interview recorded on a phone or tablet.
The interviewer still reads every question aloud. The phone takes care of the skips, refuses an age of 340 and records where the interview took place, and there is no paper left to key afterwards.
In TIInterviews run on the interviewer app or with ODK Collect, and both send into the same study.
CATI, computer-assisted telephone interviewing, is an interview done over a phone call, with the interviewer reading the questions from a screen.
A team can reach 1,200 people in a week without leaving the office. Everyone without a phone, or who never answers an unknown number, is left out, and the results have to be read with that in mind.
In TIInterviewers work call lists with each call's outcome recorded.
A USSD survey asks questions through a short code that works on any mobile phone, with no smartphone or data bundle needed.
A grandmother with a basic phone can rank her ward's five projects by pressing numbers. The session times out quickly, so the questions must be few and the choices short.
In TIUSSD is one of the nine collection channels, and the questionnaire builder names any question USSD cannot carry.
An SMS survey sends questions by text message and reads the replies as answers.
After a public forum the county texts everyone who signed in and asks them to rate the meeting from 1 to 5. Two or three questions like that is about as far as a text survey can go.
In TIText messages are one of the nine collection channels, and their replies land in the same study as interviews and web answers.
A web link survey is a questionnaire people open from a link and fill in themselves.
Shared in WhatsApp groups and on a county's Facebook page, a link can bring in thousands of answers in a day. They come only from people who are online, under half the country, so the link is usually paired with interviews or USSD.
In TIA study's web link opens once it is launched, and answers appear as they arrive.
Data entry is typing the answers on paper forms into a database, exactly as they were written.
Three hundred handwritten feedback forms from a sub-county forum count for nothing until someone keys them. A good clerk types what is on the page, leaves a number they cannot read blank and logs it, and never improves an answer.
In TIPaper forms are keyed on the data entry screen, and the Data Entry Clerk course trains people to do it.
Second keying is entering the same form twice and comparing the two, to catch typing mistakes.
On polling day a campaign's agents photograph each Form 34A, and two people key its figures. A station counts only when both keyings agree on every candidate; where they differ, someone keys it again from the photo.
In TIPolling day results are keyed twice by design, and the polling day count uses a station only when both keyings agree.
Interviews, discussions and documents, and how their text is coded.
A focus group discussion is a guided conversation about one topic with six to ten people who have something in common.
Before writing a survey on fishing permits in Kilifi, a team might sit separately with young fishermen, with boat owners and with the women who sell the catch. Each group has its own words for the same problem, and the survey is then written in those words.
In TIRecordings are uploaded to the study, and each speaker can be given attributes so codes can be compared by group.
A key informant interview is a one-to-one conversation with someone who knows a subject because of their position.
To understand why school feeding in Marsabit stalls every term, you would talk to the county education officer, a head teacher and the cook. None of them speaks for residents; each tells you how one part of the programme really works.
In TIInterview notes and recordings are prepared for coding beside the study's other text.
An in-depth interview is a long, open conversation with one person about their own experience.
A researcher trying to understand why mothers in a Nairobi settlement stop clinic visits after the first vaccine might spend an hour with each of fifteen mothers, following whatever each one raises.
In TITranscripts are coded passage by passage, and memos hold what the researcher makes of them.
A transcript is the word-for-word text of a recorded interview, discussion or broadcast.
A two-hour focus group in Kiswahili and Dholuo becomes about 15,000 words of text, every speaker marked. Every finding is coded from that text, so a mistake in the transcript travels all the way into the report.
In TIA machine transcript is checked against its recording before it is coded.
A coding frame is the list of themes, each one defined, used to sort open answers or passages of text.
Ask 2,000 residents what the county should fix first and you get 2,000 different sentences. With codes for water, roads, health and jobs, each defined with an example, two people can sort them the same way and count them.
In TIQualitative researchers also call it a codebook, which is the name of the screen where it is built.
Coder agreement measures how often two people, coding the same text on their own, give it the same codes.
Two monitors code the same 50 radio clips about a county budget. If they agree barely more often than chance, the codes are too loose to count. Krippendorff's alpha is the usual measure, and 0.8 is the usual bar for relying on the result.
