The numbers we use in development, and most of what we think of as facts, are actually estimates. It’s time for a data revolution
The Guardian, 31 January 2014
You know a lot less than you think you do. Around 1.22 billion people live on less than a $1.25 (75p) day? Maybe, maybe not. Malaria deaths fell by 49% in Africa between 2000 and 2013? Perhaps. Maternal mortality in Africa fell from 740 deaths per 100,000 births in 2000 to 500 per 100,000 in 2010? Um … we’re not sure.
These numbers, along with most of what we think of as facts in development, are actually estimates. We have actual numbers on maternal mortality for just 16% of all births, and on malaria for about 15% of all deaths. For six countries in Africa, there is basically no information at all.
In the absence of robust official systems for registering births and deaths, collecting health or demographic data, or the many other things that are known by governments about people in richer countries, the household survey is the foundation on which most development data is built. Numbers from the surveys are used to estimate almost all the things we think we know – from maternal mortality to school attendance to income levels. Household surveys are run by governments or by external agencies such as the World Bank, USAid or Unicef.
But it’s a shaky foundation. First, to make the survey representative of the population, you need to know a lot about the population to make a good sampling frame. This knowledge comes from a population census. But only around 12 of the 49 countries in sub-Saharan Africa have held a census in the past 10 years. So there might be large population groups missing – especially in countries undergoing rapid change. There are likely to be big urban informal settlements, for example, which are not included in the most recent census, and therefore don’t exist for sampling purposes. They also don’t happen very often – 21 African countries haven’t had a survey in the past seven years.
And they’re not all done in the same way, which makes comparing countries or combining data from different countries very difficult – and illustrates how hard it is to know the “real” number. There are, for example, seven perfectly acceptable ways of asking questions in surveys about how much people eat. A recent experiment by World Bank researchers in Tanzania, comparing results from the different methods, found that estimates of how many people in the country are hungry varied from just under 20% to nearly 70%, depending on the method chosen.