Survey Statistics · Small Area Estimation · Data Science & AI

I turn survey and multi-source data into estimates people can trust, and build the software that makes them reproducible.

Twenty years of small-area estimation, survey methodology, and official statistics: from the WHO/UNICEF immunization estimates for 195 countries to state- and county-level crime estimates for the Bureau of Justice Statistics. I consult through Samplics LLC, and every number I deliver can be re-run from its inputs.

Taking engagements — Q4 2026/Q1 2027
Small area estimation Data fusion Survey methodology Multi-source data integration Reproducible analytics Causal inference Python R svy Rust
svylab.com / lab / nhanes_2017
Question
"Mean systolic BP among adults with diabetes, by age group."
svy-agents, translation
import svy

design = svy.Design(stratum="sdmvstra", psu="sdmvpsu", wgt="wtmec2yr")
sample = svy.Sample(data=nhanes, design=design)

# mean SBP among adults with diabetes, by age group
sbp = sample.estimation.mean(
  y="sys_bp",
  by="age_group",
  domain="diabetes == 1",
)
Result · Taylor-linearization 95% CI
18–39
122.4
40–59
131.7
60–79
138.2
80+
142.9
provenance: model · prompt · code · weights · CI audited
20
Years turning multi-source data into local estimates
195
Countries: multi-source data integration (WUENIC)
400K+
svy & Samplics downloads (Python)
8
National HIV surveys designed (PHIA, Sub-Saharan Africa)
UNICEF WHO Westat NORC CDC / PEPFAR Statistics Canada Gates Foundation Gavi

About

Statistician by training.
Builder by instinct.

I'm Mamadou S. Diallo, a Ph.D. statistician based in Hoboken, NJ. For two decades I've turned sparse, multi-source data into defensible local estimates: small-area estimation, survey methodology, and the integration of survey, administrative, census, and geospatial data, across immunization, HIV, chronic disease, and official statistics.

Designed the sampling and estimation strategy for the Population-based HIV Impact Assessment surveys (8 countries, with ICAP and CDC/PEPFAR). Led the WHO/UNICEF immunization estimates (WUENIC) for 195 countries at UNICEF, and developed sampling methodology for Statistics Canada's national health surveillance system (CCHS) earlier in my career. Creator of the open-source svy ecosystem (400K+ downloads with its predecessor, Samplics, JOSS).

I also build the infrastructure evidence runs on. Solo architect of svyLab, a multi-tenant analytics platform with architectural sandbox isolation, lifecycle and classification governance, and AI-assisted analysis with full provenance. Built so that survey-correct estimation is reproducible and auditable by construction: every result carries its model, code, weights, and CI.

My long-term goal is to modernize how survey data gets analyzed: open, reproducible, survey-correct tooling as a genuine alternative to SAS, SPSS, and Stata. The svy libraries are open source and a career-long commitment — I maintain them the way statisticians maintain CRAN packages. Day to day I consult through Samplics LLC: sample design review, weighting and variance estimation, small-area estimation, and capacity building for statistics offices, health agencies, foundations, and research institutes. I am taking engagements for Q4 2026/Q1 2027.

Education

Ph.D. Statistics, Carleton University
(advisor: J.N.K. Rao)
M.Sc. Statistics, Université Laval

Affiliations

ASA (since 2010) · AAPOR
Guest Editor, JSSAM 2025

Languages

English & French: fluent in both speaking and writing

Based in

Hoboken, NJ: open to remote and on-site engagements

Taking engagements for Q4 2026/Q1 2027: small-area estimation, survey design, weighting and variance estimation, and reproducible survey analysis.

Services

How I work

I work with health agencies, statistics offices, foundations, research institutes, and survey firms running household and population surveys, in defined engagements, each with a fixed scope and timeline. These are the seven I take most often; if your problem sits between them, describe it and I'll scope it.

Sample Design Review

Up to 2 weeks

Independent review of a planned or in-field design: frame, stratification, sample size calculation and allocation, precision under realistic response. Written memo with prioritized, defensible recommendations.

Discuss this engagement →

Weighting & Variance Estimation

Up to 3 weeks

Base weights, nonresponse adjustment, calibration, and replicate or linearized variance estimation — delivered as documented, re-runnable code with a methodology annex fit for publication or audit.

Discuss this engagement →

Small-Area Estimation Pilot

4–6 weeks

Subnational estimates for one priority indicator: Fay–Herriot, unit-level, or spatial models, with model diagnostics, uncertainty, and disclosure review. Scoped to prove the approach before committing to a programme.

Discuss this engagement →

Proposal & Bid Statistical Support

3–5 days

Sampling and statistical methodology sections for competitive bids, with named-personnel credentials and power calculations. Built for firms responding to federal or multilateral solicitations on deadline.

Discuss this engagement →

Training & Capacity Building

2–5 days

Workshops for statistics offices and research teams: sampling design, weighting, small-area estimation, reproducible survey analysis in Python or R. Delivered in English or French.

Discuss this engagement →

Embedded Senior Statistician

6+ weeks

Two to three days per week inside a team to unblock a live estimation workstream, or to hold methodological continuity during a hire.

Discuss this engagement →

Modernization: Legacy to Open Source

Scoped per program

Move a survey program's weighting and estimation from SAS, SPSS, or Stata to open, reproducible Python or R, validated cell by cell against your published tables, and handed over as documented code your team can run, as in my Python port of AHRQ's MEPS workshop. AI-assisted only where its output can be checked.