In TIDraw a double-coded share and measure agreement before the codes are counted.
A desk review reads what is already written about a question before any new data is collected.
Before a survey on youth jobs in Kakamega you would read the county's CIDP, the latest Economic Survey from KNBS and any recent studies. Some of your questions may already be answered, and the survey can then ask only what is missing.
In TIDocuments are read, coded and set against each other when you run a desk review.
Triangulation is checking a finding against more than one independent source.
If a survey says most homes in a ward lack piped water, the water company's records and a talk with the ward administrator should say the same. When they disagree, you have found something worth following up.
In TIThe platform shows what several sources agree on and where they part.
How a figure is worked out, and how sure it is.
A dataset is a table of records, usually one row per respondent and one column per question.
A survey of 800 households becomes 800 rows. It tells you more when you set it beside public data, such as the IEBC's registered voters by ward.
In TIYou can bring in a dataset from SPSS, Stata, CSV or Excel and analyse it with the platform's tools.
A codebook explains every column of a dataset: its name, the question behind it, and what each number in it stands for.
Without one, a column called Q7 holding 1, 2 and 99 is a puzzle. With one, it reads: main source of drinking water, where 1 is piped into the dwelling, 2 a public tap and 99 no answer.
In TIExports to SPSS or Stata carry each variable's question and value labels, and an Excel export adds a sheet of variables.
Weighting adjusts how much each interview counts, so the sample matches the population on facts such as age, sex and county.
Young men are hard to find at home, so a county poll may end up with them as 10 per cent of its interviews when they are 18 per cent of the adults. Weighting gives each of their interviews a little more say. It cannot fix a leading question, or the views of people who refused.
In TIA study can be weighted to county population from the 2019 census, and the screen shows how much precision the weights cost.
The margin of error tells you how far a poll's figure could be from the truth simply because a sample was interviewed and not everyone.
Say a poll of 1,000 voters puts one aspirant on 48 per cent and her rival on 44. Each figure carries a margin of about 3 points, so the rival could in fact be ahead, and a headline saying she leads claims more than the poll can. In a ward poll of 400 the margin grows to about 5 points. None of this covers a leading question or the people nobody reached; the margin measures sampling and nothing else.
In TIEstimates are shown with their interval, and figures below the minimum sample sizes are held back from reports.
A confidence interval is the range the true figure probably lies in, at a stated level of confidence such as 95 per cent.
A survey finds 40 per cent of households use a pit latrine, with an interval of 35 to 45. A sanitation budget built on that figure should still work anywhere in that range.
In TICharts can show each figure's interval when you chart a result.
The design effect measures how much a survey's design, such as clustering interviews in a few villages or weighting them, widens its margin of error.
Twenty interviews in each of forty villages costs far less than eight hundred scattered homes, but neighbours tend to think alike. With a design effect of 2, those 800 interviews are worth about 400 random ones.
In TIThe weighting screen shows the design effect its weights add.
The effective sample size is how many truly random interviews your sample is worth once its design effect is counted.
A poll of 1,200 with a design effect of 1.5 is worth 800. Its margin of error should be worked out from 800, or the poll promises more precision than it has.
In TIThe weighting screen reports the effective sample size after weighting.
A difference is statistically significant when it is bigger than chance alone would be likely to produce.
A candidate moves from 41 to 44 per cent between two polls of 600 people each. Chance alone could produce that, so it is not yet news. A twelve-point move would be.
In TIEach round can be compared with the last to see whether a change is significant when you measure change between rounds.
A crosstab breaks the answers to one question down by the answers to another.
Trust in the county government might be 55 per cent among people over 50 and 30 per cent among those under 25. The crosstab shows where opinion splits, and a test then tells you whether the split is real.
In TIAny two questions can be cross-tabulated in the study's analysis.
Regression estimates how much each of several factors matters to an outcome, with the others held steady.
Older voters say they are more likely to vote, and older voters are also more likely to own land. Regression pulls the two apart, so a campaign can see whether it is age or land doing the work.
In TIModels for counts, choices and rating scales run in the workbench.
Measuring whether a programme or a message changed anything.
An indicator is a measure, agreed in advance, that shows whether things are moving the way you intended.
