What district planners get from a household survey

A planning officer at a district headquarters has the national household survey open on his desk, because the district development plan is due and the budget conference is three weeks away. He needs one number: how many households in his district are poor. His district appears twice — once in the list of districts the survey covered, and once in the table filing it into a subregion with a dozen neighbours of very different fortunes. That subregional figure is the only one he may quote, and it is not what he was asked for.

Multiply that desk across the country. The district health officer deciding where the next outreach team goes, the programme manager choosing which sub-county a nutrition project enters, the council working out where the vendors it is about to license actually sleep.

The machinery behind that report is genuinely good, and it deserves saying plainly first. The Uganda National Household Survey 2023/24, the eighth the Uganda Bureau of Statistics has run since 1999/2000, drew 1,735 enumeration areas, selected 17,160 households and interviewed 15,813, a response rate of 92 per cent (UBOS, 2025). The 2022 Uganda Demographic and Health Survey did better still: of 20,481 households selected, 19,758 were interviewed, 98.6 per cent (UBOS, 2023). Fieldwork is not where this breaks.

WHAT THE SAMPLE WAS BUILT TO ANSWER

A survey answers the question its sample was designed for and no other. The UNHS 2023/24 covered all 135 districts and 11 cities, yet its appendix states it "is designed to produce representative estimates for the poverty indicators for the country as a whole, for the urban and rural areas separately, for each of the 15 statistical regions" (UBOS, 2025). The 2022 UDHS made the same choice, with 14 sub-regions plus a refugee domain (UBOS, 2023). Coverage and estimability are different properties, and most disappointed users stand in the gap between them.

Precision thins even at the level the surveys do support. In the UNHS's own sampling error tables, the national estimate for people who sought treatment when ill or injured carries a coefficient of variation of 1.4 per cent; for Kampala the same indicator carries 7.96 per cent, on an interval of 15.2 to 20.7 per cent (UBOS, 2025). One is a figure a ministry can plan against; the other holds two opposite conclusions. A district estimate from that sample would be wider still, which is why slicing results below the design domain manufactures data rather than reads it.

THE QUESTIONNAIRE IS THE MEASUREMENT

The second thing deciding whether a survey can be acted on is fixed before anyone reaches a village: the wording and length of the questions. The clearest demonstration in the region is a World Bank field experiment in Tanzania, reported by Beegle, De Weerdt, Friedman and Gibson in 2010, which assigned eight consumption questionnaires at random to 4,000 households between September 2007 and August 2008. Same households, same year, same poverty line applied afterwards. What varied was the instrument, and measured poverty varied with it by more than nineteen percentage points.

The questionnaire is part of the measurement, which bears on how Ugandan rounds get compared. The UNHS food list grew from 73 items in 2012/13 to 144 by 2023/24, the report's main adjustment for the round (UBOS, 2025), and the Tanzanian experiment found that collapsing a 58-item food list into 11 categories, recall period held constant, cut reported food consumption by almost 32 per cent. None of that indicts Uganda's numbers, which came from the more detailed instrument; it is a reason to establish what changed in the questions before announcing what changed in the country.

TIMING DECIDES WHETHER ANYONE READS IT

Ugandan household surveys tend to arrive after the decisions they were built to inform. Fieldwork for the UNHS 2023/24 ran from March 2023 to February 2024; the report is dated May 2025, fifteen months after the last interview (UBOS, 2025). It collected its data for monitoring "the National Development Plan III (NDP III)", which ended with FY2024/25, while NDP IV, covering FY2025/26 to FY2029/30, was already set by the National Planning Authority. Editing, weighting and poverty-line construction are slow because they are done properly, so haste is not the answer; working backwards from the plan the data must serve is, as is releasing the indicators that decision turns on ahead of the full volume.

WHO THE FINDINGS BELONG TO AFTERWARDS

The UNHS lists among its objectives providing data "for research, academia, training institution and Civil Society Organization, and public debate", and its documented process opens with user needs assessment meetings held before the questionnaire is written (UBOS, 2025). That intent becomes real only when the microdata leaves the building, because an institution confined to published tables is confined to somebody else's questions. The alternative exists here already: the Uganda National Panel Survey, run by UBOS within the World Bank's Living Standards Measurement Study, is published as a public use file any researcher can download (World Bank Microdata Library, 2020).

RWANDA SIZED ITS SAMPLE FOR THE DISTRICT

Rwanda's seventh Integrated Household Living Conditions Survey is almost exactly the size of Uganda's: the National Institute of Statistics of Rwanda drew 1,674 enumeration areas and nine households in each, targeting 15,066, against Uganda's 1,735 areas, nineteen of them never listed, and ten households each for 17,160. What differed was the design goal. Rwanda's strata are its 30 districts rather than grouped regions, and NISR sized the survey by computing "the sampling errors and 95% confidence intervals for district level poverty rate estimates" from the previous round, then raising it "to further improve the precision of the district-level results" (NISR, 2024). Its frame came from the census of 2022, a year old at fieldwork, where both Ugandan flagship surveys drew on the census of 2014.

Execution matched intent. Of the 15,066 targeted households, 15,054 were interviewed, above 99 per cent, reached by replacing non-responders during fieldwork rather than weighting around them afterwards, and the questionnaire was validated in June 2023 with ministries and international partners (NISR, 2024). The results were published as national and thematic reports and as a district presentation for each of the 30 districts (NISR, 2025). A Rwandan mayor is handed a document about the place they are answerable for, while Uganda's planning officer is handed a subregion.

WHAT IS THE WAY FORWARD?

Several changes would close the distance. UBOS should state on the cover of every survey, not in an appendix, the smallest geography at which its estimates hold. The National Planning Authority and sector ministries should commission surveys against the calendar of the decisions they feed, and accept phased release of key indicators. Districts that genuinely need their own numbers should stop slicing national samples and fund booster samples instead. Organisations commissioning their own research should treat question design as the measurement decision Tanzania's experiment shows it to be. And every survey paid for with public money should end with anonymised microdata in a public catalogue, because a dataset only its author can open is a report, not evidence.

None of this needs a new institution or a larger sample; the response rates are already extraordinary and the field teams trained. What the planning officer needs is not a better survey than the one lying open on his desk, but one designed, from the first sampling decision onward, to answer the question he is actually asked.