Where Things Stand: Afghanistan 2010 - Note on Methodology

PHOTO Afghanistan Poll

This survey was conducted for ABC News, the BBC, ARD and The Washington Post by the Afghan Center for Socio-Economic and Opinion Research (ACSOR) in Kabul, a subsidiary of D3 Systems Inc. of Vienna, Va. Interviews were conducted in person, in Dari or Pashto, among a random national sample of 1,691 Afghan adults Oct. 29-Nov. 13, 2010. Project design, management and analysis for ABC News were provided by Langer Research Associates.

In sampling, 148 sampling points were distributed proportional to population size in each of Afghanistan's 34 provinces, stratified by urban/nonurban status. Sampling points then were distributed to randomly selected districts within provinces, also proportionate to population size; and lastly to randomly selected villages or neighborhoods within those districts, by simple random sampling. Sources for population parameters were population projections from the Afghan Central Statistics Office.

Half the sampling points were designated for male interviews, half for female interviews. Male respondents were interviewed only by male interviewers, female respondents only by female interviewers. Residences were selected within each settlement by random route/random interval and respondents were selected within residence by Kish grid.

In addition to the national sample, booster samples were drawn in Balkh, Bamiyan, Farah, Ghazni, Ghor, Helmand, Kandahar, Khost, Kunar, Kunduz, Nimroz, Nuristan, Paktia, Panjshir and Wardak provinces via a total of 79 additional sampling points (also stratified by urban/nonurban status), for a combined total of 227 sampling points. The final sample was weighted by population of province and sex by region.

In order to reduce the design effect due to clustering, where randomly drawn male/female sampling points fell within close proximity to each other in districts with fewer than 20,000 residents, the number of sampling points was doubled, also by random selection, and the number of interviews per point was halved, from 10 to five. Sampling points also were halved in Kabul. Of the 227 sampling points, five interviews were conducted in each of 101 points, 10 interviews in each of the remaining 126.

Of the 398 districts in Afghanistan's 34 provinces, 59 were inaccessible for security reasons at the time of the field work, and female interviewers could not work in an additional 75, including all of Paktika and Uruzgan provinces, which comprise 1.7 and 1.3 percent of the Afghan population, respectively. This inaccessibility produced noncoverage of 10 percent of men, 21 percent of women and 15 percent of the national population overall.

The 116 districts initially drawn in the sample included 18 that were inaccessible and so were randomly replaced with other districts in the same province. Fifteen were replaced due to security reasons, two because of weather issues and one because of transportation difficulties. At the settlement level, 41 of the 227 sampling points were replaced, 22 for security reasons, 14 because no such places were found and 5 for inaccessibility. These were randomly substituted with settlements in the same districts.

Interviews, which averaged 36 minutes, were conducted by 209 interviewers (98 female and 111 male) in 33 supervised teams. All interviewers were trained and all had experience on previous ACSOR field projects. Eight percent of interviews were directly observed by field supervisors and 20 percent were back-checked in person afterwards. Questionnaires were all subjected to logical controls conducted at ACSOR offices in Kabul. Eighty of the total 1,771 questionnaires collected were rejected for quality-control reasons.

The survey had a contact rate of 86 percent and a cooperation rate of 94 percent for a net response rate of 81 percent. The sampling procedures have an estimated design effect of 2.2, for a total margin of sampling error of 3.5 percentage points at the 95 percent confidence level.

Join the Discussion
blog comments powered by Disqus
You Might Also Like...