Making the census count

Upwards of 15 million more children have been born, pushing the population close to the 50 million mark. We will know – not the exact figure, but something approximating it – in a few months. And this time around the counting is happening on tablets, not sheets of paper. 

How close will this approximation be? The short answer; it depends, on many variables. 

Many concerns have been raised about the exercise: logistical challenges, elusive enumerators, and the length of the questionnaire.  

In trying to put these concerns into context, we must remind ourselves that no census is perfect. There will always be people not counted, inaccurate answers given, data accidentally lost, and enumerators committing errors both intended and otherwise. The question is how, given resource constraints, to collect the essentials for the largest number of people, as accurately as possible, minimising systematic omissions or errors, such that these data can be used by the government, the public, and researchers. How well is the 2024 census achieving these aims?

Let’s start with content, and why we need censuses in the first place. Some of the earliest censuses from ancient Greece and Rome were designed primarily for state control, documenting the number of males who could serve in the military and the extent and location of property that could be taxed. Today, the goal is, in theory, to serve rather than control the public.

The stated aim of most statistical agencies who collect census data – which may or may not be believed by the public, presenting its own challenge – is to use these data for the distribution and management of resources, policy planning, and, in democratic societies, to facilitate relatively even political representation. 

Census data are also used as a sampling frame – a complete list of all units (for instance, individuals) in a population of interest – to help researchers and governments draw samples for surveys that are representative of some population. Unlike a census, these sample surveys do not enumerate every individual in the population of interest. 

Rather, a sample survey provides an estimate of a larger population, which could be as small as a neighbourhood or as large as the entire country. These surveys are smaller and less expensive than a census, covering many topics, from health (the Demographic and Health Surveys) to political attitudes and behaviour (the Afrobarometer surveys).

If there are high-quality census data on the distribution and basic demographics of the population, we can use a representative sample to construct a good estimate of say, the child mortality rate, without asking every single family in the country how many children they have lost, which is a time-consuming series of sensitive and painful questions. 

Too many questions?
These sample surveys are helpful because there are only so many questions that can be asked in a census. There are many more important topics than there is space on the census questionnaire, given constraints to time (including the person-hours of both enumerators and respondents) and resources.

If you have been enumerated in the ongoing census, you’ll know that this questionnaire feels rather long. There are close to 60 questions asked of anyone above 18, with a dozen or so more for any woman who has at least one child. In this census, a 25-year-old woman with two children would be asked around 70 questions. Of her 3-year-old child, there are more than 40 questions. Is this a lot?

By comparison, the 2020 U.S. census asked a total of seven questions per person, regardless of age. But countries like the U.S. can afford to keep the questionnaire short because they can supplement census data with a large tranche of high-quality administrative and industry data on topics ranging from banking to home ownership to birth and death registries.

Regionally, the questionnaire length looks more similar, but even then, Uganda stands out. For an adult male, the Uganda 2024 census is 40 to 80 percent longer than the census for Kenya 2019, Rwanda 2022, Tanzania 2022 or Ghana 2021.

There are about two to three times as many questions for a Ugandan three-year-old than for her regional counterparts. And this is not even counting the more than 50 household-level questions (including whether the household owns a motor engine for boats!).

The length of the census questionnaire has also grown over the decades. In the 2002 and 2014 censuses, there were about 20 questions for each child up to age 10, around 25 questions for those over 10, and about 30 questions for women between the ages of 12 to 54, with a few more for those women with children.  

The number of questions has almost doubled in the current census due to the inclusion of, and additional detail on, topics such as mental and physical health, migration, and access to government services.

These are important topics, but must they be asked of every Ugandan? Are there any for which there are other sources of data that would serve the purpose? Would a representative sample serve the same purpose? Most importantly, what is the opportunity cost of including these questions? 

There are no right answers regarding which topics to include, as the questions asked reflect the priorities of each country. For example, Tanzania asks relatively more questions than Uganda about the economic activities of each person, and from a younger age, on top of recording the phone number of every mobile phone in the household. Recent censuses in the region have moved toward including more detail about topics like disabilities, the use of ICT, and migration, all of which have become more salient or important for policy in the past decade.

However, increasing the number of questions and the complexity of the questionnaire also comes with costs in terms of time, money, and quality, as Ubos itself has noted in the postmortems of past censuses.

For instance, the administrative report for the 2002 census, when the questionnaire was less than half as long and the population half as big, noted that “the questionnaire was too overloaded” which created “excessive work without careful planning” and “increased enumerator workload.”

Tempting as it might be, adding questions on important topics may come at the cost of lower-quality data on the essentials.

The United Nations recommendations for population censuses notes, “The selection of topics should be carefully considered in relation to the total resources available for the census. An efficient collection of accurate data for a limited number of topics, followed by prompt tabulation and publication, is more useful than the collection of data for an overambitious list of topics that cannot be properly processed and disseminated in a timely, reliable and cost-effective manner.”

After providing a list of core topics, the report cautions, “It should be stressed that no country should attempt to cover all the topics included in the list of population topics…” which looks quite a bit like what this census has done.

Surprisingly, despite the 2024 questionnaire being longer than other recent censuses in the region, the cost doesn’t appear to be higher.

Comparing the exact amount each country spends on a census is difficult: the budget may cover several years, disbursement may not match official budget figures, and there may be last-minute budget allocations. However, a back-of-the-envelope calculation suggests Uganda is spending substantially less per person than its peers. 

Kenya spent a reported $180m – nearly $100m more than Uganda, for a slightly larger population, implying a cost of $3.80 compared to $1.70 per person.

Rwanda spent Rwf30b for a population of just over 13 million, for a cost of about $2.20 per person, while Ghana and Tanzania spent about $2.90 and $2.30 per person respectively.

The irony is that too many questions – perhaps acquired with too thin a budget – may so hamper the quality of the data collected that the census is not able to serve as a sampling frame for surveys that could cover these important topics more cost-effectively and accurately. Why does quality suffer?

Undertaking a census, much like conducting an election, is a massive logistical operation. It is one of the most complex activities the state undertakes – reaching every one of its citizens and non-citizen residents and having an extended interaction, in person, in a matter of days. 

This year, the government hired over 100,000 enumerators, each of whom is expected to cover an average of 145 households over the 10 days, amounting to nearly 15 households per day. There is a hierarchical organisational structure in place but managing so many temporary workers within a temporary administrative apparatus poses unusual challenges which are exacerbated by a complex questionnaire.

A long questionnaire, and uneven effort required depending on the type of respondent (for instance, men versus women, with the number of questions different for a 2, 5-, 10-, 16-, or 18-year-old), leads to an increase in the likelihood of enumerator error and systematic omissions. Completing a questionnaire of this length – much less 14 in a row, in a day, for 10 days! – is exhausting. 

Imagine you are an enumerator. It is 5pm, you are on household seven of your daily target, running behind, hungry and tired, and you realise the home you have just entered has three females between the ages of 10 and 59, each of which will require at least 60 questions. Or you realise the household has migrated from elsewhere and speaks a language neither you nor your supervisor speaks. Humans being humans, you can imagine what happens next. Recent research shows that across countries, enumerators conducting household surveys tend to systematically omit “high cost” (read time-consuming) household members. These systematic omissions are worse than random data loss because they lead to biases in our estimates of the population. 

Ideally, data are checked as they are coming in, which is now possible with electronic as opposed to paper-based data capture. If we know how long it takes to enumerate the average household, we can figure out if and when enumerators are spending much less time on average on the entire questionnaire or on particular questions, or are submitting questionnaires with suspiciously small households.

Poor performers should be replaced and backchecks conducted. This of course requires a team of data analysts and a fairly tight command-and-control structure. 

The good news is that the move to electronic data capture should improve the overall quality of the data. Paper sheets won’t be gone with the wind and there will be no mistakes made in manually entering data from paper to digital records, a previously time-consuming and error-prone process.