For a water programme it might be the share of households within 500 metres of a safe water point, counted the same way every year. Its definition is fixed before the first count, so nobody can move the goalposts later.
In TIIndicators are defined with their targets and counted from partners' reports by period.
A baseline is the first measurement, taken before a programme starts, that later change is measured against.
Before a county starts school meals, a baseline survey records attendance and children's height and weight, both in the schools that will get meals and in some that will not. Without it, nobody can say later what the meals changed.
In TIA study can declare its baseline and endline rounds and measure the change between them.
An endline is the measurement at the end of a programme, repeating the baseline's questions in the same way.
If 62 per cent of households treated their drinking water at the baseline and 74 per cent do at the endline, the change is 12 points. How much of that the programme caused depends on what happened in places it never reached.
In TIBaseline and endline rounds are compared, with the people lost between them shown, when you measure change between rounds.
Monitoring and evaluation means tracking a programme while it runs and judging, at set points, whether it worked.
A county health programme's monthly report on clinic visits and drugs in stock is the monitoring. The evaluation comes at the middle and the end and asks the harder question: did maternal deaths fall, and why?
In TIA programme can be started from a monitoring pack, with its indicators, partners and reporting periods in place.
A control group is people or places that do not get a programme or message, kept aside so its effect can be measured against them.
A campaign texts 20,000 supporters and holds back 5,000 chosen at random. If turnout among those who got the text is 4 points higher, the message moved about 4 points; without the held-back group there would be no way to tell.
In TIWhen you test a message, the platform draws the held-back group and compares the two; a held-back group is agreed per study on request.
A message test sends different versions of a message to comparable groups and measures which one changes what people do.
A county wants more residents at its budget hearings. Half its list gets a text about money for their ward, the other half a text about their right to be heard, and the county counts who comes.
In TIYou can test a message and see what moved, sending through your own text-message account.
Impact is the change a programme or message actually caused, beyond what would have happened anyway.
Clinic visits rose 20 per cent in the year a county ran radio adverts. In the next county, with no adverts, they rose 15 per cent. The adverts' impact is nearer five points than twenty.
In TIThe platform reports a message's effect only against a held-back group, when you test a message.
The plans, papers and polls of Kenyan public life.
Public participation is the duty of Kenyan public bodies to involve the people in the laws, budgets and plans that affect them, and to consider what they say.
Holding a meeting is not enough on its own. In June 2025 the High Court, in a petition no respondent opposed, declared the Nairobi City County Finance Act 2023 unconstitutional, finding that residents had been shown schedules of figures with no explanation of the policy behind them.
In TIEach view is recorded with a decision and a reason, and the consultation is filed as a public participation record.
A memorandum is a written submission setting out your views on a Bill, a budget or a policy, sent to Parliament, a county assembly or another public body.
A residents' association unhappy about a new parking fee can send a two-page memorandum saying what it wants changed and why. Committees receive them in numbers: the one that handled the Finance Bill 2023 received 1,080, as the Supreme Court recorded.
In TIMemoranda are uploaded to the consultation, read, and given a decision each.
A public forum is a meeting where a public body presents a plan, Bill or budget and residents respond in person.
Held on a market day in a ward's social hall, with the draft plan handed out a week before, a forum can gather real views. Held without the documents in advance, it leaves residents little to respond to.
In TIForums, barazas and meetings are planned, publicised by SMS and USSD, and checked in when you run an event.
The Annual Development Plan is a county's plan for the coming financial year, drawn from its CIDP, which the county executive must submit to the county assembly by 1 September.
It lists the projects the county means to fund in the coming year, ward by ward, with their costs. A road or a dispensary that is not in the ADP is unlikely to reach the budget.
In TIThe public participation system carries the county budget calendar, with each document's legal deadline.
The County Fiscal Strategy Paper sets a county's priorities and spending limits for the next budget, and must be submitted to the county assembly by 28 February each year.
This is where a county decides how much goes to health, roads or water before the detailed budget is written. The County Treasury must seek the public's views on it, which is why CFSP hearings come early in the year.
In TIViews on the CFSP are recorded with a decision each in the public participation system.
