Tuesday 16th July
Wednesday 17th July
Thursday 18th July
Friday 19th July
Download the conference book
Survey research in developing countries 2 |
|
| Convenor | Dr Evelyn Ersanilli (University of Oxford) |
| Coordinator 1 | Dr Melissa Siegel (Maastricht Graduate School of Governance ) |
Research on survey methodology is burgeoning. However, most research on survey methodology is conducted in developed countries and it remains unclear to what extent best practices developed there are also valid in developing countries.
On the one hand survey researchers in developing countries do not always share the problems of their colleagues in developed countries. Response rates are for instance rarely an issue in rural Africa where social trust is high and surveys are sometimes seen as a welcome break from the monotony of daily life rather than an infringement on personal time. On the other hand the demographic structure and level of development of these countries can pose a number of specific challenges. A high number of different local languages can complicate questionnaire translation, low quality roads make it hard to reach remote villages and obtain a representative sample, ethnic tensions can make the selection of interviewers an arduous task. Some challenges such as the lack of up-to-date data on population size and composition are not unique to the development context, but are more common.
This session aims to explore the challenges involved in conducting survey research in developing countries and discuss best practices. We welcome papers on all phases of survey design and data collection. Papers may address topics such as:
- Creating sampling frames with good coverage
- Consequences of illiteracy/low literacy for questionnaire and answer scale design
- Development of standardised question wording in areas with high linguistic diversity or non-standardised scripts
- Conducting surveys in non-democratic countries
- Conducting surveys in countries with (recent) ethnic tensions or civil war
- The role of gender in survey research in patriarchal cultures
- Use of mobile phone and other modern communication technology in survey research
- Cultural challenges in partnerships with local researchers and authorities (hierarchy, post-colonialism)
Livestock ownership is an important component of rural livelihoods, particularly in developing countries, but high quality data on livestock are difficult to collect due to the nomadic and semi-nomadic nature of many pastoralists. Most national household surveys use the most recent census as the sampling frame and thus exclude those without permanent dwellings, leading to undercoverage error. In addition, we worry about the quality of the livestock data provided by those who can be reached via a household sample: household based family members may not be able to provide accurate data on livestock that they seldom see, leading to measurement error in data collected from households as well. In this study, we explore the use of a random geographic cluster sample (RGCS) as an alternative to the census frame-based sample. In an RGCS design, points in the survey area are randomly selected within strata defined by geographic characteristics, and all persons living inside circles drawn around the selected points are interviewed for the survey. Properly implemented, this design eliminates undercoverage resulting from mobile populations. This paper discusses the results of an RGCS survey to measure livestock ownership in the Afar region of Ethiopia carried out in 2012 (n=750). Benchmarking our results against a recent dwelling-based survey, we discuss the ways in which RGCS improved data quality, as well as the implementation challenges encountered by the interviewers in Afar.
Abstract
The Practicalities and Innovative Techniques of Conducting Surveys in Informal Contexts: Lessons from South Africa.
Conducting a survey is a complex process and more so in developing countries where researchers face a myriad of challenges. The absence of up-to-date and reliable data, informality of settlement patterns, local political dynamics, and responses are some of the challenges researchers need to address in order to develop a more sound methodological approach that is adaptive to the particular contexts without compromising reliability. This paper shares the experience of conducting multiple surveys in multiple contexts in South Africa as part of the African Centre for Migration and Society's (ACMS), research undertakings. It will highlight the key challenges we faced throughout the duration of surveys and how we addressed those challenges, including some innovative sampling techniques. More emphasis will be given to the development of sampling strategy in the absence or inadequate reliable existing data. The test is more critical when it comes to conducting household surveys in informal settlements (slums) in urban areas; implementing any strategy is even more complicated in this context. What options are available to sample from a densely populated informal settlement (slum)? What are the practicalities or how do we implement the available options? All these questions and more will be dealt in the paper.
For over 60 years, researchers have documented multiple sources of survey error yet discussions of survey quality often still focus on sampling error alone. We investigate problems that this may pose even for well designed probability surveys in developing countries, especially after a natural disaster like the 2010 earthquake in Haiti.
We conduct a secondary analysis of Haitian presidential election surveys published by the Bureau de Recherche en Informatique et en Developpement Economique et Social (BRIDES) in 2010 and 2011. We interviewed the director and staff of BRIDES in Haiti about their sophisticated sampling methodology. Their probability samples were unusually large (6,000) with a margin of error of just 1.25 percentage points. This allowed BRIDES to correctly predict Michel Martelly to win over establishment candidate Mirlande Manigat who was favored to win by other survey organizations. However, the prediction of 53% for Martelly fell significantly short of his 67.6% landslide victory, a discrepancy of more than 14 percentage points. We utilize district-level variation in prediction accuracy to compare the impact of the earthquake (distance to the epicenter) from other factors (e.g., urban vs. rural population, age, and gender structure).
The paper's purpose is assessing the potential impact of various sources of error on survey accuracy with the goal of broadening the discussion of survey quality in post-disaster areas. Our aim is to produce actionable recommendations that improve survey accuracy in areas with limited telephone coverage and unreliable residency records as is often