research

my research lies at the intersection of bayesian methods, spatial statistics, small-area estimation, survey methodology, and official statistics. one question runs through all of it: how do we measure health, poverty, and access to services where data are sparse, contested, or expensive to collect, and how do we build tools that ministries of health and statistical offices can actually own and use?

i collaborate with the world health organization, unaids, gavi, unicef, the world bank, and national programmes, most often on hiv and immunisation coverage across african countries. i am a member of the unaids reference group on estimates, modelling and projections.

calibrated estimation from routine data

for two decades, subnational estimates of health in many african countries have rested on large, donor-funded household surveys. as that funding contracts, a pressing question is whether the routine data that countries already collect and own, through their health information systems, can carry more of the load. routine facility data are abundant and timely but biased, because the people who appear in them are not a random sample; surveys are unbiased by design but sparse and infrequent. with li-chun zhang, i develop a register-based framework that treats routine data as the primary signal and uses a sparse probability survey to calibrate its bias, rather than the other way around. the same structural idea covers two very different problems: selection bias in routine hiv testing, and denominator bias in administrative immunisation coverage. a recurring finding is that the calibration uncertainty, not the routine sample size, is what binds; that turns the design of a small anchoring survey into the high-leverage decision. this work was presented at sae 2026 and is being written up as a methods paper.

hiv epidemiology and the care cascade

i build models that combine population-based surveys, programmatic data, and routine health information systems to describe the hiv epidemic and the care cascade. recent work covers disparities in hiv burden and care engagement across 43 african countries, temporal trends in antenatal hiv testing, and the scale-up of hiv self-testing across eight national programmes. a current strand uses calibrated routine testing data to estimate undiagnosed prevalence at district level in mozambique, lesotho, zimbabwe and malawi, where probability surveys are valid only at coarser resolution.

immunisation coverage and geographic inequalities

geographic inequalities in childhood vaccination remain large, and they are sensitive to how we measure them. i work on small-area methods for immunisation coverage that draw on dhs, mics and administrative data, with attention to the assumptions hidden inside common indicators, and in particular to the denominator: the target population that administrative coverage divides by is often wrong, sometimes badly enough to push reported coverage above 100%. a recent scoping review, with colleagues at who, gavi and unicef, maps how spatial methods are used across this literature and where they remain bound to survey data.

ai and the production of official statistics

i’m interested in what happens to official statistics when machine-learning methods enter their production: at coding and classification, at the integration of administrative and alternative data, and at estimation. a methodological strand, with paul smith, frames “algorithm-assisted inference” as a continuity with model-assisted estimation rather than a clean break. a companion strand studies the same transition ethnographically, asking what concepts like accuracy, comparability and reproducibility come to mean inside statistical offices as parts of the production chain are handed to learned functions.

survey methodology and the measurement of wealth

a recurring question in my work is whether widely-used survey instruments measure what we think they measure. i’ve examined the dhs wealth index as a proxy for socio-economic status in low-resource settings, comparing it against income and consumption measures in mozambique and across north africa and the middle east, and looking at when asset-based indices and monetary poverty measures disagree.

the politics of quantification

i am increasingly drawn to questions about how numbers govern. one project, with colleagues in criminology and survey methodology, asks how excluding incarcerated people from household surveys, and reassigning them in the census, distorts official statistics on deprivation, inequality and health. from september 2026 i also co-supervise a phd project on data-driven governance in global health, studying how donor targets and model-based estimates travel into national and local decisions.