Statistical Modelling Consultant, Data and Analytics, DAPM, NYHQ, remote. Req# 590596

About UNICEF

If you are a committed, creative professional and are passionate about making a lasting difference for children, the world's leading children's rights organization would like to hear from you. For 70 years, UNICEF has been working on the ground in 190 countries and territories to promote children's survival, protection and development. The world's largest provider of vaccines fordeveloping countries, UNICEF supports child health and nutrition, good water and sanitation, quality basic education for all boys and girls, and the protection of children from violence, exploitation, and AIDS. UNICEF is funded entirely by the voluntary contributions of individuals, businesses, foundations and governments. UNICEF has over 12,000 staff in more than 145 countries.

Consultancy: Statistical Modelling Consultant

Duty Station: Data and Analytics, DAPM

Duration: February 1, 2026 – May 31, 2026

Home/ Office Based: Remote

 

BACKGROUND

Purpose of Activity/ Assignment:

The Maternal, Newborn, Child, and Adolescent Health (MNCAH) portfolio of the Data and Analytics Section of UNICEF New York Headquarters maintains and updates the global MNCAH database which is a key source for country- and regional-level MNCAH data. This database is used by a wide range of stakeholders for activities related to SDG monitoring, global, regional, and country programme planning, and policy advocacy. The MNCAH team is expanding the MNCAH database to align with the emerging global health landscape and priorities, incorporating administrative data, modeled data, and subnational data. This expansion will improve the timeliness and availability of the data, necessitating strategic enhancements to the global database workflow.  In addition to maintaining the global database, this consultancy will support maintenance and development of data products and data reporting processes such as SDG monitoring, Strategic Plan reporting, and other global reporting processes related to MNCAH.

The purpose of this assignment is to develop and validate country-level hierarchical Bayesian models for several MNCAH indicators for UNICEF’s reporting and monitoring. These models will evaluate the uncertainty and integrate data from country routine systems and survey data, provide annualized estimates for indicators at the country level, and forecast indicators within short time horizons.

Scope of Work:

Under the supervision and guidance of Statistical and Monitoring Specialist (MNCAH), the scope of the consultancy is to advance the estimation processes used by the MNCAH team, including developing the methodology, toolset, estimates, and validation approaches. The scope includes:

1. Model Review and Development

• Review the existing modelling approach and finalize recommendations on the temporal structures and short-term deviations, autocorrelation structures (AR(1)/ARMA), and hierarchical random effects.
• Compare time-only models, covariate-driven models, and multi-source models, assessing their suitability for different indicators and data contexts.
• Implement systematic covariate selection strategies, including Bayesian shrinkage (horseshoe priors) and hybrid methods (screening such as LASSO followed by Bayesian estimation).

2. Scaling & Validation

• Extend the modelling framework to handle both data-rich and data-sparse indicators, possibly through a dual or unified structure.
• Explore random-walk and intervention-sensitive models for indicators influenced by programmatic changes rather than covariate trends.
• Develop and apply rigorous validation strategies (out-of-sample prediction, sensitivity analysis, performance comparisons).
• Recommend the most suitable model specifications for each MNCAH indicator, ensuring adaptability across contexts.
• Identify a framework to evaluate model results, including constraing models against empirical data, and flagging anomalous results.

3. Codebase & Tools

• Convert existing JAGS models into brms/cmdstanr equivalents on Databricks to enhance efficiency, reproducibility, and integration with modern Bayesian workflows.
• Refactor existing code into a modular, reusable, and transparent codebase, including functions for estimation, prediction, and visualization.
• Develop a visualization toolkit with reusable plotting utilities for country-level estimates, raw data overlays, covariate diagnostics, and model performance plots.
• Set up a structured GitHub repository (or equivalent) with documented scripts, reproducible workflows, and annotated vignettes.