Exploring the Hyak Real-time Health and Population Measurement Platform in Tanzania using DHS Surveys and HDSS Data

Hyak is an idea for a statistical data collection and analysis platform that leverages the advantages of traditional sample survey and demographic surveillance methods. The overall goal is to produce a varying and diverse array of indicator values in near real- time that describe large populations and can identify differences across relatively small distances in space. To accomplish this Hyak combines and builds on existing knowledge and experience with sampling, follow-up, fieldwork, spatial modeling, estimation and other statistical methods. This project explores the foundational ideas of the Hyak system. The key aim is to inves- tigate one of the important pre-conditions necessary to make Hyak possible – whether or not traditional sample surveys and demographic surveillance methods produce similar estimates of a key demographic indicator, and further, if the differences can be exploited to improve the usefulness of the indicator. Addition aims address spatial models, a sim- ulation study of the whole Hyak idea and further development of detailed ideas about innovative sampling methods and statistical approaches to integrating data from several structurally difference sources to estimate demographic indicators. In our main comparison we focused on child mortality (5q0) using data from the 2010 de- mographic and health survey (DHS) of Tanzania and from two health and demographic surveillance system (HDSS) sites in Tanzania – Ifakara and Rufiji. Over the period 1990 – 2000 estimates of child mortality from the two data sources are generally similar but dif- ferent in useful ways. The HDSS estimates are accurate (low bias) and precise (small vari- ance) measurements for comparatively small, geographically-defined populations, and the DHS estimates are less accurate and much less precise but representative of large populations. Altogether this is exactly what we need for Hyak to work. We developed several spatio-temporal smoothing models of the DHS data for the regions of Tanzania, and we developed one ‘merged’ model that combines data from DHS and HDSS in the regions where they are close to each other. These clearly demonstrate the utility of smoothing and integration of data from multiple sources. The key result for the regions of Tanzania is to dampen noisy fluctuations in time and space and greatly reduce variance, and in areas near HDSS sites, to adjust overall estimates to more closely match the HDSS. We present additional ideas relating to spatio-temporal modeling of survey data, var- ious sampling methods, informed sampling in particular, and statistical approaches to integrating data from several sources into unified models for demographic indicators.