The County Integrated Development Plan is a county's five-year plan, which every county must have under the County Governments Act.
It sets out what a county means to achieve over five years, and each year's ADP and budget should follow from it. Public participation in county planning is mandatory under section 115 of the same Act.
In TIA desk review of the CIDP sets a study's questions against what the county has already planned.
An opinion poll asks a sample of people their views and reports what the wider population is likely to think; in Kenyan electoral law it is a survey of public opinion, or of a sample of it, on an election, candidates or political issues.
A ward poll of 400 voters before the party primaries can show an aspirant which issues matter and how well known she is, within about 5 points. The Publication of Electoral Opinion Polls Act 2012 bans publishing results on election day or in the five days before it.
In TIWard and constituency polls run on the election and campaign management software, weighted to the register or the census.
A tracker poll asks the same questions at regular intervals to show how opinion moves.
A governor's team polls 600 voters every month until the August 2027 election. One month's change is often within the margin of error; three or four months moving the same way is a trend worth acting on.
In TIEach round is compared with the last, and a short tracker between rounds can re-ask people who left a number.
A polling station is the place where a set group of registered voters casts their ballots, with its own register, its own presiding officer and its own results forms.
Kenya voted at 46,229 polling stations in 2022, none with more than 700 registered voters. A school used as a polling centre often holds several stations, which campaigns call streams, and each produces its own Form 34A for the presidential result.
In TIAgents' copies of each station's forms are photographed and keyed on polling day, and the command centre follows them station by station.
The Data Protection Act 2019, as it applies to research.
A data controller decides why and how personal data is processed; the Data Protection Act 2019 defines it as a person, public authority, agency or other body which, alone or jointly with others, determines the purpose and means of processing of personal data.
When a county commissions a household survey from a research firm, the county decides what is asked and why, so the county is the controller. If a respondent complains, it is the county that answers, to them and to the Data Protection Commissioner.
In TIThe organisation running a study is its controller, and Takwimu processes the data on its instructions under the data processing agreement.
A data processor handles personal data for a data controller; the Act defines it as a person, public authority, agency or other body which processes personal data on behalf of the data controller.
The research firm collecting a county's survey and the platform storing the answers are both processors. The Act requires a written contract keeping them to the county's instructions, and a processor that goes further is treated as a controller for that processing.
In TITakwimu is a processor for every study on the platform, and its subprocessors are listed by name.
Consent is a person's free and informed agreement to what will be done with their personal data, which the Data Protection Act 2019 requires to be given by a clear statement or action.
In a survey it is a real moment at the door. The enumerator says who is running the study, what the answers are for and how long they will be kept, then asks plainly whether the person agrees. If the answer is no, the interview stops there, and a person who said yes can change their mind later.
In TIEvery interview records which version of the consent statement it agreed to, and each study records its retention period and research permit.
The Office of the Data Protection Commissioner is the body the Data Protection Act 2019 establishes to oversee how personal data is handled in Kenya, led by the Data Commissioner.
No one may act as a data controller or processor without registering with the Data Commissioner, within thresholds the Commissioner sets. Anyone unhappy with how their data was handled can complain to the Commissioner.
In TIEach study's consent, retention period and research permit are recorded on its compliance screen.
A data protection impact assessment is an assessment of the impact of planned processing on the protection of personal data, which the Act requires before processing that is likely to result in high risk to people's rights and freedoms.
A study that asks about HIV status, ethnicity or violence at home is likely to be high risk. Before fieldwork the controller writes down what could go wrong and how it will be prevented; if the risk stays high, the Data Commissioner is consulted, and the report goes in sixty days before the work starts.
In TISensitive answers can be sealed where they are given: encrypting a questionnaire means only the holder of the key file can read them.
Anonymisation is the removal of personal identifiers from personal data so that the person it describes is no longer identifiable.
Before a county publishes a survey dataset, names, phone numbers and exact locations come out, and small groups are merged so the only teacher in a village cannot be picked out by age and job. Replacing names with codes while keeping the key elsewhere is pseudonymisation, which the Act treats separately.
In TIA phone number asked with a contact question is kept encrypted and apart from the answers, in no dataset or export, and used only for the study's tracker between rounds.