Bridging gaps in school completion statistics with the ABC (Adjusted Bayesian Completion Rates) model

• In Event: Better data for SDG 4: recent methodological developments

Proposal

SDG target 4.1 on school completion and target 4.5 on equal access increased interest in using household survey or census data to monitor education indicators. Indeed, there is no alternative for some indicators, such as completion rates for specific population groups, since administrative data on graduates by population characteristics are rarely available. Survey data bring their own challenges. Many household surveys are conducted every three to five years and the results released at least one year later, leading to the ‘latest available’ data often being several years old.

For several indicators, multiple surveys are available and may provide conflicting information. The 2016 Global Education Monitoring (GEM) Report raised the question of reconciling the different sources (UNESCO, 2016, Box 14.2). Simply averaging estimates ignores relevant information: Some sources may systematically result in lower or higher estimates relative to others, reflecting differences in sampling frames or how questions are asked. Some sources may show greater variability due to small sample size or other, non-statistical issues that make them less reliable. Some respondents provide information retrospectively and the time that has lapsed increases the risk of errors that need to be corrected.

The international health community faced a similar challenge in measuring indicators, such as under-5 mortality or maternal mortality rates, based on multiple sources. The UN Inter-agency Group for Child Mortality Estimation adopted a consensus model to generate annual estimates for under-5 and neo-natal mortality in each member state. The Inter-Agency Group for Maternal Mortality Rates followed a similar process.

Here, we introduce an approach that builds on these models for health indicators, but is fully adapted to estimating school completion rates. We present the structure of the model and how it addresses a number of specific challenges arising in the context of the completion rate indicator, present a first set of estimates and demonstrate their robustness and superiority over more simplistic approaches. A key output of the model are ‘nowcasts’ of current school completion rates that consistently take advantage of all available survey information.

Author

• Bilal Barakat, Global Education Monitoring Report (GEMR)