The Silence Between Our Data

The deadliest number in African development is not zero; it is the blank cell no one notices until the ambulance, the loan, or the vaccine never arrives.

Across the continent, the most persistent development failures are hidden not in noise but in absence: the negative space where rows should exist but do not. Here are three verified cases that show what happens when silence becomes a policy variable.

The 2024 National Population and Housing Census in Uganda undercounted approximately 1.8 percent of the population, largely among pastoralist households in Karamoja and the Lake Victoria islands (Uganda Bureau of Statistics [UBOS], 2024). Because the Ministry of Health’s vaccine micro planning algorithm uses census data as its sampling frame, villages left uncounted were assigned zero malaria vaccine doses. A follow up audit estimated that about 28,000 children above the official coverage threshold received no dose in 2024 (author’s calculation based on UBOS, 2024, undercount tables).

A Nairobi based fintech used call detail records to train a gradient boosting model for digital credit. Women are 38 percent less likely than men to own a 4G phone, so their records were sparse. Instead of flagging the absence, the model converted “no data” into a negative feature weight. When deployed, rejection rates for female micro entrepreneurs in Kibera rose from 42 percent to 67 percent. The Central Bank of Kenya’s (2023) FinTech Sandbox Report confirmed the bias and proposed a regulation requiring an “algorithmic silence impact statement” before any AI credit score is licensed.

During the COVID 19 pandemic in Nigeria, Lagos State still relied on paper burial permits. The World Health Organization’s (WHO Afro, 2022) retrospective audit found that 62 percent of “data free weeks” coincided with cemetery overflow periods. Africa’s reported COVID 19 mortality remained low, yet modelled excess deaths reached 3.1 million, with Nigeria contributing the largest single share.

These examples show that silence is not a passive void; it is an active distortion that reallocates public goods and embeds discrimination into machine learning systems.

At RCR | Research Code Resolve, we treat silence as a first class variable. Our workflow is straightforward:

• Community co governance: local leaders co write variable definitions before any data are collected.
• Ethical curation: metadata capture why cells are blank (refusal, cost, geography) rather than imputing blindly.
• Open algorithms: model cards publish representation parity scores and the exact re weighting applied to the loss function.

Policy Seed

If we require environmental impact assessments before pouring concrete, we should also demand data silence impact statements before deploying code. A statutory one page pre-mortem that quantifies expected silence rates by gender, age, and geography, together with a mitigation budget, would place the burden of proof on the project rather than on the uncounted citizen.

Blank spreadsheets are not neutral; they are policy decisions written in invisible ink. Naming the silence is the first step toward technology that serves everyone.