Survey research in Conflict Areas: Learnings from Case Studies |
|
| Chair | Ms Magali Rheault (Gallup ) |
Based on survey data from Angolan veterans, we study the causal impacts of conflict
exposure on post-war behavior more than a decade after the end of the Angolan Civil
War. To survey conflict exposure, we employ a novel conflict exposure module which
we developed recently (Brück et al. 2016). The survey captures information from three
different stages of a soldier’s life: just before wartime military service (incl. conflict
exposure), during wartime military service (incl. conflict exposure), and today, i.e.
twelve years after the end of the war. The pre-service and service data was necessarily
based on recall. In any study using retrospective data respondents may misreport for
instance, due to memory failure and/or other personal reasons. If this is somehow
systematic, recall bias may affect key statistics and estimated relationships. We
therefore collaborated with psychologists to design the questionnaire and pre-tested it
extensively to mitigate concerns of recall bias and optimize the reliability of the data.
It is important to note that conflict exposure is usually not randomly assigned across
individuals. Ordinary estimation of the impacts of conflict exposure on post-conflict
behavior, such as ordinary least squares, is therefore likely to recover biased estimates.
To establish causality, we draw on natural variation in conflict exposure induced by the
Angolan war and use this exogenous variation to produce instrumental variable
estimates of the impact of conflict exposure on post-war behavior.
The study sampled 759 war veterans from 34 different localities in Angola’s Huambo
province. To ensure as representative a sample as possible, the survey employed three
levels of randomization, where the first two involved the primary sampling unit (PSU)
and enumeration area (EA) levels. In the absence of systematic and reliable veteran
population data, up-to date data of the total population were used from the ongoing
Angolan census to draw PSUs and EAs. At each survey site, we engaged with local chiefs,
coordinators and formal administrations to produce listings of all former soldiers
residing in the EA. Results were cross- and double-checked to develop credibly complete
listings of the local veteran population. Conditional on the reliability of the general
population as a proxy for the ex-combatant population, as used in the first stages, the
sampling strategy is self-weighting and ensures that the geographic spread across the
province is representative. Under the assumption that we did obtain complete lists of
ex-soldiers, the EA-level sample is representative of the EA-level veteran population.
We complemented the survey with an anthropological companion study, which included
twelve months of ethnographic fieldwork preceding the survey (Spall, 2015). The
resulting insights and findings were used to assess the validity our hypotheses,
determine their relevance in the local context, refine the survey questionnaire design,
interpret quantitative results and explore underlying mechanisms.
The views expressed in the article are those of the authors and do not involve the responsibility of the European Union Electoral Observation Missions.
Sample experiences with different methodologies from Africa, America and Europe are reported. The European Union Electoral Observation Missions observe democratic elections all over the world (extra EU). EU observers during the Election Day were distributed around the observed country without a statistical design and a scientific plan. In the last years, EU has been focusing on data collection complying with the procedures during the Election Day in polling stations. The random sample was experimented in Honduras, Kosovo and Tunisia. The random sample reduces the sample size, nevertheless the improvement of quality of the sample and the estimation of sample error enable the evaluation of the quality data.
In Kenya, Tunisia, Kosovo, Nigeria, Honduras and Peru, it could occur entire regions or neighborhoods were removed from the population list for logistic and security issues. In addition, the limited time of the opening hours of the polling stations worsen the logistic situation.
A pilot of the random sample was implemented in four regions of Honduras 2013, one of the most dangerous countries in the world. The success of the pilot was replicate in Kosovo.
In Kosovo 2014, the turnout was 41.54%, the estimation of the mission was 41.58%. Implementing random sampling to the election observation aims at reaching a more representative and wider capillarity and instilling an element of surprise and transparency. Random sampling provides a more accurate representation of observed centres and an effective picture of the reality on the ground during Election Day. In Kosovo, the Mission’s report was representative of all Kosovo thanks to this new implementation. The Mission has determined no security restrictions for the whole territory of Kosovo, except from the municipality of Zubin Potok. Logistic suffers from lack of maps and the observers needed an extra logistic support.
In Tunisia, we covered 25 out of 27 regions, except for the desert zone and the island Kerkennah due to logistical issues as well as Le Kef and Jendouba for security issues. We could reach the 89% of the total number of polling stations in Tunisia. The margin of error was 3.6% with 95% level of confidence and we observed 243 polling stations for voting. We can see the extreme precision of our estimation thanks to the random sample. The electoral results were for Essebsi 55.68% and Marzouki 44.32%. The estimations were respectively 55.83% and 44.17%.
Despite of the context, the ‘quick count’ can reach accurate forecast of the electoral results. The ‘quick count’ is a sample survey applied to the official result (administrative data) of the selected polling stations.
This methodology could be an extra deterrent against fraud. Nevertheless, political and ethical issues could arise in the management of this sensitive information. In a peaceful post electoral environment, is it reasonable to create suspicions of electoral frauds? And to risk armed conflicts and people life.
Conducting survey research in South Sudan presents a unique set of methodological, logistical and ethical challenges. Researchers from Ipsos MORI, Tango Consult and the National University of Ireland are working on behalf of the UK Department for International Development to measure the social and economic costs of violence against women. As part of this research, a random probability survey was carried out in South Sudan. Face-to-face interviews were conducted with women (aged 18 or older) and with the head of each selected household (male or female). We faced a number of challenges when conducting this research. These were methodological – in terms of accessing accurate population data and maps for sampling as well as logistical – in terms of the unstable security situation in Juba and other areas of the country in addition to the weather and national infrastructure. We also faced challenges that arose from the sensitive nature of the survey topic and the need to assure both the privacy of respondents and the safety of interviewers working in potentially dangerous, unstable areas. In this paper we will describe the different challenges faced during the design and implementation of the survey as well as the measures taken to overcome these. We will reflect on our experiences and the lessons learnt in order to help researchers understand how populations living in conflict areas like South Sudan can be surveyed.