The issues that hinder development in sub-Saharan Africa are many and complex, but one factor that stands out for scientists is the dearth of reliable data on the decades of development projects there.

When a project fails, the tendency has been to move straight on to the next idea.

A lack of information on what has worked and what hasn't has contributed to a lack of accountability among donor nations, host nations and even development professionals. Donors in particular have learnt little from past mistakes, and are impatient. When a project fails, as so many do, the tendency has been to move straight on to the next idea.

Development specialists know this, and today data and analysis are prized. In this issue (page 22) we examine the early progress of one notable experiment in Africa. It involves the support of 12 African Millennium Research Villages, which are receiving a package of interventions, at a maximum cost of US$110 per person per year, tailored to lift them out of poverty and onto a sustainable path.

The approach has won support from the African governments involved and from private philanthropists, who have pledged $100 million to a charity, called Millennium Promise, that aims to expand the programme to an additional 78 villages in the next year.

The administrators of the village projects intend to measure 27 important indicators of project performance, mainly by closely monitoring the progress of some 300 households in each village.

They hope to learn three things: whether each intervention works, whether the links between various interventions can be exploited, and whether the community is ultimately better placed to manage its own future. This last involves 'softer' measures of capacity and sustainability, and will be the hardest both to monitor and to achieve.

It is early days yet — the longest-running project, at Sauri in Kenya, is just two years old — and few hard data are available so far. But it is crucial that the schemes deliver on their research goals and that they absorb lessons, positive or negative, from the data.