Monday, June 15, 2009

Misleading Op-ed Tries to Support Election Results

Nate Silver has weighed in on the controversial oped in today's Washington Post by two representatives of an organization called Terror Free America that claim their polling results correspond to Friday's election results. Silver's analysis doesn't help relieve the traditional media of the scathing critique it's received for its coverage of the protests over the weekend. (See here and here.)

The authors argue that their survey, conducted three weeks before the election, showed a two-to-one victory for Ahmedinejad in accordance with the final tally. What they fail to mention, however, is the precise percentage breakdown of the poll's results: 34 percent for Ahmedinejad to 14 percent for Mousavi, with the remainder of respondents claiming they supported none (8 percent), didn't know who they would vote for yet (27 percent), or simply refused to answer (15 percent). Silver suggests the pollsters published the wrong story: it isn't that Ahmedinejad garners a two-to-one margin over Mousavi -- hardly a responsible conclusion given the high percentage of unallocated respondents -- but that those not supporting Ahmedinejad were likely hesitant to report their true preferences.

Despite this incredibly low rate of response, the authors of the Post article claim that their respondents felt comfortable enough to express their opinions straightforwardly since they answered other controversial questions, such as on Ahmedinejad's performance. But Silver says that method isn't fail safe -- and, furthermore, that the lack of enthusiasm for Ahmedinejad that the poll also shows suggests many of those not responding to the vote question would be unlikely to support the incumbent.

Allocating undecideds is a notoriously tricky and controversial practice, especially in countries that suffer from political repression and intimidation. The poll might well have been accurate, as the Post piece's authors go to great lengths to demonstrate. The problem is with their analysis. Simply scaling up the results, as the survey's authors seem to do, is quite poor practice.

No comments: