Data collection where the network ends

It is a little after five in a parish in Kotido, and a field officer has finished her last household. The answers are on the tablet, and the tablet is telling her what it has been telling her since the middle of the morning — that there is no network. A day of interviews sits in local storage on a device whose battery will not see the evening out, and her supervisor in Kampala will see none of it tonight. So she does what the job quietly requires, and walks back up the track towards the trading centre, where somebody once got a signal on the rise behind the shops.

Multiply that walk across the country, and across every programme that runs on what field teams carry home.

The village health team member in Amudat writing referrals into a book because the application stopped responding. The extension officer at the edge of the lake in Buliisa. The nutrition monitor moving between blocks of a refugee settlement in Kiryandongo. The enumerator in Kaabong whose supervisor sees no completed form until Thursday. None of them chose to work offline — the places they were sent to chose it for them, and the tools they were issued were designed somewhere that never has to think about it.

What Uganda can already do here is considerable, and it deserves to be said first. The 2024 National Population and Housing Census was the country's first digital census, and the Uganda Bureau of Statistics ran it on 121,240 handheld tablets carrying the enumeration application and digital area maps, with data synchronised from the device straight to the server as each interview finished — which, UBOS reports, removed the data entry and preliminary editing stages altogether. The Uganda National Household Survey 2023/24 was collected the same way, on tablets running Survey Solutions, with consistency checks running in the background during the interview. A country that can do that does not have a technology problem.

WHERE THE MAP STOPS AGREEING WITH THE PLAN

Coverage is reported in the units of a telecoms market; field work happens in the units of a map. The Uganda Communications Commission's annual communications sector report for 2024 sets both side by side.

Ninety-eight per cent population coverage is the figure that gets repeated, and the Commission is careful to note that it is an aggregate across all operators rather than any single network, fully realisable only once national roaming is enforced. The land area column is the one a programme manager should be reading. Against a total land area of 241,038 square kilometres, the Commission puts the 2G coverage gap at 54,234 square kilometres and the 4G gap at 91,112. Surveys, immunisation outreaches and monitoring visits are allocated by enumeration area and parish, not by population density — so a sampling frame drawn properly keeps sending teams into exactly the fraction of the country the coverage headline leaves out.

POWER IS THE CONSTRAINT NOBODY BUDGETS FOR

Underneath the signal problem sits an older one. The same report records that more than 780 network-ready towers in Uganda cannot be switched on for want of electricity, and that power costs the telecom sector about USD 60 million a year because operators fall back on generators and solar rather than the grid. The grid that will not reach a tower will not reach a field team either. The 2024 census found 25.3 per cent of Ugandan households using grid electricity for lighting and 28.1 per cent using solar, with the rural and urban grid figures far apart — 9.1 per cent against 49.0 per cent. For a team working a week out of a district town, charging is not a convenience question. It is the data pipeline.

A DROPPED UPLOAD COSTS MORE THAN A DAY

The fallback, when the tool will not hold, is paper, and paper is measurably worse. In a randomised comparison of electronic and paper questionnaires in an African field setting, Dillon and colleagues reported in the Journal of Clinical Epidemiology in 2014 that the electronic instrument produced 0.17 errors per 100 questions against 0.73 on paper, and no major errors at all against 0.59 per 100 questions, while the higher start-up cost was recovered inside six months. The cost is not only in the data — it is in the double entry that follows, and in the fact that nobody at headquarters can act on a figure still travelling in a rucksack.

Where the design assumes disconnection instead, the record holds. Kalibbala and colleagues, in PLOS Global Public Health in 2022, reported on a birth defects surveillance system across four Kampala hospitals that documented 110,752 births from August 2015 to December 2018, of which 110,573, or 99.8 per cent, went straight into the electronic system and only 179 onto paper. Their account of why is exactly to the point: despite poor connectivity, the software let staff collect on the tablets and submit later once signal was restored. The paper is plain about what wears out, noting that touch screen sensitivity, working speed and battery life all declined over time and slowed collection down.

KENYA STAFFED THE TECHNOLOGY AS A JOB OF ITS OWN

Kenya's 2019 Population and Housing Census, the first the country ran without paper, is instructive less for the tablets than for what was built around them. The Kenya National Bureau of Statistics records a pilot census in August 2018, a full year before enumeration, and two pre-tests in April and May 2019 — held to test map accuracy, the data capture application and data transmission, not the questionnaire alone. The field structure carried the same logic. Alongside 138,572 enumerators and 22,268 content supervisors, KNBS recruited 2,467 ICT supervisors as a separate cadre — one for roughly every fifty-six enumerators — and ran a toll-free contact centre for field teams. The devices were assembled by local universities, and data was encrypted before transmission and backed up off site. Transmission was treated as an operation with its own staff, its own rehearsal and its own failure modes.

WHAT IS THE WAY FORWARD?

The United Nations Statistics Division's guidelines on electronic data collection in censuses, published in 2019, already say most of what needs saying, and say it at the planning stage rather than the procurement stage: identify and map the places that lack electricity and internet before anything is bought, plan interviewer routing around the opportunities to transfer, establish collection stations with better connectivity where the network is poor, and provide power banks and solar chargers. Regulators could publish coverage at parish granularity, since a national percentage is not a planning instrument. Ministries and funders should budget power and devices as line items rather than overheads. Those building these systems should treat the disconnected case as the normal one — the record written locally first, validation running on the device, and transfer resuming by itself rather than starting again.

None of that is expensive next to what it protects. It would also mean the field officer on the rise behind the shops in Kotido is standing there because the view is good, and not because a day of somebody's work is still trapped in her hand.