But tablets bring their own challenges, especially in an environment with patchy network coverage (made worse by inopportune undersea cable cuts) and power outages. Anyone with a phone or laptop is familiar with the hoarding of power when it is available – now imagine walking around for 12 hours a day using your device non-stop. 

Assuming the tablets are kept charged and connected to the network, Ubos should be able to receive and monitor the data in a near real-time fashion, catching and correcting errors as they go.

According to Dr Madina Guloba, a statistician at the Economic Policy Research Centre, a think-tank at Makerere University, who is also a member of the National Census Commission, Ubos is using a live dashboard to track all enumeration data as it comes in from the field.

But we probably won’t know how well data collection went – apart from anecdotal click-baity mishaps about dogs and enumerators indebted to respondents – for many months.

Eventually, another administrative report will be drafted, outlining the challenges faced and recommendations for census 2034, and by the time it comes out we will all be caught up by something else entirely.

Show me the data
Even then, the quality of the data will be difficult to assess, because it is notoriously difficult to access census data in Uganda whether by government institutions, independent researchers or interested members of the public.

After several years Ubos posted parish-level aggregates for many of the variables from the 2014 census, but while the microdata exists at a lower level – say at the level of the LC1 or even individual – these are hard to access, and certainly not available countrywide.

It is entirely up to government to determine whether and how to release anonymised microdata, and census data should be treated with great care, as enough information is collected to make individuals and households identifiable. 

Researchers can request access to subsets of the census data at the village or individual level, but anecdotal evidence from research firms in the country suggests these efforts are only sometimes successful, and often after a fee has been requested.

A more transparent process would help, with the terms for payment clearly stated, if they are indeed legally mandated.

Perhaps even more important is the sharing of census data within the government and the bureaucratic capacity to use these data for planning at the national and sub-national levels.

Does your city or local government use census microdata for planning the location of new schools or roads, or for emergency relief? Were these data used, for example, to identify the areas most in need of government assistance during Covid-19 relief efforts?

Do local governments even have access to village-level census data? The answers to these questions will help us determine the returns to and efficiency of this costly but essential data collection effort.

At the moment, the answers depend on whom you ask. Mr Nathan Itungo, the NRM MP for Kashari South Constituency, says the 2014 census revealed a large student population in Bukiro and Wanyamayembe sub-counties in Mbarara Municipality, which prompted the government to build two new schools. 

Mr Godfrey Luyomba, the NUP Speaker of Nakawa Division in Kampala District, has a different take. He says despite the last census showing that 380,000 people lived in the division, it has only one public secondary school apart from the less accessible one inside Luzira Prison, and few public health centres. 

“This means that service delivery in terms of health and education within Nakawa Division remained unchanged after the 2014 census,” he said. “That’s why some members of the community are even hesitant to take part in this year’s census. Many are asking how they will benefit from it.”

Counting priorities 
The census is not just a set of statistics about those living within a particular set of borders. It is also a reflection of the state’s priorities and capabilities. It reveals how the state defines social categories – for example, the set of ethnic and religious groups that it seeks to quantify.

It shines a light on the social, economic, and political priorities of the government. It also reflects the extent of learning – or mistake repeating – that has taken place, from past censuses and population enumeration in other contexts. It is a sort of diagnostic of the functioning of the state itself. 

In a few months we will presented with exact figures and percentages, say the number of inhabitants of Kotido District, the percentage of households that have gone a whole day without eating, and even the percentage of people who experience suicidal thoughts. These numbers and figures will seem precise and definitive. 

Statistics have a way of appearing authoritative and unassailable. But bear in mind that these figures are in fact approximations whose proximity to the truth is a function of many factors upon which counting stands or falls.

How well the census is conducted, the quality of the data produced, and by whom they are used and for what, are not questions relevant only to the statistician, but to every Ugandan for whom being counted could serve the public interest, or amount to an hour of lost time.

It is less likely that the materials used in this year’s census will be found being used as food wrappers but the integrity of the data, like the taste of the rolex, will be in the eating.

Additional reporting by Joyce Diana Nakato. 

Dr Platas is an assistant professor of political science at the New York 
University in Abu Dhabi

[email protected]