[Paper Review] Space-Time Smoothing of Survey Outcomes using the R Package SUMMER

Zehang Li, Bryan D Martin+6 more|arXiv (Cornell University)|Jul 10, 2020

demographic modeling and climate adaptation23 references4 citations

TL;DR

This paper introduces the R package SUMMER, which enables Bayesian spatio-temporal smoothing of complex survey data—particularly for demographic and health indicators like child mortality—in low- and middle-income countries. It integrates survey design features with INLA-based fast Bayesian computation to produce reliable, subnational estimates with proper uncertainty quantification, validated through applications to DHS and MICS data including official UN IGME estimates for 22 countries.

ABSTRACT

The increasing availability of complex survey data, and the continued need for estimates of demographic and health indicators at a fine spatial and temporal scale, which leads to issues of data sparsity, has led to the need for spatio-temporal smoothing methods that acknowledge the manner in which the data were collected. The open source R package SUMMER implements a variety of methods for spatial or spatio-temporal smoothing of survey data. The emphasis is on small-area estimation. We focus primarily on indicators in a low and middle-income countries context. Our methods are particularly useful for data from Demographic Health Surveys and Multiple Indicator Cluster Surveys. We build upon functions within the survey package, and use INLA for fast Bayesian computation. This paper includes a brief overview of these methods and illustrates the workflow of accessing and processing surveys, estimating subnational child mortality rates, and visualizing results with both simulated data and DHS surveys.

Motivation & Objective

• Address data sparsity and unstable estimates in subnational demographic and health surveys, especially in low- and middle-income countries (LMICs).
• Develop a computational framework that accounts for complex survey designs—such as multi-stage cluster sampling—during smoothing and estimation.
• Enable accurate, reliable, and uncertainty-quantified estimation of health indicators like under-5 mortality rates (U5MR) at fine spatial and temporal resolutions.
• Support integration with existing survey analysis tools (e.g., R's survey package) and scalable Bayesian inference via INLA for efficient computation.
• Facilitate official use in global health monitoring, including the UN IGME’s subnational U5MR estimates (1990–2018) for 22 countries.

Proposed method

• Use area-level and unit-level small area estimation (SAE) models tailored for complex survey data with known sampling weights and design effects.
• Implement Bayesian hierarchical models with conditional autoregressive (CAR) priors for spatial smoothing and random walk (RW2) priors for temporal trends.
• Leverage the INLA (Integrated Nested Laplace Approximation) package for fast, accurate Bayesian inference without MCMC, enabling efficient computation on large survey datasets.
• Integrate survey design features (e.g., strata, clusters, sampling weights) via the R package survey to ensure design-based inference and reduce selection bias.
• Apply the smoothDirect function to combine direct survey estimates with spatial and temporal smoothing, using a flexible time-scale specification (e.g., five-year periods or yearly levels).
• Use the SUMMERproj class to organize and visualize posterior estimates, including 95% credible intervals, with built-in plotting functions for spatial and temporal trends.

Experimental results

Research questions

• RQ1 How can Bayesian spatio-temporal smoothing improve the reliability of subnational estimates for health indicators when survey data are sparse?
• RQ2 To what extent does accounting for complex survey design (e.g., multi-stage cluster sampling) reduce bias and improve uncertainty estimation in small-area mortality estimation?
• RQ3 Can the SUMMER package produce accurate and credible subnational estimates of under-5 mortality rates (U5MR) comparable to or better than national-level models?
• RQ4 How effective is the integration of INLA with survey design features for scalable, fast, and accurate estimation in large-scale demographic surveys?
• RQ5 What is the impact of temporal modeling (e.g., RW2) on the stability and trend detection of health indicators over time?

Key findings

• The SUMMER package successfully produced official UN IGME subnational (admin-2 level) U5MR estimates for 22 low- and middle-income countries from 1990 to 2018, extending beyond previous national-only modeling.
• The method significantly reduced estimation instability in small areas by borrowing strength across space and time, particularly where direct survey estimates had high variance due to small sample sizes.
• The use of INLA enabled fast Bayesian computation, making it feasible to fit complex spatio-temporal models on large survey datasets with proper uncertainty quantification.
• The package demonstrated strong performance in simulated data, with smoothed estimates showing improved coverage and precision compared to direct estimates.
• The temporal modeling component (e.g., RW2) effectively captured trends in U5MR and NMR over time, even when direct estimates were only available in five-year periods.
• Visualization tools within SUMMER, including faceted plots and credible interval overlays, enabled clear communication of results across regions and time periods.