In February 2006, the first international meeting to discuss survey methods in complex settings was held in London. Following that meeting, the conference organisers, with support from the open access journal Emerging Themes in Epidemiology, published 10 papers about survey methodology.
On 4 and 5 June 2007, a second meeting on this topic was organised jointly by the London School of Hygiene and Tropical Medicine and the Centre for Research on the Epidemiology of Disasters. About 70 delegates from many countries met in Brussels. Speakers considered challenges for conducting cluster surveys. Methodological issues faced by a recently completed survey of the Iraqi population in Jordan were presented and how adaptive cluster sampling (ACS) was used to take into account extreme geographical clustering. Adaptive ACS designs are frequently used in ecology and use the selection of initial random (unrestricted or restricted) samples of plots, areas or transects that allow for inclusion of all relevant observations in the vicinity of the initial sample.
For more traditional cluster designs used in health research, sample size estimation is particularly important as such surveys are expensive and time consuming. However, this requires an estimate of the rate of homogeneity as a measure of the level of cluster within the population. We were shown an approach whereby information about clustering from previous surveys was used to improve the efficiency of a cluster survey in Uganda .
Retrospective mortality surveys during humanitarian crises are an increasingly important component of the humanitarian response as they provide crucial information about the health status of the affected population. However, due to their retrospective nature, they are only able to provide historical mortality rates, typically for two or more month previously. A novel approach was proposed whereby population estimation and exhaustive sampling of selected clusters could be used as a rapid way of estimating exiting mortality rates. A pilot study using this approach in Darfur, Sudan, gave similar results when compared to the standard EPI (30 clusters of 30 individuals: 30×30) design, although a there remain some theoretical concerns about the interpretation and further field testing of this design was proposed. While surveys are necessary in many contexts, prospective surveillance is often preferable and the experience of establishing an early warning disease surveillance system in Banda Aceh, Indonesia, following the South Asian Tsunami was presented as an example of a developing field of disaster epidemiology.
Like mortality surveys, nutritional status during humanitarian emergencies is often done using the EPI two-stage cluster sampling design. However, the FANTA project used mathematical modelling which indicated certain advantages of a 33×6 and 67×3 Lot Quality Assurance Sampling design over this tradition 30 X30 design. We heard about field testing of these designs and how they gave similar estimates to the traditional design, but at much lower cost and time.
In the non-emergency setting, there was a session about Respondent Driven Sampling. This is a new and rapidly developing method created specifically to sample difficult to reach populations. Key individuals from target populations are identified and asked to recruit a number of other similar individuals from their social network. Not only does this achieve deep penetration into communities, the method also results in a probabilistic sample. Examples of this methodology were presented from a study of injecting drug users in India and health problems among child labourers in alluvial diamond mines in Sierra Leone. Further aspects of survey implementation were also discussed, such as the use of audio-computer assisted self-interviewing (ACASI) for studying risk behaviours among adolescents in Mexico  and the use of qualitative evaluation methods to enrich the interpretation of quantitative methods.
It is hoped that following the success of these meetings, similar events will be held each year.
1. Kish L (1965) Survey Sampling. New York: Chichester, John Wiley and Son.
2. Cooley P, Turner C (1998) Implementing Audio-CASI on Windows Platforms Computers in Human Behavior 14: 195-207.