The perils of polling in Afghanistan

By Sameer Lalwani

On the eve of the Afghan presidential election, there are very few public polls on which analysts and commentators can base their forecasts. So it's worth taking a moment to evaluate how useful these are and consider some of the biases that confound accurate polling.

First, polling will suffer from tremendous sampling bias, in which the sample is not representative of the national population. Afghanistan is still one of the poorest countries in the world with an extremely low telecommunications penetration rate. According to the International Telecommunication Union, there is less than one fixed phone line per thousand people. The world average is about nineteen per thousand.

Afghan mobile phone usage has increased exponentially in recent years to 29 per thousand but this still remains half the global average. This telecom divide means phone polling tends to favor wealthier urban constituents and under-sample rural areas. Pollsters can correct for this with face-to-face interviews (as the IRI and Glevum Associates did in their recent election polls) though they are more time consuming and expensive. But polling in a conflict-zone incurs new sampling biases by tending to over-sample safer and less-conflict ridden areas.

Second, polling in America is far different from polling in Afghanistan. A colleague who extensively traveled and studies the Middle East once warned me about putting much stock in foreign polling numbers, particularly out of American firms. His reason was this: the United States is an open society with a long history of polling and consumer surveys that acculturates people to be more responsive and forthright in their opinions. In conflict-ridden societies with a history of repressive governments, people have learned to guard their beliefs and are more likely to mislead polling interviewers out of fear retribution from the central government, local elites, or insurgents.

Finally, polling will predict inaccuracies if it doesn't match up with actual turnout. Specifically in Afghanistan's case, the two real election polls by IRI and Glevum both equally sample men and women. While appropriate in theory, the fact is in Afghanistan's 2004 presidential election, women were only 40% of the turnout. And with the resurgence of the Taliban in the south and east, as well as the deal cutting President Karzai has done to win the support of conservative Shia men, if anything women's turnout will likely be further depressed.

These biases also need to be accounted for when it comes to polls of Afghan support for the central government or US/NATO forces. And while these biases might introduce just a few points of error, it is those few points that could be the difference between a Karzai majority victory and a runoff election.

Sameer Lalwani is a research fellow with the New America Foundation and a Ph.D. student in political science at the Massachusetts Institute of Technology.