Discuss this engagement →

Every engagement is defined in writing before it starts — deliverable, timeline, fee — and every number delivered can be re-run from its inputs.

That last part is not a promise, it's infrastructure. I build svy, the open-source Python library for survey statistics (400K+ downloads with its predecessor, Samplics), and svyLab, where every result persists with its model, code, weights, and CI. Clients get documented, re-runnable code — not a spreadsheet of final numbers.

Selected work

Platforms, methods, and the studies they shape

A selection of the platforms I've built, the methods I've developed, and the population-based studies that ground both.

Small area estimation

Small area estimation: research and practice

Ph.D. thesis on small area estimation under skew-normal nested error models (Carleton, advisor J.N.K. Rao), extended in the Scandinavian Journal of Statistics (2018). Applied small area estimation to food insecurity (FIES) in Senegal with ANSD, and taught to national statistics offices in Burundi, Senegal, Ethiopia, and Vietnam: Fay–Herriot and unit-level models for subnational poverty and food-security estimates.

Machine learning

Population estimation from satellite imagery

At UNICEF, led research in Chad and Mozambique estimating small-area target populations for immunization from satellite imagery, GRID3 and WorldPop layers, and microcensus data, with representation learning and uncertainty quantification. Published in Scientific Reports (2022), with the University of Edinburgh.

Analytics engine

HILDX: Healthy Illinois analytics engine

At NORC, led development of the analytical engine behind the Healthy Illinois Analytics Platform, which serves county- and community-level small area health estimates for all 102 Illinois counties alongside American Community Survey and survey microdata, under disclosure control.

Multi-source data fusion

WUENIC: global immunization estimates

At UNICEF, led the annual production of WHO/UNICEF immunization coverage estimates for 14 vaccines across all 195 countries. Integrated administrative reporting (DHIS2), household-survey microdata (DHS, MICS), and programme surveillance under GATHER reporting standards, with fit-for-purpose assessments across data modalities.

Open source · 400K+ downloads

svy

Python library for the whole survey workflow: design, weighting, estimation, and regression, with a Rust variance engine and Polars data layer, validated against R's survey package to six significant digits and by reproducing published statistics from NHANES, the Ethiopia DHS, and Statistics Canada's Labour Force Survey. Successor to Samplics (JOSS, 2021); together they have passed 400K downloads.

Case studies →
Open source · Small area estimation

svy-sae

Small area estimation at census scale: Fay–Herriot, unit-level EBLUP, and Molina–Rao EBP for poverty and inequality indicators, on JAX and validated against R's sae package. A 2.5-million-household census runs in about 16 seconds.

Documentation →
AI · LLM agents

svy-agents

Turns a plain-language question about a survey into a typed, reproducible analysis. Generated code runs in a sandbox, independent runs and model providers are compared numerically, and every reported number is filled from computed results, never typed by the model. Its evaluation suite caught a structured-output setting that collapsed plan quality from 20/20 to 2/20 while every output stayed schema-valid.

U.S. federal · BJS release

NCVS subnational crime estimates

At Westat, with R.E. Fay, developed model-based crime-rate estimates from 15 years of National Crime Victimization Survey data for states (1999–2013) and for the 65 largest counties and 51 largest metro areas, combining time-series and small area estimation. The Bureau of Justice Statistics released them as developmental subnational estimates in December 2015.

Population-based research

PHIA: HIV biomarker surveys, 8 countries

At Westat, designed the sampling and estimation strategy for the Population-based HIV Impact Assessment surveys across Cameroon, Côte d'Ivoire, Malawi, Namibia, Tanzania, Uganda, Zambia, Zimbabwe, in collaboration with ICAP at Columbia and CDC/PEPFAR. Multi-stage probability samples for HIV prevalence, incidence, viral-load suppression, and ART coverage at national and subnational levels.

Platform · 5 design partners

svyLab

Sole architect and developer of a multi-tenant analytics platform for health-data evidence generation: architectural sandbox isolation, lifecycle and classification governance with audit logs, and AI-assisted analysis where every output persists with full provenance — model, prompt, code, tokens, cost, timestamp. Opened to a first cohort of five design partners in September 2026. This is the infrastructure behind the reproducibility guarantee on my engagements.

Platform deep-dive →

Publications

Selected publications

Peer-reviewed work in observational and survey methodology, small area estimation, population health, and machine learning.

Population Health & Immunization

Get in touch

Let's talk.

I consult through Samplics LLC on sample design, weighting and variance estimation, small-area estimation, and statistical capacity building — for statistics offices, health agencies, foundations, research institutes, and firms bidding on federal or multilateral solicitations. Taking engagements for Q4 2026/Q1 2027, remote or on-site, in English or French.

Tell me what you're trying to estimate and by when. If it's a fit I'll come back with a scope, a timeline, and a fee; if it isn't, I'll say so. I respond to all inquiries within two business days.

I am also open to senior full-time roles in the US: principal statistician, survey methodologist, or data scientist, especially where statistical rigor meets production ML and AI systems.

Consulting one-pager (PDF)

Engagements, methods, and selected work on one page. Full CV available if you need it.

Send a message

I'll respond within two business days.