How to identify the furthest behind?
This note gives a brief explanation of how to identify population groups belonging to the furthest behind (and furthest ahead) as well as how to interpret the results. The statistical method used for this purpose is called the classification and regression tree (CART) methodology. In short, an algorithm identifies population groups with the lowest and highest access to a number of basic services and dimensions at both individual and household levels. At the individual level, the following areas are analysed: completion of education; access to full-time employment; women’s access to health care; and children’s nutrition. At the household level the following dimensions are studied: access to basic water and sanitation; access to clean energy; and ownership of a bank account. These dimensions and services are all linked to specific SDGs.
The classification and regression tree (CART) methodology uses an algorithm to split the population into population groups with significantly different access levels based on predetermined individual or household characteristics. It does so by looking at each characteristic, separating households or individuals into groups, and repeating the analysis to consider the remaining characteristics until no further information gain can be generated by a new partition.
These characteristics are based on a combination of shared circumstances that households or individuals have in common, but no control over. The following characteristics are used in the analysis:
At the household level: wealth (belong to the bottom 40 or top 60 of a country’s wealth distribution); residence (living in a rural or urban area); and the highest educational attainment of the household head (no education, primary, secondary or tertiary education).
At the individual level: education level (no education, primary, secondary and higher); wealth (belong to the bottom 40 or top 60 of a country’s wealth distribution); age group (being aged 15-24 years, 25-34 years, over 35 years old); marital status (currently in union/married, formerly in union/married); number of children under the age of 5 (numerical variable); sex (male or female); residence (living in a rural or urban area); and mother’s education (no education, primary, secondary or tertiary education), as relevant or appropriate.
When data is available, additional characteristics such as ethnicity, religion, caste and language, are also included in the analysis.
Interpreting the classification and regression tree – access to electricity in Myanmar
The classification tree shows that 56 per cent of households in Myanmar have access to electricity. Disaggregating the sample by household wealth, it appears that those belonging to the bottom 40 of the wealth distribution have an access rate of 27 per cent, while those at the top 60 of the wealth distribution have an access rate of 75 per cent. Highest educational level in the household and residence represent the characteristics of the remaining branches of the tree. The green box shows the best-off group: urban households belonging to the top 60 of the wealth distribution, who report an access to electricity of 97 per cent. The red box, instead, shows the furthest behind group: households in the bottom 40 of the wealth distribution with at most primary education, with an access rate to electricity of only 23 per cent.
Data sources
The analysis uses the Demographic and Health Surveys (DHS) and the Multiple Indicator Cluster Surveys (MICS). The datasets are selected because of their comparability across countries, accessibility, and the rich set of questions on health, demographic and socioeconomic data that refer both to the household and individual level. DHS and MICS are publicly available for 8 countries in Asia and the Pacific region (Cambodia, Indonesia, Lao People’s Democratic Republic, Mongolia, Myanmar, Philippines, Thailand, and Viet Nam), and 12 countries in Latin America and the Caribbean region (Bolivia, Brazil, Colombia, Dominican Republic, Ecuador, El Salvador, Guatemala, Honduras, Mexico, Nicaragua, Paraguay, and Peru).
Further queries
For more information on the project, please contact: escap-sdd@un.org